Front cover
IBM

Developing and Porting
C and C++ Applications
on AIX
Detailed explanations about 32- and 64-bit
process models
Effective management of shared
objects and libraries
Exploring parallel
programming using OpenMP
Keigo Matsubara
Edison Kwok
Inge Rodriguez
Murali Paramasivam
ibm.com/redbooks


International Technical Support Organization
Developing and Porting C and C++ Applications on
AIX
June 2003
SG24-5674-01

Note: Before using this information and the product it supports, read the information in
“Notices” on page xvii.
Second Edition (June 2003)
This edition applies to C for AIX (program number 5765-F57) and VisualAge C++ for AIX Version
6.0 (product number 5765-F56) installed on AIX 5L Version 5.2 (product number 5765-E62).
© Copyright International Business Machines Corporation 2000, 2003. All rights reserved.
Note to U.S. Government Users Restricted Rights -- Use, duplication or disclosure restricted by GSA ADP
Schedule Contract with IBM Corp.

Contents
Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiii
Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xv
Notices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xvii
Trademarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xviii
Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xix
The team that wrote this redbook. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xix
Become a published author . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xx
Comments welcome . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xxi
Summary of changes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xxiii
June 2003, Second Edition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xxiii
September 2000, First Edition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xxiv
Chapter 1. C and C++ compilers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 C for AIX Version 6.0. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.1.1 New or improved optimization features . . . . . . . . . . . . . . . . . . . . . . . . 2
1.1.2 ISO C Standard conformance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.1.3 GNU C compatibility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.1.4 Enhanced language level support . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.2 VisualAge C++ for AIX Version 6.0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.2.1 New or improved optimization features . . . . . . . . . . . . . . . . . . . . . . . 16
1.2.2 OpenMP support . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
1.2.3 Automatic parallelization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
1.2.4 Improved template handling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
1.2.5 C99 features . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
1.2.6 GNU G++ compatibility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
1.3 Installing the compilers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
1.3.1 Install compiler filesets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
1.3.2 Retaining a previous version of the compiler . . . . . . . . . . . . . . . . . . 22
1.4 Activating the compilers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
1.4.1 What is LUM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
1.4.2 Configuring LUM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
1.5 Activating the LUM server . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
1.6 Enrolling a product license . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
1.6.1 Enrolling a concurrent license . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
1.6.2 Enrolling a simple nodelock license . . . . . . . . . . . . . . . . . . . . . . . . . 29
© Copyright IBM Corp. 2000, 2003. All rights reserved.
iii

1.7 Invoking the compilers. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
1.7.1 Default compiler drivers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
1.8 Where to find help . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
1.8.1 Online documentations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
1.8.2 Viewing online documentation remotely . . . . . . . . . . . . . . . . . . . . . . 31
1.8.3 Where to find help on the Web . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
1.8.4 Applying fixes and service updates . . . . . . . . . . . . . . . . . . . . . . . . . . 35
Chapter 2. Compiling and linking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
2.1 32- and 64-bit development environments . . . . . . . . . . . . . . . . . . . . . . . . 38
2.1.1 The 64-bit advantage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
2.1.2 Compiler support . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
2.1.3 Utility commands support . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
2.2 Compiling and linking: A quick overview . . . . . . . . . . . . . . . . . . . . . . . . . . 43
2.2.1 Building C and C++ programs with system libraries . . . . . . . . . . . . . 43
2.2.2 Objects and libraries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
2.2.3 Difference between shared object and library on AIX . . . . . . . . . . . . 49
2.2.4 Difference between shared and static objects on AIX. . . . . . . . . . . . 51
2.3 Resolving symbols at link-time . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
2.3.1 The -L linker option . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
2.3.2 Searching objects and libraries at link-time. . . . . . . . . . . . . . . . . . . . 56
2.3.3 LIBPATH environment variable . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
2.3.4 Link-time and load-time . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
2.4 Supported link methods on AIX . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
2.4.1 AIX default linking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
2.4.2 Static linking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
2.4.3 Lazy loading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
2.5 Run-time linking. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
2.5.1 How to use run-time linking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
2.5.2 Examining the executable and shared objects using dump . . . . . . . 75
2.5.3 Enabling the main program object as run-time linking . . . . . . . . . . . 78
2.5.4 Rebinding symbols at the program load-time . . . . . . . . . . . . . . . . . . 79
2.5.5 Extended search order with the -brtl linker option . . . . . . . . . . . . . . . 81
2.6 Dynamic loading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
2.7 Commands when manipulating objects and libraries . . . . . . . . . . . . . . . . 85
2.7.1 dump . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
2.7.2 genkld . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
2.7.3 ldd . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
2.7.4 nm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
2.7.5 rtl_enable . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
2.7.6 slibclean . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
2.8 Creating shared objects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
2.8.1 Import and export files. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
iv
Developing and Porting C and C++ Applications on AIX

2.8.2 A self-contained shared object . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
2.8.3 Interdependent shared objects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
2.8.4 Initialization and termination routines . . . . . . . . . . . . . . . . . . . . . . . . 99
2.9 Shared libraries in a development environment . . . . . . . . . . . . . . . . . . . . 99
2.9.1 Production and development environments . . . . . . . . . . . . . . . . . . 100
2.9.2 Private shared objects. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
2.9.3 NFS consideration. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
2.9.4 Sufficient free disk space on the target directory and /tmp . . . . . . . 102
Chapter 3. Understanding user process models . . . . . . . . . . . . . . . . . . . 105
3.1 User process models on AIX. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
3.1.1 How to determine hardware bit mode . . . . . . . . . . . . . . . . . . . . . . . 107
3.1.2 How to determine kernel bit mode . . . . . . . . . . . . . . . . . . . . . . . . . 107
3.1.3 How to determine user process bit mode . . . . . . . . . . . . . . . . . . . . 108
3.2 The 32-bit user process model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
3.2.1 Default memory model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
3.2.2 Large memory model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
3.2.3 Very large memory model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
3.2.4 Using the large and very large memory model . . . . . . . . . . . . . . . . 121
3.2.5 Checking large memory model executables . . . . . . . . . . . . . . . . . . 124
3.2.6 Resource limits in 32-bit model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
3.2.7 Large file support in a 32-bit model . . . . . . . . . . . . . . . . . . . . . . . . . 129
3.3 The 64-bit user process model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
3.3.1 The first 16 segments (0 - 4 GB) . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
3.3.2 Application text, data, and heap (4 GB - 448 PB) . . . . . . . . . . . . . . 133
3.3.3 Default shared memory segments (448 - 512 PB) . . . . . . . . . . . . . 135
3.3.4 Privately loaded objects (512 - 576 PB) . . . . . . . . . . . . . . . . . . . . . 135
3.3.5 Shared text and data (576 - 640 PB) . . . . . . . . . . . . . . . . . . . . . . . 135
3.3.6 System reserved (640 - 960 PB) . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
3.3.7 User process stack (960 PB - 1 EB) . . . . . . . . . . . . . . . . . . . . . . . . 136
3.3.8 Resource limits in 64-bit mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
3.3.9 Large file support in 64-bit model . . . . . . . . . . . . . . . . . . . . . . . . . . 139
3.4 Introduction to shared memory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
3.4.1 The shmat services . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
3.4.2 The mmap services . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
3.4.3 Difference between shmat and mmap services . . . . . . . . . . . . . . . 147
3.4.4 Shared memory limits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
3.5 Shared memory segments allocation order. . . . . . . . . . . . . . . . . . . . . . . 149
3.5.1 Order in the 32-bit default memory model. . . . . . . . . . . . . . . . . . . . 153
3.5.2 Order in the 32-bit very large memory model with DSA . . . . . . . . . 154
3.5.3 Extended mode shared memory segments . . . . . . . . . . . . . . . . . . 155
3.5.4 Order in the 64-bit memory model . . . . . . . . . . . . . . . . . . . . . . . . . 156
3.6 Large page support . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
Contents
v

3.6.1 Large page support overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
3.6.2 Large page application usage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
3.6.3 Large page usage security capability . . . . . . . . . . . . . . . . . . . . . . . 161
3.6.4 Configuring system to use large pages . . . . . . . . . . . . . . . . . . . . . . 162
3.6.5 Other system changes for large pages . . . . . . . . . . . . . . . . . . . . . . 164
3.6.6 Large page usage considerations . . . . . . . . . . . . . . . . . . . . . . . . . . 164
Chapter 4. Managing the memory heap . . . . . . . . . . . . . . . . . . . . . . . . . . 165
4.1 Malloc subsystem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166
4.1.1 malloc(), calloc(), valloc(), and alloca() . . . . . . . . . . . . . . . . . . . . . . 167
4.1.2 mallopt(), mallinfo, and mallinfo_heap() . . . . . . . . . . . . . . . . . . . . . 168
4.1.3 disclaim() . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
4.2 Memory allocators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
4.2.1 The 3.1 memory allocator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
4.2.2 The default memory allocator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172
4.2.3 The default memory allocator with the malloc buckets extension . . 173
4.2.4 The debug malloc allocator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176
4.2.5 User-defined malloc replacement . . . . . . . . . . . . . . . . . . . . . . . . . . 178
4.2.6 Malloc multiheap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
4.3 Use of MALLOCDEBUG options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182
4.3.1 MALLOCDEBUG with the debug memory allocator . . . . . . . . . . . . 182
4.3.2 MALLOCDEBUG with memory allocators other than debug. . . . . . 190
4.4 Heap management using MEMDBG . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199
4.4.1 How to handle a user-created heap . . . . . . . . . . . . . . . . . . . . . . . . 200
4.4.2 A user-defined heap allocated from shared memory segments . . . 202
Chapter 5. Creating DLPAR-aware applications . . . . . . . . . . . . . . . . . . . . 207
5.1 Dynamic logical partitioning overview . . . . . . . . . . . . . . . . . . . . . . . . . . . 208
5.2 The process flow of a DLPAR operation . . . . . . . . . . . . . . . . . . . . . . . . . 210
5.3 DLPAR-safe and DLPAR-aware applications . . . . . . . . . . . . . . . . . . . . . 214
5.3.1 DLPAR-safe . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
5.3.2 DLPAR-aware . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215
5.4 Integrating a DLPAR operation into the application . . . . . . . . . . . . . . . . 217
Chapter 6. Programming hints and tips . . . . . . . . . . . . . . . . . . . . . . . . . . 219
6.1 Programming recommendations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220
6.1.1 Variables and data structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220
6.1.2 Functions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221
6.1.3 Pointers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 222
6.1.4 Arithmetic operations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223
6.1.5 Selection and iteration statements . . . . . . . . . . . . . . . . . . . . . . . . . 223
6.1.6 Expression . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 224
6.1.7 Memory usage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225
6.1.8 Built-in functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225
vi
Developing and Porting C and C++ Applications on AIX

6.1.9 Virtual functions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 226
6.2 Diagnosing compile-time errors. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227
6.2.1 Anatomy of a message . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227
6.2.2 Useful options and compiler aids . . . . . . . . . . . . . . . . . . . . . . . . . . 228
6.2.3 Migrating from 32-bit to 64-bit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 230
6.3 Diagnosing link-time errors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 239
6.3.1 Unresolved symbols . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 240
6.3.2 Duplicate symbols . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241
6.3.3 Insufficient memory for the linker process. . . . . . . . . . . . . . . . . . . . 243
6.3.4 The c++filt utility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243
6.4 Diagnosing run-time errors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 244
6.4.1 Uninitialized variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 244
6.4.2 Run-time checking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245
6.4.3 Unsignedness preservation in C . . . . . . . . . . . . . . . . . . . . . . . . . . . 246
6.4.4 ANSI aliasing. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 247
6.4.5 #pragma option_override . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 248
Chapter 7. Debugging your applications . . . . . . . . . . . . . . . . . . . . . . . . . 249
7.1 Working with core files . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 250
7.1.1 Core file naming . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 250
7.1.2 Creating core files with assert(). . . . . . . . . . . . . . . . . . . . . . . . . . . . 251
7.1.3 Creating core files with coredump() . . . . . . . . . . . . . . . . . . . . . . . . 251
7.1.4 Including shared memory information in the core file . . . . . . . . . . . 253
7.1.5 Gathering core files . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 253
7.1.6 AIX error log entry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 255
7.1.7 Lightweight core file support . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 256
7.2 Using the printf()-debug method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 258
7.3 Preparing your application for debugging . . . . . . . . . . . . . . . . . . . . . . . . 259
7.4 Using dbx . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 259
7.4.1 Starting a dbx session . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 260
7.4.2 Customizing a dbx session . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 261
7.4.3 Working with breakpoints: The stop subcommand . . . . . . . . . . . . . 263
7.4.4 Redirection of library location in object files with the -p flag . . . . . . 263
7.4.5 Using dbx with gcc . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 264
7.5 Debugging with the truss command . . . . . . . . . . . . . . . . . . . . . . . . . . . . 264
7.6 Using the trace facility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 267
7.6.1 Introduction to trace . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 267
7.6.2 Tracing an application on the command line. . . . . . . . . . . . . . . . . . 268
7.6.3 Tracing an application with subroutine calls . . . . . . . . . . . . . . . . . . 271
Chapter 8. Introduction to POSIX threads. . . . . . . . . . . . . . . . . . . . . . . . . 275
8.1 Overview of threads . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 276
8.1.1 Relationship between a process and a user thread . . . . . . . . . . . . 276
Contents
vii

8.2 POSIX threads (Pthreads) on AIX . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 278
8.2.1 Advantages of using Pthreads . . . . . . . . . . . . . . . . . . . . . . . . . . . . 279
8.2.2 The POSIX threads API . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 279
8.2.3 Multi- and single-threaded processes . . . . . . . . . . . . . . . . . . . . . . . 280
8.3 Pthread management . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 284
8.3.1 Creating and terminating Pthreads . . . . . . . . . . . . . . . . . . . . . . . . . 284
8.3.2 Joining Pthreads . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 289
8.3.3 Detaching a Pthread . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 291
8.3.4 Thread stack . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 292
8.4 Data synchronization between Pthreads . . . . . . . . . . . . . . . . . . . . . . . . . 296
8.4.1 Synchronizing Pthreads with mutexes . . . . . . . . . . . . . . . . . . . . . . 297
8.4.2 Synchronizing Pthreads with condition variables . . . . . . . . . . . . . . 303
8.4.3 Synchronizing Pthreads with read-write locks. . . . . . . . . . . . . . . . . 310
8.5 Thread-specific data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 312
8.5.1 Allocating thread-specific data . . . . . . . . . . . . . . . . . . . . . . . . . . . . 314
8.5.2 Accessing thread-specific data . . . . . . . . . . . . . . . . . . . . . . . . . . . . 315
8.5.3 Deleting thread-specific data. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 316
8.5.4 Thread-safe and reentrant functions . . . . . . . . . . . . . . . . . . . . . . . . 319
8.6 Pthread cancellation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 320
8.7 Pthread priority and scheduling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 321
8.7.1 Thread models in AIX . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 322
8.7.2 Scheduling Pthreads . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 324
8.7.3 Scheduling limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 328
8.8 Pthread specific environment variables in AIX . . . . . . . . . . . . . . . . . . . . 328
8.9 User API for Solaris threaded applications . . . . . . . . . . . . . . . . . . . . . . . 332
8.9.1 Application binary interface (ABI) . . . . . . . . . . . . . . . . . . . . . . . . . . 333
Chapter 9. Program parallelization using OpenMP . . . . . . . . . . . . . . . . . 335
9.1 Introduction to OpenMP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 336
9.2 The OpenMP programming model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 336
9.3 Classification of OpenMP directives . . . . . . . . . . . . . . . . . . . . . . . . . . . . 337
9.3.1 The OpenMP directive format . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 337
9.4 Parallel region construct . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 338
9.5 Work-sharing constructs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 340
9.5.1 for construct . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 341
9.5.2 sections construct . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 344
9.5.3 single construct . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 345
9.6 Combined parallel work-sharing constructs . . . . . . . . . . . . . . . . . . . . . . 347
9.6.1 parallel for construct . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 347
9.6.2 parallel sections construct . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 348
9.7 Synchronization constructs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 348
9.7.1 master construct . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 349
9.7.2 critical construct . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 349
viii
Developing and Porting C and C++ Applications on AIX

9.7.3 barrier directive . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 350
9.7.4 atomic construct . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 352
9.7.5 flush directive . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 352
9.7.6 ordered construct . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 353
9.8 Data environment: The threadprivate directive . . . . . . . . . . . . . . . . . . . . 354
9.9 Data-sharing attribute clauses. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 355
9.9.1 private clause . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 355
9.9.2 firstprivate clause . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 356
9.9.3 lastprivate clause . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 357
9.9.4 shared clause . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 357
9.9.5 default clause . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 358
9.9.6 reduction clause . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 358
9.9.7 copyin clause . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 362
9.9.8 copyprivate clause . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 363
9.10 Run-time library functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 363
9.10.1 Execution environment functions . . . . . . . . . . . . . . . . . . . . . . . . . 364
9.10.2 Lock functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 368
9.10.3 Example usage of run-time library functions . . . . . . . . . . . . . . . . . 370
9.11 Environment variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 373
9.11.1 OMP_SCHEDULE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 373
9.11.2 OMP_NUM_THREADS. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 374
9.11.3 OMP_DYNAMIC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 374
9.11.4 OMP_NESTED . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 374
Chapter 10. Dealing with C++ templates . . . . . . . . . . . . . . . . . . . . . . . . . . 377
10.1 What is a template. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 378
10.2 AIX template implementations. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 378
10.2.1 Generated function bodies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 380
10.3 Simple code layout method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 381
10.3.1 Disadvantages of the simple method . . . . . . . . . . . . . . . . . . . . . . 381
10.4 Template instantiation file method. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 383
10.4.1 The -qtempinc option . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 384
10.4.2 Contents of the tempinc directory . . . . . . . . . . . . . . . . . . . . . . . . . 385
10.4.3 Forcing template instantiation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 386
10.5 Template registry: The preferred method . . . . . . . . . . . . . . . . . . . . . . . 387
10.5.1 The -qtemplateregistry option . . . . . . . . . . . . . . . . . . . . . . . . . . . . 387
10.5.2 The -qtemplaterecompile option . . . . . . . . . . . . . . . . . . . . . . . . . . 388
10.6 Standard C++ Library and STL . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 388
10.6.1 Standard Template Library . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 389
10.6.2 A STL example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 392
Chapter 11. Creating shared objects from C++ source codes . . . . . . . . 393
11.1 Creating shared objects from C++ source codes . . . . . . . . . . . . . . . . . 394
Contents
ix

11.1.1 Creating a C++ shared object . . . . . . . . . . . . . . . . . . . . . . . . . . . . 395
11.1.2 Generating an export file . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 395
11.1.3 The -qmkshrobj option . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 396
11.1.4 Mixing C and C++ object files . . . . . . . . . . . . . . . . . . . . . . . . . . . . 396
11.1.5 Order of initialization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 397
11.2 Shared objects with templates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 402
11.2.1 Templates and makeC++SharedLib . . . . . . . . . . . . . . . . . . . . . . . 402
11.2.2 Templates and -qmkshrobj . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 404
Chapter 12. Packaging your applications . . . . . . . . . . . . . . . . . . . . . . . . . 405
12.1 Understanding the AIX standard packaging . . . . . . . . . . . . . . . . . . . . . 406
12.1.1 Filesets and package files . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 406
12.1.2 Bundles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 406
12.1.3 Managing filesets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 408
12.1.4 Viewing the TOC file (.toc) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 412
12.1.5 Viewing package files . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 415
12.2 Packaging applications using mkinstallp . . . . . . . . . . . . . . . . . . . . . . . . 418
12.2.1 mkinstallp . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 418
12.2.2 Packaging examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 419
12.2.3 Verification of packages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 424
12.2.4 Optional installation control executable files . . . . . . . . . . . . . . . . . 426
Appendix A. Previous versions of C and C++ compiler products . . . . . 429
Compiler product similarities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 430
Multiple command line drivers. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 430
Installation directory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 431
IBM C compilers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 431
IBM XL C Version 3. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 431
IBM C for AIX Version 4.1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 432
IBM C for AIX Version 4.3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 433
IBM C for AIX Version 4.4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 434
IBM C for AIX Version 5.0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 434
C compiler summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 435
IBM C++ compilers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 435
IBM C Set ++ for AIX Version 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 436
IBM C and C++ compilers Version 3.6 . . . . . . . . . . . . . . . . . . . . . . . . . . . 436
IBM VisualAge C++ Professional for AIX Version 4 . . . . . . . . . . . . . . . . . 437
IBM VisualAge C++ Professional for AIX Version 5 . . . . . . . . . . . . . . . . . 438
C++ compiler summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 439
Appendix B. Useful information for linking and loading on AIX. . . . . . . 441
A brief history of UNIX programming development . . . . . . . . . . . . . . . . . . . . 442
Historical view of linking and loading in AIX . . . . . . . . . . . . . . . . . . . . . . . . . . 443
Definitions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 443
x
Developing and Porting C and C++ Applications on AIX

Appendix C. Subroutine references for shmat and mmap services. . . . 447
References for shmat services . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 448
The ftok() subroutine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 448
The shmat() routine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 449
The shmctl() subroutine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 454
The shmget() routine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 457
The shmdt() subroutine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 460
References for mmap services. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 461
The mmap() subroutine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 461
The mprotect() subroutine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 467
The msync() subroutine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 469
The munmap() subroutine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 471
Appendix D. Subroutine references for POSIX threads. . . . . . . . . . . . . . 473
Subroutines defined in the POSIX thread standard . . . . . . . . . . . . . . . . . . . . 474
Subroutines defined in the UNIX 98 Specification . . . . . . . . . . . . . . . . . . . . . 479
Extensions to POSIX thread. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 481
Appendix E. Supported IBM SMP directives . . . . . . . . . . . . . . . . . . . . . . 483
IBM SMP directives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 484
The IBM SMP directives syntax . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 485
Appendix F. Sample compiler listing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 489
Compiler listing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 490
Abbreviations and acronyms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 495
Related publications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 497
IBM Redbooks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 497
AIX official publications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 497
pSeries hardware related publications . . . . . . . . . . . . . . . . . . . . . . . . . . . 498
C for AIX official publications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 498
VisualAge C++ for AIX official publications . . . . . . . . . . . . . . . . . . . . . . . . 499
Other publications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 499
Referenced Web sites . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 499
How to get IBM Redbooks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 500
IBM Redbooks collections . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 500
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 501
Contents
xi

xii
Developing and Porting C and C++ Applications on AIX

Figures
1-1
Core and orthogonal extensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2-1
Object files and a library archive . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
2-2
Compiling and linking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
2-3
An XCOFF format executable file and exec() . . . . . . . . . . . . . . . . . . . . 63
2-4
Static and shared text code in the executable file . . . . . . . . . . . . . . . . . 65
2-5
The -bdynamic and -bstatic linker options . . . . . . . . . . . . . . . . . . . . . . . 66
2-6
Function calling relationship. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
2-7
Function calling relationship after rebinding symbols . . . . . . . . . . . . . . 81
2-8
Function calling relationship for an interdependent shared object . . . . . 98
3-1
Hardware, kernel, and user process relationships . . . . . . . . . . . . . . . . 106
3-2
Default memory model (segment usage) . . . . . . . . . . . . . . . . . . . . . . . 110
3-3
Default memory model (detail) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
3-4
Large memory model (segment usage) . . . . . . . . . . . . . . . . . . . . . . . . 116
3-5
Very large memory model (0 < maxdata < 0xB0000000) . . . . . . . . . . 118
3-6
Very large memory model (0xB0000000 =< maxdata < 0xD0000000) 119
3-7
Very large memory model (maxdata = 0). . . . . . . . . . . . . . . . . . . . . . . 120
3-8
Data and stack resource limits (default 32-bit process model) . . . . . . 126
3-9
The 64-bit memory model (1EB) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
3-10 The 64-bit memory model (4 GB, the first 16 segments) . . . . . . . . . . . 132
3-11 Shared memory segments between two processes . . . . . . . . . . . . . . 143
5-1
Process flow of a DLPAR operation. . . . . . . . . . . . . . . . . . . . . . . . . . . 211
6-1
Structure padding in 32-bit mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237
6-2
Structure padding in 64-bit mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 238
6-3
Structure with user-defined paddings in both 32-bit and 64-bit mode . 239
7-1
Definition of HkWord . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 269
8-1
Two user threads in a process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 277
8-2
Five Pthreads created by pthread_create() . . . . . . . . . . . . . . . . . . . . . 285
8-3
Thread stacks (default 32-bit process model) . . . . . . . . . . . . . . . . . . . 293
8-4
Thread models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 323
9-1
Concept of barrier . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 350
11-1 Illustration of objects in fish.o and animals.o . . . . . . . . . . . . . . . . . . . . 399
12-1 Relationship among filesets, packages, and bundles . . . . . . . . . . . . . 406
12-2 State diagram between applied and committed state . . . . . . . . . . . . . 410
12-3 Relationship between APARs and update fileset. . . . . . . . . . . . . . . . . 411
12-4 Sample .toc file . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 413
12-5 Directory structure for packaging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 419
© Copyright IBM Corp. 2000, 2003. All rights reserved.
xiii

xiv
Developing and Porting C and C++ Applications on AIX

Tables
1-1
C for AIX Version 6.0 packages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
1-2
VisualAge C++ for AIX Version 6.0 packages . . . . . . . . . . . . . . . . . . . . 19
1-3
License certificate locations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
1-4
Compiler driver extensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2-1
C and C++ data type sizes in bits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
2-2
XCOFF headers and loading target segments . . . . . . . . . . . . . . . . . . . 62
2-3
Supported link methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
2-4
LDLAZYDEBUG environment variable values . . . . . . . . . . . . . . . . . . . . 68
2-5
Linker options equivalent to -G . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
2-6
IMPid values for imported symbols . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
2-7
Directive lines for import files . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
3-1
Shared memory limits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
4-1
Memory allocators and MALLOCTYPE . . . . . . . . . . . . . . . . . . . . . . . . 170
4-2
Default configuration values for malloc buckets. . . . . . . . . . . . . . . . . . 176
4-3
User-defined replacement subroutines . . . . . . . . . . . . . . . . . . . . . . . . 179
4-4
MALLOCDEBUG options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
5-1
Applications that should be DLPAR-aware . . . . . . . . . . . . . . . . . . . . . 216
6-1
ISO C99 integer constant type selection . . . . . . . . . . . . . . . . . . . . . . . 236
8-1
Filesets for Solaris user thread library . . . . . . . . . . . . . . . . . . . . . . . . . 333
9-1
OpenMP components . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 336
9-2
OpenMP directive categories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 337
11-1 Order of initialization of objects in prriolib.a . . . . . . . . . . . . . . . . . . . . . 401
12-1 Fileset state . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 408
12-2 Fields description of the .toc file. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 414
12-3 Definition of entries in <fileset_name>.inventory . . . . . . . . . . . . . . . . . 417
12-4 Template file keywords . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 422
12-5 Optional installation control executable files . . . . . . . . . . . . . . . . . . . . 427
A-1
IBM C compilers for AIX. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 435
A-2
C++ compiler products . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 439
B-1
Definitions of terms regarding linking and loading process on AIX . . . 443
C-1
Values for the third parameter of shmat() . . . . . . . . . . . . . . . . . . . . . . 452
C-2
Values for the third parameter of shmctl() . . . . . . . . . . . . . . . . . . . . . . 455
C-3
Values for the third parameter of shmget() . . . . . . . . . . . . . . . . . . . . . 458
C-4
Values for the sixth parameter of mmap() . . . . . . . . . . . . . . . . . . . . . . 465
C-5
The third parameter of msync() . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 470
D-1
Thread management sub-routines . . . . . . . . . . . . . . . . . . . . . . . . . . . . 474
D-2
Execution scheduling sub-routines . . . . . . . . . . . . . . . . . . . . . . . . . . . 476
D-3
Synchronization sub-routines. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 477
© Copyright IBM Corp. 2000, 2003. All rights reserved.
xv

D-4
Thread-specific data sub-routines . . . . . . . . . . . . . . . . . . . . . . . . . . . . 479
D-5
Read-write lock sub-routines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 480
D-6
Additional POSIX threads sub-routines defined in UNIX 98 . . . . . . . . 481
D-7
Non-portable thread routines in AIX. . . . . . . . . . . . . . . . . . . . . . . . . . . 482
E-1
Regular expressions for countable loops . . . . . . . . . . . . . . . . . . . . . . . 484
E-2
Supported IBM pragma directives . . . . . . . . . . . . . . . . . . . . . . . . . . . . 486
xvi
Developing and Porting C and C++ Applications on AIX

Notices
This information was developed for products and services offered in the U.S.A.
IBM may not offer the products, services, or features discussed in this document in other countries. Consult
your local IBM representative for information on the products and services currently available in your area.
Any reference to an IBM product, program, or service is not intended to state or imply that only that IBM
product, program, or service may be used. Any functionally equivalent product, program, or service that
does not infringe any IBM intellectual property right may be used instead. However, it is the user's
responsibility to evaluate and verify the operation of any non-IBM product, program, or service.
IBM may have patents or pending patent applications covering subject matter described in this document.
The furnishing of this document does not give you any license to these patents. You can send license
inquiries, in writing, to:
IBM Director of Licensing, IBM Corporation, North Castle Drive Armonk, NY 10504-1785 U.S.A.
The following paragraph does not apply to the United Kingdom or any other country where such provisions
are inconsistent with local law: INTERNATIONAL BUSINESS MACHINES CORPORATION PROVIDES
THIS PUBLICATION "AS IS" WITHOUT WARRANTY OF ANY KIND, EITHER EXPRESS OR IMPLIED,
INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF NON-INFRINGEMENT,
MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE. Some states do not allow disclaimer
of express or implied warranties in certain transactions, therefore, this statement may not apply to you.
This information could include technical inaccuracies or typographical errors. Changes are periodically made
to the information herein; these changes will be incorporated in new editions of the publication. IBM may
make improvements and/or changes in the product(s) and/or the program(s) described in this publication at
any time without notice.
Any references in this information to non-IBM Web sites are provided for convenience only and do not in any
manner serve as an endorsement of those Web sites. The materials at those Web sites are not part of the
materials for this IBM product and use of those Web sites is at your own risk.
IBM may use or distribute any of the information you supply in any way it believes appropriate without
incurring any obligation to you.
Information concerning non-IBM products was obtained from the suppliers of those products, their published
announcements or other publicly available sources. IBM has not tested those products and cannot confirm
the accuracy of performance, compatibility or any other claims related to non-IBM products. Questions on
the capabilities of non-IBM products should be addressed to the suppliers of those products.
This information contains examples of data and reports used in daily business operations. To illustrate them
as completely as possible, the examples include the names of individuals, companies, brands, and products.
All of these names are fictitious and any similarity to the names and addresses used by an actual business
enterprise is entirely coincidental.
COPYRIGHT LICENSE:
This information contains sample application programs in source language, which illustrates programming
techniques on various operating platforms. You may copy, modify, and distribute these sample programs in
any form without payment to IBM, for the purposes of developing, using, marketing or distributing application
programs conforming to the application programming interface for the operating platform for which the
sample programs are written. These examples have not been thoroughly tested under all conditions. IBM,
therefore, cannot guarantee or imply reliability, serviceability, or function of these programs. You may copy,
modify, and distribute these sample programs in any form without payment to IBM for the purposes of
developing, using, marketing, or distributing application programs conforming to IBM's application
programming interfaces.
© Copyright IBM Corp. 2000, 2003. All rights reserved.
xvii

Trademarks
The following terms are trademarks of the International Business Machines Corporation in the United States,
other countries, or both:
AIX 5L™
IBM®
pSeries™
AIX®
Open Class®
Redbooks (logo)™
C Set ++®
OS/2®
Redbooks™

PartnerWorld®
RISC System/6000®
^™
POWER4™
VisualAge®
IBM eServer™
PowerPC®
zSeries™
The following terms are trademarks of other companies:
Java and all Java-based trademarks are trademarks of Sun Microsystems, Inc. in the United States, other
countries, or both.
Microsoft, Windows, Windows NT, and the Windows logo are trademarks of Microsoft Corporation in the
United States, other countries, or both.
Intel, Intel Inside (logos), MMX and Pentium are trademarks of Intel Corporation in the United States, other
countries, or both.
UNIX is a registered trademark of The Open Group in the United States and other countries.
SET and the SET Logo are trademarks owned by SET Secure Electronic Transaction LLC.
Other company, product and service names may be trademarks or service marks of others.
xviii
Developing and Porting C and C++ Applications on AIX

Preface
This IBM Redbook will help experienced UNIX application developers who are
new to the AIX operating system. The book explains the many concepts in detail,
including the following:
Enhancements and new features provided by the latest C and C++ compilers
for AIX
Compiling and linking tasks required to effectively use and manage shared
libraries and run-time linking
Use of process heap and shared memory in the 32- and 64-bit user process
models
A new programming paradigm in a partitioned environment where resources
can be dynamically changed
Parallel programming using POSIX threads and OpenMP
The following chapters are also useful for system administrators who are
responsible for the software problem determination and application software
release level management on AIX systems:
Chapter 3, “Understanding user process models” on page 105
Chapter 7, “Debugging your applications” on page 249
Chapter 12, “Packaging your applications” on page 405
This publication expands on the information found in the AIX 5L Porting Guide,
SG24-6034.
The team that wrote this redbook
This redbook was produced by a team of specialists from around the world
working at the International Technical Support Organization, Austin Center.
Keigo Matsubara is an advisory IT specialist at the International Technical
Support Organization (ITSO), Austin Center. Before joining the ITSO, he worked
in the System and Web Solution Center in Japan as a Field Technical Support
Specialist (FTSS) for pSeries™. He has been working for IBM® for 11 years.
Edison Kwok is a senior Software Developer in the IBM Toronto Software
Laboratory in Canada. He has eight years of experience in C and C++ compiler
development on the zSeries™ and pSeries platforms. He holds a degree in
© Copyright IBM Corp. 2000, 2003. All rights reserved.
xix

electrical engineering from the University of Victoria. His area of expertise
include compiler construction, C language standard, and C and C++
programming on various UNIX operating systems and the mainframe.
Inge Rodriguez is an IT specialist from IBM Germany. She has 20 years of
experience in UNIX application development. She has been working for IBM for
three years. Her main responsibility is support for ISVs regarding application
development and porting. She holds a MSc in Medical Computer Science of
University Heidelberg.
Murali Paramasivam is a Software Engineer from IBM India. He has nearly
three years of application development experience in C and C++ on various
UNIX operating systems. He holds an engineering degree in Material Science
and Metallurgy. His areas of expertise include shared libraries, multi-threaded
programming, and C/C++ compilation and linking concepts in AIX®.
Thanks to the following people for their contributions to this project:
International Technical Support Organization, Austin Center
Scott Vetter and Wade Wallace
IBM Austin
Alfredo Mendoza, Ann Wigginton, Betty Riggle, Donald Stence, David Hepkin,
Gary Hook, Joel H Schopp, Julie Craft, Kedron J Touvell, Kenji Kindo, Kevin W
Monroe, Luke Browning, Mark Rogers, Michael Mall, Nathan Fontenot, Randy
Swanberg, Richard Cutler, Sara D Epsztein, Steven Molis
IBM Japan
Hajime Mita and Tomoyuki Niijima
IBM Toronto
Steven E. Hikida, Wang Chen, Sean Perry, Roger E. Pett.
Become a published author
Join us for a two- to six-week residency program! Help write an IBM Redbook
dealing with specific products or solutions, while getting hands-on experience
with leading-edge technologies. You'll team with IBM technical professionals,
Business Partners and/or customers.
Your efforts will help increase product acceptance and customer satisfaction. As
a bonus, you'll develop a network of contacts in IBM development labs, and
increase your productivity and marketability.
xx
Developing and Porting C and C++ Applications on AIX

Find out more about the residency program, browse the residency index, and
apply online at:
ibm.com/redbooks/residencies.html
Comments welcome
Your comments are important to us!
We want our Redbooks™ to be as helpful as possible. Send us your comments
about this or other Redbooks in one of the following ways:
Use the online Contact us review redbook form found at:
ibm.com/redbooks
Send your comments in an Internet note to:
redbook@us.ibm.com
Mail your comments to:
IBM Corporation, International Technical Support Organization
Dept. JN9B Building 003 Internal Zip 2834
11400 Burnet Road
Austin, Texas 78758-3493
Preface
xxi

xxii
Developing and Porting C and C++ Applications on AIX

Summary of changes
This section describes the technical changes made in this edition of the book and
in previous editions. This edition may also include minor corrections and editorial
changes that are not identified.
Summary of Changes
for SG24-5674-01
for Developing and Porting C and C++ Applications on AIX
as created or updated on March 25, 2009.
June 2003, Second Edition
This revision reflects the addition, deletion, or modification of new and changed
information described below.
New information
The following chapters are new:
Chapter 2, “Compiling and linking” on page 37
Chapter 3, “Understanding user process models” on page 105
Chapter 4, “Managing the memory heap” on page 165
Chapter 5, “Creating DLPAR-aware applications” on page 207
Chapter 6, “Programming hints and tips” on page 219
Chapter 7, “Debugging your applications” on page 249
Chapter 9, “Program parallelization using OpenMP” on page 335
Chapter 12, “Packaging your applications” on page 405
Changed information
The following chapters were rewritten in order to cover new features and
enhancements provided by the latest products:
Chapter 1, “C and C++ compilers” on page 1
Chapter 8, “Introduction to POSIX threads” on page 275
© Copyright IBM Corp. 2000, 2003. All rights reserved.
xxiii

Unchanged information
The following chapters and appendixes are unchanged but reviewed again:
Chapter 10, “Dealing with C++ templates” on page 377
Chapter 11, “Creating shared objects from C++ source codes” on page 393
Appendix A, “Previous versions of C and C++ compiler products” on page 429
Appendix E, “Supported IBM SMP directives” on page 483
September 2000, First Edition
The first version of this book, C and C++ Application Development on AIX,
SG24-5674 was written by the following authors:
Richard Cutler, Francois Armingaud, Eduardo Conejo, Kumaravel Nagarajan
The following list shows contributors for the first version of this book, C and C++
Application Development on AIX, SG24-5674:
IBM Toronto
Derek Truong, Mark Changfoot, Paul Pacholski, Rene Matteau
xxiv
Developing and Porting C and C++ Applications on AIX

1
C and C++ compilers
Chapter 1.
This chapter focuses on the latest versions of the IBM C and C++ compiler
products for AIX: C for AIX Version 6.0 and VisualAge C++ for AIX Version 6.0.
The latest compiler products offer enhanced support in optimizations,
POWER4™ architecture exploitation, the latest ISO C and C++ Standard
conformance, as well as compatibility features targeted to GNU C/C++ portability,
explained in the first two sections:
Section 1.1, “C for AIX Version 6.0” on page 2
Section 1.2, “VisualAge C++ for AIX Version 6.0” on page 16
The other sections provide a comprehensive guide to installing and configuring
the compiler products on your AIX systems:
Section 1.3, “Installing the compilers” on page 19
Section 1.4, “Activating the compilers” on page 23
Section 1.5, “Activating the LUM server” on page 26
Section 1.6, “Enrolling a product license” on page 27
Section 1.7, “Invoking the compilers” on page 29
Section 1.8, “Where to find help” on page 31
For a description of the previous versions of IBM C and C++ compiler products,
please refer to Appendix A, “Previous versions of C and C++ compiler products”
on page 429 for details.
© Copyright IBM Corp. 2000, 2003. All rights reserved.
1

1.1 C for AIX Version 6.0
The C for AIX Version 6.0 compiler is the latest IBM C compiler product available
on AIX. It offers several new enhancements over the previous versions,
particularly in the area of optimization features and new PowerPC® architecture
support. This compiler is supported on AIX Version 4.3.3 or later.
C programs written using Version 4 or 5 of IBM C for AIX are source compatible
with IBM C for AIX Version 6.0. C programs written using either Version 2 or 3 of
IBM Set ++ for AIX or the XL C compiler component of AIX Version 3.2 are
source compatible with IBM C for AIX Version 6.0 with exceptions to detect
invalid programs or areas where results are undefined. Source compatibility,
however, does not guarantee a program will perform in an identical manner; new
option defaults can sometimes influence how a program behaves. Always consult
the official documentation when migrating to a new version of the product.
If installed, the compiler is installed under /usr/vac by default and uses the
/etc/vac.cfg configuration file. To install to an alternate directory, or to retain the
installation of a previous version of C for AIX compiler, refer to 1.3.2, “Retaining a
previous version of the compiler” on page 22.
The C for AIX Version 6.0 compiler uses the LUM licensing system, which is
explained in the following sections, to control usage of the product.
Section 1.4, “Activating the compilers” on page 23
Section 1.5, “Activating the LUM server” on page 26
Section 1.6, “Enrolling a product license” on page 27
1.1.1 New or improved optimization features
A number of optimization features have been introduced or improved in C for AIX
Version 6.
Interprocedural analysis
Interprocedural analysis, or IPA, is an optimization performed across function
boundaries. In a traditional compilation, only intraprocedural analysis is done,
where each function is optimized individually within a single compilation unit. IPA
takes optimization one step further by analyzing all functions in the entire
application.
2
Developing and Porting C and C++ Applications on AIX

In addition to the usual optimizations performed by the optimizer, IPA also
performs many optimizations interprocedurally, including:
Inlining across compilation units
Program partitioning
Global variables coalescing
Code straightening
Dead code elimination
Constant propagation
Copy propagation
Keep in mind that because the compiler is performing extra processing with
-qipa, compilation time is expected to increase. However, with C for AIX Version
6.0, a new suboption, -qipa=threads, has been introduced to take advantage of
multi-threaded interprocedural analysis. You can also specify -qipa=threads=N,
where N is the number of threads used by the compiler for IPA analysis and code
generation. Please refer to C for AIX Compiler Reference, SC09-4960 for more
details.
Profile-directed feedback
With profile-directed feedback, or PDF, special instrumentation code is inserted
in the executable to gather information about the program’s execution pattern.
Information gathered from the execution is then fed back to the compiler for a
second compilation, and the compiler performs optimization based on the code
execution frequency and conditional branch pattern.
In order to gain the most using this feature, make sure the program execution is
performed as close to the intended conditions as possible. That is, choose input
parameters and a data set that are representative and meaningful.
Only use PDF towards the very end of a development cycle, where the program
is fully functional at a high optimization level.
New options and pragmas
C for AIX Version 6.0 introduces several new performance related options and
pragmas:
-qarch=pwr4
-qtune=pwr4
-qhot
-qlargepage
Chapter 1. C and C++ compilers
3

-qsmallstack
-qunwind
-qtocmerge
-qreport
-qipa=threads=N
#pragma execution_frequency
#pragma pack
#pragma snapshot
Note: The -qarch=pwr4 and -qtune=pwr4 options are used to generate
executable files optimized and tuned for the POWER4 processor.
For a detailed description of these options, please refer to the C for AIX Compiler
Reference, SC09-4960. For detailed explanation about the tuning considerations
on the POWER4 processor, please refer to The POWER4 Processor Introduction
and Tuning Guide, SG24-7041.
New built-in functions
Compiler built-in functions are often provided to allow programmers direct access
to features or machine code instructions on the hardware architecture. They are
directly mapped to hardware instructions, hence any overhead associated with
function calls (for example, parameter passing, stack allocation and adjustment,
and so on) are completely eliminated. Please refer to C for AIX Compiler
Reference, SC09-4960 for a list of built-in functions supported in C for AIX
Version 6.0.
1.1.2 ISO C Standard conformance
The IBM C for AIX Version 6.0 compiler supports the latest ISO/IEC 9899:1999
International Standard, commonly referred to as C99. C99 includes many new
features and enhancements to the original ISO/IEC 9899:1990 International
Standard (C89), which extends the capability of the C language. We will discuss
some useful features defined in this standard.
_Bool type
Similar to the C++ bool type, C99 supports _Bool type in addition to the long list
of type specifiers already in the original C89 standard. Applications no longer
need to define their own macros, as the system header file stdbool.h already
defines the macros true and false.
4
Developing and Porting C and C++ Applications on AIX

long long data type
Unlike the -qlonglong option in the previous versions of the C for AIX compilers,
the addition of the long long data type in C99 changes the semantics for integer
constants. In C89, the type of an unsuffixed integer constant is either int, long int,
or unsigned long int, whichever is large enough to represent the constant.
However in C99, unsuffixed integer constants have type int, long int, or long long
int instead. To illustrate this difference, consider the following example:
#include <stdio.h>
#include <stdlib.h>
int main(int argc, char *argv[])
{
printf("sizeof(2147483648) = %d\n", sizeof(2147483648));
exit(0);
}
In the above example, the constant 2147483648 is one greater than
LONG_MAX. When compiled with -qlanglvl=stdc89, the execution of this
example will print the value 4, since the type selected for the constant is
unsigned long int. With -qlanglvl=stdc99, on the other hand, the result will be 8,
and the constant will have long long int type.
Complex data types
C99 introduces native complex data types to the C language. There are three
complex types: float _Complex, double _Complex, and long double _Complex, as
well as three pure imaginary types: float _Imaginary, double _Imaginary, and long
double _Imaginary. Collectively they are called complex floating types. Each
complex type is logically the same as an array of two elements of the
corresponding real floating type, where the first element is the real part of the
complex number, and the second element the imaginary part. Therefore, the size
of a complex type is double the size of its corresponding floating type.
The following shows how to declare a complex variable c, and initialize it to
{1.0, 2.0i}:
double _Complex c = 1.0 + 2.0 * __I;
Basic arithmetic operators are supported natively in the language. There are also
mathematical functions provided by the run-time library, by including the system
header file complex.h. For more information on the semantics of the complex
data types, please refer to C for AIX C/C++ Language Reference, SC09-4958.
inline function specifier
Function inlining reduces function call overhead, as well as allowing the optimizer
to perform better optimizations at or near the function call site. The compiler
Chapter 1. C and C++ compilers
5

already does inline optimization with the -O option, so the use of this feature may
not provide any further performance benefits.
restrict qualifier
If an object is modified through a restrict qualified pointer, than all access to that
object must be based on, directly or indirectly, the same pointer, that is, no other
pointers will access the object. This allows the compiler to perform better
optimization. Please refer to the C for AIX C/C++ Language Reference,
SC09-4958 for more details.
static keyword in array declaration
In a function declaration, array parameters are generated as pointers to the array
element type. For example:
void func(int arr[])
{
...
}
and
void func(int *arr)
{
...
}
are equivalent declarations. In C99, you can use the storage class specifier static
in an array parameter declaration, to indicate to the compiler that the argument in
the function call is guaranteed to be non-NULL, and contains at least the
specified number of elements. For example:
void vector_add(int a[static 10], const int b[static 10])
{
int i;
for (i = 0; i < 10; i++)
a[i] += b[i];
}
With this extra information, the compiler will be able to apply better optimization
analysis and generate faster performing code.
Universal character name
Universal characters support, already available since C for AIX Version 5 with the
-qlanglvl=ucs option, is now part of the C99 standard. It is used to write
characters that are not in the basic character set. You can have universal
characters in identifiers, string literals, and comments.
6
Developing and Porting C and C++ Applications on AIX

__func__
Similar to the C for AIX compiler predefined macro __FUNCTION__, __func__ is
a compiler generated internal variable that has the following declaration:
static const char __func__[] = “function_name”;
where function_name is the name of the current function where __func__ is
referenced. This is useful in writing debug code. See “Function-like macros with
variable arguments” on page 9 for an example usage.
Hexadecimal floating point constant
Just as you can use hexadecimal integer constants to represent exactly the
binary format of an integer, C99 allows you to have floating point constant
specified in hexadecimal format. For example:
double d = 0x123.abcp+10;
Variable length arrays
The size of local automatic objects is determined at compile time, and the
duration and scope of these objects end when you leave the function body where
the object is declared. If the size requirement of an object is unknown at compile
time, for example, an array of unknown number of elements, the programmer is
responsible for dynamically allocating storage at run time, and freeing the
storage before exiting the current scope. C99 introduces variable length arrays,
where its usage removes the burden from the programming for allocating and
remembering to free local automatic storage.
In the following example, the local array, new, is a variable length array:
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
void reverse(const char *str, int n)
{
int i;
char new[n];
for (i = 0; i < n; i++)
new[i] = str[n-i-1];
printf("%.*s\n", n, new);
}
int main(int argc, char *argv[])
{
reverse("Hello World", strlen("Hello World"));
exit(0);
}
Chapter 1. C and C++ compilers
7

Without variable length array support, storage for the local array new would have
to be dynamically allocated as follows:
char *new = (char *)malloc(n);
and freed explicitly at the end of function reverse:
free(new);
Compound literals
A compound literal is an unnamed object of type specified in parentheses. The
value of the object is given in a braced initializer list. It is mainly used in situations
where a temporary object would otherwise be required. In the following example,
the emphasized line shows an example usage of compound literals:
#include <stdio.h>
typedef struct {
short serial;
char *name;
} Record;
void show(Record rec)
{
printf("Employee serial: %d\n", rec.serial);
printf("Employee name: %s\n", rec.name);
}
int main(int argc, char *argv[])
{
show((Record){ 12345, "Elizabeth" });
exit(0);
}
Without a compound literal, a temporary object of type Record would be needed
to be used in the function call:
Record tmp_rec = { 12345, “Elizabeth” };
show(tmp_rec);
Designated initialization
In C89, initializers must be specified in the order and sequence of the elements
or members to be initialized. Although for static storage duration objects, where
they are implicitly initialized to zero already, if you need to initialize only specific
members or elements of the object, you have to supply enough initializers. For
example, for the following structure declaration:
typedef struct {
short serial;
8
Developing and Porting C and C++ Applications on AIX

char *name;
int salary;
char addr[40];
char city[20];
char state[2];
char zip[5];
short location;
} Record;
If all members are to be initialized with the default value of zero, except the last
member, location, all initializers must still be supplied:
Record emp1 = { 0, 0, 0, "", "", "", "", 649 };
With a designated initializer, on the other hand, initializers are only needed for
members that are required to be explicitly initialized:
Record emp2 = { .location = 649 };
This greatly reduces the risk of errors that are proven to be hard to debug.
Non-constant initializers for automatic aggregates
With C99, you can now initialize automatic storage duration aggregate members
with non-constant initializers. For example:
#include <stdlib.h>
void func()
{
struct {
short serial;
char *name;
} rec = { .name = (char*)malloc(30) };
}
Function-like macros with variable arguments
Functions with variable arguments, for example:
extern int printf(const char *, ...);
eliminate the need for many versions of the same function that accepts different
numbers of arguments. C99 extends the concept further and allows variable
arguments in function-like macros. As shown in the following example, debug
code can now be handled more elegantly:
#include <stdio.h>
#if !defined(DEBUG)
#define DBGMSG(fmt, ...) ((void)0)
#else
Chapter 1. C and C++ compilers
9

#define DBGMSG(fmt, ...) ( \
fprintf(stderr, "In %s: ", __func__), \
fprintf(stderr, fmt, __VA_ARGS__) )
#endif
int main(int argc, char *argv[])
{
int rc = 55;
DBGMSG("return code = %d\n", rc);
return rc;
}
Pragma operator
The _Pragma operator allows you to code pragma directives in macros. In the
following example, the declaration of struct S and the definition of its instance s
are surrounded by two macros, PACK_1 and PACK_POP:
#define PACK_1 _Pragma("pack(1)")
#define PACK_POP _Pragma("pack(pop)")
PACK_1 struct S {
char ch;
int i;
} s; PACK_POP
Mixed declarations and code
C99 allows declarations mixed with code similar to C++. For example:
void func()
{
int i;
i = 10;
int j;
j = 20;
...
}
1.1.3 GNU C compatibility
There are plenty of programs developed using the GNU C compiler (also known
as gcc). The proliferation of the open source concept, together with the far
reaching nature of the Web, spawns a whole new group of developers who
collaborate across physical boundaries, and the choice of compiler for this group
is the gcc compiler. This does not necessarily mean that gcc is superior; on the
contrary, the IBM compiler optimization technologies are among the best in the
industry. The main reason for the gcc compiler’s wide acceptance has to do with
its many useful extensions, and the fact that it is freely available. Also, the GNU C
10
Developing and Porting C and C++ Applications on AIX

and C++ compilers are available on various operating systems running on the
many different types of hardware, providing cross platform development
capabilities that are rivaled by no other.
The C for AIX Version 6 compiler supports many of the gcc extensions, and they
allow you to port programs written for gcc to AIX more easily. The availability of
each extension is indicated by a compiler predefined macro of the form
__IBM_feature, where feature is the gcc feature name.
For example, the following code fragment written for gcc:
#if defined(__GNUC__)
extern char *func(const char *__s, int __c) __attribute__((__pure__));
#endif
will be successfully compiled with C for AIX Version 6, if the code is modified as
follows:
#if defined(__IBM_ATTRIBUTES) || defined(__GNUC__)
extern char *func(const char *__s, int __c) __attribute__((__pure__));
#endif
To compile this code with previous versions of C for AIX compilers, it must be
modified as follows:
#if defined(__IBM_ATTRIBUTES) || defined(__GNUC__)
extern char *func(const char *__s, int __c) __attribute__((__pure__));
#else
extern char *func(const char *__s, int __c);
#pragma isolated_call(func)
#endif
The identifier, __pure__, is one of function attributes supported by C for AIX
Version 6 (see “Function attributes” on page 13).
Please refer to C for AIX Compiler Reference, SC09-4960 for the full list of
supported GNU C compatibility.
Local labels
Ordinary labels has function scope, that is, they can only be defined once within
a function body. This prevents the use of labels and goto statements inside
macros, when the macro is expected to be expanded more than once in a
function. As shown in the following example, a local label, on the other hand, is
visible only within the block where it is declared, as well as in all nested blocks. A
local label also hides the function scope label of the same name:
#include <stdio.h>
#include <stdlib.h>
Chapter 1. C and C++ compilers
11

#define FIND(str, val, len)
\
{
\
__label__ done;
\
int i;
\
for (i = 0; i < len; i++)
\
if (str[i] == val) goto done;
\
printf("%c not found in %s\n", val, str); \
done:
\
printf("found %c in [%d]\n", val, i);
\
}
int main(int argc, char *argv[])
{
FIND("hello", 'l', 5);
FIND("world", 'l', 5);
exit(0);
}
__typeof__ operator
Similar to the sizeof operator, the __typeof__ operator takes an expression or a
type as an operand, but returns the type of the operand instead of the size. It can
be used anywhere a typedef name is used. This operator is particularly useful in
marcos, where the type of the macro argument is not known before hand. For
example, instead of writing several versions of a SWAP_type macro for the
different integral types:
#define SWAP_char(a,b) { char temp = a; a = b; b = temp; }
#define SWAP_short(a,b) { short temp = a; a = b; b = temp; }
...
you can use one SWAP macro that handles all types:
#define SWAP(a,b) { __typeof__(a) temp = a; a = b; b = temp; }
__alignof__ operator
Use the __alignof__ operator to find out the alignment of a type or an object. Due
to the different alignment rules and packing supported by the C for AIX compiler,
the alignment of an object may change depending on options or pragmas used,
as shown in the following example:
#include <stdio.h>
#include <stdlib.h>
#pragma pack(2)
struct {
int i;
double d;
} s;
12
Developing and Porting C and C++ Applications on AIX

int main(int argc, char *argv[])
{
printf("alignment of double is %d\n", __alignof__(double));
printf("alignment of s.d is %d\n", __alignof__(s.d));
exit(0);
}
will yield:
alignment of double is 8
alignment of s.d is 2
Function attributes
Function attributes are used to help the compiler apply better optimization on
function calls. The C for AIX Version 6.0 compiler supports three function
attributes: noreturn, const, and pure.
The noreturn attribute indicates to the compiler that the function does not return
control to the statement following the function call. The C library already has
several functions, such as abort and exit, that behave as if the function is
declared with the noreturn attribute; the compiler is already aware of these
functions and is able to optimize the function calls accordingly (when the
-qlibansi option is in effect). User defined functions that do not return control to
the calling side can be declared with this attribute for better performance. This
attribute is functionally equivalent to #pragma leaves.1
The const and pure attributes are equivalent, and are used to indicate to the
compiler that the function does not have or rely on any side effects. The return
value only depends on the parameters, and pointer arguments are not examined
in the function body. The function does not call any non-const function, nor
access any global or external storage. This attribute is functionally equivalent to
#pragma isolated_call.2
To use the pure function attribute, see the example in 1.1.3, “GNU C
compatibility” on page 10.
Variable attributes
Variable attributes (aligned, mode, or packed) are used to modify variable
declarations.
1 The #pragma leaves directive specifies that a function never returns.
2 The #pragma isolated_call directive specifies that a function does not have or rely on side effects.
Chapter 1. C and C++ compilers
13

The aligned attribute causes the complier to use a different alignment for variable
or structure members. It specifies the minimum number of bytes to use for
aligning the declaration, as shown in the following example:
#include <stdio.h>
#include <stdlib.h>
struct {
int i;
double __attribute__((__aligned__(16))) d;
} s;
int main(int argc, char *argv[])
{
printf("alignment of double is %d\n", __alignof__(double));
printf("alignment of s.d is %d\n", __alignof__(s.d));
exit(0);
}
will yield the result:
alignment of double is 8
alignment of s.d is 16
To change the packing of aggregate or structure members, or to use the smallest
possible alignment, use the packed attribute, as shown in the following example:
#include <stdio.h>
#include <stdlib.h>
struct {
int i;
double __attribute__((__packed__)) d;
} s;
void main()
{
printf("alignment of double is %d\n", __alignof__(double));
printf("alignment of s.d is %d\n", __alignof__(s.d));
}
will yield the result:
alignment of double is 8
alignment of s.d is 1
The mode attribute lets you select an integer based on width, as shown in the
following example (the supported widths are byte, word, and pointer):
#include <stdio.h>
#include <stdlib.h>
14
Developing and Porting C and C++ Applications on AIX

int __attribute__((__mode__(byte))) x;
void main()
{
printf("sizeof(int) is %d\n", sizeof(int));
printf("sizeof(x) is %d\n", sizeof(x));
}
will yield the result:
sizeof(int) is 4
sizeof(x) is 1
1.1.4 Enhanced language level support
Not only does the C for AIX Version 6.0 compiler support the latest ISO C
language standard (C99), it also has several language extension modes that
make the use of the C language ever more powerful. These extensions add
flexibility and allow the programmer to achieve a programming task more easily.
Language levels are supported by the -qlanglvl compiler option. The two core
levels, -qlanglvl=stdc89 and -qlanglvl=stdc99, strictly enforces the standards, and
are used mainly for compiling standard conforming programs. To use extensions
to the standard language levels, -qlanglvl=extc89 and -qlanglvl=extc99 add to the
core levels orthogonal features that do not interfere with the core standard
features (see Figure 1-1), that is, a standard conforming program will be
compiled just as successfully with -qlanglvl=extc89 or -qlanglvl=extc99 as it will
be with -qlanglvl=stdc89 or -qlanglvl=stdc99. Therefore, it is recommended that
you use the extension language levels to compile programs that are written using
other compilers such as gcc.
Orthogonal language extensions
Core language level
Figure 1-1 Core and orthogonal extensions
Chapter 1. C and C++ compilers
15

1.2 VisualAge C++ for AIX Version 6.0
VisualAge C++ for AIX Version 6.0 offers a command-line compiler that supports
the ISO/IEC 14882:1998 C++ International Standard. This latest version of the
C++ compiler features a conforming C++ Standard Library, including the
Standard Template Library (STL), enhanced optimization options that exploit the
POWER4 architecture, OpenMP support, and more. It is supported on AIX
Version 4.3.3 or later.
Note: The Standard Template Library is discussed in more detail in 10.6,
“Standard C++ Library and STL” on page 388.
C++ programs written using Version 4 or 5 of IBM VisualAge C++ Professional
for AIX are source compatible with VisualAge C++ for AIX Version 6.0. However,
programs written using IBM C and C++ Compilers for AIX Version 3.6 and earlier
are not source compatible, because the former compilers were based on the ISO
C++ Draft. Also, the IBM Open Class® library has been removed and is no longer
supported.
If installed, the compiler is installed under /usr/vacpp by default and uses the
/etc/vacpp.cfg configuration file. To install to an alternate directory, or to retain the
installation of a previous version of VisualAge C++ for AIX compiler, refer to
1.3.2, “Retaining a previous version of the compiler” on page 22.
The VisualAge C++ for AIX Version 6.0 compiler uses the LUM licensing system
to control usage of the product. Refer to the following sections:
Section 1.4, “Activating the compilers” on page 23
Section 1.5, “Activating the LUM server” on page 26
Section 1.6, “Enrolling a product license” on page 27
1.2.1 New or improved optimization features
A number of optimization features have been introduced in VisualAge C++ for
AIX Version 6.
Interprocedural Analysis
Interprocedural Analysis, or IPA, is now available in VisualAge C++ for AIX
Version 6.0. IPA is an optimization performed on the entire application across
function boundaries. See “Interprocedural analysis” on page 2 for more details.
16
Developing and Porting C and C++ Applications on AIX

New options and pragmas
In addition to the “New options and pragmas” on page 3 supported by C for AIX
Version 6.0, VisualAge C++ for AIX Version 6.0 now supports the following
options:
-qsmp (previously supported by C for AIX only)
-O4 and -O5 (previously supported by C for AIX only)
-qipa (previously supported by C for AIX only)
-qcache (previously supported by C for AIX only)
-qkeepinlines
-qtemplateregistry
-qtemplaterecompile
-qalign=bit_packed
-qoldpassbyvalue
-qlanglvl=ansiinit
#pragma unroll (previously supported by C for AIX only)
For a detailed description of these options, please refer to the VisualAge C++ for
AIX Compiler Reference, SC09-4959.
1.2.2 OpenMP support
New in VisualAge C++ for AIX Version 6.0, the compiler now supports the
OpenMP Version 1.0 specification for shared memory parallel programming.
OpenMP is an Application Program Interface specification that provides a simple
and flexible interface, including a number of pragma directives, data scope
attributes, library functions, and environment variables, for parallel application
development. Applications that conform to the OpenMP specification is easily
portable to other platforms that support the specification.
OpenMP support is enabled in the compiler by the -qsmp option, which was
previously available with the C for AIX compiler only. To ensure strict compliance
to the OpenMP specification, use the -qsmp=omp option. Bear in mind that
-qsmp=omp disables automatic parallelization performed by the compiler. See
1.2.3, “Automatic parallelization” on page 18. In either case, you must use one of
the thread-safe compiler driver with the -qsmp option (see Table 1-4 on page 30).
For further information about OpenMP support, see Chapter 9, “Program
parallelization using OpenMP” on page 335.
Chapter 1. C and C++ compilers
17

1.2.3 Automatic parallelization
In addition to strict OpenMP specification support, the -qsmp option enables
automatic parallelization for program loops. Each parallel portion of the program
is executed in its own thread, perhaps in its own processor of a multi-processor
machine. To use the -qsmp option, you must use one of the thread-safe compiler
drivers, as described in Table 1-4 on page 30.
1.2.4 Improved template handling
VisualAge C++ for AIX Version 6.0 introduces two new options,
-qtemplateregistry and -qtemplaterecompile, which completely eliminate the
need to structure your template code for reduced compilation time.
The -qtemplateregistry option works by storing template instantiation information
in a registry as compilation occurs. The registry is read for each compilation, and
a check is done when an instantiation is encountered. If the instantiation has
already been seen, nothing will happen; otherwise, the template will be
instantiated in the object file. In either case, a record is added to the registry to
keep track of the information about use and instantiation for each compilation
unit.
Should a change to a compilation unit remove a template instantiation,
recompiling the compilation will result in undefined symbols at link time, since
other files now may require the missing template instantiation. In this case, use
the -qtemplaterecompile option to cause a recompilation of all source files that
rely on the template.
The -qtemplateregistry and -qtemplaterecompile options are discussed in more
detail in 10.5, “Template registry: The preferred method” on page 387.
1.2.5 C99 features
The following C99 features are available in VisualAge C++ for AIX Version 6.0
with the -qlanglvl=extended option:
“restrict qualifier” on page 6
“Function-like macros with variable arguments” on page 9
“Pragma operator” on page 10
1.2.6 GNU G++ compatibility
The VisualAge C++ for AIX Version 6.0 compiler supports the same set of GNU
C extension features described in 1.1.3, “GNU C compatibility” on page 10.
18
Developing and Porting C and C++ Applications on AIX

1.3 Installing the compilers
The installation of the C for AIX Version 6.0 and VisualAge C++ for AIX Version
6.0 compilers is a very simple task. There are a number of steps that need to be
performed to end up with correctly installed and working compilers.
1.3.1 Install compiler filesets
The first step in the installation process is to install the compiler product filesets
onto the system. The filesets to be installed will vary, depending on the compiler
product and the desired configuration.
Selecting the required filesets
The compiler products are either delivered on CD-ROM media, or downloaded
from the official IBM Web site, and are accompanied by a license certificate for
the number of licenses purchased. The CD-ROM media or download includes
the compiler filesets along with a number of other filesets, some of which are
optionally installable, and some of which are co-requisites of the compiler filesets
and are installed automatically. Table 1-1 lists the main packages in the C for AIX
Version 6.0 product, and Table 1-2 lists the main packages in the VisualAge C++
for AIX Version 6.0 product.
Table 1-1 C for AIX Version 6.0 packages
Group
Description
IMNSearch
Search engine for HTML documentation
idebug
Debugger with graphical user interface
memdbg
Memory debugging toolkit
vac
C for AIX compiler
xlopt
Optimization library and run time
xlsmp
Parallelization run-time component
Table 1-2 VisualAge C++ for AIX Version 6.0 packages
Group
Description
IMNSearch
Search engine for HTML documentation
idebug
Debugger with graphical user interface
memdbg
Memory debugging toolkit
vac
C for AIX compiler
Chapter 1. C and C++ compilers
19

Group
Description
vacpp
VisualAge C++ for AIX compilera
vatools
VisualAge® Tools Help
xlC
C++ Application Development Toolkit
xlopt
Optimization library and run time
xlsmp
Parallelization run-time component
a. The vacpp group contains installp packages for part of the C++ compiler only;
it requires packages from the vac group.
Note: A single package can contain multiple filesets and a fileset can be
included in only one package. For further information about AIX software
packaging terminology, see Chapter 12, “Packaging your applications” on
page 405.
In all cases, the target AIX system should already have the bos.adt.include fileset
installed, which contains the system provided header files. The other filesets in
the bos.adt package contain useful tools and utilities often used during
application development, so it is a good idea to install the entire package. If your
system does not have the filesets installed, you will need to locate your AIX
installation media and install them prior to installing the compilers, since these
filesets are AIX version specific and are not supplied with the compiler product.
When installing the C for AIX Version 6.0 product, installing the vac.C fileset will
automatically install the minimum of additional required filesets. Installing the
vacpp.cmp fileset will automatically include the minimum required filesets for
VisualAge C++ for AIX. The additional filesets you may wish to install are the
documentation filesets. Ensure that the vac.lic and vacpp.lic filesets are installed,
as they contain the license files required when activating the compiler.
Regardless of the product or required configuration, the filesets can be installed
using one of two methods, as discussed in the following sections.
Install using the Web-based System Manager
If your system has a graphical user interface, the filesets can be installed using
the wsm command. The procedure is as follows:
1. Log in as the root user.
2. Insert the product CD-ROM media in the CD or DVD device.
3. Start the software installation taskguide with the following command:
# wsm install
20
Developing and Porting C and C++ Applications on AIX

4. From the Software drop-down menu, select New Software (Install/Update)
Install Additional Software Advanced Method.
5. In the Install Software dialog, select the CD-ROM device as the software
source. Then, select to install the specific software available from the software
source.
6. Select the Browse button to generate a list of software on the media.
7. Select the desired filesets from the dialog. Press and hold down the control
key while dragging the mouse cursor to select one or more additional objects.
8. Select the OK button once you have selected the desired filesets to return to
the Software Install dialog.
9. Select the OK button to start the install.
10.Select the YES button to continue with the install. A pop-up window will
appear and show the output of the installation process.
11.Select the Close button once the installation has completed.
Install using SMIT
If your system does not have a graphical user interface, or you do not wish to use
a Web-based System Manger, you can install the required filesets using the smit
command as follows:
1. Log in as the root user.
2. Insert the product CD-ROM media in the CD or DVD device.
3. Start the SMIT dialog with the following command:
# smit install_latest
4. Press the F4 key to generate a list of possible input devices.
5. Select the CD-ROM device.3
6. Press the F4 key to generate a list of available filesets.
7. Select the required filesets by highlighting them and then pressing the F7 key.
8. Press the Enter key once the required filesets have been selected.
9. Press the Enter key to start the install.
10.Press the Enter key to continue the install.
11.Press the F10 key to exit once the installation has completed.
3 A DVD-RAM or DVD-ROM device is also shown as CD-ROM device in the selection panel.
Chapter 1. C and C++ compilers
21

Note: The compiler products cannot be used immediately after installation.
Prior to invoking the compiler, a product licence must be enrolled with the
License Use Management (LUM) system. See 1.4, “Activating the compilers”
on page 23.
1.3.2 Retaining a previous version of the compiler
Before installing the compiler, it is recommended that you uninstall any previous
version of C for AIX and VisualAge C++ for AIX compilers already installed.
However, you can install the compilers in another directory, and retain the
previous installation. This is done by using the supplied Perl scripts, vacndi and
vacppndi.
Ensure the Perl run-time environment fileset, perl.rte, is installed on your system.
Install the vac.ndi fileset from the C for AIX CD-ROM media, or the vacpp.ndi
fileset from the VisualAge C++ for AIX CD-ROM media, or both. You can then
use the /usr/vac/bin/vacndi and /usr/vacpp/bin/vacppndi scripts to install just the
compiler, or the compiler with help documentation and samples, to a location of
your choice. Do not move or rename the directory or any of its components after
installation; you must reinstall to a new location if you want to change the
installed directory.
To install C for AIX compiler with help documentation and samples, run:
# perl /usr/vac/bin/vacndi -d source_path -b target_directory
where source_path is where the C for AIX product filesets are located, and
target_directory is the installation directory. If the -b option is omitted, the default
installation directory is used (that is, /usr/vac).
To install C for AIX compiler only without help documentation and samples, run:
# perl /usr/vac/bin/vacndi -d source_path -b target_directory -m
To install the VisualAge C++ for AIX compiler product, simply replace
/usr/vac/bin/vacndi in the above commands with /usr/vacpp/bin/vacppndi.
After installation, check the ./vacndi.log or ./vacppndi.log log file and make sure
the installation is performed successfully. To remove the installation, delete the
target_directory directory as specified in the installation step.
22
Developing and Porting C and C++ Applications on AIX

1.4 Activating the compilers
Once you have installed the desired compiler filesets onto the system, the next
step in the process is to enroll a license for the product into the LUM system. This
section describes the process of configuring a LUM server and enrolling a
product license. If you already have a LUM environment enabled, you may go
directly to 1.6, “Enrolling a product license” on page 27.
1.4.1 What is LUM
IBM License Use Management Runtime, referred to hereafter as License Use
Management (LUM), contains the tools needed in an end-user environment to
manage product licenses and to get up-to-date information about license usage.
LUM is the replacement for the iFOR/LS and Net/LS systems that were used in
previous versions of AIX and with previous versions of the IBM compilers.
The LUM run time is included with AIX Version 4.3 and higher and is
automatically installed. A comprehensive description of the functionality of LUM
can be found in the LUM online documentation supplied on the AIX Version 4.3
and higher product media in the ifor_ls.html.en_US.base.cli fileset, which is not
automatically installed when installing AIX.
1.4.2 Configuring LUM
Normally, one or more LUM license servers need to be configured. However, no
license server needs to be configured if the licensed product supplies a simple
nodelock license certificate. Both the C for AIX Version 6.0 and VisualAge C++
for AIX Version 6.0 compiler products supply a simple nodelock license
certificate.
The simplest method of licensing the latest compiler products is to use the simple
nodelock license certificate. When this is done, there is no need to configure a
LUM server; however, the installation of the certificate in large numbers of
machines can be cumbersome.
If you wish to use the simple nodelock certificate, you can skip directly to 1.6,
“Enrolling a product license” on page 27. If you wish to use the additional
functionality available when using a license server, then the first step is to decide
which server type is best suited for your environment.
There are two types of license servers:
Concurrent nodelock license server
Concurrent network license server
Chapter 1. C and C++ compilers
23

A concurrent nodelock license server supports concurrent nodelock product
licenses. A concurrent nodelock license is local to the node where the LUM
enabled product has been installed. It allows a limited number of simultaneous
users to invoke the enabled licensed product on the local system.
A concurrent network license server supports concurrent network product
licenses. A concurrent network license is a network license that can temporarily
grant a user on a client system the authority to run a LUM enabled product.
Either or both of the above license servers may be configured on a single
system. The number of concurrent users for the product is specified during the
enrollment of the product license certificate, as described in 1.6, “Enrolling a
product license” on page 27.
The advantage of using a concurrent nodelock license server is that the server is
installed on the same machine as the compiler, and, therefore, users can obtain
compiler licenses even if the machine is temporarily disconnected from the
network. The disadvantage, however, is that installation of licenses is
cumbersome in environments with a large number of client machines.
The main advantage of using a central network license server is that the
administration of product licenses is very simple. The disadvantage is that client
machines must be able to contact the license server in order to use the licensed
products.
Configuring LUM requires answering several questions on how you would like to
set up the LUM environment. It is recommended that users read the LUM
documentation supplied with the AIX product media prior to configuring LUM.
A LUM server can be configured in several different ways. You can issue
commands on the command line with appropriate arguments to configure the
LUM server. You can issue a command that starts a dialog and asks a number of
questions to determine the appropriate configuration, or you can configure the
server using a graphical user interface.
Configuring a nodelock server
For small numbers of client machines (typically 10 or less), using a nodelock
license server on each machine is the simplest method of configuring LUM.
Log in as the root user and perform the following commands to configure a
machine as a nodelock license server:
# /usr/opt/ifor/ls/os/aix/bin/i4cfg -a n -S a
# /usr/opt/ifor/ls/os/aix/bin/i4cfg -start
24
Developing and Porting C and C++ Applications on AIX

The first command configures the local machine as a nodelock license server
and sets the option that the LUM daemons should be started automatically when
the system boots. The second command starts the LUM daemons.
Using the interactive configuration tool
As an alternative to using the above commands, you can use the
/usr/opt/ifor/ls/os/aix/bin/i4config interactive configuration script to perform the
same actions.
1. Log in as user ID root on the system where the license server will be installed.
2. Invoke the LUM configuration tool by entering the
/usr/opt/ifor/ls/os/aix/bin/i4config command. This is the command line
version of the LUM configuration tool.
3. Answer the LUM configuration questions as appropriate. The answers to the
configuration questions are dependent on the LUM environment you wish to
create.
The following are typical answers to the configuration questions of LUM in order
to configure both concurrent nodelock and concurrent network license servers on
a single system. You may change the various answers accordingly to suit your
preferred system environment. For details on configuring LUM, please read the
documentation that comes with LUM.
1. Select 4 “Central Registry (and/or Network and/or Nodelock) License Server”
on the first panel.
2. Answer y to “Do you want this system be a Network License Server too?”
3. Answer y to “Do you want this system be a Nodelock License Server too?”
4. Answer n to “Do you want to disable remote administration of this Network
License Server?”
5. Answer n to “Do you want to disable remote administration of this Nodelock
License Server?”
6. Select 2 “Direct Binding only” as the mechanism to locate a license server.
7. Answer n to “Do you want to change the Network License Server ip port
number?”
8. Answer n to “Do you want to change the Central Registry License Server ip
port number?”
9. Answer n to “Do you want to change the Nodelock License Server ip port
number?”
10.Select 1 “Default” as the desired server(s) logging level.
11.Enter blank to accept the default path for the default log file(s).
Chapter 1. C and C++ compilers
25

12.Answer y to “Do you want to modify the list of remote Nodelock and/or
Network License Servers this system can connect to in direct binding mode
(both for administration purposes and for working as Network License
Client)?”
13.Select 3 “Create a new list” to the direct binding list menu.
14.Enter the host name, without the domain, of the system you are configuring
LUM when prompted for the “Server network name(s).”
15.Answer n to “Do you want to change the default ip port number?”
16.Select 3 “Create a new list” to the direct binding list menu.
17.Enter the host name, without the domain, of the system you are configuring
LUM when prompted for the “Server network name(s).”
18.Answer n to “Do you want to change the default ip port number?”
19.Answer y to “Do you want the License Server(s) automatically start on this
system at boot time?”
20.Answer y to continue the configuration setup and write the updates to the
i4ls.ini file.
21.Answer y to “Do you want the License Server(s) start now?”
Both concurrent nodelock and concurrent network license servers should now be
configured on your system.
For more information on configuring and using LUM, refer to the LUM online
documentation supplied with AIX. As an alternative, the LUM manual, Using
License Use Management Runtime for AIX, can be viewed online in PDF format
at the following URL:
ftp://ftp.software.ibm.com/software/lum/aix/doc/V4.6.0/lumusg.pdf
1.5 Activating the LUM server
After configuring and starting the LUM server, you can enroll product licenses.
Before attempting to enroll a license, you must first ensure that the LUM
daemons are active. This can be done with the following command:
# /usr/opt/ifor/ls/os/aix/bin/i4cfg -list
Depending on the type of LUM server configured, the output will be similar to the
following:
i4cfg Version 4.6.6 AIX -- LUM Configuration Tool
(c) Copyright 1995-2002, IBM Corporation, All Rights Reserved
US Government Users Restricted Rights - Use, duplication or disclosure
26
Developing and Porting C and C++ Applications on AIX

restricted by GSA ADP Schedule Contract with IBM Corp.
Subsystem Group PID Status
i4llmd iforls 24006 active
If no subsystem is listed as active, then start them with the following command:
# /usr/opt/ifor/ls/os/aix/bin/i4cfg -start
The only daemon that must be active is the Nodelock License Server Subsystem
(i4llmd) daemon. The other daemons that may be active, depending on your
configuration, are as follows:
License Sever Subsystem (i4lmd)
Central Registry Subsystem (i4gdb)
Global Location Broker Data Cleaner Subsystem (i4glbcd)
1.6 Enrolling a product license
After LUM has been configured on your system, the product license certificates
can be enrolled with the LUM license server. Three LUM product license
certificates are provided with each of the compiler products:
1. Concurrent nodelock license certificate
2. Concurrent network license certificate
3. Simple nodelock license certificate
You should enroll the appropriate license certificate for the type of LUM
environment you have configured.
The default locations of the license certificates for the compiler products are
detailed in Table 1-3.
Table 1-3 License certificate locations
Compiler
License certificate type
Location
C for AIX Version 6.0
Concurrent Network
/usr/vac/cforaix_c.lic
Concurrent Nodelock
/usr/vac/cforaix_cn.lic
Simple Nodelock
/usr/vac/cforaix_n.lic
VisualAge C++ for AIX
Concurrent Network
/usr/vacpp/vacpp_c.lic
Version 6.0
Concurrent Nodelock
/usr/vacpp/vacpp_cn.lic
Simple Nodelock
/usr/vacpp/vacpp_n.lic
Chapter 1. C and C++ compilers
27

1.6.1 Enrolling a concurrent license
To enroll a Concurrent Network or Concurrent Nodelock license certificate,
perform the following steps:
1. Log in as root on the system where the license server is installed.
2. Invoke the LUM configuration tool by entering the LUM Basic License Tool
command as follows:
/usr/opt/ifor/ls/os/aix/bin/i4blt
The i4blt tool contains both a graphical user interface and a command line
interface. Note that the LUM daemons must be running before starting the i4blt
tool. Refer to 1.5, “Activating the LUM server” on page 26 for information on how
to check the status of the LUM daemons.
If the X11 run time (X11.base.rte fileset) has been installed on your system, the
GUI version of the tool will be invoked. Otherwise, the command line version will
be invoked, and an error will occur since the appropriate command line
parameters were not specified.
Enrolling using the graphical user interface
When the GUI version of i4blt tool is available, follow these steps:
1. Select the Products pull-down and click on the Enroll Product item.
2. Click on the Import button. The Import panel should be displayed.
3. In the Filter entry prompt, enter /usr/vacpp/*.lic if you are enrolling a
license for VisualAge C++ for AIX or /usr/vac/*.lic if you are enrolling a
license for C for AIX, and press Enter. This will show the various product
license files in the Files panel. The three license files for the product, as
detailed in Table 1-3 on page 27, should be displayed.
4. Select either the prod_c.lic or prod_cn.lic (where prod is either vacpp or
cforaix) license by clicking on the entry.
5. Click OK. The Enroll Product panel should be re-displayed with information
regarding the product indicated.
6. Click on the OK button of the Enroll Product panel. The Enroll Licenses panel
should be displayed.
7. Fill in the information on the Administrator Information portion of the panel
(optional.)
8. Fill in the number of valid purchased licenses of the product under Enrolled
Licenses in the Product information portion of the panel. (mandatory.)
9. Click on the OK button of the Enroll Licenses panel. The product should be
successfully enrolled. You may terminate the i4blt tool.
28
Developing and Porting C and C++ Applications on AIX

Enrolling using the command line
When you use the command line interface of i4blt tool, follow these steps:
1. From the required product license file, as detailed in Table 1-3 on page 27,
extract the i4blt command from the top of the file.
2. Replace number_of_lics from the command with the number of valid
purchased licenses of the product (mandatory).
3. Replace admin_name with the name of the administrator (optional).
4. Invoke this command as root from /var/ifor. The product should be
successfully enrolled.
1.6.2 Enrolling a simple nodelock license
Read the instructions at the top of the simple nodelock license certificate file. In
general, this type of license will be installed when no LUM system has been
configured. This means enrolling the license is simply a case of placing the
indicated license information line into the LUM nodelock file, /var/ifor/nodelock.
1.7 Invoking the compilers
Once a compiler product license has been enrolled, you are now ready to use the
compilers. However, the compiler drivers are not installed in a directory that is
searched with the default PATH environment variable. There are a number of
methods for resolving this issue:
Create symbolic links of the compiler drivers to /usr/bin using the ln
command.
Add the directory containing the compiler drivers to the default PATH
environment variable set in the /etc/environment configuration file.
Add the directory containing the compiler drivers to the PATH environment
variable in each user’s .profile shell configuration file.
Change the Makefiles used in your development environment to configure the
compiler macro to use the absolute path. For example:
CC=/usr/vac/bin/cc
Note: Creating symbolic links is the preferred option since it resolves the
problem for all users after a simple action by the root user.
Chapter 1. C and C++ compilers
29

1.7.1 Default compiler drivers
The Version 6.0 compiler products include a number of default compiler
configurations in the /etc/vac.cfg compiler configuration file. The default C++
command line driver is /usr/vacpp/bin/xlC. The three main C compiler command
line drivers are as follows:
/usr/vac/bin/cc
Extended mode C compiler.
/usr/vac/bin/xlc
ANSI C compiler, using UNIX header files.
/usr/vac/bin/c89
ANSI C compiler, using ANSI C header files.
There are a number of additional command line drivers available, each one
based on the basic cc, xlc, and xlC drivers described above, as described in
Table 1-4.
Table 1-4 Compiler driver extensions
Command extension
Meaning
_r
Use the UNIX98 threads libraries.
_r7
Use the POSIX Draft 7 threads libraries.
_r4a
Use the POSIX Draft 4 (DCE)b threads libraries.
128
Enable 128 bit double precision floating point values and use
appropriate libraries.
128_r
Enable 128 bit double precision floating point values and use
the UNIX98 threads libraries.
128_r7
Enable 128 bit double precision floating point values and use
the POSIX Draft 7 threads libraries.
128_r4
Enable 128 bit double precision floating point values and use
the POSIX Draft 4 (DCE) threads libraries.
a. Compiler drivers with extension _r4 are not supported on AIX 5L Version 5.2
and later. AIX 5L Version 5.1 supports those compiler drivers only if DCE is in-
stalled on the system.
b. DCE stands for Distributed Computing Environment.
For example, to compile an ANSI C program using Draft 7 of the POSIX threads
standard, use the xlc_r7 compiler driver. To compile a C++ program that uses
128-bit floating point values, use the xlC128 compiler driver.
Note: The use of compiler drivers with extensions _r4 and _r7 is discouraged
when developing new applications.
30
Developing and Porting C and C++ Applications on AIX

1.8 Where to find help
The Version 6.0 compilers provide documentation in both HTML and PDF format.
The default configuration makes it very easy to view the online documentation on
the machine on which it is installed. There is also information available on the
Web that provides useful information for developers using the AIX platform.
1.8.1 Online documentations
The Version 6.0 compiler documentation is written in HTML format. The HTML
files are stored in a single file in ZIP format. The files are viewed using an HTML
browser, which uses a cgi-bin script to extract and view the required files. There
is no need to manually unpack the ZIP file.
If not already installed, the online help documentation can be found in the
vac.html.en_US and vacpp.html.en_US filesets. Once installed, you can access
the online help with the /usr/vac/bin/cforaixhelp and
/usr/vacpp/bin/vacpphelp commands. The commands start the default
Netscape browser (which is supplied on the AIX Bonus Pack media) with the
correct URL.
If you are using the AIX CDE interface, the C for AIX compiler documentation can
also be started by selecting Application Manager C for AIX Help
Homepage, or Application Manager VisualAge C++ Professional Help
Homepage for VisualAge C++ for AIX.
The official compiler documentations are also available in PDF format in
/usr/vac/pdf and /usr/vacpp/pdf.
1.8.2 Viewing online documentation remotely
By default, it is not possible to view the online documentation from a remote
machine. It can be done in a simple way by logging in to the machine that has the
documentation installed, set the DISPLAY environment variable to use a remote
X11 display, then view the documentation by invoking the same command used
to view locally.
A better solution, particularly in larger environments or where remote clients do
not have the capability to display X applications, is to configure the machine to
allow remote viewing of the documentation.
Configuring the HTTP server
Suppose the machine that has the documentation filesets installed has a fully
qualified domain name of docs.ibm.com. The following example demonstrates
Chapter 1. C and C++ compilers
31

the steps performed on that machine to allow remote clients to view the compiler
documentation using their Web browser:
1. Log in as the root user.
2. Perform the following command:
# cp /etc/IMNSearch/httpdlite/httpdlite.conf \
/etc/IMNSearch/httpdlite/vacpp.conf
3. Edit /etc/IMNSearch/httpdlite/vacpp.conf, and make the following changes:
a. Change the HostName line from:
HostName localhost
to:
HostName docs.ibm.com
If the HostName line is not present, or has a comment symbol (#) at the
start of the line, then simply add the following line to the file:
HostName docs.ibm.com
b. Change the Port line from:
Port 49213
to:
Port 49214
c. Add one or more Allow lines to specify which hosts are permitted to
access the Web server. The Allow statement has the following syntax:
Allow network-ip network-mask
A client is only granted access if the following rule is met (& is a bitwise
AND operation):
client-ip & network-mask == network-ip & network-mask
For example, if you wanted machines within a network address, such as
9.x.x.x, to be able to access the help server, you would add the following
statement to vacpp.conf:
Allow 9.0.0.0 255.0.0.0
d. Save the file and exit the editor.
4. Edit the file /etc/inittab. There is a line that executes the httpdlite command
with a configuration filename argument. The line is as follows:
httpdlite:23456789:once:/usr/IMNSearch/httpdlite/httpdlite -r
/etc/IMNSearch/httpdlite/httpdlite.conf >/dev/console 2>&1
Make a copy of this line immediately below the original line. In the new line:
a. Change the first field from httpdlite to httpdlite2.
32
Developing and Porting C and C++ Applications on AIX

b. Change the part of the line that reads httpdlite.conf to vacpp.conf
The result should be as follows:
httpdlite2:23456789:once:/usr/IMNSearch/httpdlite/httpdlite -r
/etc/IMNSearch/httpdlite/vacpp.conf >/dev/console 2>&1
Save the file and exit from the editor.
5. Reboot the system or run the following command to start the second instance
of the httpdlite process:
/usr/IMNSearch/httpdlite/httpdlite -r /etc/IMNSearch/httpdlite/vacpp.conf
>/dev/console 2>&1
The steps described above configure an instance of an HTTP server to respond
on a specific port number to requests to access compiler documentation.
The following sections detail the additional steps required to configure the
documentation for each compiler product to be served by the HTTP server.
Configuring the C++ documentation
The following steps are required to enable the online documentation for the
VisualAge C++ for AIX Version 6 compiler to be served by the HTTP server:
1. Log in as the root user.
2. Change the directory to /var/vacpp/en_US.
3. Edit the hgssrch.htm file, and change the line:
<form action=”http://localhost:49213/cgi-bin/vacsrch.exe” method=”POST”
target=”content”>
to:
<form action=”http://docs.ibm.com:49214/cgi-bin/vacsrch.exe” method=”POST”
target=”content”>
Then, save the file and exit the editor.
4. Issue the following command:
/usr/IMNSearch/cli/imndomap -u "VAC6ENUS"
"http://docs.ibm.com:49214/cgi-bin/vahwebx.exe/en_US/vacpp6/Extract/0/"
"VisualAge C++"
Chapter 1. C and C++ compilers
33

5. Users can point their browser at the following URL to browse and search the
documentation:
http://docs.ibm.com:49214/cgi-bin/vahwebx.exe/en_US/vacpp/Extract/0/index.htm
Configuring the C compiler documentation
The following steps are required to enable the online documentation for the C for
AIX Version 6 compiler to be served by the HTTP server:
1. Log in as the root user.
2. Change the directory to /var/vac/en_US.
3. Edit the hgssrch.htm file, and change the line:
<form action=”http://localhost:49213/cgi-bin/caixsrch.exe” method=”POST”
target=”content”>
to:
<form action=”http://docs.ibm.com:49214/cgi-bin/caixsrch.exe” method=”POST”
target=”content”>
Then, save the file and exit the editor.
4. Issue the following command:
/usr/IMNSearch/cli/imndomap -u "CAIXENUS"
"http://docs.ibm.com:49214/cgi-bin/vahwebx.exe/en_US/cforaix/Extract/0/" "C
for AIX"
5. Users can point their browser at the following URL to browse and search the
documentation:
http://docs.ibm.com:49214/cgi-bin/vahwebx.exe/en_US/cforaix/Extract/0/index
.htm
1.8.3 Where to find help on the Web
IBM maintains many Web sites that provide useful information for developers
using the AIX platform. The most important ones are described in the following
sections.
AIX operating system documentation
The online documentation for the AIX operating system can be viewed at the
following URL:
http://www.ibm.com/servers/aix/library/
The site contains up-to-date versions of the HTML documentation supplied with
the AIX product media.
34
Developing and Porting C and C++ Applications on AIX

As new releases of the AIX operating system become available, they generally
add new functionality. As a developer, you might wish to use some of the new
functionality, but the decision to do so may also be based on the minimum level of
AIX required to use a particular feature. The IBM Redbook, AIX 5L Differences
Guide Version 5.2 Edition, SG24-5765, is updated with each new release of AIX,
and contains information on when particular features were introduced.
Compiler product information
The latest compiler products both have support Web sites that contain useful
hints, tips, frequently asked questions, and links to other useful Web sites. The
support page for the VisualAge C++ for AIX Version 6.0 compiler is:
http://www.ibm.com/software/ad/vacpp/support.html
The support page for the C for AIX Version 6.0 compiler is:
http://www.ibm.com/software/ad/caix/support.html
Information on the availability of IBM products for the AIX operating system,
along with details of when support for products will be withdrawn, is available on
the following Web site:
http://www.ibm.com/servers/aix/products/ibmsw/list/
PartnerWorld® for developers
PartnerWorld for Developers is a worldwide program supporting developers who
build solutions using IBM technologies. The program covers all IBM platforms,
not just AIX. Its Web site contains a lot of useful information for the AIX
developer, including white papers, sample code, and technology articles. It can
be located on the Web at the following URL:
http://www.ibm.com/partnerworld/developer
1.8.4 Applying fixes and service updates
From time to time, IBM issues fixes and corrective service updates to its products
that are still being supported. You can download these updates for C for AIX
under the “Support downloads” section of the support Web site at:
http://www.ibm.com/software/ad/caix/support.html
You will also find service updates for VisualAge C++ for AIX at:
http://www.ibm.com/software/ad/vacpp/support.html
You can also access the Fix Delivery Center for AIX 5L™ at:
http://techsupport.services.ibm.com/server/aix.fdc
Chapter 1. C and C++ compilers
35

where you can search for available fixes and updates based on fileset name or
APAR/PTF number.
Once you have downloaded the service update to the AIX system, which you are
going to apply the service update, follow these steps to apply the update:
1. Log on as the root user.
2. If the downloaded files are in compressed tar format (with the .tar.Z suffix),
uncompress and untar with the following commands:
# uncompress filename.tar.Z
# tar -xvf filename
3. Start the SMIT dialog with the following command:
# smit install_latest
4. Type the directory where the downloaded files reside as the INPUT device /
directory for software and press Enter.
5. Press the F4 key to generate a list of available filesets, or press Enter to install
the full product update.
6. If F4 is selected on the previous step, select the desired filesets by
highlighting them and then pressing the F7 key.
7. Press the Enter key once the required filesets have been selected.
8. Press the Enter key to start the update.
9. Press the Enter key to continue the update.
10.Press the F10 key to exit once the update has completed.
If you installed the compiler to a non-default directory as described in 1.3.2,
“Retaining a previous version of the compiler” on page 22, follow these
instructions to apply the update:
1. Create a text file listing the filesets you want to update, one fileset per line.
2. Execute the vacndi or vacppndi script with the following command:
# perl /usr/vac/bin/vacndi -d source_path -b target_directory \
-u update_file
where source_path is where the downloaded filesets are located, and
target_directory is the directory that contains the installation. update_file is
the file created in step 1 above. To update VisualAge C++ for AIX, replace
/usr/vac/bin/vacndi with /usr/vacpp/bin/vacppndi in the command in step 2.
36
Developing and Porting C and C++ Applications on AIX

2
Compiling and linking
Chapter 2.
Developers porting code to the AIX operating system from other UNIX operating
systems might, at first, have difficulties with the compile and linking tasks on AIX.
This chapter helps the developers with the tasks on AIX by providing the
following sections:
Section 2.1, “32- and 64-bit development environments” on page 38
Section 2.2, “Compiling and linking: A quick overview” on page 43
Section 2.3, “Resolving symbols at link-time” on page 53
Section 2.4, “Supported link methods on AIX” on page 63
Section 2.5, “Run-time linking” on page 68
Section 2.6, “Dynamic loading” on page 82
Section 2.7, “Commands when manipulating objects and libraries” on
page 85
Section 2.8, “Creating shared objects” on page 92
Section 2.9, “Shared libraries in a development environment” on page 99
For further information about the compile and linking tasks on AIX and how to
manage shared libraries, please refer to AIX 5L Version 5.2 General
Programming Concepts: Writing and Debugging Programs and the ld command
section in AIX 5L Version 5.2 Reference Documentation: Commands Reference.
Note: The definitions shown in Table B-1 on page 443 are very useful in
understanding the technical details of linking and loading process on AIX.
© Copyright IBM Corp. 2000, 2003. All rights reserved.
37

2.1 32- and 64-bit development environments
AIX, together with the C and C++ compilers, offer two different programming
models:
ILP32
LP64
ILP32, which stands for integer/long/pointer 32, is the native 32-bit programming
environment for AIX. It provides a 32-bit address space, with a theoretical
memory limit of 4 GB.
LP64, or long/pointer 64, is the 64-bit programming environment for AIX. It can
address memory beyond the 4 GB limit by providing a 64-bit address space. In
general, except for the data type size and alignment difference, LP64 supports
the same programming features as the ILP32 model, and backward compatibility
with the most widely used int data type.
According to the C and C++ language standards, int and short should be at least
16 bits, and long should be at least as long as int, but not smaller than 32 bits.
This relationship among the integral data types still holds true in the LP64 model:
sizeof(char) <= sizeof(short) <= sizeof(int) <= sizeof(long)
The LP64 data model is the de facto standard on 64-bit UNIX-based systems
provided by all major system vendors. Applications that transition to the LP64
data model are therefore highly portable to other LP64 vendor platforms.
Table 2-1 on page 39 lists the basic C and C++ data types on AIX and their
corresponding sizes in bits for both the ILP32 and LP64 programming models.
38
Developing and Porting C and C++ Applications on AIX

Table 2-1 C and C++ data type sizes in bits
Data type
ILP32
LP64
char
8
8
short
16
16
int
32
32
long
32
64
long long
64
64
pointer
32
64
float
32
32
double
64
64
long doublea
64 or 128
64 or 128
a. The size of long double is controlled by the -qlongdouble option, or when you
invoke the compiler with the 128 suffix (see Table 1-4 on page 30).
2.1.1 The 64-bit advantage
The primary objective for developing in 64-bit is to take advantage of the newer,
faster 64-bit hardware1 and operating systems2, and to make complex, memory
demanding applications (for example, database and scientific computational
applications) perform more efficiently. It offers the following benefits not found in
32-bit systems:
Full 64-bit addressing that expands the address space available to
applications beyond the 4 GB limit
Large process data space mapped in a large virtual address space
Support for large data structures and executable files
Large file support using standard system library calls
Large file caches on systems with large physical memory
64-bit data elements with instructions for performing efficient arithmetic and
logical computations as operations, using full-register widths, the full-register
set, and new instructions
Greater scalability of system derived data types, for example, time_t and
dev_t
1 Refer to 3.1.1, “How to determine hardware bit mode” on page 107.
2 Refer to 3.1.2, “How to determine kernel bit mode” on page 107.
Chapter 2. Compiling and linking
39

In addition, by keeping data in memory rather than writing out to disk, I/O bound
applications can realize improved performance, since disk I/O is usually more
time-consuming than memory access.
2.1.2 Compiler support
The compiler drivers (see 1.7.1, “Default compiler drivers” on page 30) by default
invoke the compiler and linker in 32-bit mode. The following features are provided
to enable 64-bit development:
Predefined __64BIT__ macro when invoked for 64-bit compilations
OBJECT_MODE environment variable
The -q64 option
-qarch support for 64-bit suboption
The compilers can be invoked for 64-bit or 32-bit mode by setting an environment
variable, or by using a command line option to set the compilation mode of the
compiler. Inconsistent options for compilation mode are resolved in the following
order:
1. OBJECT_MODE environment variable
2. Configuration file
3. Command line options
Please refer to the VisualAge C++ for AIX Compiler Reference, SC09-4959 for
more details on the options supported.
__64BIT__ preprocessor macro
Like many of the features supported by the compilers, the preprocessor macro
__64BIT__ is defined when compiling in 64-bit mode. The purpose of the macro
is to allow the programmer to select different data structures or lines of code for
32-bit and 64-bit execution in the same source file.
The macro can be tested using conditional directives. For example:
#if defined(__64BIT__)
/* 64-bit specific data structures or code */
#else
/* 32-bit mode */
#endif
This ability to choose execution mode (of the final executable) at compile time
using the __64BIT__ macro implies that there is no need to test execution mode
at run time. It also eliminates the need of maintaining different variations of
40
Developing and Porting C and C++ Applications on AIX

common header files. All the library header files are already coded in such a way
that they can be compiled regardless of the mode.
Command line options
The compilers can be invoked for 64-bit or 32-bit mode by setting an 64-bit mode
support in the C and C++ compilers is mainly provided by the two compiler
options, -q64 and -q32, respectively.
Note: Unlike the -X option for the utility commands, the -q32 or -q64 compiler
options do not allow a white space after -q; -q64 is valid whereas -q 64 is
invalid.
When used in conjunction with the -qarch option, they determine the mode and
instruction set for the target architecture. The -q32 and -q64 options take
precedence over the setting of the -qarch option. Conflicts between the -q32 and
-q64 options are resolved by the last option wins rule. Setting -qarch=com will
ensure future compatibility for applications, whereas specific settings will be
more hardware architecture dependent.
Please refer to the “Acceptable Compiler Mode and Processor Architecture
Combinations” section of the VisualAge C++ for AIX Compiler Reference,
SC09-4959 for valid combinations of the -q32, -q64, -qarch, and -qtune compiler
options.
Note: In 64-bit mode, -qarch=com is treated the same as -qarch=ppc.
OBJECT_MODE environment variable
Having to specify an option every time you compile can become cumbersome
and prone to mistakes (for example, inadvertently mixing 32-bit and 64-bit
objects in the link edit step, which is not supported). If you are always compiling
in one mode during a development session, you can set the OBJECT_MODE
environment variable to change the default compilation mode. Permissible values
for the OBJECT_MODE environment variable are:
(unset)
Generate and/or use 32-bit objects.
32
Generate and/or use 32-bit objects.
64
Generate and/or use 64-bit objects.
32_64
Accept both 32- and 64-bit objects.
Chapter 2. Compiling and linking
41

Note: The compiler and linker do not support OBJECT_MODE=32_64, and
using this choice will generate the error message:
1501-254 OBJECT_MODE=32_64 is not a valid setting for the compiler.
The benefit of using OBJECT_MODE to control the development environment is
that other utilities that are often used during development are also sensitive to
this environment variable.
Note: The use of OBJECT_MODE to determine the default mode can cause
serious problems if a user is unaware of the current setting. For example, the
user may not be aware that OBJECT_MODE has been set to 64 and may
unexpectedly obtain 64-bit object files from a compiling source file that is not
designed for 64-bit. We strongly urge users to be aware of the setting of
OBJECT_MODE at the time and to set OBJECT_MODE to ensure that the
compiler is invoked for the correct mode.
Linker command line options
The linker, ld, also supports 64-bit link editing with the -b64 option as well as the
default 32-bit link editing with -b32. Since the compiler drivers automatically sets
the linker option properly based on the -q64 and -q32 options, it is not necessary
to specify these linker specific options.
2.1.3 Utility commands support
The following utility commands that deal with object files, by default, assumes the
object file is in 32-bit XCOFF3 format:
ar
Maintains the indexed libraries used by the linkage editor.
dump
Dumps selected parts of an object file.
lorder
Finds the best order for member files in an object library.
nm
Displays information about symbols in object files,
executable files, and object-file libraries.
ranlib
Converts archive libraries to random libraries.
size
Displays the section sizes of the XCOFF object files.
strip
Reduces the size of an XCOFF object file by removing
information used by the binder and symbolic debug
program.
3 The eXtended Common Object File Format, XCOFF, is the object format for AIX. For a detailed
description of the XCOFF format, see the AIX 5L Version 5.2 Files Reference.
42
Developing and Porting C and C++ Applications on AIX

To support the 64-bit XCOFF object format, these utility commands have been
enhanced with the -X option. The -X option specifies the type of object file the
utility should examine, and accepts one of the following values:
32
Processes only 32-bit object files
64
Processes only 64-bit object files
32_64
Processes both 32-bit and 64-bit object files
Note: These utility commands also refer to the OBJECT_MODE environment
value; however, the -X option overrides the environment value setting.
Before developing your 64-bit application, you must confirm whether the libraries
that your application depends on are provided in 64-bit. Most C and C++ libraries
provided by AIX are hybrid mode archives (both 32- and 64-bit objects are
included). However, this may not be the case with third party vendor libraries.
See “32- and 64-bit objects” on page 48 and “Hybrid mode library archives” on
page 48 for further information about this topic.
2.2 Compiling and linking: A quick overview
This section provides a quick overview for developers about compiling and linking
on AIX.
2.2.1 Building C and C++ programs with system libraries
It is quite straightforward to build (compiling and linking) C and C++ programs
with system libraries on AIX. For example, to compile and link the very simple C
program shown in Example 2-1, do the following:
$ cc helloworld.c
Example 2-1 helloworld.c
#include <stdlib.h>
#include <stdio.h>
int main(int argc, char *argv[])
{
printf("Hello World\n");
exit(0);
}
Chapter 2. Compiling and linking
43

Then, the compiler driver4 will generate an executable file in the current directory
(if the executable file name is not specified with the -o option, the compiler driver
uses the default executable file name a.out):
$ ls -l a.out
-rwx------ 1 k5 k5 4429 Apr 09 16:18 a.out
If you execute the generated executable file, it prints the following output as
expected:
$ ./a.out
Hello World
Internally, the compiler driver cc does the following processes in this example:
1. Pick up the default compiler and linker options listed in /etc/vac.cfg (see
Example 2-2), then invokes several internal programs.
2. Generate an object file (in this example, the file name would be helloworld.o).
3. Invoke the linker, ld, and it links the object file with default libraries listed in
/vac/vac.cfg, in order to generate the executable file, a.out.
4. Remove the object file helloworld.o and temporary files.
Example 2-2 The cc compiler driver stanza in /etc/vac.cfg
* C compiler, extended mode
cc: use = DEFLT
crt = /lib/crt0.o
mcrt = /lib/mcrt0.o
gcrt = /lib/gcrt0.o
libraries = -L/usr/lpp/xlopt,-lxlopt,-lc
proflibs = -L/lib/profiled,-L/usr/lib/profiled
options = -qlanglvl=extended,-qnoro,-qnoroconst
These internal processes are shown if the -v compiler option is specified, as
shown in the following example:
$ cc -v helloworld.c
exec:
/usr/vac/exe/xlcentry(/usr/vac/exe/xlcentry,-D_AIX,-D_AIX32,-D_AIX41,-D_AIX43,-
D_AIX50,-D_AIX51,-D_AIX52,-D_IBMR2,-D_POWER,-qlanglvl=extended,-qnoro,-qnorocon
st,-ohelloworld.o,helloworld.c,/tmp/xlcW0m.58Ea,/tmp/xlcW1m.58Eb,/dev/null,hell
oworld.lst,/dev/null,/tmp/xlcW2m.58Ec,NULL)
exec:
/usr/vac/exe/xlCcode(/usr/vac/exe/xlCcode,-qlanglvl=extended,-qnoro,-qnoroconst
,/tmp/xlcW0m.58Ea,/tmp/xlcW1m.58Eb,helloworld.o,helloworld.lst,/tmp/xlcW2m.58Ec
,NULL)
4 Select the appropriate compiler driver as explained in 1.7.1, “Default compiler drivers” on page 30.
44
Developing and Porting C and C++ Applications on AIX

exec:
/bin/ld(/bin/ld,-b32,/lib/crt0.o,-bpT:0x10000000,-bpD:0x20000000,helloworld.o,-
L/usr/lpp/xlopt,-lxlopt,-lc,NULL)
unlink: helloworld.o
unlink: /tmp/xlcW0m.58Ea
unlink: /tmp/xlcW1m.58Eb
unlink: /tmp/xlcW2m.58Ec
Specifying a library name with the -l linker option
If your program needs to be linked with system libraries other than libc.a, specify
the library name using the -l linker option.
Note: The standard C library, libc.a, is defined in the compiler driver stanza as
an automatically linked library in /etc/vac.cfg, as highlighted in Example 2-2 on
page 44. Therefore, you do not have to specify the -lc option in most cases.
For example, the following example shows you how to link a user program, foo.c,
with the system provided mathematical library, /usr/lib/libm.a:
$ cc foo.c -lm
If the -l linker option is used, the linker treats it as a part of a library name. In this
example, the linker automatically adds “lib” in front of “m” and adds “.a” after “m”,
in order to complete the library file name. This is a generic behavior of linker on
most UNIX operating systems.
When searching libraries to resolve symbols, the linker always looks into two
directories, /usr/lib and /lib5, where most system libraries are installed.
This option can be specified more than once in the same command line. In the
following example, libabc.a and libdef.a will be searched from the /usr/lib or /lib
directories to resolve symbols:
$ cc foo.c -labc -ldef
To find the appropriate system library name, which contains functions or
variables you are going to use in your application, please consult the AIX 5L
Version 5.2 Technical Reference: Base Operating System and Extensions.
2.2.2 Objects and libraries
Compilers normally generate an object file from an input program source file. The
term object file is a generic term for a file containing executable code, data,
relocation information, a symbol table, and other information. Objects files are
5 The /lib directory is actually a symbolic link to the /usr/lib directory on AIX.
Chapter 2. Compiling and linking
45

defined by XCOFF6 (eXtended Common Object File Format) on AIX. Multiple
object files can be archived into a single library archive file, which is sometimes
simply called library. The merit of creating a library is that it is easy to handle a
fewer number of library files than many object files. Once a library is created from
several object files, you can erase those object files, as long as the program
source files are kept.
For example, assuming that there are three C program source files, foo1.c,
foo2.c, and foo3.c, in the current directory, to generate object files from these
source files, do the following:
$ cc -c foo1.c foo2.c foo3.c
The cc compiler driver would generate three object files as follows:
$ ls *.o
foo1.o foo2.o foo3.o
To archive them into a library, named libfoo.a, do the following:
$ ar -vq lifoo.a foo1.o foo2.o foo3.o
ar: Creating an archive file lifoo.a.
q - foo1.o
q - foo2.o
q - foo3.o
Note: If the libfoo.a library does not exist, this command creates it and enters
copies of the files foo1.o, foo2.o, and foo3.o into it. If the libfoo.a library does
exist, then this command adds the new members to the end of the library
archive without checking for duplicate members. The -v option sets verbose
mode, in which the ar command displays progress reports as it proceeds.
Figure 2-2 on page 50 illustrates these processes to create a library.
6 For the complete definition of XCOFF, refer to the AIX 5L Version 5.2 Files Reference.
46
Developing and Porting C and C++ Applications on AIX

Source files
foo1.c
foo2.c
foo3.c
Compile
Object files
foo1.o
foo2.o
foo3.o
Archive
foo1.o
Library archive file: libfoo.a
foo2.o
foo3.o
Figure 2-1 Object files and a library archive
Once the library is created, it can be maintained using the ar command as
follows:
To list the table of contents of a library, enter:
ar -v -t libfoo.a
This command lists the table of contents of the libfoo.a library, displaying a
long listing similar to the output of the ls -l command. To list only the
member file names, omit the -v option.
To replace or add new members to a library, enter:
ar -v -r libfoo.a foo1.o foo4.o
This command replaces the foo1.o member and adds foo4.o to the end of the
library.
To update a member that has been changed, enter:
ar -v -r -u libfoo.a foo2.o
This command replaces the existing foo2.o member, but only if the foo2.o file
has been modified since it was last added to the library.
To extract library members, enter:
ar -v -x libfoo.a foo1.o foo3.o
This command copies the foo1.o and foo3.o members into individual files
named foo1.o and foo3.o, respectively.
To delete a member, enter:
ar -v -d libfoo.a foo2.o
Chapter 2. Compiling and linking
47

This command deletes the foo2.o member from the libfoo.a library.
Note: The object files and libraries explained in this section are static
(non-shared). Shared object files and libraries must be created and treated by
specific ways, as explained in 2.8, “Creating shared objects” on page 92.
For further information about the usage of the ar command, please consult with
AIX 5L Version 5.2 Reference Documentation: Commands Reference.
32- and 64-bit objects
As explained in 2.1, “32- and 64-bit development environments” on page 38, AIX
provides two different development environments: 32- and 64-bit. Therefore,
object files on AIX have two different bit modes: 32-bit object files and 64-bit
object files. To build executable files, all the participated object files and archive
members must be in the same bit mode, either 32 or 64.
To distinguish the bit mode of object files, use the file command as follows:
$ file *.o
foo2.o: executable (RISC System/6000) or object module not stripped
foo3.o: executable (RISC System/6000) or object module not stripped
foo4.o: 64-bit XCOFF executable or object module not stripped
In this example, foo4.o is a 64-bit object, whereas foo2.o and foo3.o are 32-bit.
Hybrid mode library archives
A library archive can contain both the 32- and 64-bit object modules as its
members on AIX. This is called a hybrid mode library archive. In fact, most
system libraries provided by AIX are hybrid mode.
As explained in 2.1.3, “Utility commands support” on page 42, utility commands
that deal with object files, such as ar, dump, and nm, have been enhanced with the
-X option in order to support the 64-bit XCOFF format object format.
For example, to list only 64-bit object modules contained in the standard C
library, libc.a, do the following:
$ ar -X 64 -t /usr/lib/libc.a
frexp_64.o
itrunc_64.o
ldexp_64.o
modf_64.o
logb_64.o
scalb_64.o
finite_64.o
... rest of output is omitted on purpose ...
48
Developing and Porting C and C++ Applications on AIX

If you omit the -X 64 option, or specify -X 32, the command lists only 32-bit object
modules contained in libc.a. If -X 32_64 is specified, the command lists both 32-
and 64-bit object modules.
To determine if the functions you require are provided in 64-bit, use the nm
command. The mathematical library (libm.a) provided by AIX supports both a
32-bit and a 64-bit version of the acos() sub-routine in the following example
output7:
$ nm -X 32 -g /usr/lib/libm.a | head -5
/usr/lib/libm.a[acos.o]:
._Errno
U
-
.acos
T
0
acos
D
704
12
guesses
U
-
$ nm -X 64 -g /usr/lib/libm.a | head -7
/usr/lib/libm.a[acos_64.o]:
._Errno
U
-
._restf27
U
-
._savef26
U
-
.acos
T
0
acos
D
744
24
guesses
U
-
2.2.3 Difference between shared object and library on AIX
If you are already familiar with other UNIX operating systems, there is nothing to
be explained in the simple example used in 2.2.1, “Building C and C++ programs
with system libraries” on page 43. However, it is worth mentioning that the
program is linked with system provided several libraries, as shown in the
following ldd8 command output:
$ ldd ./a.out
./a.out needs:
/usr/lib/libc.a(shr.o)
/unix
/usr/lib/libcrypt.a(shr.o)
The two system libraries, libc.a and libcrypt.a, are needed by the executable file
and they both contain a shared object, named shr.o; in other words, these two
objects are dependent modules of the executable file. Although both libraries
coincidently contain the same name object file in this case, the two shared
objects with the same name shr.o are different modules. Most system libraries
provided by AIX contain one or more shared objects.
7 The -g option instructs the nm command to handle all archive members contained in the specified
library archive.
8 See 2.7.3, “ldd” on page 90 for further information about the ldd command.
Chapter 2. Compiling and linking
49

Figure 2-2 illustrates the compile and link processes of our example program and
references to shared objects contained in system libraries.
/usr/lib/libc.a
Source file
shr.o
helloworld.c
shrdl.o
Compile
shrpse.o
Object module file
helloworld.o
/usr/lib/libcrypt.a
Link
References
shr.o
Executable file
a.out
Figure 2-2 Compiling and linking
As shown in the following genkld9 command output, there are three shared
object modules loaded into the system global memory from the libc.a system
library on our test system:
$ genkld | grep 'libc.a'
d2023070 1df /usr/lib/libc.a/dl.o
d012cd7a 1da /usr/lib/libc.a/pse.o
d01cfbe0 1e6257 /usr/lib/libc.a/shr.o
Although, our example program shown in Example 2-1 on page 43 needs only
shr.o out of libc.a, the other two have been loaded since other programs needed
them.
In fact, this is the marked difference between AIX and other UNIX operating
systems. On other UNIX operating systems, shared libraries (also often referred
to as dynamic link libraries or DLL) are actually shared objects, whereas
multiple shared objects can be contained in a single library on AIX.10
9 The genkld command is used to list already loaded shared objects into the system memory. See
2.7.2, “genkld” on page 88 for further information about this command.
10 Technically, a shared object contained in a shared library archive is referred to as an archive
member that has been archived into the library from a separate object file with SHROBJ.
50
Developing and Porting C and C++ Applications on AIX

The terms, shared library and shared object, are generally used interchangeably
on other UNIX operating systems, whereas there is a distinct difference between
the two terms on AIX:
Shared object
A shared object is a single object file that has the
SHROBJ flag in the XCOFF header (see Example 2-3 on
page 52). A shared object normally has a name of the
form name.o on AIX. This is the default file name
extension generated by compilers.
Shared library
A shared library refers to an ar format archive library
file11, where one or more of the archive members is a
shared object. Note that the library can also contain
regular, non-shared object files, which are handled in the
normal way by the linker. A shared library normally has a
name of the form libname.a on AIX.
The magic number of the file is used by the linker to determine whether the file is
valid object file or not. Therefore, it is possible not to use .o as the file name
extension for shared objects, though it could be misleading. As for library name
naming convention (libname.a), it is strongly recommended to use it; otherwise
the linker cannot find the library location if the -l option is used.
2.2.4 Difference between shared and static objects on AIX
The AIX linking and loading mechanism uses an unique file name convention for
the shared and static object. On many UNIX operating systems, a shared object
file normally has a file name extension “.so” (stands for shared object) and a
static object file has normally a file name extension “.o” (stands for object).
However, an object normally has a file name extension “.o” regardless of shared
or static on AIX. Therefore, you cannot determine whether an object is shared or
regular static object from the file name extension on AIX.
Note: AIX also supports shared objects with a file name extension “.so” like
other UNIX operating systems do, which are used for run-time linking (see 2.5,
“Run-time linking” on page 68).
To determine whether the object file is shared or static, use the dump command.
As highlighted in Example 2-3 on page 52, if the SHROBJ keyword is shown in
the Flags line, then the object file is shared; otherwise, it is static.
11 An ar format archive file is a file that is created by the ar command.
Chapter 2. Compiling and linking
51

Example 2-3 SHROBJ flag in the XCOFF header of a shared object
$ dump -ov shr.o
shr.o:
***Object Module Header***
# Sections Symbol Ptr # Symbols Opt Hdr Len Flags
5 0x00251764 26925 72 0x3002
Flags=( EXEC DYNLOAD SHROBJ )
Timestamp = "Feb 03 08:59:14 2003"
Magic = 0x1df (32-bit XCOFF)
***Optional Header***
Tsize Dsize Bsize Tstart Dstart
0x00171bc0 0x00045ae0 0x00045bc8 0x00000000 0x00000000
SNloader SNentry SNtext SNtoc SNdata
0x0004 0x0000 0x0001 0x0002 0x0002
TXTalign DATAalign TOC vstamp entry
0x0005 0x0003 0x00043ad0 0x0001 0xffffffff
maxSTACK maxDATA SNbss magic modtype
0x00000000 0x00000000 0x0003 0x010b RE
To determine whether archive members in a library are shared or static, use the
dump command with the -g option.12 For example, the frexp.o archive member is
static, whereas shr.o is shared in the /usr/lib/libc.a library, as shown in
Example 2-4.
Example 2-4 SHROBJ flag in the XCOFF header of archive members in a library
$ dump -gov /usr/lib/libc.a
/usr/lib/libc.a[frexp.o]:
***Object Module Header***
# Sections Symbol Ptr # Symbols Opt Hdr Len Flags
3 0x0000030c 34 28 0x0000
Flags=( )
Timestamp = "Sep 15 16:12:35 2002"
Magic = 0x1df (32-bit XCOFF)
***Optional Header***
Tsize Dsize Bsize Tstart Dstart
0x00000108 0x00000080 0x00000000 0x00000000 0x00000108
12 The -g option instructs the dump command to examine archive members in the specified library.
52
Developing and Porting C and C++ Applications on AIX

... many output lines are omitted on purpose ...
/usr/lib/libc.a[shr.o]:
***Object Module Header***
# Sections Symbol Ptr # Symbols Opt Hdr Len Flags
5 0x00250c0c 26913 72 0x3002
Flags=( EXEC DYNLOAD SHROBJ )
Timestamp = "Sep 19 00:14:43 2002"
Magic = 0x1df (32-bit XCOFF)
***Optional Header***
Tsize Dsize Bsize Tstart Dstart
0x00171360 0x00045a08 0x00045bb8 0x00000000 0x00000000
SNloader SNentry SNtext SNtoc SNdata
0x0004 0x0000 0x0001 0x0002 0x0002
TXTalign DATAalign TOC vstamp entry
0x0005 0x0003 0x00043a00 0x0001 0xffffffff
maxSTACK maxDATA SNbss magic modtype
0x00000000 0x00000000 0x0003 0x010b RE
... rest of output is omitted on purpose ...
2.3 Resolving symbols at link-time
On AIX, symbol resolution is performed at link-time and cannot be rebound at the
program load-time, except for the two exceptions explained in the following
sections:
2.5, “Run-time linking” on page 68
2.6, “Dynamic loading” on page 82
This means that once an executable file was generated, dependent shared
objects and archive members in libraries must be referenced using the same
path name all the time; otherwise, the executable cannot run.
However, this does not necessary mean that dependent shared objects and
archive members in libraries must be in the same directory all the time. If no path
information is associated those objects and members, the system loader will look
for them in several directories specified by the -L linker option (see 2.3.1, “The -L
linker option” on page 55) and the LIBPATH environment variable (see 2.3.3,
“LIBPATH environment variable” on page 58), in addition to the default library
search directories, /usr/lib and /lib.
Chapter 2. Compiling and linking
53

This mechanism simplifies the work of the system loader when a module is
loaded, and thus results in better execution performance, though it could be seen
rigid and tedious from the application programmers’ view.
To demonstrate this behavior, we have copied libc.a to /tmp and build an
executable file from the program shown in Example 2-1 on page 43 as follows:
$ cp /usr/lib/libc.a /tmp
$ ls -l /tmp/libc.a
-r-x------ 1 k5 k5 6793964 Apr 18 15:35 /tmp/libc.a
$ chmod a+r /tmp/libc.a
$ ls -l /tmp/libc.a
-r-xr--r-- 1 k5 k5 6793964 Apr 18 15:35 /tmp/libc.a
$ cc helloworld.c /tmp/libc.a
Note: Shared objects or libraries must be readable from other users;
otherwise they are treated as private shared objects (see 2.9.2, “Private
shared objects” on page 101 for more detail).
The executable file cannot run if /tmp/libc.a is removed as follows:
$ ./a.out
Hello World
$ rm -i /tmp/libc.a
rm: Remove /tmp/libc.a? y
$ ./a.out
exec(): 0509-036 Cannot load program ./a.out because of the following errors:
0509-150 Dependent module /tmp/libc.a(shr.o) could not be loaded.
0509-022 Cannot load module /tmp/libc.a(shr.o).
0509-026 System error: A file or directory in the path name does not
exist.
The reason is that the reference information to the removed /tmp/libc.a file was
statically stored in the XCOFF header of the executable file, as emphasized in
Example 2-5 on page 55.
The directory path names, except for the first (index 0) section, shown in the
PATH column are called optional path components for dependent modules. In
Example 2-5 on page 55, libc.a has the optional path component /tmp.
54
Developing and Porting C and C++ Applications on AIX

Example 2-5 dump -H output with the full directory path name
$ dump -H a.out
a.out:
***Loader Section***
Loader Header Information
VERSION# #SYMtableENT #RELOCent LENidSTR
0x00000001 0x00000007 0x00000010 0x00000031
#IMPfilID OFFidSTR LENstrTBL OFFstrTBL
0x00000002 0x00000188 0x0000002a 0x000001b9
***Import File Strings***
INDEX PATH BASE MEMBER
0 /usr/lpp/xlopt:/usr/lib:/lib
1 /tmp libc.a shr.o
Note: It is always recommended to avoid having any optional path
components when creating shared objects and libraries and building
executable files, in order to avoid unnecessary dependency to the file path
names of the dependent modules.
If the same program is built as follows:
$ cc helloworld.c
the XCOFF header will not contain any path name information for libc.a as
highlighted in the following (the compiler driver automatically add -lc before
invoking the linker in this case):
***Import File Strings***
INDEX PATH BASE MEMBER
0 /usr/lpp/xlopt:/usr/lib:/lib
1 libc.a shr.o
Since libc.a has no path information, the system loader will look for this library
from the directories (/usr/lpp/xlopt:/usr/lib/:/lib) listed in the first (index 0) loader
header section of the XCOFF header of the generated executable file.
2.3.1 The -L linker option
If the program is referencing libraries, then use the -L and -l linker options rather
than specifying library file path names directly.
Chapter 2. Compiling and linking
55

For example, if the referenced library, libabc.a, is placed in the /project/test/lib
directory, specify the options as follows:
$ cc -o a.out main.c -labc -L/project/test/lib
The linker adds the /project/test/lib directory in front of the default directory
search path in order to look for referenced shared objects and libraries.
In this case, the directory name, /project/test/lib, will be added to in front of the
first entry for the first (index 0) loader header section of the XCOFF header of the
generated executable file, as shown in Example 2-6.
Example 2-6 dump -H with the PATH information
$ dump -H a.out
a.out:
***Loader Section***
Loader Header Information
VERSION# #SYMtableENT #RELOCent LENidSTR
0x00000001 0x00000007 0x00000010 0x00000037
#IMPfilID OFFidSTR LENstrTBL OFFstrTBL
0x00000002 0x00000188 0x0000002a 0x000001bf
***Import File Strings***
INDEX PATH BASE MEMBER
0
/project/test/lib:/usr/lpp/xlopt:/usr/lib:/lib
1 libabc.a shr.o
When executing the command, if libabc.a is found in the directories shown in the
following directories, the command can run:
/project/test/lib:/usr/lpp/xlopt:/usr/lib:/lib
Note: The shared objects and libraries must not have any optional path
component information in the XCOFF loader header section to be searched
from the directories listed in the first (index 0) loader header section of the
executable file.
2.3.2 Searching objects and libraries at link-time
The linker can handle two types of files as input: object file or library. Those
object files and libraries can be specified in the linker command line explained in
this section.
56
Developing and Porting C and C++ Applications on AIX

Note: The order of libraries and objects specified on the linker command line
is not important unless run-time linking (see 2.5, “Run-time linking” on
page 68) is used.
Object files
Specify the absolute path name for the object file. For example:
$ cc -o a.out main.c /prod/obj/shr1.o
Specify the relative path name for the object file. For example:
$ cc -o a.out main.c ../../prod/obj/shr1.o
Specify the file name for the object file, if it resides in the current directory. For
example:
$ ls main.c shr.o
main.c
shr.o
$ cc -o a.out main.c shr1.o
Except for the last method, the generated executable file would have an optional
path component for its dependent shared object shr.o in its XCOFF header, if
shr.o is a shared object.
Libraries
Specify the absolute path name for the library. For example:
$ cc -o a.out main.c /prod/lib/libabc.a
Specify the relative path name for the library. For example:
$ cc -o a.out main.c ../../prod/lib/libabc.a
Specify the file name for the library, if it resides in the current directory. For
example:
$ ls main.c libabc.a
libabc.a main.c
$ cc -o a.out main.c libabc.a
Specify the library installed directory using the -L and -l linker options. For
example:
$ cc -o a.out main.c -L/prod/lib -labc
Except for the last two methods, the generated executable file would have an
optional path component for its dependent shared library libabc.a in its XCOFF
header, if libabc.a is a shared library.
Chapter 2. Compiling and linking
57

Note: The linker -bnoipath options instructs the command to not include any
file path name information in the resultant module. The default option, -bipath,
preserves file path name information.
2.3.3 LIBPATH environment variable
If the LIBPATH13 environment variable is defined, the system loader refers to it in
order to search referenced shared objects and libraries that contain shared
archive members, when the executable file is invoked. The linker does not refer to
LIBPATH in order to look for shared objects and libraries.
The syntax of LIBPATH is:
LIBPATH=/path1:/path2:/path3:…
If defined, the system loader does the following processes when loading
modules:
1. Adds the text string value of LIBPATH in front of the PATH information stored
in the first (index 0) loader header section of the XCOFF header of the
executable.
2. The system loader will search shared objects and libraries referenced by the
executable in the directories in the order of the list created in the previous
step.
Note: When a non-root user is attempting to run a setuid or setgid executable,
only the directories listed in the header section of the executable are
searched; the LIBPATH variable is ignored, even if it is set.
For example, if the LIBPATH environment variable is defined as follows when
executing the program example used in 2.3.1, “The -L linker option” on page 55:
LIBPATH=/project/build/lib
then the system loader will search for the referenced shared objects and libraries
in the following order:
1. /project/build/lib
2. /project/test/lib
3. /usr/lpp/xlopt
4. /usr/lib
5. /lib
13 On some other UNIX operating systems the LD_LIBRARY_PATH variable is used for a similar
purpose. On AIX, however, LD_LIBRARY_PATH has no meaning.
58
Developing and Porting C and C++ Applications on AIX

Note: The shared objects and libraries must not have any optional path
component information in the XCOFF loader header section to be searched
from the directories specified by the LIBPATH environment variable.
For example, assuming the following:
The shared object shr.o is linked with main.o in order to create the executable
a.out.
Both shr.o and main.o are located in the current directory.
then you can select the following two methods to specify the file name shr.o in the
command line to build the executable:
1. cc -o a.out main.o shr.o
2. cc -o a.out main.o ./shr.o
The executable file generated using the first method would have the loader
information for shr.o, as shown in Example 2-7. In this case, the system loader
will look for shr.o in the /usr/lpp/xlopt, /usr/lib, and /lib directories. If shr.o is stored
in one of these directories, there is no need to set LIBPATH. If shr.o is stored in
other than these directories, set the LIBPATH environment variable accordingly.
Example 2-7 dump -H without dot in PATH
$ dump -H a.out
a.out:
***Loader Section***
Loader Header Information
VERSION# #SYMtableENT #RELOCent LENidSTR
0x00000001 0x00000007 0x00000010 0x00000036
#IMPfilID OFFidSTR LENstrTBL OFFstrTBL
0x00000003 0x00000188 0x0000002a 0x000001be
***Import File Strings***
INDEX PATH BASE MEMBER
0 /usr/lpp/xlopt:/usr/lib:/lib
1 libc.a shr.o
2 shr.o
Chapter 2. Compiling and linking
59

For example, if shr.o is stored in the current directory, run the executable file after
setting the variable as follows:
$ LIBPATH=$PWD ./a.out
If you have moved shr.o to /project/lib, then do the following:
$ LIBPATH=/project/lib ./a.out
If the executable is generated using the second method, the last five lines in
Example 2-7 on page 59 would be:
***Import File Strings***
INDEX PATH BASE MEMBER
0 /usr/lpp/xlopt:/usr/lib:/lib
1 libc.a shr.o
2 .
shr.o
The dot character in the PATH column for shr.o, which is an optional path
component for shr.o, makes a big difference. When executing the executable
generated with the second method, the system loader will not refer to LIBPATH
nor the directories listed in the first (index 0) loader header section, in order to
search shr.o. Therefore, shr.o must be located in the current directory whenever
the executable file is invoked.
Using LIBPATH for modules could not be loaded
If a shared object cannot be found by the system loader when trying to start an
executable, an error message similar to the following will be seen:
exec(): 0509-036 Cannot load program ex1 because of the following errors:
0509-022 Cannot load library libone.so.
0509-026 System error: A file or directory in the path name does not
exist.
The missing objects will be listed with 0509-022 error messages. Use the find
command to search the system for the missing shared objects. If the object is
found, try setting the LIBPATH environment variable to include the directory that
contains the shared object and restart the application. Also, ensure that the
object or library has read permission for the user trying to start the application.
A similar error message is produced when the system loader finds the specified
shared objects, but not all of the required symbols can be resolved. This can
happen when an incompatible version of a shared object is used with an
executable. The error message is similar to the following:
exec(): 0509-036 Cannot load program ./example because of the following errors:
0509-023 Symbol func1 in ex1 is not defined.
0509-026 System error: Cannot run a file that does not have a valid
format.
60
Developing and Porting C and C++ Applications on AIX

The unresolved symbols are listed in the 0509-023 message lines. Write down
the name of the unresolved symbol (func1), and use the dump -Tv command to
determine which shared object the executable expects to resolve the symbol
from. For example:
# dump -Tv example | grep func1
[4] 0x00000000 undef IMP DS EXTref libone.a(shr1.o) func1
This indicates that the executable is expecting to resolve the symbol func1 from
the shared object shr1.o that is an archive member of libone.a. This information
can help you start the problem determination process.
Note: To solve typical link-time errors, see 6.3, “Diagnosing link-time errors”
on page 239.
2.3.4 Link-time and load-time
Shared objects and libraries are used in two stages when creating and executing
an executable on AIX:
1. At link-time, the link editor (the ld command) searches the specified shared
objects and libraries to resolve all undefined symbols that are referenced in
the generating executable file. If a shared object file or library contains the
referenced symbols, the loader section of the XCOFF header of the created
executable file should contain a reference to that shared object or library.
In other words, symbols are exported from those shared objects or libraries
and imported from the executable file.
2. At the program load-time, the system loader (the kernel component that starts
new processes) reads the XCOFF header information of the executable and
attempts to locate any referenced shared libraries. Assuming all the
referenced shared objects and libraries are found, the executable can be
started. Then, the system loader attempts to load the sections in the
executable file into the appropriate segments in the process address space,
as explained in Table 2-2 on page 62. The program text in shared objects and
libraries is loaded into the global system memory by the first program that
needs it and is shared by all programs that use it.
Chapter 2. Compiling and linking
61

Table 2-2 XCOFF headers and loading target segments
Program executable
Corresponding
Loading target segment in the
components
section header
process address space
name in the
XCOFF format
Program text
.text
Process text segment
Program data
.data and .bss
Process data segment
(initialized and
un-initialized)
Referenced shared
loader
Shared library text segment
objects
(per-process shared library data
segment will be also populated with the
necessary data for the process)
For the detailed information about the each segment usage in the process
address space, see Chapter 3, “Understanding user process models” on
page 105.
Figure 2-3 on page 63 depicts the processes done by the system loader at the
program load-time.
62
Developing and Porting C and C++ Applications on AIX

XCOFF executable file
exec()
Files header
Process address space
Auxiliary header
Sections header
.text section
Process text segment
.data section
Process data segment
.bss section
loader section(s)
Library archive #1
shr1_1.o
shr1_2.o
shr1_3.o
Library archive #2
shr2_1.o
Shared library text segment
shr2_2.o
Per-process shared
library data segment
Figure 2-3 An XCOFF format executable file and exec()
Technically, each section header in the XCOFF file provides an offset address to
the actual section in the file. However, this is not shown in Figure 2-3 to avoid
unnecessary complexity.
2.4 Supported link methods on AIX
In order to link application program code with objects and libraries, AIX supports
several link methods shown in Table 2-3 on page 64 (for the definition of terms
used in this table, see Table B-1 on page 443). These methods are not mutually
exclusive, except for the combination of lazy loading and run-time link methods.
Therefore, an executable can be generated using more than one link methods.
Chapter 2. Compiling and linking
63

Table 2-3 Supported link methods
Link
Symbol
Symbol
Module
Linker
Explained section
method
resolution
rebound
loading
option
Default
link-time
N/A
program
N/A
Section 2.4.1, “AIX default linking” on
load-time
page 64
Static
link-time
N/A
program
-bstatic
Section 2.4.2, “Static linking” on
load-timea
page 66
Lazy
link-time
N/A
run time
-blazy
Section 2.4.3, “Lazy loading” on
loadingb
page 67
Run time
link-time
program
program
-brtl
Section 2.5, “Run-time linking” on
load-time
load-time
page 68
Dynamic
N/A
Section 2.6, “Dynamic loading” on
run time
loadingc
page 82
a. Program text of statically linked objects and archive members are contained in the executable file.
b. The lazy loading is actually a variation of the default link method. Except for the timing when ref-
erenced shared modules are loaded, the behavior is quite similar to the default link method.
c. Dynamic loading is a programming scheme provided by a set of sub-routines rather than by linker
options or special object file types.
2.4.1 AIX default linking
The linker can handle two types of files as input: object file or library. On AIX,
object files and archive members in a library can be static or shared, as
explained in 2.2.4, “Difference between shared and static objects on AIX” on
page 51. Regardless of static or shared, the input file names are specified in the
same way in the linker command line in the default link method, as explained in
2.3.2, “Searching objects and libraries at link-time” on page 56.
It is imperative to understand that the default link method is used when
generating an executable file, unless the following linker options are explicitly
specified:
-bstatic
Static linking
-blazy
Lazy loading
-brtl
Run-time linking
Shared and static objects
Static objects are always statically linked and contained in the generated
executable file. Shared objects are usually dynamically linked, thus the shared
library code is not contained in the generated executable file. However, shared
objects can be linked statically (see 2.4.2, “Static linking” on page 66).
64
Developing and Porting C and C++ Applications on AIX

Figure 2-4 depicts the difference between static and shared objects or archive
members in the generated executable file. In this figure, the following commands
are used in order to build the executable file a.out:
Compile
cc -c main.c foo1.c foo2.c
Link
cc main1.o foo1.o foo2.o -labc -L/project/lib
Source files
main.c
foo1.c
foo2.c
/project/lib/libabc.a
mbr1.o
Compile:
mbr2.o
Object files
main.o
foo1.o
foo2.o
mbr3.o
/usr/lib/libc.a
Link:
References
Loader section in the header
shr.o
Executable file
main.o
foo1.o
foo2.o
mbr2.o
Static object or archive member
Shared object or archive member
Figure 2-4 Static and shared text code in the executable file
If a shared object is linked, its shared program text is loaded into the system
shared library segment and shared by all processes that reference it, as depicted
in Figure 2-3 on page 63.
LDR_CNTRL=PREREAD_SHLIB
If not already loaded, a shared object is loaded into memory when the program
that depends on the shared object is executed. The loading process is done by
the kernel virtual memory manager (VMM) on a demand-page basis, and the
actual loading page size depends on the current VMM setting defined by several
options, which are set by either the vmo command on AIX 5L Version 5.2 or the
vmtune command on other versions of AIX.
For detailed information about VMM, please refer to the AIX 5L Version 5.2
Performance Management Guide.
Chapter 2. Compiling and linking
65

In some cases, especially if shared libraries are written by C++ and there are
many references between these libraries, it may be faster to read the library into
memory rather at the program load-time than the default demand-page basis.
In this case, set the following environment variable:
LDR_CNTRL=PREREAD_SHLIB
2.4.2 Static linking
AIX supports the -bdynamic and -bstatic linker options to determine how shared
objects and libraries should be treated by the linker.14 These options are toggles
and can be used repeatedly in the same linker command line.
When -bdynamic is in effect, which is the default, shared objects are used in the
usual way, whereas when -bstatic is in effect, all referenced objects are linked
statically, even if those objects are shared objects.
For example, if a program is built using the command line shown in Figure 2-5,
the three command line arguments, func1.o, -ldef, and -ljkl, are treated as static,
whereas the other arguments, -labc and -lghi, are treated as shared. Therefore,
the object module func1.o and referenced archive members in libraries, libdef.a
and libjkl.a, are statically linked with main.o.
Static linked
Static linked
$ cc -o a.out main.o -labc -bstatic func1.o -ldef -dynamic -lghi -static -ljkl -dynamic
Figure 2-5 The -bdynamic and -bstatic linker options
Note: We are assuming all objects and libraries except for main.o are shared
in Figure 2-5. Non-shared objects or archive members are statically linked
regardless of the effectiveness of -bdynamic.
If you use the -bstatic option, the -bdynamic option should be specified as the
last option on the link line to ensure that the system libraries are treated as
shared objects by the linker. Otherwise, all the object members in the system
libraries are treated as static, and the executable produced will be larger than
normal and may not work on future versions of AIX since it is statically linked with
a specific version of system libraries. The -bdynamic added at the end of the
command line ensures that the system libraries, such as libc.a, are processed as
shared objects.
14 The -bdynamic and -bstatic linker options has been supported by AIX, starting from Version 4.3.
66
Developing and Porting C and C++ Applications on AIX

Note: Statically linking shared objects or shared archive members in the 64-bit
development environment is not supported.
2.4.3 Lazy loading
Lazy loading,15 which is a variation of the default linking method, is a mechanism
for deferring the loading of modules until one of its functions is required to be
executed. By default, the system loader automatically loads all of the module’s
dependants at the same time.
By linking a module with the -blazy linker option, the module is loaded only when
a function within it is called for the first time; however, symbol resolution is
performed at link-time.
Note: Lazy loading only works if the run-time linker (-brtl option) is not
specified when building executables. A module is lazy loaded when all
references to the module are function calls. If variables in the module are
referenced, the module is loaded in the normal way.
Let us assume, for example, that main() calls myfunc1() while myfunc1() is in the
libone.so shared module. If myfunc1() calls myfunc2() conditionally in shared
module libtwo.so, then libtwo.so is a candidate for lazy loading. If myfunc2() is
never called, then libtwo.so is not loaded at all. If myfunc2() is called, the lazy
loading code executes the load() function to load libtwo.so, patches a function
descriptor to make sure subsequent calls simply go directly to the function itself,
and then invokes the called function. If for some reason the module cannot be
loaded, the lazy loader's error-handler is invoked.
Using lazy loading does not usually change the behavior of a program, but there
are the following exceptions:
Any program that relies on the order that modules are loaded in is going to be
affected, because modules can be loaded in a different order, and some
modules might not be loaded at all.
Be careful while comparing function pointers if you are using lazy loading.
Usually a function has a unique address in order to compare two function
pointers to determine whether they refer to the same function. When using
lazy loading to link a module, the address of a function in a lazy loaded
module is not the same address computed by other modules. Programs that
depend upon the comparison of function pointers should not use lazy loading.
15 The lazy loading function has been supported by AIX, starting from Version 4.3.
Chapter 2. Compiling and linking
67

If any modules are loaded with relative path names and if the program
changes working directories, the dependent module might not be found when
it needs to be loaded. When you use lazy loading, you should use only
absolute path names when referring to dependent modules at link-time.
The decision to enable lazy loading is made at link-time on a module-by-module
basis. In a single program, you can mix modules that use lazy loading with
modules that do not. When linking a single module, a reference to a variable in a
dependent module prevents that module from being loaded lazily. If all references
to a module are to function symbols, the dependent module can be loaded lazily.
Tracing the lazy loading execution
The environment variable LDLAZYDEBUG can be used to trace the lazy loading
activity as it takes place. The value of this variable is the sum of one or more of
the values shown in Table 2-4.
Table 2-4 LDLAZYDEBUG environment variable values
Value
Description
1
Show load or look-up errors. If a requested symbol is not available in the
loaded referenced module, a message is displayed before the error handler is
called.
2
Write tracing messages to stderr instead of stdout.
4
Display the name of the module that is getting loaded.
8
Display the name of the called function.
2.5 Run-time linking
As explained in 2.4.1, “AIX default linking” on page 64, all referenced symbols
must be resolved at link-time when building executable files using the default,
static, and lazy loading link methods on AIX. Those pre-resolved symbols cannot
be rebound after the executable files are created.
The run-time link method, or run-time linking, enables a program to resolve its
referenced symbols at the program load-time rather than link-time. It is the ability
to resolve undefined and non-deferred symbols in shared modules after the
program execution has already began. It is a mechanism for providing run-time
definitions (for example, function definitions that are not available at the program
link-time) and symbol rebinding capabilities. For example, if main() calls func1()
in libfunc1.so, which then calls func2() in libfunc2.so, assuming that libfunc1.so
and libfunc2.so were built to enable run-time linking, then the main application
68
Developing and Porting C and C++ Applications on AIX

could provide an alternate definition of func2() that would override the one
originally found in libfunc2.so.
Please note that it is the main application that has to be built to enable run-time
linking. Simply linking a module with the run-time link library is not enough. This
structure allows a module to be built to support run-time linking, yet continue to
function in an application that has not been so enabled.
Note: In AIX, even if the run-time link method is used, all symbols except for
the deferred symbols (see “Displaying symbol definition with dump -Tv” on
page 87 for the definition of deferred symbols) must be resolved at the
program load-time. However, in some other UNIX operating systems,
resolution of function symbols is deferred until the function is first called
(references to variables must be resolved at load-time). This allows the
definition for a function to be loaded after the module referring to that symbol
is loaded on those operating systems.
In order to use the run-time link method, the following points must be understood:
The run-time link method is enabled by the run-time link library
(/usr/lib/librtl.a) specified by the -brtl option when generating the executable
file. The -brtl option is mutually exclusive with the -blazy option.
Only run-time linking shared objects are run-time linked. Objects or archive
members other than run-time shared objects are linked in the default link
method.
Run-time linking shared objects must be created before linking them with the
main program, as explained in “Creating run-time linking shared objects” on
page 71.
Run-time linking shared objects should use the file name convention
libname.so. If this naming convention is used, it is easy to distinguish run-time
linking shared objects from other types of objects and easy to specify the file
path name on the command line (see 2.5.5, “Extended search order with the
-brtl linker option” on page 81).
An advantage of using run-time linking is that developers do not need to maintain
a list of module interdependencies and import/export lists (see 2.8.1, “Import and
export files” on page 92). By using the -bexpall linker option, all shared objects
can export all symbols, and the run-time linker can be used to resolve the
inter-module dependencies.
Chapter 2. Compiling and linking
69

2.5.1 How to use run-time linking
This section explains how to use run-time linking by providing several simple
examples.
Sample program source files
Our example program is composed of four C program source files, main.c
(shown in Example 2-8), func1.c (Example 2-9), func2.c, and func3.c
(Example 2-10). The source file of func2.c is the same as func3.c, except that the
function name is func2, not func3.
Example 2-8 Run-time linking example (main.c)
#include <stdio.h>
#include <stdlib.h>
extern void func1(void);
int main(int argc, char *argv[])
{
func1();
}
Example 2-9 Run-time linking example (func1.c)
#include <stdio.h>
#include <stdlib.h>
void func1(void)
{
printf("within function %s at line number: %d in %s\n"
, __FUNCTION__, __LINE__, __FILE__);
func3();
}
Example 2-10 Run-time linking example (func3.c)
#include <stdio.h>
#include <stdlib.h>
void func3(void)
{
printf("within function %s at line number: %d in %s\n"
, __FUNCTION__, __LINE__, __FILE__);
}
70
Developing and Porting C and C++ Applications on AIX

If this simple application is statically built as follows:
$ cc -c main.c func1.c func2.c func3.c
main.c:
func1.c:
func2.c:
func3.c:
$ ls main.o func1.o func2.o func3.o
func1.o func2.o func3.o main.o
$ cc main.o func1.o func2.o func3.o
then the generated executable file a.out prints the following output:
$ ./a.out
within function func1 at line number: 7 in func1.c
within function func3 at line number: 7 in func3.c
Please note the function func2 in func2.c is not called at all in this application.
This function is intentionally included in this example in order to show the
important aspect of run-time linking in later sections.
Creating run-time linking shared objects
In order to use run-time linking, run-time linking shared objects must be created
before linking them with the main program. To create a run-time linking shared
object, do the following:
1. Compile the source file and create an object file.
2. Re-link the object file with the -G linker option and create a run-time linking
shared object (.so).
Note: Multiple object files can be combined into a single run-time linking
shared object file.
3. Archive the created run-time linking shared object into a library archive
(libname.a). This is an optional step.
In fact, these steps are very similar to the steps to create regular (run-time linking
disabled) shared objects except for the specified linker options (see 2.8,
“Creating shared objects” on page 92).
For example, to create three run-time linking shared objects from func1.c,
func2.c, and func3.c, do the following:
$ cc -c func1.c func2.c func3.c
func1.c:
func2.c:
func3.c:
Chapter 2. Compiling and linking
71

$ ld -G -o func1.so func1.o -bnoentry -bexpall
$ ld -G -o func2.so func2.o -bnoentry -bexpall -lc
$ ld -G -o func3.so func3.o -bnoentry -bexpall -lc
$ ls *.o *.so
func1.o func1.so func2.o func2.so func3.o func3.so
The following linker options are specified in this example:
-bnoentry
Indicates that the output file has no entry point. To retain
any needed symbols, specify them with the -u flag or with
an export file. You can also use the -r flag or the -bnogc or
-bgcbtpass options to keep all external symbols in some
or all object files. If neither the -bnoentry nor the -bnox
option is used and the entry point is not found, a warning
is issued.
-bexpall
Exports all global symbols, except imported symbols,
unreferenced symbols defined in archive members, and
symbols beginning with an underscore (_). You may
export additional symbols by listing them in an export file.
This option does not affect symbols exported by the
-bautoexp option. This option only applies to AIX Version
4.2 and later.
When you use this option, you may be able to avoid using
an export file. On the other hand, using an export file
provides explicit control over which symbols are exported,
and allows you to use other global symbols within your
shared object without worrying about conflicting with
names exported from other shared objects. The default is
-bnoexpall.
-lc
Specifies referenced libraries. As for func2.o and func3.o,
the standard C library must be specified to resolve the
symbol of printf(). In the case of func1.o, there is no need
to specify any libraries, since it does not call any
functions, except for func3.
-G
The -G linker option is equivalent to specifying all the
options shown in Table 2-5 on page 73.
72
Developing and Porting C and C++ Applications on AIX

Table 2-5 Linker options equivalent to -G
Option
Description
-berok
Enables creation of the object file, even if there are unresolved
references
-brtl
Enables run-time linking. All shared objects listed on the command line
(those that are not part of an archive member) are listed in the output
file. The system loader loads all such shared modules when the
program runs, and the symbols exported by these shared objects may
be used by the run-time linker.
-bsymbolic
Assigns this attribute to most symbols exported without an explicit
attribute.
-bnortllib
Removes a reference to the run-time linker libraries. This means that
the module built with the -G option (which contains the -bnortllib option)
will be enabled for run-time linking, but the reference to the run-time
linker libraries will be removed. Note that the run-time libraries should
be referenced to link the main executable only.
-bnoautoexp
Prevent automatic exportation of any symbol.
-bM:SRE
Build this module to be shared and reusable.
For the detailed information about these linker options, please refer to the ld
command section in the AIX 5L Version 5.2 Reference Documentation:
Commands Reference.
Note: Compiler drivers also have the -G option. Do not use the compiler -G
option to create run-time linking shared objects from object files.
Creating an executable with run-time linking shared objects
In order to create an executable linked with run-time linking shared objects, do
the following:
1. Compile the main program source file, which contains main(), and create an
object file.
2. Re-link this object file with the -G linker option and create a run-time linking
shared object (.so).
3. Link the object file compiled from the main program source file with run-time
linking shared objects as well as the -brtl linker option specified.
Chapter 2. Compiling and linking
73

If the object file created from the main program source file does not have be
run-time linking enabled, then the required task is shorten as follows:
Use an appropriate compiler driver, not the linker, to compile the main program
source file, which contains main(), then link the object file with run-time linking
shared objects by specifying the -brtl option.
As for our program example, we have built the executable a.out as follows using
the later simple method:
$ ls *.o *.so
func1.o func1.so func2.o func2.so func3.o func3.so
$ cc main.c func1.so func2.so func3.so -brtl
$ ls a.out
a.out
If executed, the program would print the following error message and fail to
execute, since the dependent module func1.so could not be found by the system
loader:
$ ./a.out
exec(): 0509-036 Cannot load program ./a.out because of the following errors:
0509-150 Dependent module func1.so could not be loaded.
0509-022 Cannot load module func1.so.
0509-026 System error: A file or directory in the path name does not
exist.
If LIBPATH is set as follows, it executes as expected:
$ LIBPATH=$PWD ./a.out
within function func1 at line number: 7 in func1.c
within function func3 at line number: 7 in func3.c
If the linker command (ld) was used instead of the compiler driver (cc) as follows:
$ cc -c main.c
$ ld main.o func1.so func2.so func3.so -brtl
then the program would print the following error message, since the generated
executable would not contain the necessary start up routine:
$ LIBPATH=$PWD ./a.out
exec(): 0509-036 Cannot load program ./a.out because of the following errors:
0509-151 The program does not have an entry point or
the o_snentry field in the auxiliary header is invalid.
0509-194 Examine file headers with the 'dump -ohv' command.
74
Developing and Porting C and C++ Applications on AIX

2.5.2 Examining the executable and shared objects using dump
By using the dump command, you can examine whether the generated executable
is linked using run-time linking or not. For example, the dump command with the
-H option shows the header information for our example application a.out, as
shown in Example 2-11.
Example 2-11 includes two important lines (highlighted):
The run-time linking shared object func2.so is included in the dependent
module list, though any symbols in this module are not referenced in the
application at all.
Note: If this executable was compiled without -brtl, func2.so would not be
included in the dependent module list.
The run-time linker, the archive member shr.o in librtl.a, is included in the
dependent module list.
Example 2-11 dump -H a.out
$ dump -H a.out
a.out:
***Loader Section***
Loader Header Information
VERSION# #SYMtableENT #RELOCent LENidSTR
0x00000001 0x00000009 0x00000011 0x0000005e
#IMPfilID OFFidSTR LENstrTBL OFFstrTBL
0x00000006 0x000001c4 0x0000002a 0x00000222
***Import File Strings***
INDEX PATH BASE MEMBER
0 /usr/lpp/xlopt:/usr/lib:/lib
1 func1.so
2 func2.so
3 func3.so
4 libc.a shr.o
5 librtl.a shr.o
Chapter 2. Compiling and linking
75

Note: The order of run-time linking shared objects listed in the executable’s
header is same with the order of object files specified in the linker command
line when generating the executable file:
$ cc main.c func1.so func2.so func3.so -brtl
The dump command with the -Tv option shows the referenced symbols
information for our example application a.out, as shown in Example 2-12.
A very important fact in Example 2-12 is that there is no entry for the function
func3, which is called from func1 in func1.c. As explained in 2.3, “Resolving
symbols at link-time” on page 53, all symbols must be resolved at the program
link-time, and all the resolved symbol information will be shown in the XCOFF
loader section of the generated executable on AIX by default. However, in the
case of run-time linking, symbols to be resolved by the run-time linker will not be
shown in the XCOFF loader section of the generated executable, thus func3 is
not shown in Example 2-12.
Example 2-12 dump -Tv a.out (1)
$ dump -Tv a.out
a.out:
***Loader Section***
***Loader Symbol Table Information***
[Index] Value Scn IMEX Sclass Type IMPid Name
[0] 0x20000448 .data RW SECdef [noIMid] __rtinit
[1] 0x00000000 undef IMP RW EXTref libc.a(shr.o) errno
[2] 0x00000000 undef IMP DS EXTref libc.a(shr.o) exit
[3] 0x00000000 undef IMP DS EXTref libc.a(shr.o) __mod_init
[4] 0x00000000 undef IMP BS EXTref libc.a(shr.o) __crt0v
[5] 0x00000000 undef IMP BS EXTref libc.a(shr.o) __malloc_user_defined_name
[6] 0x00000000 undef IMP DS EXTref func1.so func1
[7] 0x00000000 undef IMP DS EXTref librtl.a(shr.o) __rtld
[8] 0x20000458 .data ENTpt DS SECdef [noIMid] __start
Example 2-13 on page 77 shows the header information for the dependent
shared object module func1.so. The double dot characters (..) in the last line
indicate that at least one run-time linking shared objects are required in order to
resolve unresolved symbols in this module.
76
Developing and Porting C and C++ Applications on AIX

Example 2-13 dump -H func1.so
$ dump -H func1.so
func1.so:
***Loader Section***
Loader Header Information
VERSION# #SYMtableENT #RELOCent LENidSTR
0x00000001 0x00000003 0x00000006 0x00000015
#IMPfilID OFFidSTR LENstrTBL OFFstrTBL
0x00000002 0x000000b0 0x00000000 0x00000000
***Import File Strings***
INDEX PATH BASE MEMBER
0 /usr/lib:/lib
1 ..
Example 2-14 shows the loader section information for the dependent shared
object module func1.so. There are two imported symbols, printf and func3, in this
module.16 The double dot characters (..) in the IMPid column indicate that the
symbol will be resolved by the run-time linker at the program load-time.
Example 2-14 dump -Tv func1.so
$ dump -Tv func1.so
func1.so:
***Loader Section***
***Loader Symbol Table Information***
[Index] Value Scn IMEX Sclass Type IMPid Name
[0] 0x00000050 .data EXP DS SECdef [noIMid] func1
[1] 0x00000000 undef IMP DS EXTref .. printf
[2] 0x00000000 undef IMP DS EXTref .. func3
Figure 2-6 on page 78 depicts the function calling relationship for our application
built in this section. The run-time linking shared object func2.so is referenced by
the executable file (a.out) and will be loaded into the system memory by the
system loader at the program execution time, even if a symbol in this object is not
referenced by the executable.
16 See “Displaying symbol definition with dump -Tv” on page 87 how to interpret columns shown in
the dump -Tv output.
Chapter 2. Compiling and linking
77

main.o
func1.so
func2.so
func3.so
main() {
func1() {
func2() {
func3() {
func1();
func3();
printf();
printf();
}
}
}
}
Unreferenced
Figure 2-6 Function calling relationship
2.5.3 Enabling the main program object as run-time linking
If you closely examine Example 2-12 on page 76, you notice that func1.so was
not linked as a run-time linking shared object. The IMPid column of the func1
symbol shows the file name of func1.so, thus it is treated as a regular shared
object. The reason is that the object file created from the main program source
file, which includes main(), had not been run-time linking enabled, therefore all
dependent modules that were required to resolve symbols referenced by main.o
had to be loaded at the program load-time by the system loader.
In order to enable the main program object as a run-time linking shared object,
we could have done the following (highlighted lines and command options are
different):
$ cc -c main.c func1.c func2.c func3.c
main.c:
func1.c:
func2.c:
func3.c:
$ ld -G -o main.so main.o -bexpall
ld: 0711-327 WARNING: Entry point not found: __start
$ ld -G -o func1.so func1.o -bnoentry -bexpall
$ ld -G -o func2.so func2.o -bnoentry -bexpall -lc
$ ld -G -o func3.so func3.o -bnoentry -bexpall -lc
$ cc -o a.out main.so func1.so func2.so func3.so -brtl
After specifying LIBPATH, the program prints the following output as expected:
$ LIBPATH=$PWD ./a.out
within function func1 at line number: 7 in func1.c
within function func3 at line number: 7 in func3.c
Example 2-15 on page 79 shows the referenced symbol information for a.out
generated by the compiler driver. Please note that Example 2-15 on page 79
does not contain the symbol information for either func1 or func3.
78
Developing and Porting C and C++ Applications on AIX

Example 2-15 dump -Tv a.out (2)
$ dump -Tv a.out
a.out:
***Loader Section***
***Loader Symbol Table Information***
[Index] Value Scn IMEX Sclass Type IMPid Name
[0] 0x200003e8 .data RW SECdef [noIMid] __rtinit
[1] 0x00000000 undef IMP RW EXTref libc.a(shr.o) errno
[2] 0x00000000 undef IMP DS EXTref libc.a(shr.o) exit
[3] 0x00000000 undef IMP DS EXTref libc.a(shr.o) __mod_init
[4] 0x00000000 undef IMP BS EXTref libc.a(shr.o) __crt0v
[5] 0x00000000 undef IMP BS EXTref libc.a(shr.o) __malloc_user_defined_name
[6] 0x00000000 undef IMP DS EXTref main.so main
[7] 0x00000000 undef IMP DS EXTref librtl.a(shr.o) __rtld
[8] 0x200003f8 .data ENTpt DS SECdef [noIMid] __start
Although, it is technically possible to link all dependent shared objects in an
application using run-time linking, it should be avoided in order for the application
performance. In general, to take advantage of the AIX architecture, the shared
modules should be as self contained as possible. Run-time linking should be
used only when necessary. Application program code ported from other UNIX
operating systems often does not have this sort of organization and can therefore
require extra effort to enable it on AIX. It is important to emphasize the fact that
the performance and efficiency of AIX is best explained by a well organized
application structure with a well defined interface between modules.
2.5.4 Rebinding symbols at the program load-time
Example 2-16 on page 80 and Example 2-17 on page 80 respectively show the
symbol information for func2.so and func3.so used in 2.5.1, “How to use run-time
linking” on page 70. Although these objects themselves are run-time linking
enabled, they have no dependent run-time linking shared object.
Chapter 2. Compiling and linking
79

Example 2-16 dump -Tv func2.so
$dump -Tv func2.so
func2.so:
***Loader Section***
***Loader Symbol Table Information***
[Index] Value Scn IMEX Sclass Type IMPid Name
[0] 0x00000000 undef IMP DS EXTref libc.a(shr.o) printf
[1] 0x00000050 .data EXP DS SECdef [noIMid] func2
Example 2-17 dump -Tv func3.so
$ dump -Tv func3.so
func3.so:
***Loader Section***
***Loader Symbol Table Information***
[Index] Value Scn IMEX Sclass Type IMPid Name
[0] 0x00000000 undef IMP DS EXTref libc.a(shr.o) printf
[1] 0x00000050 .data EXP DS SECdef [noIMid] func3
If func2.so and func3.so are swapped as follows, the new func2.so will contain
the function func3 and the new func3.so will contain func2:
$ mv func3.so temp.so
$ mv func2.so func3.so
$ mv temp.so func2.so
After func2.so and func3.so swapped, the run-time linker looks for the symbol
definition for func3 and found in the new func2.so in the order of func1.so,
func2.so, and func3.so. The referenced symbol func3 is rebound at the program
load-time without re-linking the application, thus the executable prints the
following output:
$LIBPATH=$PWD ./a.out
within function func1 at line number: 7 in func1.c
within function func3 at line number: 7 in func3.c
Figure 2-7 on page 81 depicts the function calling relationship for our application
after swapping func2.so and fun3.so. Now, new func2.so, which was originally
func3.so, contains func3, and new func3.so, which was originally func2.so,
contains func2.
80
Developing and Porting C and C++ Applications on AIX

main.o
func1.so
func2.so
func3.so
main() {
func1() {
func3() {
func2() {
func1();
func3();
printf();
printf();
}
}
}
}
Unreferenced
Figure 2-7 Function calling relationship after rebinding symbols
The run-time linking shared object func3.so is required by the executable file
(a.out) and will be loaded into the system memory by the system loader at the
program execution time, even if no symbol in this object is not referenced by the
executable.
2.5.5 Extended search order with the -brtl linker option
When an executable file is linked in the run-time link method by specifying the
-brtl linker option, the linker uses the extended search order to look for the shared
objects and libraries to resolve symbols. In order to exploit this extended search
order, run-time linking shared objects must follow the file name convention
libname.so.
If the -brtl linker option is specified, the linker looks for libname.so, as well as
libname.a, in the directories specified by the -L linker option in addition to the
default directory search path (/usr/lib and /lib). The -lname linker option can be
also used to specify libname.so as well as libname.a.
For example, if a run-time linking shared object named libfunc1.so, which is
stored in the /project/lib directory, is linked when generating the executable file
a.out, there are several methods to specify the object:
cc main.c /project/lib/libfunc1.so -brtl
This method almost invalidates the merit of the run-time link method.
Although libfunc1.so is loaded at the program load-time, it must reside in the
/project/lib directory when the executable is invoked.
cp /project/lib/libfunc1.so .; cc main.c libfunc1.so -brtl
This method is useful if all the run-time linking shared objects are located in
the current directory. However, once a stable version is developed, those
objects are most likely deployed in a specific directory. Therefore, it is less
useful compared to the third method.
Chapter 2. Compiling and linking
81

cc main.c -lfunc1 -L/project/lib -brtl
This is the recommended method to generate an executable file using the
run-time link method. The linker looks for the specified run-time shared
objects in the order explained in 2.3.1, “The -L linker option” on page 55.
Note: This method can be used only if the run-time shared object’s file
name follows the libname.so naming convention.
To follow the libname.so naming convention, either rename the run-time linking
shared object files or archive them into a library.
2.6 Dynamic loading
Dynamic loading is the process in which one can attach a shared library to the
address space of the process during execution, look up the address of a function
in the library, call that function, and then detach the shared library when it is no
longer needed. It is implemented as an interface to the services of the dynamic
linker. This gives programmers extra control in managing the memory allocated
to the shared library segment in an effective manner.
The dlopen() family of subroutines is supported on the AIX operating system.
The functions include:
dlopen()
dlclose()
dlsym()
dlerror()
dlopen()
The dlopen() subroutine is used to open a shared object, and dynamically map it
into the running programs address space. The syntax of the subroutine is as
follows:
#include <dlfcn.h>
void *dlopen (FilePath, Flags);
const char *FilePath;
int Flags;
The FilePath parameter is the full path to a shared object, for example, shrobj.o,
or libname.so. It can also be a pathname to an archive library that includes the
82
Developing and Porting C and C++ Applications on AIX

required shared object member name in parenthesis, for example,
/lib/libc.a(shr1.o).
The Flags parameter specifies how the named shared object should be loaded.
The Flags parameter must be set to RTLD_NOW or RTLD_LAZY. If the object is
a member of an archive library, the Flags parameter must be OR’ed with
RTLD_MEMBER.
The subroutine returns a handle to the shared library that gets loaded. This
handle is then used to with the dlsym subroutine to reference the symbols in the
shared object. On failure, the subroutine returns NULL. If this is the case, the
dlerror() subroutine can be used to print an error message.
Note: On AIX, the handle from dlopen() is specific to an instance of dlopen().
The dlclose() call only closes the instance of the object to which the handle
refers.
dlsym()
The dlopen() subroutine is used to load the library. If successful, it returns a
handle for use with the dlsym() routine to search for symbols in the loaded
shared object. Once the handle is available, the symbols (including functions and
variables) in the shared object can be found easily. For example:
lib_func = dlsym(lib_handle, "locatefn");
error = dlerror();
if (error) {
fprintf(stderr, "Error:%s \n",error);
exit(1);
}
The dlsym() subroutine accepts two parameters. The first is the handle to the
shared object returned from the dlopen() subroutine. The other is a string
representing the symbol to be searched for.
If successful, the dlsym() subroutine returns a pointer that holds the address of
the symbol that is referenced. On failure, the dlsym subroutine returns NULL.
This, again, can be used with the dlerror subroutine to print an error message as
shown above.
dlclose()
The dlclose() subroutine is used to remove access to a shared object that was
loaded into the processes’ address space with the dlopen subroutine. The
subroutine takes as its argument the handle returned by dlopen().
Chapter 2. Compiling and linking
83

dlerror()
The dlerror() subroutine is used to obtain information about the last error that
occurred in a dynamic loading routine (that is, dlopen, dlsym, or dlclose). The
returned value is a pointer to a null-terminated string without a final newline.
Once a call is made to this subroutine, subsequent calls without any intervening
dynamic loading errors will return NULL.
Applications can avoid calling the dlerror() subroutine, in many cases, by
examining errno after a failed call to a dynamic loading routine. If errno is
ENOEXEC, the dlerror() subroutine will return additional information. In all other
cases, it will return the string corresponding to the value of errno.
Note: The dlerror() subroutine is not thread-safe.
Using dynamic loading subroutines
In order to use the dynamic loading subroutines, an application must be linked
with the libdl.a library. The shared objects used with the dynamic loading
subroutines can be traditional AIX shared objects or shared objects that have
been enabled for run-time linking with the -G linker option.
When the dlopen() subroutine is used to open a shared object, any initialization
routines specified with the -binitfini option, as described in 2.8.4, “Initialization
and termination routines” on page 99, will be called before dlopen returns.
Similarly, any termination routines will be called by the dlclose subroutine.
Advantages of dynamic loading
Use of dynamic loading allows several benefits for application developers:
The ability to share commonly-used code across many applications, leading
to disk and memory savings.
It allows the implementation of services to be hidden from applications.
It allows the re-implementation of services, for example, to permit bug and
performance fixes or to allow multiple implementations selectable at run time.
The following example program is used to demonstrate the use of dynamic
loading subroutines:
/* File: main.c */
#include <stdio.h>
#include <stdlib.h>
#include <dlfcn.h>
int main(int argc, char *argv[])
{
84
Developing and Porting C and C++ Applications on AIX

void *handle;
void (*fct)();
printf("hello from main()\n");
if ((handle = dlopen("libfoo.so", RTLD_NOW)) == NULL) {
perror("dlopen");
exit(1);
}
if ((fct = (void (*)())dlsym(handle, "foo_")) == NULL) {
perror("dlsym");
exit(1);
}
(*fct)();
printf("exit from main()\n");
dlclose(handle);
}
/* File : foo.c - Shared library libfoo.so */
#include <stdio.h>
#include <stdlib.h>
void foo_(void)
{
printf("hello from foo_()\n");
}
Compile the program as follows:
$ cc -c foo.c
$ ld -o libfoo.so foo.o -bM:SRE -bnoentry -bexpall -lc
$ cc -o main main.c
The program will print the following output if libfoo.so is successfully found and
loaded by the dynamic linker:
hello from main()
hello from foo_()
exit from main()
2.7 Commands when manipulating objects and libraries
This section explains how to use the following commands, which are necessary
to manipulate objects and libraries on AIX by providing the following subsections:
Section 2.7.1, “dump” on page 86
Section 2.7.2, “genkld” on page 88
Chapter 2. Compiling and linking
85

Section 2.7.3, “ldd” on page 90
Section 2.7.4, “nm” on page 90
Section 2.7.5, “rtl_enable” on page 91
Section 2.7.6, “slibclean” on page 91
2.7.1 dump
The dump command is used to examine the header information of executable files
and shared objects. The main options that are useful when working with shared
libraries are the -H option and the -Tv options.
Displaying header information with dump -H
Use the dump -H command to determine which shared objects an executable or
shared object depends on for symbol resolution at run time. The interesting
information is in the last section of output, as shown under the ***Import File
Strings*** header in Example 2-18.
Example 2-18 Sample dump -H output
$ dump -H func1.so
func1.so:
***Loader Section***
Loader Header Information
VERSION# #SYMtableENT #RELOCent LENidSTR
0x00000001 0x00000003 0x00000006 0x00000023
#IMPfilID OFFidSTR LENstrTBL OFFstrTBL
0x00000003 0x000000b0 0x00000000 0x00000000
***Import File Strings***
INDEX PATH BASE MEMBER
0 /usr/lib:/lib
1 libc.a shr.o
2 ..
The number of INDEX entries will depend on how many shared objects the target
depends on for symbol resolution.
The INDEX 0 entry is a colon separated list of directories shown as /usr/lib:/lib in
Example 2-18. If the LIBPATH environment variable is not set when the
executable is started, the directories listed in the first (index 0) entry will be used
in order to look for the shared objects listed in subsequent entries. If LIBPATH is
86
Developing and Porting C and C++ Applications on AIX

defined, the directories listed in it will be used before the directories in the first
(index 0) entry.
The format of the interesting columns is as follows:
PATH
Optional path name component of the shared object or library. A
pathname will be present if a path name was specified on the
linker command line. It is recommended to avoid having this
optional path name. See 2.3.1, “The -L linker option” on page 55
and 2.3.3, “LIBPATH environment variable” on page 58 for further
detailed information how this optional path name component will
be selected depending on the linker command line file name
selections.
BASE
The name of the archive library containing the shared archive
member, or the name of the shared object itself.
MEMBER
The archive member name of the shared library. In the case of
non-archived shared objects, this column will be blank.
Note: If there is a .. entry in the BASE column, the shared object is enabled for
the run-time linking (see 2.5, “Run-time linking” on page 68).
Displaying symbol definition with dump -Tv
Use the dump -Tv command to examine the symbol information of a shared object
or executable. It displays information on the symbols the object is exporting. It
also lists the symbols the object or executable will try and import at load time
and, if known, the name of the shared object that contains those symbols. The
interesting columns are IMEX, IMPid, and Name, as highlighted in Example 2-19.
Example 2-19 Sample dump -Tv output
$ dump -Tv func1.so
func1.so:
***Loader Section***
***Loader Symbol Table Information***
[Index] Value Scn IMEX Sclass Type IMPid Name
[0] 0x00000000 undef IMP DS EXTref libc.a(shr.o) printf
[1] 0x00000050 .data EXP DS SECdef [noIMid] func1
[2] 0x00000000 undef IMP DS EXTref .. func3
Chapter 2. Compiling and linking
87

The values in these columns have the following meaning:
Name
The symbol name.
IMEX
Determines whether a symbol is exported from the object
file or archive member (EXP) or imported from the other
object files or archive members (IMP).
IMPid
For exported symbols, the value is not used and is always
[noIMid]. For imported symbols, there are three patterns
for the value, as shown in Table 2-6.
Table 2-6 IMPid values for imported symbols
Value
Description
File name
If the file name is in the form of libabc.a(def.o), then the symbol is imported
from the def.o member of the libabc.a library. If the file name is without
parentheses, then the symbol is imported from that shared object file. In
both cases, the dependent shared object or archive member will be loaded
by the system loader upon the program execution time.
..
The symbol will be resolved by the run-time linker. After resolving the
symbol, the run-time linker tells the system loader to load an appropriate
shared object or archive member.
[noIMid]
The symbol is a deferred symbol and will not be resolved by the run-time
linker. It is the application’s responsibility to resolve deferred symbols in
order to avoid the application abend due to the unresolved symbols. In
order for the programatic symbol resolution, use the dynamic loading (see
2.6, “Dynamic loading” on page 82).
2.7.2 genkld
The genkld command is used to list the shared objects that are loaded in the
system shared library segment. The output is quite lengthy and has three
columns, the virtual address of the object within the system segment, the size of
the object, and the name of the file that was loaded, as shown in Example 2-20.
Example 2-20 genkld output example (1)
$ genkld | pg
Virtual Address Size File
d1febce0 19907 /usr/lib/libcurses.a/shr.o
... rest of the output is omitted on purpose ...
If the file name does not have a slash character at its end, then it is a shared
archive member of a library archive. In Example 2-20, shr.o is a shared archive
member of /usr/lib/libcurses.a.
88
Developing and Porting C and C++ Applications on AIX

If the file name has a slash character at its end, it is a shared object file, as
shown in Example 2-21.
Example 2-21 genkld output example (2)
$ genkld | head -1; genkld | grep NIS
Virtual Address Size File
d00e7000 2237 /usr/lib/security/NIS/
$ ls -l /usr/lib/security/NIS
-r-xr-xr-x 1 root system 8760 Sep 15 2002 /usr/lib/security/NIS
$ file /usr/lib/security/NIS
/usr/lib/security/NIS: executable (RISC System/6000) or object module
The virtual address of a 32-bit shared object starts from 0xD, whereas the
address of a 64-bit shared object typically starts from 0x90000000, as shown in
Example 2-22.
Example 2-22 genkld output example (3)
$ genkld | head -1; genkld | grep 'libc.a'
Virtual Address Size File
d01bb3b8 1df /usr/lib/libc.a/dl.o
d014d0c2 1da /usr/lib/libc.a/pse.o
d01d0be0 1e5907 /usr/lib/libc.a/shr.o
90000000022aca0 20d75e /usr/lib/libc.a/shr_64.o
The 0xD segment is the system-wide 32-bit shared text segment, which is
shared by all the 32-bit processes on the system. As for 64-bit processes, shared
objects can be loaded into the system-wide 64-bit shared text segments, which
can be located from 0x9000_0000_0000_0000 to 0x9FFF_FFFF_FFFF_FFFF
(576 - 640 PB); however, they are most likely loaded into the first segment in this
address range.
For further information about the system-wide shared text segments, see 3.2,
“The 32-bit user process model” on page 109 and 3.3, “The 64-bit user process
model” on page 130.
Chapter 2. Compiling and linking
89

2.7.3 ldd
The ldd17 command lists the shared objects and archive members that will be
loaded to start the executable. The following example shows what shared object
modules and shared libraries (dependencies) are required to run the executable
file helloworld:
# ldd helloworld
helloworld needs:
./libone.so
./libtwo.so
/usr/lib/libc.a(shr.o)
/usr/lib/librtl.a(shr.o)
The command reports dependencies by traversing valid XCOFF header
information of the specified executable file.
2.7.4 nm
The nm command displays information about symbols in the specified file, which
can be an object file, an executable file, or shared object. For example:
$ nm libone.so
.myfunc1 T 144
.myfunc1 t 144 40
.myfunc2 T 0
.myfunc2 T 184
.myfunc2 t 184 40
.printf T 104
.printf t 104 40
TOC d 52
_$STATIC d 0 40
_$STATIC d 52 4
glink.s f -
glink.s f -
glink.s f -
myfunc1 U -
myfunc1 d 60 4
myfunc2 D 40 12
myfunc2 d 64 4
printf U -
printf d 56 4
source1.c f -
17 The ldd command is supported on AIX 5L Version 5.2 and later.
90
Developing and Porting C and C++ Applications on AIX

If the file contains no symbol information, the nm command reports the fact, but
does not interpret it as an error condition. The nm command reports numerical
values in decimal notation by default. Unlike the dump command, nm does not
display the shared object or archive member name that is expected to supply the
symbol.
2.7.5 rtl_enable
The rtl_enable command is used to convert a shared library that is not enabled
for run-time linking into a run-time linking enabled one.
2.7.6 slibclean
The slibclean command can be used by the root user to unload all shared
objects with a use count value of zero from the system shared library segment.
This command is useful in an environment when shared libraries are under
development. You can run the slibclean command followed by the genkld
command to ensure that the shared objects under development are not loaded in
the system shared library segment. This means that any application started after
this will automatically use the latest version of the shared objects since the
system loader will search for and load them. It also prevents multiple versions of
the same objects existing in the system segment.
During the development of shared objects, you may sometimes see an error
message similar to the following when creating a new version of an existing
shared object:
# make libone.so
cc -O -c source1.c
cc -berok -G -o libone.so source1.o
ld: 0711-851 SEVERE ERROR: Output file: libone.so
The file is in use and cannot be overwritten.
make: 1254-004 The error code from the last command is 12.
The error message means that the target shared object file is in use and it has
been loaded into the system shared library segment. Once loaded, the file is
marked as in use, even if the use count is already zero. Running the slibclean
command will unload all of the unused shared objects from the system. An
alternative (and simpler) method of avoiding this problem is to use the rm -f
command to remove the shared object before creating it.
However, frequent slibclean invocation on production systems should be
avoided, because it may affect the system performance by unloading frequently
used, but unused when the command is issued, shared objects and libraries on
the system. It is recommended to issue slibclean on production systems that
Chapter 2. Compiling and linking
91

are in a software maintenance phase, especially before the deinstallation of no
longer required applications or the updating of installed applications.
Note: The root authority is required to use the slibclean command.
2.8 Creating shared objects
This section explains how to create shared objects on AIX. Once shared objects
are created, you can archive them into a library using the ar command (see
2.2.2, “Objects and libraries” on page 45).
In order to create a shared object, the following tasks are required:
1. Compile the source code files and create object files. Use the appropriate
compiler driver to do so, for example, cc -c foo.c.
2. Create an export file to explicitly control exported symbols from the shared
object, which you are going to create.
3. Re-link the created object file(s) to a shared object file. It is possible to
combine multiple object files into a single shared object file. The following
linker options are used in this step:
-bE:export_file
Specifies the export file name created in the previous
step.
-bM:SRE
Marks the resultant object file as a re-entrant shared
object module.
-bnoentry
Indicates that the output object file has no entry point.
4. Create an import file to explicitly control how exported symbols in the shared
object will be imported by other modules. This is an optional step.
Note: Steps 2 and 4 can be skipped if the -bexpall linker option is used.
However, if this option is used, you cannot precisely control the symbol
exporting (see “The -bexpall linker option” on page 94).
2.8.1 Import and export files
An export file is a text file containing a list of symbols. It is used to control which
symbols are visible outside the shared object. The symbols not specified in the
export file are only visible to other routines within the shared object. The use of
export files allows a developer to create a shared object that has a well defined
interface. Only the symbols listed in the export file can be referenced by
92
Developing and Porting C and C++ Applications on AIX

executables and other shared objects that are linked with the object. Export files
are normally identified by the file name extension .exp.
An import file is a text file that lists the names of symbols that the shared object
may reference. It allows the object to be created without the source of those
symbols being available. Once an export file is created, an import file can be
created from the export file by adding several directive lines starting from “#!”.
Table 2-7 shows the supported directives for import files.
Table 2-7 Directive lines for import files
Directive
Description
#!
(Nothing after the #!) Use null path, null file, and null number. This is treated as
a deferred import by the system loader.
#!()
Use -bipath, -bifile, and -bimember. This line can be used if the import file is
specified as an InputFile parameter on the command line. The file must begin
with #! in this case. This line can also be used to restore the default name if it
was changed by another #! line.
#!path/file(member)
Use the specified path, file, and archive member.
#!path/file
Use the specified path and file, and a null member.
#!file
Use a null path, the specified file, and a null member. At run time, a list of
directories is searched to find the shared object.
#!(member)
Use -bipath, -bifile, and the specified member. At run time, a list of directories is
searched to find the shared object.
#!file (member)
Use a null path and the specified file and member. At run time, a list of
directories is searched to find the shared object.
#!.1
(A single dot) This name refers to the main executable. Use this file name when
you are creating a shared object that imports symbols from multiple main
programs with different names. The main program must export symbols
imported by other modules, or loading will fail. This import file name can be
used with or without the run-time linker.
#!..1
(Two dots) Use this name to list symbols that will be resolved by the run-time
linker. Use this file name to create shared objects that will be used by programs
making use of the run-time linker. If you use a module that imports symbols from
.. in a program that was not linked with the -brtllib option, symbols will be
unresolved, and references to such symbols will result in undefined behavior.
1. These directive lines have been supported on AIX since Version 4.2.
Chapter 2. Compiling and linking
93

The creation and use of import files are not mandatory except for the following
situation:
Limiting accessible symbols exported from shared objects so that only
specified symbols listed in the import file will be imported.
Creating multiple shared objects that have dependencies on each other (see
2.8.3, “Interdependent shared objects” on page 96).
Accessing special symbols such as symbols in the main program. In this
case, use the #!. directive line in the import file before the symbols that will be
imported from the main program.
If run-time linking (explained in 2.5, “Run-time linking” on page 68) or dynamic
loading (explained in 2.6, “Dynamic loading” on page 82) is not used, all
referenced external symbols must be accessible and resolvable when the
module is linked.
If symbols are listed after the #! directive in an import file, the linker considers
those symbols to be created declared differently, so they must be programatically
resolved and handled by a set of library calls provided by dynamic loading. If
symbols are listed after the #!.. directive in an import file, the linker considers
those symbols to be declared to be handled by the run-time linker.
The -bexpall linker option
If you are creating a shared object and want all symbols to be exported, then you
do not need to create an export file. You can use the -bexpall linker option, which
will automatically export all global symbols (except imported symbols,
unreferenced symbols defined in archive members, and symbols beginning with
an underscore). Additional symbols may be exported by listing them in an export
list. Any symbol with a leading underscore will not be exported by this option.
These symbols must be listed in an exports list to be exported.
The -bexpfull option
The new -bexpfull18 link option operates like -bexpall, except that it actually
exports all symbols and does not skip symbols that begin with an underscore.
Creating an export list using CreateExportList
You can use the /usr/vac/bin/CreateExportList shell script supplied with the C for
AIX Version 6 compiler to automatically generate the symbols that should be
included in an export list. It can save a considerable amount of time if you need to
create shared objects.
18 The -bexpfull linker option is supported on AIX 5L Version 5.2 and later and on AIX 5L Version 5.1
with 5100-02 Recommended Maintenance Level and later.
94
Developing and Porting C and C++ Applications on AIX

To use this command, do the following:
1. Compile all of the source files that will be included in the shared object.
2. Create a single file that lists the names of all of the object files that will be
included in the shared object. For example, create a file called object.list that
contains the following lines:
source1.o
source2.o
3. Invoke CreateExportList as follows:
/usr/vac/bin/CreateExportList file.exp -f object.list
where file.exp is the name of the export file you want to create, and object.list
is the file that contains the list of object file names.
4. Edit the resulting export file to remove the symbol names you wish to keep
private within the shared object.
5. Edit the export file to include the #!pathname(member) lines at the
appropriate positions in order to use the file as an import file.
2.8.2 A self-contained shared object
The scenario described in this section for creating a self-contained shared object
uses the following source code files:
The file source1.c is as follows:
/* source1.c : First shared library source */
void private(void)
{
printf(“private\n”);
}
int addtot(int a, int b)
{
int c;
c = a + b;
return c;
}
The file source2.c is as follows:
/* source2.c : Second shared library source */
#include <stdio.h>
int disptot(int a)
{
printf(“The total is : %d \n”,a);
}
Chapter 2. Compiling and linking
95

To create a self-contained shared object, do the following:
1. Create the object files that will be combined together to create the shared
object. This is achieved by using the -c compiler option. For example:
cc -c source1.c source2.c
2. Create an export file that lists the symbol names that should be visible outside
the shared object. In this example, the symbols addtotal and displaytotal are
the names of the functions that will be called by the main application. The
symbol names can also include variable names in addition to function names.
Create the libadd.exp export file as follows:
addtot
disptot
3. Create the shared object shrobj.so with the following command:
cc -o shrobj.o source1.o source2.o -bE:libadd.exp -bM:SRE -bnoentry
Use the dump command to examine the exported symbols from this shared
object, as shown in the following example:
$ dump -Tv shrobj.o
shrobj.o:
***Loader Section***
***Loader Symbol Table Information***
[Index] Value Scn IMEX Sclass Type IMPid Name
[0] 0x00000000 undef IMP DS EXTref libc.a(shr.o) printf
[1] 0x2000020c .data EXP DS SECdef [noIMid] addtot
[2] 0x20000218 .data EXP DS SECdef [noIMid] disptot
Two symbols, addtot and disptot, are exported from the created shared object
shrobj.o as we expected (see “Displaying symbol definition with dump -Tv” on
page 87 how to interpret the command output).
Note: If you do not have to precisely control which symbols will be exported,
you can skip the step 2 and use the following command line in step 3:
$ cc -o shrobj.o source1.o source2.o -bexpall -bM:SRE -bnoentry
2.8.3 Interdependent shared objects
The process for creating interdependent shared objects is similar to the process
of creating a self-contained shared object but requires the use of an import file.
Suppose there are two shared objects, shr1.o and shr2.o, and each references
symbols in the other. When creating the first shared object (shr1.o), the second
96
Developing and Porting C and C++ Applications on AIX

shared object may not exist. This means that when the command to create the
first shared object is executed, there will be unresolved symbols since, at this
point, the second shared object does not exist. To avoid this paradox, create and
use import files for these object files.
Consider the following files for use in this example scenario:
The file source1.c is as follows:
/* source1.c : First shared library source */
int function1(int a)
{
int c;
c = a + function2(a);
return c;
}
int function3(int a)
{
int c;
c = a / 2;
return c;
}
The file source2.c is as follows:
/* source2.c : Second shared library source */
int function2(int a)
{
int c;
c = function3(a + 5);
return c;
}
In this example, each source file needs to be made into a separate, shared
object. Note that the resulting shared objects are interdependent, since:
Subroutine function1 in source1.c calls function2 in source2.c.
Subroutine function2 in source2.c calls function3 in source1.c.
Create the export file libone.exp to define exporting symbols of the shared object
shr1.o, which we are going to create from source1.c:
function1
function3
then insert the following directive at the beginning of libone.exp in order to use
the file as an import file:
#!libone.a(shr1.o)
Chapter 2. Compiling and linking
97

Create the export file libtwo.exp to define exporting symbols of the shared object
shr2.o, which we are going to create from source2.c:
function2
then insert the following directive at the beginning of libtwo.exp in order to use the
file as an import file:
#!libtwo.a(shr2.o)
To create two libraries, libone.a and libtwo.a, which contain only one archive
member, shr1.o and shr2.o respectively, do the following:
cc -c source1.c source2.c
cc -o shr1.o source1.o -bE:libone.exp -bI:libtwo.exp -bM:SRE -bnoentry
ar rv libone.a shr1.o
cc -o shr2.o source2.o -bE:libtwo.exp -bI:libone.exp -bM:SRE -bnoentry
ar rv libtwo.a shr2.o
Note the use of the file libone.exp as an export file when creating the first shared
object and as an import file when creating the second. If the file was not used
when creating the second shared library, the creation of the second shared
object would have failed with an error message complaining of unresolved
symbols as follows:
cc -o shr2.o source2.o -bE:libtwo.exp -bM:SRE -bnoentry
ld: 0711-317 ERROR: Undefined symbol: .function3
ld: 0711-345 Use the -bloadmap or -bnoquiet option to obtain more information.
Figure 2-8 depicts the function calling relationship between these two shared
objects.
libone.a(shr1.o)
libtwo.a(shr2.o)
function1 {
function2 {
function2();
function3();
}
}
function 3 {
;
}
Figure 2-8 Function calling relationship for an interdependent shared object
A single import file can be used to list all symbols that are imported from different
modules. In this case, the import file is just a concatenation of the individual
export files for each of the shared objects. For example, Example 2-23 on
page 99 could have been used for the combined import file when linking two
shared objects in the previous example.
98
Developing and Porting C and C++ Applications on AIX

Example 2-23 Combined import file
#!libone.a(shr1.o)
function1
function3
* a comment line starts with an asterix character. blank lines are ignored.
#!libtwo.a(shr2.o)
function2
Note: It is recommended to create a single self-contained shared object
instead of multiple interdependent shared objects whenever possible, in order
to avoid the complexity and the possible application performance degrade.
2.8.4 Initialization and termination routines
Optional shared object initialization and termination routines can be specified
when creating the shared object. You can use one or the other, or both. The
routines may be useful for initializing dynamic data structures or reading
configuration information. The initialization routines are called by the program
startup code and are performed before the application main routine is started.
Termination routines are called when the program makes a graceful exit. They
will not be called if the program exits due to receipt of a signal.
The -binitfini linker option is used to specify the names of the routines along with
a priority number. The priority is used to indicate the order that the routines
should be called in when multiple shared objects with initialization or termination
routines are used.
2.9 Shared libraries in a development environment
When an application is started, the system loader reads the loader section of the
header of the executable file. It reads the dependency information for any shared
objects the executable requires and attempts to load the code for those shared
objects into the system shared library segment if they are not already loaded.
Shared objects that are loaded into the system shared library segment have an
attribute called the use count. Each time an application program that uses the
shared object is started, the use count is incremented. When an application
terminates, the use count for any shared objects it was using is decreased by
one. When the use count for a shared object in the system shared library
segment reaches zero, the shared object is not unloaded, but instead remains in
memory. This reduces the overhead of starting any applications that use the
shared object since they will not have to load the object into the system shared
segment again.
Chapter 2. Compiling and linking
99

To relinquish the unused shared objects or library archive members residing in
the system memory, you need to either explicitly call the slibclean command
(requires root authority on the system) or remove the shared objects or libraries
using the rm command.
However, this default behavior of shared objects and libraries on AIX can be
difficult to control, especially in the development environment where shared
objects or libraries can be constantly changed and altered so that new versions
should be used. Furthermore, multiple application developers might be working
with their own version of a single shared library in a large development
environment.
In order to effectively manage shared objects and libraries in the development
environment, there are several methods, including the following:
Use the -L and -l linker options and the LIBPATH environment variable.
Use private shared objects and libraries.
Note: To solve typical link-time and run-time errors, see 6.3, “Diagnosing
link-time errors” on page 239 and 6.4, “Diagnosing run-time errors” on
page 244.
2.9.1 Production and development environments
If your application development directory structure does not match the directory
structure used when your application is installed in a production environment,
then, potentially, you need to adjust the arguments used with the linker to ensure
that the resulting executables have the desired library search path.
When the application is installed in a production environment, for example, after
being installed on the production system, the directory structure may be different.
The method to use when compiling the executables will depend on the degree of
freedom the customer is permitted when installing the application. For example,
some applications expect that the executables and libraries must be installed in a
specific directory, such as /opt/productname. Some applications allow the
binaries and libraries to be installed in any directory structure.
If the libraries for the application need to be installed in a specific directory,
then you can either:
– Create the shared libraries and then copy them to the same directory
structure to be used when the product is installed in a production
environment. In this case, use the -L option to specify the directory where
the linker looks for shared objects and libraries. For example:
cc -o ../bin/app1 main.c -llibone -L/product/lib
100
Developing and Porting C and C++ Applications on AIX

– Create the shared libraries, but leave them in the development directory
structure. When compiling the applications, use absolute pathnames to
specify the shared libraries along with the -bnoipath linker option to
prevent the pathname being included in the header section of the final
executable. At the same time, use the -L option to specify the directory
where the libraries will exist on a production system. For example:
cc -o ../bin/app1 main.c -bnoipath ../lib/libone.a -L/product/lib
Note: In either case, all dependent shared objects and libraries should be
installed in /product/lib (or /usr/lib, /lib) on the production system.
If your application allows the executables and libraries to be installed in any
directory structure, then the LIBPATH environment variable must be used
accordingly on the production system to look for the dependent shared
objects and libraries.
Note: It is always recommended to avoid having any optional path
components (see “Displaying header information with dump -H” on page 86)
when creating shared objects and libraries and building executable files, in
order to avoid unnecessary dependency to the file path names of the
dependent modules.
2.9.2 Private shared objects
When used under normal circumstances, a shared object is loaded into the
system global shared object segment. Subsequent executables that use the
shared object benefit from the fact that it is already loaded.
In a development environment, particularly on a system with multiple developers,
it may be preferable to use a private copy of a shared object. This may be useful
when developing and testing new functionality in a shared object that is specific
to a particular version of the application that a single developer is working on.
If the shared object or library has the access permissions modified as detailed
below, then when the system loader starts an application that uses this shared
object, the shared object text will be loaded into the process private segment
rather than the system shared object segment. The shared object data will also
be loaded into the process private segment instead of its normal location of the
process shared object data segment. This means every application that uses
private shared objects and libraries will have its own private copy of the shared
object text and data while sacrificing the smaller size of a process private data
segment.
Chapter 2. Compiling and linking
101

When the program exits, all private shared objects required by the program will
be released from the memory.
Note: Private shared objects are not shown in the genkld command output,
even if they are loaded into memory.
To use a private copy of the shared text and data, modify the access permissions
as follows:
Remove the read permission for other of the shared object as follows:
chmod o-r foo.o
Remove the read permission for other of the shared library as follows:
chmod o-r libfoo.a
Note: If the read permission for other of a library is removed, all the shared
archive members in the library are considered private.
2.9.3 NFS consideration
Often, an application runs on NFS clients, whereas the actual executable file for
the application and referenced shared objects or libraries stored on the NFS
server. AIX tries to maintain the reliability of the running process by copying
modules to the NFS client’s paging space.
Although the likelihood of a shared module changing from a running process is
minimal, this behavior avoids the potential of a process abend due to an NFS
client and server losing synchronization in their respective understandings of the
contents of a particular file. If a shared object referenced by the application is
updated on the NFS server, and the inode does not change, then the kernel
believes that the existing copy is current, and therefore will not pick up the new
file. In this case, the slibclean command must be executed on the NFS client to
pick up the new module on the NFS server. This is only effective if the module in
question is no longer in use by any running process on the client.
2.9.4 Sufficient free disk space on the target directory and /tmp
The linker19 maps the output file into its shared memory segment using shmat()
(see “Mapping files with shmat()” on page 144 for detailed information about the
shmat() services). The pages are flushed to disk when the linker exits. If the
output file is on a remote file system, or on a file system that does not support
mapping, a temporary file is created locally, which is then copied to the remote
19 The linker, ld, is a 32-bit program.
102
Developing and Porting C and C++ Applications on AIX

file system. The temporary file will be created in the directory defined by the
TMPDIR environment value, or in /tmp, if TMPDIR is not defined. Therefore, it is
recommended to have enough free disk space on the target directory and /tmp in
a development environment for this action.
Chapter 2. Compiling and linking
103

104
Developing and Porting C and C++ Applications on AIX

3
Understanding user process
Chapter 3.
models
This chapter explains several memory models available on AIX and how these
models affect the heap and shared memory usage in your applications by
providing the following sections:
Section 3.1, “User process models on AIX” on page 106
Section 3.2, “The 32-bit user process model” on page 109
Section 3.3, “The 64-bit user process model” on page 130
Section 3.4, “Introduction to shared memory” on page 140
Section 3.5, “Shared memory segments allocation order” on page 149
Section 3.6, “Large page support” on page 157
For further information about the topics explained in this chapter, please refer to
the following sections in AIX 5L Version 5.2 General Programming Concepts:
Writing and Debugging Programs:
“Shared Libraries and Shared Memory”
“System Memory Allocation Using the malloc Subsystem”
© Copyright IBM Corp. 2000, 2003. All rights reserved.
105

3.1 User process models on AIX
Starting with AIX Version 4.3, AIX supports two user process models, 32- and
64-bit, which we explain in the following sections:
Section 3.2, “The 32-bit user process model” on page 109
Section 3.3, “The 64-bit user process model” on page 130
Before explaining these models in detail, we provide a short introduction in this
section. As shown in Figure 3-1, there are four combinations of kernel and
hardware support of 32- and 64-bit support.
A
B
C
D
User process
32
64
32
64
32
64
Kernel
32
64
32
64
Hardware
32
32
64
64
Figure 3-1 Hardware, kernel, and user process relationships
If the hardware bit mode is 32-bit, then 64-bit user processes cannot run on the
32-bit kernel (represented as A in Figure 3-1), also the 64-bit kernel cannot run
on the 32-bit hardware (B).
If the hardware bit mode is 64-bit, then both 32- and 64-bit user processes can
run regardless of the kernel type, though 64-bit user processes performance can
be slightly degraded on the 32-bit kernel (C).
Note: AIX has been supporting the 64-bit user process model starting with
Version 4.3. 64-bit application programs developed on AIX Version 4.3 are
source-compatible with AIX 5L Version 5.1 and later; however, they are not
binary-compatible. Therefore, those programs must be recompiled on AIX 5L
Version 5.1 and later before execution.
106
Developing and Porting C and C++ Applications on AIX

3.1.1 How to determine hardware bit mode
To determine the hardware bit mode on your system, run either of the following
commands, depending on the operating system versions:
AIX 5L Version 5.2:
getconf HARDWARE_BITMODE
AIX 5L Version 5.1 and AIX Version 4.31:
bootinfo -y
The command requires the root user authority.
These commands return either 32 (32-bit hardware) or 64 (64-bit hardware). The
following example shows the command output executed on an AIX 5L Version
5.2 partition on the pSeries 690:
# oslevel; getconf HARDWARE_BITMODE
5.2.0.0
64
If you need to know the hardware bit mode where your application is running, you
can use the sysconf() routine with the _SC_AIX_HARDWARE_BITMODE
parameter, as shown in the following code fragment:
#include <unistd.h>
long bit_mode;
bit_mode = sysconf(_SC_AIX_HARDWARE_BITMODE);
If the application is running on the 32-bit hardware, it returns 32. If running on the
64-bit hardware, it returns 64.
Note: In general, you do not have to know the hardware bit mode within your
application source code, since the difference of hardware bit-modes does not
affect user processes.
3.1.2 How to determine kernel bit mode
To determine the kernel bit mode on your system, run either of the following
commands, depending on the operating system versions:
AIX 5L Version 5.2:
getconf KERNEL_BITMODE
1 Prior to Version 4.3, AIX did not support 64-bit hardware models.
Chapter 3. Understanding user process models
107

AIX 5L Version 5.12:
bootinfo -K
The command requires the root user authority.
These commands return either 32 (32-bit kernel) or 64 (64-bit kernel). The
following example shows the command output executed on an AIX 5L Version
5.2 partition with the 64-bit kernel on the pSeries 690:
# oslevel; getconf KERNEL_BITMODE
5.2.0.0
64
If you need to know the kernel bit mode, where your application is running, you
can use the sysconf() routine with the _SC_AIX_KERNEL_BITMODE parameter
as shown in the following code fragment:
#include <unistd.h>
long bit_mode;
bit_mode = sysconf(_SC_AIX_KERNEL_BITMODE);
If the application is running on the 32-bit kernel, it returns 32. If running on the
64-bit kernel, it returns 64.
Note: In general, except for developing kernel extensions, you do not have to
know the kernel bit mode within your application source code, since the
difference of kernel bit-modes does not affect user processes.
3.1.3 How to determine user process bit mode
To determine the user process bit mode, run the file command with the
executable file of the corresponding user process.
If the executable file is 64-bit, then the command prints 64-bit XCOFF
executable, as shown in the following example:
$ cc -o helloworld32 helloworld.c
$ cc -q64 -o helloworld64 helloworld.c
$ file helloworld*
helloworld.c: C program text
helloworld32: executable (RISC System/6000) or object module not stripped
helloworld64: 64-bit XCOFF executable or object module not stripped
2 Prior to Version 5.1, AIX supported 32-bit kernels (UP or MP) only.
108
Developing and Porting C and C++ Applications on AIX

Note: The -q64 option instructs the compiler to generate a 64-bit executable
file (see 2.1.2, “Compiler support” on page 40). Although you can generate
64-bit executable files on 32-bit hardware systems, you cannot run them on
32-bit hardware systems.
3.2 The 32-bit user process model
This section provides a brief explanation about 32-bit user process models.
Every 32-bit user process has its own address space, whose maximum size is 4
GB. An address space is prepared by the kernel as part of the process
initialization tasks as a subset of virtual memory address on the system. A 32-bit
address space is composed of up to sixteen 256 MB memory chunks called
segments. Each segment has a specific purpose illustrated in several figures in
the following sections. A segment is divided into smaller size memory chunks
called pages. A page is an atomic unit, with which the virtual memory manager
(VMM) in the kernel uses to satisfy many different memory requests.
On AIX, the default page size is 4 KB, unless the large page support3 is used. If
the default page size, 4 KB, is used, a segment contains 256 MB / 4 KB = 256 / 4
x 1024 = 64 K = 65,536 pages.
In the 32-bit user process model, there are several variations, called large
memory model and very large memory model, depending on the required heap
size for the process. We also explain these models in the following sections:
Section 3.2.2, “Large memory model” on page 116
Section 3.2.3, “Very large memory model” on page 117
3.2.1 Default memory model
Figure 3-3 on page 111 illustrates the segment usage of the default 32-bit
memory model. This is the default memory model, unless you explicitly set a
linker option (-b:maxdata) when generating executables or set an environment
value (LDR_CNTRL) when executing 32-bit executables.
3 See 3.6, “Large page support” on page 157 for more detail about the large page support.
Chapter 3. Understanding user process models
109

Segment
Usage
number
0x0
Kernel text and data
0x1
User text
0x2
User stack, data
0x3
0x4
0x5
0x6
0x7
Available for the user process
if shmat() or mmap() is called.
0x8
0x9
0xA
0xB
0xC
0xD
Shared library text
0xE
0xF
Per-process shared library data
Figure 3-2 Default memory model (segment usage)4
The usage of each segment is briefly described in the following:
The first segment, 0x0, is reserved by system for the kernel text and data.
Therefore, access from the user process to the segment 0x0 is prohibited.5
The 0x1 segment contains user process text. If another process shares the
same executable file, there will be only one instance of the text pages for that
executable file on the system.
The 0x2 segment contains user data, heap, and stack. We provide detailed
information about this segment in “Process private segment (0x2)” on
page 111.
The 0xD segment contains shared library text. This segment is shared by all
32-bit user processes. The virtual addresses of loaded shared text objects
can be examined using the genkld command. For further information about
the usage of genkld, see 2.7.2, “genkld” on page 88.
The 0xF segment contains per-process shared library data.
4 The access to the 0xE segment from a user process has been supported on AIX Version 4.2.1 and
later.
5 User processes can access to kernel data only through the system calls with specific ways.
110
Developing and Porting C and C++ Applications on AIX

Segments 0x3 - 0xC and 0xE are attached to the address space if the
process called shmat() or mmap() routines to allocate shared memory
segments. These segments are allocated in order, starting from 0x3 towards
0xE excluding 0xD, unless DSA (Dynamic Segment Allocation) is used.
Further detailed explanation is provided in 3.5.1, “Order in the 32-bit default
memory model” on page 153 and 3.5.2, “Order in the 32-bit very large
memory model with DSA” on page 154.
Process private segment (0x2)
The process private segment, 0x2, is the most interesting segment for
application developers on AIX. In this segment, many memory components that
are mandatory for a user process are allocated (see Figure 3-3). In this figure,
white rectangles represent un-allocated virtual address pages; if a process
touches these address ranges, it receives the SIGSEGV6 signal and dumps a
core file. Highlighted rectangles represent allocated virtual address pages. You
may notice that a 32-bit address space is quite vacant in most cases.
Process private segment (0x2)
User text:
0 MB (offset)
User data
0x0
0x1
0x2
0x3
0x4
0x5
Shared memory:
0x6
0x7
User stack
0x8
errno
User
errnop
0x9
0xA
Shared library text:
Primary thread
kernel stack
Kernel
0xB
ublock
0xC
uthread block
0xD
0xE
Per-process loader heap
256 MB (offset)
0xF
Shared library data:
: Allocated virtual address pages
: Un-allocated virtual address pages
Figure 3-3 Default memory model (detail)
6 Signal for segmentation violation
Chapter 3. Understanding user process models
111

First of all, some pages within this segment are prepared by the kernel as a part
of the process initialization tasks. The access to those pages from the user
process is prohibited (represented as several highlighted rectangles under the
Kernel in the right most rectangle). The information contained in this area are
only accessible from the user process through system calls.
The rest of the pages within this segment can be accessed by the process,
though there is a big hole between the process heap (represented as “User data”
in the figure) and stack. If the process touches pages in this hole, it receives the
SIGSEGV signal and dumps a core file.
User data contains three different memory components: initialized data,
un-initialized data,7 and heap. To demonstrate how they are mapped into virtual
pages, we have prepared a simple program shown in Example 3-2 on page 113.
When compiled and executed, it prints addresses of several predefined symbols
in the XCOFF format, as shown in Example 3-1. Those symbols, starting from
the underscore character, are well explained in “XCOFF Object (a.out) File
Format”, AIX 5L Version 5.2 Files Reference. We only explain _data and _edata
here; the _data symbol defines the starting address of the initialized data in the
executable file, while the _edata symbol defines the first location address after
the initialized data in the executable file.
The size command tells you the size of the .text, .data, and .bss segments of the
specified executable file.8
Example 3-1 Addresses of several predefined symbols
$ size -fx a.out
a.out: 428(.text) + 234(.data) + 8(.bss) + 2db(.loader) = 0x93f
$ a.out
_text = 0x10000128
_etext = 0x10000550
_data = 0x20000550
_edata = 0x20000784
argv[] = 0x2ff22a14
environ[] = 0x2ff22ff4
errnop = 0x2ff22ffc
errno = 0x2ff22ff8
i_initialized = 0x200006f0
l_uninitialized = 0x20000788
str_buf = 0x2ff219b0
heap_mem = 0x20000798
7 The un-initialized data is generally referred to as BSS (block started by symbol) in computer
science terminology.
8 The -x option instructs the command to display data in the hexadecimal format.
112
Developing and Porting C and C++ Applications on AIX

As for the addresses of _data and _edata, we can easily subtract 0x20000550
from 0x20000784 using the bc command, as shown in the following example:
$ bc
ibase=F
obase=F
20000784 - 20000550
234
^D
To quit the command, type Control-D. The ibase=F and obase=F keywords
instruct the command to treat both input and output as hexadecimal numbers.
The answer is 0x234, which is equivalent to the .data segment size in the
executable file reported by the size command.
Note: You need to use upper-case characters (A - F) to represent
hexadecimal 0xA - 0xF with the bc command.
After the initialized data, un-initialized data (BSS) is loaded into the memory as
shown by the address of l_uninitilized. Then heap starts as shown by the address
of heap_mem. It grows toward the last address of this segment (0x2FFFFFFF).
On the other hand, the stack grows from a relatively higher address toward the
first address of this segment (0x20000000). The user process stack contains
several predefined symbols, such as argv, environ, errno, and so on. The user
process stack grows, if the process calls functions or allocates auto storage class
symbols.
Note: Thread stacks for Pthreads, except the initial thread within a
multi-threaded processes, are allocated in the process heap. See 8.3.4,
“Thread stack” on page 292 for further information about the thread stack for
multi-threaded processes.
Example 3-2 underscore_symbols_32.c9
#include <stdlib.h>
#include <stdio.h>
#include <errno.h>
extern int errnop;
extern int errno;
extern _text;
extern _etext;
extern _data;
extern _edata;
9 Extern symbols start with underscore are run-time symbols prepared by the system loader.
Chapter 3. Understanding user process models
113

extern char *environ[];
int i_initialized = 1; /* initialized global variable. */
long l_uninitialized; /* uninitialized global variable. */
int
main(int argc, char *argv[])
{
char str_buf[BUFSIZ]; /* auto storage class variable. */
char *heap_mem; /* heap memory. */
if ((heap_mem = malloc(BUFSIZ)) == (void *)NULL) {
sprintf(str_buf, "malloc() failed with errno = %d", errno);
perror(str_buf);
}
printf("_text =\t\t\t0x%08p\n", &_text);
printf("_etext =\t\t0x%08p\n", &_etext);
printf("_data =\t\t\t0x%08p\n", &_data);
printf("_edata =\t\t0x%08p\n", &_edata);
printf("argv[] =\t\t0x%08p\n", argv);
printf("environ[] =\t\t0x%08p\n", environ);
printf("errnop =\t\t0x%08p\n", &errnop);
printf("errno =\t\t\t0x%08p\n", &errno);
printf("i_initialized = \t0x%08p\n", &i_initialized);
printf("l_uninitialized = \t0x%08p\n", &l_uninitialized);
printf("str_buf = \t\t0x%08p\n", str_buf);
printf("heap_mem = \t\t0x%08p\n", heap_mem);
exit(0);
}
Because the process heap and stack share the same segment, the heap size
must be smaller than 256 MB in the default memory model. This is the reason
why the default definition of user soft data limit is 131,072 KB = 128 MB (see
Example 3-6 on page 125).
We provide detailed information about how heap and stack grows in this segment
in 3.2.6, “Resource limits in 32-bit model” on page 125.
114
Developing and Porting C and C++ Applications on AIX

Environment variables
Environment variables are accessible from a process using either of the
following, regardless of user process models, including 64-bit:
The environ extern symbol (see Example 3-2 on page 113).
The third parameter of the main() routine. This parameter is commonly
represented as envp, as shown in the following example:
int main(int argc, char *argv[], char *envp[]);
The getenv() sub-routine call.
On AIX, environment variables are copied into the process stack by the system
loader as a part of the process initialization tasks. The maximum size of
environment variables per process are defined by a system-wide configuration
value, ncargs, as shown in the following example:10
# lsattr -El sys0 -a ncargs
ncargs 6 ARG/ENV list size in 4K byte blocks True
The default value 6 means 6 pages x 4 [KB/page] = 24 KB. If a process is given
environment variables that require more than ncargs x 4 KB, those environment
values that exceeded the value will not be copied into virtual pages and cannot
be accessed by the process.
If you need to increase ncargs, do either of the following as the root user:
Use the chattr command:
# chattr -El sys0 -a ncargs=N
where N is the number of pages that can be used for environment variables
per-process basis.
Use SMIT
Select System Environments Change / Show Characteristics of
Operating System, then specify an appropriate number in the “ARG/ENV list
size in 4K byte block” field, and then press Enter.
In both ways, the minimum value is 6, and the maximum value is 128 (512 KB).
This change takes affect immediately and is preserved over boot.
Note: The default value ncargs = 6 is sufficient for most processes. Change it
only when your applications absolutely require a larger size of environment
variables. If there are many processes that require a large size of environment
variables, those processes may consume a significant amount of memory.
10 The ncargs system parameter has been supported on AIX since Version 5L Version 5.1.
Chapter 3. Understanding user process models
115

3.2.2 Large memory model
Although many 32-bit applications do not require a heap size more than 256 MB,
there has been a huge demand to run 32-bit user processes with a larger size of
initialized data, process heap, or process stack. To address this demand, AIX has
been supporting another memory model called large memory model, as a
variation of the 32-bit user process model.
The concept of the large memory model is simple. By sacrificing segments
available for shared memory segments, the user process running in the large
memory model can place its heap on those segments (see Figure 3-4).
In Figure 3-4, four segments (0x3 - 0x6) are reserved for heap, so seven
segments (0x7 - 0xC and 0xE) are available for shared memory segments.
Because the process heap and initialized and un-initialized data are moved from
the 0x2 segment to 0x3, the user stack can enjoy more room in the 0x2 segment
in this model.
Segment
Usage
number
0x0
Kernel text and data
0x1
User text
0x2
User stack
0x3
Process heap
1
0x4
2
0x5
3
0x6
4
0x7
0x8
0x9
Available for the user process
0xA
if shmat() or mmap() is called.
0xB
0xC
0xD
Shared library text
0xE
0xF
Per-process shared library data
Figure 3-4 Large memory model (segment usage)
To apply the large memory model to your application programs, see 3.2.4, “Using
the large and very large memory model” on page 121.
116
Developing and Porting C and C++ Applications on AIX

Note: In the large memory model, the soft data limit of a process has a slightly
different effect (see 3.2.6, “Resource limits in 32-bit model” on page 125).
3.2.3 Very large memory model
In addition to the large memory model, AIX 5L Version 5.2 supports another
model, called very large memory model, as a variation of the 32-bit user process
model. AIX 5L Version 5.1 also supports some of the features provided by the
very large memory model; it does support up to 2 GB heap memory, but not up to
3.25 GB.
Although the concept of the very large memory model remains the same, there
are three different allocation mechanisms, depending on the maxdata value
setting explained in 3.2.6, “Resource limits in 32-bit model” on page 125:
“Heap size less than 2.5 GB” on page 117
“Heap size greater than 2.5 GB” on page 118
“Heap size less than 256 MB” on page 119
To apply the very large memory model to your application programs, see 3.2.4,
“Using the large and very large memory model” on page 121.
Note: In the very large memory model, the soft data limit of a process has a
slightly different effect (see 3.2.6, “Resource limits in 32-bit model” on
page 125).
Heap size less than 2.5 GB
The big difference between the very large memory model and the large memory
model is that segments can be dynamically allocated. In the large memory
model, the allocation of segments are rigid, and there is no way to alter this
segment allocation while executing the application program. The very large
memory model relaxes this limitation by providing a new function, called dynamic
segment allocation (DSA).
With DSA, segments are dynamically allocated to either the process heap or
shared memory segments in the address space. Once a segment is allocated to
either the process heap or shared memory segments, it cannot be returned to
the free pool of segments, even if all virtual pages in the segment are
relinquished.
In Figure 3-5 on page 118, we are assuming that a process is running with
maxdata=0xA0000000 and the DSA option specified, which means the process
can acquire heap memory up to 10 segments (2.5 GB).
Chapter 3. Understanding user process models
117

Another big difference between the very large and large memory models is the
direction of how shared memory segments are allocated. In this figure, shared
memory segments will be allocated in the segments, starting from 0xE, 0xC, 0xB,
and on towards 0x3, if the target segments are not already allocated to the
process heap. To understand how these segments are allocated, see 3.5.2,
“Order in the 32-bit very large memory model with DSA” on page 154.
Segment
Usage
number
0x0
Kernel text and data
0x1
User text
0x2
User stack
0x3
1
11
0x4
2
10
Available for user heap
0x5
(Dynamic Segment Allocation).
3
9
0x6
4
8
0x7
5
7
0x8
6
6
0x9
7
5
0xA
8
4
0xB
9
3
0xC
10
2
Available for the user process
0xD
Shared library text
if shmat() or mmap() is called
(Dynamic Segment Allocation).
0xE
1
0xF
Per-process shared library data
Figure 3-5 Very large memory model (0 < maxdata < 0xB0000000)
Because of the process heap, initialized and un-initialized data are moved from
the 0x2 segment to 0x3, and the user stack can enjoy more room in the 0x2
segment in this model.
Note: AIX 5L Version 5.1 supports this model only when maxdata is less than
or equal to 0x80000000.
Heap size greater than 2.5 GB
If a process is running with maxdata=0xB0000000 and greater with DSA, then
the address space has a significant change; there are no shared text segments
available, as shown in Figure 3-6 on page 119. Therefore, all the shared text and
data that are referenced by the process will be loaded into the 0x2 segment, as
private shared objects, even if the other processes has already loaded shared
118
Developing and Porting C and C++ Applications on AIX

objects into the system memory. The process performance might be affected by
this change.
However, by sacrificing segments 0xD - 0xF, the process can acquire heap
memory up to 13 segments (3.25 GB). Also, shared memory segments will be
allocated in the segments, starting from 0xF, 0xE, 0xD, and on towards 0x3, if the
target segments are not already allocated to the process heap. To understand
how these segments are allocated, see 3.5.2, “Order in the 32-bit very large
memory model with DSA” on page 154.
Segment
Usage
number
0x0
Kernel text and data
0x1
User text
0x2
User stack
Shared library text and data are
also loaded into segment 0x2.
0x3
1
13
0x4
2
12
0x5
3
11
Available for user heap
0x6
(Dynamic Segment Allocation).
4
10
0x7
5
9
0x8
6
8
0x9
7
7
0xA
8
6
0xB
9
5
0xC
10
4
Available for the user process
0xD
11
3
if shmat() or mmap() is called
0xE
12
2
(Dynamic Segment Allocation).
0xF
13
1
Figure 3-6 Very large memory model (0xB0000000 =< maxdata < 0xD0000000)
Heap size less than 256 MB
If a process is running with maxdata=0 with DSA, then all the following memory
components are packed into the 0x2 segment:
Initialized data
Uninitialized data
Process heap
Process stack
Shared library text
Shared library data
Chapter 3. Understanding user process models
119

Apparently, this does not provide much room for the process heap; in fact, it is
smaller than the default memory model. However, it ensures that all 13 segments
(3.25 GB) are available for shared memory segments in the address space.
Some application programs that manage their own memory requests on shared
memory segments can exploit this memory model. To understand how these
segments are allocated, see 3.5.2, “Order in the 32-bit very large memory model
with DSA” on page 154.
Note: If you decide to apply this memory model to your application, it is your
responsibility to ensure that the process heap and stack usage are well
managed. You must be aware that the 0x2 segment can be easily corrupted in
this model.
Segment
Usage
number
0x0
Kernel text and data
0x1
User text
0x2
User stack, data
Shared library text and data are
0x3
13
also loaded into segment 0x2.
0x4
12
0x5
11
0x6
10
0x7
9
0x8
8
0x9
7
0xA
6
0xB
5
0xC
4
Available for the user process
0xD
3
if shmat() or mmap() is called
0xE
2
(Dynamic Segment Allocation).
0xF
1
Figure 3-7 Very large memory model (maxdata = 0)
Note:
On AIX 5L Version 5.1, shared library text and data will be loaded into
segments 0xD and 0xF in this model (maxdata=0 with DSA).
This model does not support large pages (see 3.6, “Large page support”
on page 157).
120
Developing and Porting C and C++ Applications on AIX

3.2.4 Using the large and very large memory model
To use the large memory model, any one of the following methods are supported:
Specify the -bmaxdata:0xN0000000 linker option when linking your source
codes (where N is hexadecimal from 0 to 8).
After compiling the source code, binary-edit the XCOFF header file of the
generated executable file using the /usr/ccs/bin/ldedit command. For
example:
$ /usr/ccs/bin/ldedit -bmaxdata:0x60000000 a.out
Specify the LDR_CNTRL environment value when running the executable.
For example:
$ LDR_CNTRL=MAXDATA=0x80000000 a.out
To use the very large memory model, the following methods are supported:
Specify the -bmaxdata:0xN00000000/dsa compiler option when compiling
your source codes (where N is hexadecimal from 0 to D).
After compiling the source code, binary-edit the XCOFF header file of the
generated executable file using the /usr/ccs/bin/ldedit command. For
example:
$ /usr/ccs/bin/ldedit -bmaxdata:0x80000000/dsa a.out
Specify the LDR_CNTRL environment value when running the executable.
For example:
$ LDR_CNTRL=MAXDATA=0xC0000000@DSA a.out
Note: When using the LDR_CNTRL environment variable, the keyword DSA
must be upper case character, while dsa must be lower case characters in the
other two methods.
To demonstrate the usage, we have prepared the short example program shown
in Example 3-3. When executed, it takes an integer as a command line
parameter, which will be used as an index to allocate memory with a size of
multiplies of 256 MB from the process heap.
Example 3-3 grabheap_32.c
#include <stdlib.h>
#include <stdio.h>
#include <unistd.h>
#include <errno.h>
#define ONE_SEG (256 * 1024 * 1024)
Chapter 3. Understanding user process models
121

int main(int argc, char *argv[])
{
char *p;
char buf[BUFSIZ];
int rc;
size_t sz, sz_7;
if (argc != 2) {
fprintf(stderr, "Usage: %s digit-number\n", argv[0]);
exit(1);
}
sz = atoi(argv[1]);
if (sz > 7) {
sz_7 = sz - 7;
if ((p = malloc(7 * ONE_SEG)) == (void *)NULL) {
sprintf(buf, "1st malloc() with 7 x 256 MB failed with errno = %d"
, errno);
perror(buf);
} else {
printf("1: starting address of 7 x 256 MB memory heap is 0x%08p\n"
, p);
printf("1: ending address of 7 x 256 MB memory heap is 0x%08p\n"
, p + (7 * ONE_SEG));
}
if ((p = malloc(sz_7 * ONE_SEG)) == (void *)NULL) {
sprintf(buf, "2nd malloc() with %d x 256 MB failed with errno = %d"
, sz_7, errno);
perror(buf);
} else {
printf("2: starting address of %d x 256 MB memory heap is 0x%08p\n"
, sz_7, p);
printf("2: ending address of %d x 256 MB memory heap is 0x%08p\n"
, sz_7, p + (sz_7 * ONE_SEG) - 1);
}
} else {
if ((p = malloc(sz * ONE_SEG)) == (void *)NULL) {
sprintf(buf, "malloc() with %d x 256 MB failed with errno = %d"
, sz, errno);
perror(buf);
} else {
printf("starting address of %d x 256 MB memory heap is 0x%08p\n"
, sz, p);
printf("ending address of %d x 256 MB memory heap is 0x%08p\n"
, sz, p + (sz * ONE_SEG) - 1);
}
}
122
Developing and Porting C and C++ Applications on AIX

exit(0);
}
The following example shows the output when we run Example 3-3 on page 121
using the very large memory model on AIX 5L Version 5.1:
$ oslevel -r
5100-03
$ cc grabheap_32.c
$ LDR_CNTRL=MAXDATA=0xD0000000@DSA ./a.out 7
starting address of 7 x 256 MB memory heap is 0x30000968
ending address of 7 x 256 MB memory heap is 0xa0000967
$ LDR_CNTRL=MAXDATA=0xD0000000@DSA ./a.out 8
1: starting address of 7 x 256 MB memory heap is 0x30000968
1: ending address of 7 x 256 MB memory heap is 0xa0000967
2nd malloc() with 1 x 256 MB failed with errno = 12: Not enough space
The second call of malloc() failed with errno = 12 (ENOMEM), because it
requested another 256 MB in addition to the already acquired 7 x 256 MB (1.75
GB) heap memory. Although AIX 5L Version 5.1 supports the heap size up to 2
GB, there is always a very small amount of overhead memory (initialized and
un-initialized data) required in the beginning of segment 0x3 in the large and very
large memory model regardless of operating system versions.
The following example shows the output when we run Example 3-3 on page 121
using the very large memory model on AIX 5L Version 5.2:
$ oslevel
5.2.0.0
$ cc grabheap_32.c
$ LDR_CNTRL=MAXDATA=0xD0000000@DSA ./a.out 12
1: starting address of 7 x 256 MB memory heap is 0x30000988
1: ending address of 7 x 256 MB memory heap is 0xa0000987
2: starting address of 5 x 256 MB memory heap is 0a0000998
2: ending address of 5 x 256 MB memory heap is 0xf0000997
The first call of malloc() acquired memory from the segment 0x3 to 0xA and the
second call acquired memory from the segment 0xA to 0xF; the outcome is clear.
This process successfully acquired 12 x 256 MB = 3 GB memory in total. In fact,
the program could have requested more memory, roughly 256 MB; therefore, it
could have acquired roughly 3.25 GB memory in total.
If your application needs a heap size larger than 2 GB in the very large memory
model, then it must call system heap allocation routines, such as malloc() and
calloc(), more than once, like we did in Example 3-3 on page 121. The reason for
this limitation is that the sbrk() system call, which is internally called from system
heap allocation routines, takes a signed integer value as a parameter. Therefore,
it cannot extend the break value more than 2 GB within a single call.
Chapter 3. Understanding user process models
123

For further information about the use of the system heap, see Chapter 4,
“Managing the memory heap” on page 165.
3.2.5 Checking large memory model executables
After compiling your source code with the -bmaxdata:0xN0000000 option, use
the file command with the modified /etc/magic file to check if the o_maxdata
field in the XCOFF header of executable is correctly set. On AIX, instead of the
/etc/magic file, the file command consults with a locale message file (depending
on the current locale setting) by default. Therefore, to have the command refer to
the modified /etc/magic file, do the following:
1. Log in to the system as the root user.
2. Create a backup of the original /etc/magic:
# cp -p /etc/magic /etc/magic.orig
3. Edit the file and add highlighted lines in Example 3-4 using a text editor, then
save the file and exit.
4. Run the file command against the executable file with the -m /etc/magic
option.
Example 3-4 Excerpt of modified /etc/magic
0 short 0x01df executable (RISC System/6000) or object module
>12 long >0 not stripped
>18 byte 0x14 LP_TEXT
>18 byte 0x18 LP_DATA
>19 byte >0x3F /DSA
>76 long 0x00000000 (maxdata=0)
>76 long 0x10000000 (maxdata=1)
>76 long 0x20000000 (maxdata=2)
>76 long 0x30000000 (maxdata=3)
>76 long 0x40000000 (maxdata=4)
>76 long 0x50000000 (maxdata=5)
>76 long 0x60000000 (maxdata=6)
>76 long 0x70000000 (maxdata=7)
>76 long 0x80000000 (maxdata=8)
>76 long 0x90000000 (maxdata=9)
>76 long 0xA0000000 (maxdata=A)
>76 long 0xB0000000 (maxdata=B)
>76 long 0xC0000000 (maxdata=C)
>76 long 0xD0000000 (maxdata=D)
0 short 0x01f7 64-bit XCOFF executable or object module
>18 byte 0x14 LP_TEXT
>18 byte 0x18 LP_DATA
>19 byte >0x3F /DSA
124
Developing and Porting C and C++ Applications on AIX

>20 long >0 not stripped
Example 3-5 shows the sample output from the executable file compiled with the
-bmaxdata:0x80000000/dsa option.
Example 3-5 Checking large memory executable with modified /etc/magic
$ cc -bmaxdata:0x80000000 helloworld.c
$ file a.out
a.out: executable (RISC System/6000) or object module not stripped
$ file -m /etc/magic a.out
a.out: executable (RISC System/6000) or object module not stripped /DSA (maxdata=8)
For further information about the /etc/magic file and XCOFF format, please refer
to the AIX 5L Version 5.2 Files Reference.
3.2.6 Resource limits in 32-bit model
The resource limits are used to regulate several resources consumed by a
process on UNIX operating systems. Although the resource limits are set on a
per-user basis, they are applied on a per-process basis. Therefore, if a user is
executing hundreds of processes, the user may consume huge amounts of
resources, even if the resource setting values for the user are relatively small
numbers.
To display the current user’s resource limits, use the ulimit command (see
Example 3-6).
Example 3-6 Default resource limits setting
$ ulimit -Ha
time(seconds) unlimited
file(blocks) 2097151
data(kbytes) unlimited
stack(kbytes) unlimited
memory(kbytes) unlimited
coredump(blocks) unlimited
nofiles(descriptors) unlimited
$ ulimit -Sa
time(seconds) unlimited
file(blocks) 2097151
data(kbytes) 131072
stack(kbytes) 32768
memory(kbytes) 32768
coredump(blocks) 2097151
nofiles(descriptors) 2000
Chapter 3. Understanding user process models
125

The -H option instructs the command to display hard resource limits, while the -S
option instructs the command to display soft resource limits.11 The hard resource
limit values are set by the root user using the chuser command for each user.
The soft resource limit values can be relaxed by the individual user using the
ulimit command, as long as the values are smaller than or equal to the hard
resource limit values.
Although there are seven types of resource limits displayed in Example 3-6 on
page 125, we only discuss the data and stack resource limits, which regulate
process heap and stack usage. For further information about resource limits,
please refer to the command reference of ulimit, found in the AIX 5L Version 5.2
Reference Documentation: Commands Reference.
Resource limits in the default memory model
In the default memory model, soft data and stack resource limits are strictly
enforced. During the process initialization tasks, the kernel sets soft data and
stack resource limits in the process address space (see Figure 3-8 on page 126).
While the process is running, the heap and stack can grow up to the soft data
limits from opposite directions, as represented by arrows in the figure.
Process private segment (0x2)
Initialized data
0 MB (offset)
Uninitialized data
/etc/security/limits
user:
Heap
fsize = XXXX
core = XXXX
Current heap size
cpu = XXXX
(break value)
data= XXXX
Soft data limit
rss = XXXX
stack = XXXX
Soft stack limit
Current stack size
User stack
256 MB (offset)
: Accessible virtual memory pages
: Access-prohibited virtual memory pages
Figure 3-8 Data and stack resource limits (default 32-bit process model)
11 Resource limits are defined in the /etc/security/limits file on a per-user basis.
126
Developing and Porting C and C++ Applications on AIX

If either the data or stack soft limit is set to unlimited before the program
execution, or if the setrlimit() routine is called to set these limit values to unlimited
while the process is running, it is your responsibility to prevent the process stack
from being overwritten by the over-grown process heap.
Resource limits in the very large and large memory model
When you execute a program that uses the large memory model, the operating
system attempts to modify the soft limit on the data size to match the maxdata
value. If the maxdata value is greater than the current hard limit on the data size,
either the program will not execute, if the environment variable XPB_SUS_ENV
has the value set to ON, or the soft limit will be set to the current hard limit. If the
maxdata value is smaller than the size of the program’s static data, the program
will not execute.
To demonstrate this behavior, we have prepared a simple program, shown in
Example 3-10 on page 128. When compiled and executed, it prints the soft and
hard resource limit values by calling the getrlimit() sub-routine.
When we have executed this program in the default memory model, it prints the
values shown in Example 3-7 on page 127.
Example 3-7 Verifying resource limits in default memory model
$ a.out
Resource name Soft Hard
RLIMIT_CORE 1073741312 2147483647
RLIMIT_CPU 2147483647 2147483647
RLIMIT_DATA 134217728 2147483647
RLIMIT_FSIZE 1073741312 1073741312
RLIMIT_NOFILE 2000 2147483647
RLIMIT_STACK 33554432 2147483647
RLIMIT_RSS 33554432 2147483647
You will find that this output resembles Example 3-6 on page 125.
On AIX, all hard resource limit values are set to unlimited by default, except for
the file resource limit. The value unlimited is defined as 0x7FFFFFFF in the
/usr/include/sys/resource.h include file when the program is running in the 32-bit
user process model (see Example 3-8).
Example 3-8 RLIM_INFINITY definition
#if defined(__64BIT__) && !defined(__64BIT_KERNEL)
#define RLIM_INFINITY 0x7fffffffffffffffL
#else
#define RLIM_INFINITY 0x7FFFFFFF
Chapter 3. Understanding user process models
127

#endif /* __64BIT__ */
When we have executed this program in the large memory model, it prints the
values shown in Example 3-10 on page 128. As shown in this example, the soft
data limit value is set to RLIM_INFINITY without using either the ulimit
command or the setrlimit() sub-routine.
Example 3-9 Verifying resource limits in large memory model without DSA
$ LDR_CNTRL=MAXDATA=0x80000000 a.out
Resource name Soft Hard
RLIMIT_CORE 1073741312 2147483647
RLIMIT_CPU 2147483647 2147483647
RLIMIT_DATA 2147483645 2147483647
RLIMIT_FSIZE 1073741312 1073741312
RLIMIT_NOFILE 2000 2147483647
RLIMIT_STACK 33554432 2147483647
RLIMIT_RSS 33554432 2147483647
In fact, when we executed the program shown in Example 3-3 on page 121 using
the large and very large memory models, we did not increase the soft data limit
value.
Example 3-10 is a program that calls the getrlimit() routine to print current soft
and hard resource limit values.
Example 3-10 printlimits.c
#include <stdlib.h>
#include <stdio.h>
#include <errno.h>
#include <sys/time.h>
#include <sys/resource.h>
#define pr_limit(XXX) print_limit(#XXX, XXX)
void
print_limit(const char *name, int resource)
{
int rc;
char buf[BUFSIZ];
struct rlimit rlimit;
if ((rc = getrlimit(resource, &rlimit)) < 0) {
sprintf(buf, "getrlimit() failed with errno = %d, at %d in %s"
, errno, __LINE__, __FILE__);
perror(buf);
} else {
128
Developing and Porting C and C++ Applications on AIX

printf("%-14s\t%20ld\t%20ld\n"
, name, rlimit.rlim_cur, rlimit.rlim_max);
}
}
int
main(int argc, char *argv[])
{
printf("%-14s\t%20s\t%20s\n", "Resource name", "Soft", "Hard");
pr_limit(RLIMIT_CORE);
pr_limit(RLIMIT_CPU);
pr_limit(RLIMIT_DATA);
pr_limit(RLIMIT_FSIZE);
pr_limit(RLIMIT_NOFILE);
pr_limit(RLIMIT_STACK);
pr_limit(RLIMIT_RSS);
}
3.2.7 Large file support in a 32-bit model
In the computer industry, the term large file is generally considered to refer to a
file larger than 2 GB. In the 32-bit user process model, the data type off_t is
defined as signed long type. Therefore, by default, 32-bit processes can handle
files up to 2 GB -1 byte, which is addressable by the off_t type. If a 32-bit process
attempts to seek a file pointer more than 2 GB, then the system call fails with
errno EOVERFLOW. This limitation exists on all UNIX operating systems that
conform to the POSIX standards.
Beginning with AIX Version 4.2.1, there are two methods to work around the
addressability of large files in the 32-bit user process model:
If the -D_LARGE_FILE option is specified when compiling the program, the
off_t type is redefined as long long (64-bit signed integer). Also, most system
calls and library routines are redefined to support large files.
In some circumstances, your existing 32-bit user programs might be
corrupted by defining the _LARGE_FILE macro, especially if off_t is
interchangeably used with the int or long type within your source codes. In
this case, you can modify your source codes to explicitly call 64-bit versions of
file I/O related system calls and library routines, such as fopen64(), lseek64(),
and so on. Also, variables defined as the off_t type must be carefully
converted to off64_t.
Using one of the above methods, 32-bit processes can handle large files as long
as the following conditions are met:
Chapter 3. Understanding user process models
129

The file hard and soft limits have been relaxed. The default value of file limit is
2,097,151 disk blocks = 1 GB (see Example 3-6 on page 125).
JFS2 or large file enabled JFS is used.
For further information about large file support of the 32-bit user process, please
refer to the “Writing Programs That Access Large Files” section in AIX 5L Version
5.2 General Programming Concepts: Writing and Debugging Programs.
3.3 The 64-bit user process model
This section provides a brief explanation about the 64-bit user process model.
The 64-bit user process model shares the same concept of segments and pages
with the 32-bit user process model. The difference is the number of available
segments in the address space. In the 64-bit user process model, an address
space is composed of 232 segments, while it is composed of 24 = 16 segments in
the 32-bit user process model.
Therefore, the 64-bit user process can address up to 1 EB (exabytes), which is
easily returned by the following simple calculation:
232 segments x 256 [MB/segment] = 232 x 28 x 220 bytes = 232+8+20 = 260 = 1 EB
To address this tremendous huge space, the pointer type is defined as 64-bit in
the 64-bit user process model. The program shown in Example 3-11 prints 8
bytes in the 64-bit user process model, while it prints 4 bytes in the 32-bit user
process model:
$ cc -q64 sizeofpointer.c
$ a.out
size of pointer data type is 8 byte.
$ cc sizeofpointer.c
$ a.out
size of pointer data type is 4 byte.
Example 3-11 sizeofpointer.c
#include <stdlib.h>
#include <stdio.h>
#include <unistd.h>
int
main(int argc, char *argv[])
{
char *p;
printf("size of pointer data type is %d byte.\n", sizeof(p));
exit(0);
130
Developing and Porting C and C++ Applications on AIX

}
Figure 3-9 on page 131 illustrates the segment usage in the 64-bit user process
model. The address space is divided into the areas explained in the following
sections:
Section 3.3.1, “The first 16 segments (0 - 4 GB)” on page 132 (see
Figure 3-10 on page 132)
Section 3.3.2, “Application text, data, and heap (4 GB - 448 PB)” on page 133
Section 3.3.3, “Default shared memory segments (448 - 512 PB)” on
page 135
Section 3.3.4, “Privately loaded objects (512 - 576 PB)” on page 135
Section 3.3.5, “Shared text and data (576 - 640 PB)” on page 135
Section 3.3.6, “System reserved (640 - 960 PB)” on page 136
Section 3.3.7, “User process stack (960 PB - 1 EB)” on page 136
The first 16 segments (4 GB)
0x0000 0000 0000 0000
0x0100 0000 0000 0000
0x0200 0000 0000 0000
0x0300 0000 0000 0000
Application text,
8 PB
Application data and heap
0x0400 0000 0000 0000
44
0x0500 0000 0000 0000
0x0600 0000 0000 0000
0x0700 0000 0000 0000
64 PB
Shared memory (default)
1 EB
0x0800 0000 0000 0000
64 PB
Privately loaded objects
0x0900 0000 0000 0000
64 PB
Shared library
0x0A00 0000 0000 0000
text and data
0x0B00 0000 0000 0000
0x0C00 0000 0000 0000
0 PB
0x0D00 0000 0000 0000
32
0x0E00 0000 0000 0000
0x0F00 0000 0000 0000
64 PB
User process stack
0x1000 0000 0000 0000
: System reserved (access prohibited)
Figure 3-9 The 64-bit memory model (1EB)
Chapter 3. Understanding user process models
131

3.3.1 The first 16 segments (0 - 4 GB)
These segments are exempt from general use in order to keep the compatibility
with the 32-bit user process model. Therefore, access to the segments 0x0, 0x1,
0xD, and 0xE are prohibited; these segments are reserved by the system. The
0x2 segment contains a few pages that are set up by the system loader upon the
exec() time. Necessary data, such as command line argument values,
environment variables, and errno, are stored in these pages (see Example 3-12
on page 134). The access to the rest of the 0x2 segment is prohibited.
Segments 0x3 - 0xC and 0xE are only accessible if you specify the attaching
memory address with the shmat() routine. You may use these segments to share
shared memory segments between 32-bit and 64-bit processes.
In general, hardcoding the attaching memory address with the shmat() routine is
bad programming. Unless it is absolutely required, your 64-bit application should
not specify the attaching memory address with the shmat().
Figure 3-10 illustrates the usage of first 16 segments (4 GB) in the 64-bit user
process model.
0x0000 0000 0000 0000
Kernel text and data
0x0000 0000 1000 0000
0x0000 0000 2000 0000
Only part of segment is visible
0x0000 0000 3000 0000
Shared memory (when address specified)
0x0000 0000 4000 0000
0x0000 0000 5000 0000
0x0000 0000 6000 0000
0x0000 0000 7000 0000
4 GB
0x0000 0000 8000 0000
0x0000 0000 9000 0000
0x0000 0000 A000 0000
0x0000 0000 B000 0000
0x0000 0000 C000 0000
0x0000 0000 D000 0000
0x0000 0000 E000 0000
Shared memory (when address specified)
0x0000 0000 F000 0000
0x0000 0001 0000 0000
: System reserved (access prohibited)
Figure 3-10 The 64-bit memory model (4 GB, the first 16 segments)
132
Developing and Porting C and C++ Applications on AIX

3.3.2 Application text, data, and heap (4 GB - 448 PB)
When a 64-bit program is executed, the user text is mapped into the first segment
in this area. Also, user data is mapped into another segment in this area.
In both cases, if a segment is not sufficient to contain text or data, another
segment will be contiguously attached to the process address space.
To demonstrate this behavior, we prepared a program listed in Example 3-13 on
page 134. We compiled and ran this program, as shown in Example 3-12 on
page 134. In this example, the user text is loaded into the first segment
(0x0000000100000000) in this area. The user data is loaded into the second
segment (0x0000000110000000) in this area. The heap memory acquired by
malloc() is also allocated in this segment.
Chapter 3. Understanding user process models
133

Example 3-12 Segment mapping in the 64-bit user process model
$ cc -q64 underscore_symbols_64.c
$ file a.out
a.out: 64-bit XCOFF executable or object module not stripped
$ a.out
_text = 0x00000001000001f8
_etext = 0x00000001000005e8
_data = 0x00000001100005e8
_edata = 0x0000000110000838
argv[] = 0x00000000200fe8d0
environ[] = 0x00000000200fefd8
errnop = 0x00000000200fefe8
errno = 0x00000000200fefe0
&heap_mem = 0x0fffffffffffff70
heap_mem = 0x0000000110000850
Example 3-13 is a program simplified from the one listed in Example 3-2 on
page 113 to demonstrate the segment mapping in the 64-bit user process model.
Example 3-13 underscore_symbols_64.c12
#include <stdlib.h>
#include <stdio.h>
#include <errno.h>
extern int errnop;
extern _text;
extern _etext;
extern _data;
extern _edata;
extern char *environ[];
int
main(int argc, char *argv[])
{
char buf[BUFSIZ];
char *heap_mem; /* heap memory. */
if ((heap_mem = malloc(BUFSIZ)) == (void *)NULL) {
sprintf(buf, "malloc() failed with errno = %d", errno);
perror(buf);
}
printf("_text =\t\t0x%016p\n", &_text);
printf("_etext =\t0x%016p\n", &_etext);
printf("_data =\t\t0x%016p\n", &_data);
12 Extern symbols that start with underscore are run-time symbols prepared by the system loader.
134
Developing and Porting C and C++ Applications on AIX

printf("_edata =\t0x%016p\n", &_edata);
printf("argv[] =\t0x%016p\n", argv);
printf("environ[] =\t0x%016p\n", environ);
printf("errnop =\t0x%016p\n", &errnop);
printf("errno =\t\t0x%016p\n", &errno);
printf("&heap_mem =\t0x%016p\n", &heap_mem);
printf("heap_mem =\t0x%016p\n", heap_mem);
exit(0);
}
As we demonstrate in the later section, the 64-bit user process model gives you a
very flat memory model that is easy to use, as long as there are enough physical
pages to be allocated on the system (see 3.3.8, “Resource limits in 64-bit mode”
on page 136).
3.3.3 Default shared memory segments (448 - 512 PB)
If a 64-bit process calls shmat() or mmap() routines without specifying the
attaching memory address, segments in this area will be attached to the address
space contiguously.
See 3.4.4, “Shared memory limits” on page 147 for the IPC limitation in the 64-bit
user process model.
3.3.4 Privately loaded objects (512 - 576 PB)
If objects are loaded into the address space in the following cases, those objects
will be loaded into the segments in this area:
Objects are explicitly loaded by load() and dlopen().
The file permission modes of shared objects are r-xr-x--- (no read and
execute permission bits are set for others).
All the global shared text and data segments are full (very unlikely in the
64-bit user process model).
For further information about the private shared objects, see 2.9.2, “Private
shared objects” on page 101.
3.3.5 Shared text and data (576 - 640 PB)
On the first load of a shared library object, the shared library text is loaded into a
segment in this area. This segment will be shared by all 64-bit user processes on
Chapter 3. Understanding user process models
135

the system. Also, shared library data will be created in another segment in this
area per process basis at the same time.
In both cases, if there is a segment that has enough free space to contain shared
text or shared library data, that segment will be used. Otherwise, another
segment will be attached to the process address space.
The virtual addresses of loaded shared text objects can be examined using the
genkld command. For further information about the usage of genkld, see 2.7.2,
“genkld” on page 88.
3.3.6 System reserved (640 - 960 PB)
All the segments in this area are reserved by the system and prohibited from the
user process access.
3.3.7 User process stack (960 PB - 1 EB)
By default, a 64-bit user process uses the last segment in this area for the user
process stack. The stack grows from the last address,
0x0FFF_FFFF_FFFF_FFFF, toward the first address in this area. The address of
the heap_mem pointer depicts this behavior (see Example 3-12 on page 134).
To use more than one segment for user process stack, you need to specify the
-bmaxstacksize linker option. For example, to specify two segments for the user
process stack, you need to compile your program as follows:
$ cc -q64 -bmaxstacksize:0x20000000 hello.c
Note: Thread stacks for Pthreads, except the initial thread within a
multi-threaded processes, are allocated in the process heap. See 8.3.4,
“Thread stack” on page 292 for further information about the thread stack in
multi-threaded processes.
3.3.8 Resource limits in 64-bit mode
Unlike the 32-bit user process model, the data resource limit value always
defines the actual upper limit of allocatable heap memory in the 64-bit user
process memory model.
With the default soft data limit of 128 MB (see Example 3-6 on page 125), your
application can only allocate memory from a process heap up to 128 MB. To
demonstrate this, we have simplified grabheap_32.c (Example 3-3 on page 121)
to get rid of the complexity, because the 64-bit user process memory model is flat
136
Developing and Porting C and C++ Applications on AIX

(Example 3-14). If we compile and run this program under the default soft data
limit, it fails to acquire a 256 MB heap:
$ cc grabheap_64.c
$ file a.out
a.out: 64-bit XCOFF executable or object module not stripped
$ a.out 1
malloc() with 1 x 256 MB failed with errno = 12: Not enough space
Example 3-14 grabheap_64.c
#include <stdlib.h>
#include <stdio.h>
#include <unistd.h>
#include <errno.h>
#define ONE_SEG (256 * 1024 * 1024)
int main(int argc, char *argv[])
{
char *p;
char buf[BUFSIZ];
size_t sz;
if (argc != 2) {
fprintf(stderr, "Usage: %s digit-number\n", argv[0]);
exit(1);
}
sz = atoi(argv[1]);
if ((p = malloc(sz * ONE_SEG)) == (void *)NULL) {
sprintf(buf, "malloc() with %d x 256 MB failed with errno = %d"
, sz, errno);
perror(buf);
} else {
printf("starting address of %d x 256 MB memory heap is 0x%016p\n"
, sz, p);
printf("ending address of %d x 256 MB memory heap is 0x%016p\n"
, sz, p + (sz * ONE_SEG) - 1);
}
exit(0);
}
Once the soft data limit is relaxed, the program can acquire a 256 MB heap:
$ ulimit -Sd unlimited
$ ulimit -Sa
time(seconds) unlimited
file(blocks) 2097151
Chapter 3. Understanding user process models
137

data(kbytes) unlimited
stack(kbytes) 32768
memory(kbytes) 32768
coredump(blocks) 2097151
nofiles(descriptors) 2000
$ file a.out
a.out: 64-bit XCOFF executable or object module not stripped
$ a.out 1
starting address of 1 x 256 MB memory heap is 0x0000000110000850
ending address of 1 x 256 MB memory heap is 0x000000012000084f
If the soft data limit is set to unlimited, it is set to 9,223,372,036,854,775,807 byte
= 0x7FFF_FFFF_FFFF_FFFF byte in the 64-bit user process model.13
Example 3-15 shows the soft resource limit value in the 64-bit user process
model (the source code is the same one listed in Example 3-10 on page 128).
Example 3-15 Soft data limit set to RLIM_INIFINITY in 64-bit mode14
$ cc -q64 printlimits.c
$ file a.out
a.out: 64-bit XCOFF executable or object module not stripped
$ a.out
Resource name Soft Hard
RLIMIT_CORE 1073741312 9223372036854775807
RLIMIT_DATA 134217728 9223372036854775807
RLIMIT_DATA 134217728 9223372036854775807
RLIMIT_FSIZE 1073741312 1073741312
RLIMIT_NOFILE 2000 9223372036854775807
RLIMIT_STACK 33554432 9223372036854775807
RLIMIT_RSS 33554432 9223372036854775807
$ ulimit -Sd unlimited
$ a.out | egrep '^(Resource|RLIMIT_DATA)'
Resource name Soft Hard
RLIMIT_DATA 9223372036854775807 9223372036854775807
Therefore, a 64-bit user process can acquire as much heap memory as it
requests once the soft data limit is set to unlimited. In the following example, we
specified 409600 as the command line parameter and the program successfully
acquired 409,600 x 256 MB = 100 TB:
$ a.out 409600
starting address of 409600 x 256 MB memory heap is 0x0000000110000850
ending address of 409600 x 256 MB memory heap is 0x0000640110000850
13 The value 0x7FFF_FFFF_FFFF_FFFF is defined as RLIM_INIFINITY for the 64-bit user process
model in /usr/include/sys/resource.h (see Example 3-8 on page 127).
14 The actual numerical limits might be changed in the future release of AIX. These values are
applicable on AIX starting from Version 4.3 to 5.2.
138
Developing and Porting C and C++ Applications on AIX

Although the malloc() routine does not actually allocate virtual pages (virtual
pages are allocated when the program touches them the first time), requesting
this huge amount of memory puts unnecessary stress on the system.
The 64-bit user process model gives you a very flat memory model that is easy to
use, but it is your responsibility to request the proper size of heap memory in your
application. You may consider selecting one of the following methods to place a
safety mechanism on your 64-bit programs:
Specify the -bmaxdata:0xNNNNNNNNNNNNNNNN linker option when
compiling your source codes.
After compiling the source code, binary-edit the XCOFF header file of the
generated executable file using the /usr/ccs/bin/ldedit command. For
example:
$ /usr/ccs/bin/ldedit -bmaxdata:0xNNNNNNNNNNNNNNNN a.out
Specify the LDR_CNTRL environment value when running the executable.
For example:
$ LDR_CNTRL=MAXDATA=0xNNNNNNNNNNNNNNNN a.out
Call the setrlimit() sub-routine in your application to explicitly set the soft data
limit.
The appropriate value of 0xNNNNNNNNNNNNNNNN varies, depending on your
application’s needs and the physical memory size; however, the following
numbers can be good starting points:
0x0000000040000000
1 GB
0x0000001000000000
64 GB
0x0000002000000000
128 GB
3.3.9 Large file support in 64-bit model
In the 64-bit user process model, the data type off_t is always defined as long
long. Therefore, 64-bit processes can handle large files as long as the following
conditions are met:
The file’s hard and soft limits have been relaxed. The default value of file limit
is 2,097,151 disk blocks = 1 GB (see Example 3-6 on page 125).
JFS2 or large file enabled JFS is used.
Chapter 3. Understanding user process models
139

3.4 Introduction to shared memory
The Inter-Process Communication (IPC) facilities are used by processes to
communicate with each other and to synchronize their activities.15 IPC facilities,
semaphores, message queues, and shared memory are quite common services
in the modern UNIX operating systems, including AIX. In this section, we focus
on the shared memory on AIX.
The shared memory is usually used for the following purposes:
Sharing memory segments between processes
Mapped shared memory segments can serve as a large pool for exchanging
data among processes. The mechanism does not provide locks or access
control among the processes by default. Therefore, processes using shared
memory areas must set up a signal or semaphore control method to prevent
access conflicts and to keep one process from changing data that another is
using. Shared memory segments can be most beneficial when the amount of
data to be exchanged between processes is too large to transfer with
message queues or when many processes have to share common data.
Mapping files into memory segments
Memory mapped files provide a mechanism for a process to access files by
directly incorporating file data into the process address space. The use of
mapped files can significantly reduce I/O data movement because the file
data does not have to be copied into process data buffers, as is done by the
read and write subroutines. When more than one process maps the same file,
its contents are shared among them, therefore providing a low-overhead
mechanism by which processes can synchronize and communicate.
However, once shared memory segments are attached to the address space of
your application process, those segments do not have to be shared with other
processes. Therefore, some applications exploit shared memory segments as if
they are an extended memory heap (see 4.4, “Heap management using
MEMDBG” on page 199 for detail).
AIX supports the following two well-known services for memory mapping:
Section 3.4.1, “The shmat services” on page 141
Section 3.4.2, “The mmap services” on page 145
Both services address the same type of memory usage, but shmat is normally
used to create and use shared memory segments within a program, and the
15 Signals are also commonly used for IPC in the UNIX operating systems. However, we do not
discuss signals in this redbook, because the signal management on AIX conforms very well to many
UNIX standards, such as POSIX.
140
Developing and Porting C and C++ Applications on AIX

mmap is mostly used for mapping files into the process address space, although
it can be used for creating shared memory segments as well as mapping files.
The term shared memory segments is widely used to refer to the memory chunks
allocated by these two services on UNIX operating systems, including AIX.
Although a shared memory segment smaller than 256 MB usually consumes a
256 MB segment on AIX, if the EXTSHM=ON environment variable is not defined,
the term is not the same concept referenced by segments (explained in 3.2, “The
32-bit user process model” on page 109).
For further information about the topics explained in this section, please refer to
the “Shared Libraries and Shared Memory” section in AIX 5L Version 5.2 General
Programming Concepts: Writing and Debugging Programs.
3.4.1 The shmat services
The shmat services supported by AIX include the following routines:
shmat()
Attaches a shared memory segment to a process.
shmctl()
Controls shared memory operations.
shmget()
Gets or creates a shared memory segment.
shmdt()
Detaches a shared memory segment from a process.
ftok()
Provides the key that the shmget() subroutine uses to
create the shared segment.
These subroutine references are provided in Appendix C, “Subroutine references
for shmat and mmap services” on page 447.
Terms used in the shmat services
In the shmat services, the following terms are frequently used:
key
The unique identifier of a particular shared segment. It is
associated with the shared segment as long as the
shared segment exists. In this respect, it is similar to the
file name of a file.
shmid
The identifier assigned to the shared segment for use
within a particular process. It is similar in use to a file
descriptor for a file.
attach
Specifies that a process must attach a shared segment in
order to use it. Attaching a shared segment is similar to
opening a file.
Chapter 3. Understanding user process models
141

detach
Specifies that a process must detach a shared segment
once it is finished using it. Detaching a shared segment is
similar to closing a file.
There are some well-established programming disciplines about the use of these
routines. We introduce a few of them in this section.
Mapping memory with shmat()
The following pseudo-code shows how to use shared memory segments with
shmat services:
1. Create a key to uniquely identify the shared segment. Use the ftok()
subroutine to create the key. For example, to create the mykey key using a
variable, proj (integer type), and a file name of /path/some_file, insert the
following statement:
mykey = ftok(“/path/some_file”, proj);
2. Either:
a. Create a shared memory segment with the shmget() subroutine. For
example, to create a shared segment that contains 4096 bytes and assign
its ID to an integer variable, mem_id, insert the following statement:
mem_id = shmget(mykey, 4096, IPC_CREAT | permission_modes);
You must define appropriate permission_modes ORing with S_ macros
(see Appendix C, “Subroutine references for shmat and mmap services”
on page 447).
b. Get a previously created shared segment with the shmget() subroutine.
For example, to get a shared segment that is already associated with the
mykey key and assign the ID to an integer variable, mem_id, insert the
following statement:
mem_id = shmget(mykey, 0, IPC_ACCESS);
3. Attach the shared segment to the process with the shmat() subroutine. For
example, to attach a previously created segment, insert the following
statement:
ptr = shmat(mem_id, 0, SHM_MAP);
If you specify address 0 as the second parameter of shmat(), then the
operating system will select the attaching target address. This is the
recommended programming manner.
The variable ptr is a pointer to a structure that defines the fields in the shared
segment. Use this structure to store and retrieve data in the shared segment.
The definition of the structure should be the same for all processes using the
segment.
142
Developing and Porting C and C++ Applications on AIX

4. Work with the data in the segment through ptr.
5. Detach from the segment using the shmdt() subroutine:
shmdt(ptr);
6. If the shared segment is no longer needed, remove it from the system with the
shmctl() subroutine:
shmctl(mem_id, IPC_RMID, ptr);
If a shared memory segment is created accordingly, multiple processes can
attach it into their address space, as illustrated in Figure 3-11. In this figure, two
processes, A and B, are sharing the sheared memory segment, which is
attached to 0x3 in the process A’s address space and attached to 0x4 in the
process B’s address space. If the process A writes data in this memory area, the
process B can instantly read the data value from it.
Although it is a very efficient way to transfer data between processes, a data
synchronization algorithm must be used between processes. Semaphores are
commonly used to implement this data synchronization algorithm.
User text:
User text:
Process A
Process B
0x0
Per-process private
Per-process private
0x0
0x1
data:
data:
0x1
0x2
0x2
0x3
0x3
0x4
0x4
0x5
Shared memory:
0x5
0x6
0x6
0x7
0x7
0x8
0x8
0x9
0x9
0xA
Shared library text:
0xA
0xB
0xB
0xC
0xC
0xD
0xD
0xE
Per-process shared
Per-process shared
0xE
0xF
library data:
library data:
0xF
: Allocated virtual address pages
: Un-allocated virtual address pages
Figure 3-11 Shared memory segments between two processes
Chapter 3. Understanding user process models
143

The shared memory segments created by the following methods can be shared
between processes:
The MAP_SHARED flag is specified with mmap().
Appropriate permission modes are specified with shmget().
After creating a shared memory segment with shmget(), its permission modes
are altered accordingly with the IPC_SET command flag with shmctl().
Mapping files with shmat()
The following pseudo-code shows how to map files with shmat services. To do
so, you need to map the file first, then set the end of the mapped file using
lseek(). The shmat services do not provide any mechanisms to detect the end of
mapped file.
1. To create the mapped data file:
a. Open (or create) the file and save the file descriptor:
if ((fildes = open(filename, O_RDWR)) < 0) {
printf("cannot open file\n");
exit(1);
}
b. Map the file to a segment with the shmat subroutine:
file_ptr = shmat(fildes, 0, SHM_MAP);
The SHM_MAP constant is defined in the /usr/include/sys/shm.h file. This
constant indicates that the file is a mapped file. Include this file and the
other shared memory header files in a program with the following
directives:
#include <sys/shm.h>
2. To detect the end of the mapped file:
a. Use the lseek() subroutine to go to the end of file:
eof = file_ptr + lseek(fildes, 0, SEEK_END);
This example sets the value of eof to an address that is 1 byte beyond the
end of file. Use this value as the end-of-file marker in the program.
b. Use file_ptr as a pointer to the start of the data file, and access the data as
if it were in memory:
while (file_ptr < eof) {
.
.
.
(references to file using file_ptr)
}
144
Developing and Porting C and C++ Applications on AIX

Note:
A file system object should not be simultaneously mapped using
both the shmat and mmap services.
Unexpected results may occur when references are made beyond
the end of the object.a In the current implementation, mapping files
with the shmat services allocate only enough 256 MB segments to
hold the existing file. If the process attempts to access the memory
area beyond the allocated shared memory segments using a
pointer, the SIGSEGV signal will be delivered to the process.
a. When a process maps a file using shmat(), only enough 256 MB seg-
ments to hold the file will be allocated in the current AIX implementation. If
the memory area beyond the allocated shared memory segments is refer-
enced by a pointer, the SIGSEGV signal will be delivered to the process.
c. Close the file when the program is finished working with it:
close(fildes);
3.4.2 The mmap services
The mmap services supported by AIX include the following routines:
mmap()
Maps an object file into virtual memory.
madvise()
Advises the system of a process' expected paging behavior.
mincore()
Determines residency of memory pages.
mprotect()
Modifies the access protections of memory mapping.
msync()
Synchronizes a mapped file with its underlying storage device.
munmap()
Unmaps a mapped memory segment.
These subroutine references are provided in Appendix C, “Subroutine references
for shmat and mmap services” on page 447.
There are some well-established programming disciplines for the use of these
routines. We introduce a few of them in this section.
Mapping files with mmap()
The following pseudo-code shows how to map files with the mmap services:
1. Create a file descriptor, fd, for a file system object using the open() routine:
fd = open(pathname, permissions);
2. Determine the file length by using the lseek() routine. For example:
len = lseek(fd, 0, SEEK_END)
Chapter 3. Understanding user process models
145

3. Map the file into the process address space with the mmap subroutine. For
example, to map the file for the file descriptor fd, starting at address addr,
using len bytes of size, with the access permissions defined by prot and 0
bytes of offset, use the statement:
ptr = mmap(addr, len, prot, MAP_FILE, fd, 0)
This specifies the creation of a new mapped file segment by mapping the file
associated with the fd file descriptor. The mapped segment can extend
beyond the end of the file, both at the time when the mmap subroutine is
called and while the mapping persists. This situation could occur if a file with
no contents was created just before the call to the mmap subroutine, or if a
file was later truncated.
If you specify address 0 as the first parameter of mmap(), then the operating
system will select the attaching target address. This is the recommended
programming manner.
4. Work with the data in the segment.
5. The file descriptor can be closed by using:
close(fd);
6. Detach from the segment using the munmap subroutine:
munmap(addr, len);
Note: A file system object should not be simultaneously mapped using both
the shmat and mmap services.
Mapping memory with mmap()
The following pseudo-code shows how to map memory segments with mmap
services:
1. If you specify MAP_ANONYMOUS instead of MAP_FILE for the mmap()’s
fourth parameter, then the file descriptor parameter must be specified as -1.
This is called an anonymous memory segment, because there is no file system
object associated. All pages in the anonymous memory segment are 0-filled:
ptr = mmap(addr, len, prot, MAP_ANONYMOUS, -1, 0)
This memory segment pointed to by ptr can be shared only with the
descendants of the current process.
2. Work with the data in the segment.
3. Detach the segment from the address space using the munmap() subroutine:
munmap(addr, len);
146
Developing and Porting C and C++ Applications on AIX

3.4.3 Difference between shmat and mmap services
Although both the shmat and mmap services can be used for the purposes
explained in 3.4, “Introduction to shared memory” on page 140, there are
significant differences between these two services:
The address for a shared memory segment mapped by shmat() must be
SHMLBA (256 MB) aligned if the EXTSHM=ON environment variable is not
set, while mmap() works on a page-size alignment (4 KB).
The mmap() memory mappings are a process resource and thus are cleaned
up at process exit-time. The shmget() shared memory segments are a
system-wide resource and are not cleaned up at process exit-time.
Therefore, use the shmat services under the following circumstances:
When mapping shared memory regions that need to be shared among
unrelated processes (no parent-child relationship).
When mapping entire files.
Use the mmap services under the following circumstances:
Portability of the application is a concern.
Many files are mapped simultaneously.
Only a portion of a file needs to be mapped.
Page-level protection needs to be set on the mapping.
Private mapping (MAP_PRIVATE) is required.
Note: Section 3.4.4, “Shared memory limits” on page 147 and 3.5, “Shared
memory segments allocation order” on page 149 are only applicable to the
shared memory segments created by shmat().
3.4.4 Shared memory limits
Shared memory limits vary, depending on AIX versions, as shown in Table 3-1 on
page 148.
Chapter 3. Understanding user process models
147

Table 3-1 Shared memory limits
Description
AIX V4.3.0
4.3.1
4.3.2, 4.3.3
5.1
5.2
Maximum segment size
256 MB
2 GB
2 GB
2 GB
2 GB
(32-bit)
Maximum segment size
256 MB
2 GB
2 GB
64 GB
1 TB
(64-bit)
Minimum segment size
1
1
1
1
1
Maximum number of shared
4096
4096
131072
131072
131072
memory IDs
Maximum number of
11
11
11
11
11
segments per process
(32-bit)
Maximum number of
268435456
268435456
268435456
268435456
268435456
segments per process
(64-bit)
To confirm system constants regarding shared memory, it is recommended to
call the vmgetinfo()16 routine instead of hardcoding numerical values (see the
program example shown in Example 3-17 on page 149).
Example 3-16 shows the output from this program. Please compare the values in
this example with values for AIX 5L Version 5.2 in the Table 3-1.
Example 3-16 Output from vmgetinfo() routine
$ a.out
/* Shared memory limits */
max # of shared memory id's = 131072
64bit proc max shm segment size = 1099511627776
32bit proc max shm segment size = 2147483648
min shared memory segment size = 1
max # of shm segs per 64bit proc = 268435456
max # of shm segs per 32bit proc (without EXTSHM=ON) = 11
Example 3-17 on page 149 shows an example program to call the vmgetinfo()
routine.
16 The vmgetinfo() routine is supported on AIX 5L Version 5.2 and later.
148
Developing and Porting C and C++ Applications on AIX

Example 3-17 vmgetinfo.c17
#include <stdlib.h>
#include <stdio.h>
#include <unistd.h>
#include <errno.h>
#include <sys/vminfo.h>
int main(int argc, char *argv[])
{
int rc;
char buf[BUFSIZ];
struct ipc_limits ipc_lim;
if ((rc = vmgetinfo(&ipc_lim, IPC_LIMITS, sizeof(struct ipc_limits))) != 0)
{
sprintf(buf, "vmgetinfo() at %d in %s failed with errno = %d"
, __LINE__, __FILE__, errno);
perror(buf);
exit(1);
}
printf("/* Shared memory limits */\n");
printf("max # of shared memory id's =\t\t\t\t%lld\n", ipc_lim.shmmni);
printf("64bit proc max shm segment size =\t\t\t%lld\n", ipc_lim.shmmax64);
printf("32bit proc max shm segment size =\t\t\t%u\n", ipc_lim.shmmax32);
printf("min shared memory segment size =\t\t\t\t%u\n", ipc_lim.shmmin);
printf("max # of shm segs per 64bit proc =\t\t\t%lld\n", ipc_lim.shmseg64);
printf("max # of shm segs per 32bit proc (without EXTSHM=ON) =\t%u\n"
, ipc_lim.shmseg32);
exit(0);
}
3.5 Shared memory segments allocation order
This section explains how shared memory segments are allocated in the process
address space. To demonstrate the allocation order, we have prepared the
program shown in Example 3-19 on page 151. To compile this program, type the
following command:
$ cc -DCALL_SHMDT grabshm.c
If -DCALL_SHMDT is not specified, then the shmdt() and shmctl() routines will
not be called in the program; it then leaves shared memory segments after the
17 The structure type ipc_limits is defined in the /usr/include/sys/vminfo.h header file.
Chapter 3. Understanding user process models
149

execution. If the program is executed by the ausres01 user, it then leaves the
shared memory segments highlighted in Example 3-18.
Example 3-18 ipcs -m
$ ipcs -m
IPC status from /dev/mem as of Sun Feb 9 18:21:48 CST 2003
T ID KEY MODE OWNER GROUP
Shared Memory:
m 0 0xe4663d62 --rw-rw-rw- imnadm imnadm
m 1 0x9308e451 --rw-rw-rw- imnadm imnadm
m 2 0x52e74b4f --rw-rw-rw- imnadm imnadm
m 3 0xc76283cc --rw-rw-rw- imnadm imnadm
m 4 0x298ee665 --rw-rw-rw- imnadm imnadm
m 5 0xffffffff --rw-rw---- root system
m 6 0xffffffff --rw-rw---- root system
m 48159 0xffffffff --rw------- ausres01 itsores
m 48160 0xffffffff --rw------- ausres01 itsores
m 48161 0xffffffff --rw------- ausres01 itsores
m 48162 0xffffffff --rw------- ausres01 itsores
To remove stale shared memory segments, type ipcrm -m shared_memory_ID.
The shared_memory_IDs are shown in the second column in the example. If you
need to remove many shared memory segments, do the following:
$ ipcs -m | grep user_name | awk '{print $2}' | while read id
> do
> ipcrm -m $id
> done
Where user_name is the user name of the application process that acquired the
shared memory segments.
Example 3-19 on page 151 is a sample program to allocate shared memory
segments. When executed, it requires two arguments:
Usage: a.out [-m|-s] digit-number
If you specify the -m option, then it tries to allocate multiplies of a 1 MB shared
memory segment. If -s specified, it tries to allocate multiplies of a 256 MB shared
memory segment. The argument digit-number specifies how many shared
memory segments should be allocated.
In this program, the created shared memory segments cannot be shared with
other processes, because the IPC_PRIVATE flag is specified with the shmget()
routine. Also, please note that the attaching target address is specified as
(void *)0 with the shmat() routine, so that we instruct the system to select the
attaching target address of shared memory segments.
150
Developing and Porting C and C++ Applications on AIX

Example 3-19 grabshm.c
#include <stdlib.h>
#include <stdio.h>
#include <unistd.h>
#include <strings.h>
#include <errno.h>
#include <sys/ipc.h>
#include <sys/shm.h>
#define MAX_SHM_SEGMENTS 131072
#define ONE_MB (1024 * 1024)
#define ONE_SEG (256 * 1024 * 1024)
char *size_of(int i)
{
if (i == ONE_MB) {
return("1 MB");
} else if (i == ONE_SEG) {
return("256 MB");
} else {
exit(1);
}
}
int main(int argc, char *argv[])
{
char *shmptr[MAX_SHM_SEGMENTS];
char buf[BUFSIZ];
int shmid[MAX_SHM_SEGMENTS];
int size_flag;
int cnt, max, rc;
if (argc != 3) {
fprintf(stderr, "Usage: %s [-m|-s] digit-number\n", argv[0]);
exit(1);
} else {
if (!strcmp(argv[1], "-m")) {
size_flag = ONE_MB;
} else if (!strcmp(argv[1], "-s")) {
size_flag = ONE_SEG;
} else {
fprintf(stderr, "Usage: %s [-m|-s] digit-number\n", argv[0]);
exit(1);
}
max = atoi(argv[2]);
}
for (cnt = 0; cnt < max; cnt++) {
Chapter 3. Understanding user process models
151

if ((shmid[cnt] = shmget(IPC_PRIVATE, size_flag
, IPC_CREAT | S_IRUSR | S_IWUSR)) < 0) {
sprintf(buf
, "[%2d] shmget(\"%s\") at %d in %s failed with errno = %d"
, cnt, size_of(size_flag), __LINE__, __FILE__, errno);
perror(buf);
}
if ((shmptr[cnt] = shmat(shmid[cnt], (void *)0, 0)) == (void *) -1) {
sprintf(buf, "[%2d] shmat() at %d in %s failed with errno = %d"
, cnt, __LINE__, __FILE__, errno);
perror(buf);
} else {
printf(
"[%2d] beginning address of %s shared memory segment is 0x%016p\n"
, cnt, size_of(size_flag), shmptr[cnt]);
printf(
"[%2d] ending address of %s shared memory segment is 0x%016p\n"
, cnt, size_of(size_flag), shmptr[cnt] + (size_flag - 1));
}
}
/* if you comment out the following code block,
* you need to explicitly remove the allocated shared memory segment
* by calling ipcrm -m ID.
*/
#if defined(CALL_SHMDT)
for (cnt = 0; cnt < max; cnt++) {
if ((rc = shmdt(shmptr[cnt])) == -1) {
sprintf(buf, "[%2d] shmdt() at %d in %s failed with errno = %d"
, cnt, __LINE__, __FILE__, errno);
perror(buf);
}
if (shmctl(shmid[cnt], IPC_RMID, 0) < 0) {
sprintf(buf, "[%2d] shmctl() at %d in %s failed with errno = %d"
, cnt, __LINE__, __FILE__, errno);
perror(buf);
}
}
#endif
exit(0);
}
Please note that this program is written in order to demonstrate the order of
shared memory segments allocation. If you need to allocate a larger size of
shared memory segment, you can get it with the one shmget() call.
152
Developing and Porting C and C++ Applications on AIX

3.5.1 Order in the 32-bit default memory model
If the program shown in Example 3-19 on page 151 is executed in the 32-bit
default memory model, then segments 0x3 - 0xE will be sequentially used,
excluding 0xD.
Example 3-20 shows the output from this program, when we have attempted to
allocate twelve 256 MB shared memory segments. Because the 32-bit user
process model supports up to 11 shared memory segments by default, if the
EXTSHM=ON environment variable is not specified, the last attempt to allocate
the 12th shared memory segment would fail.
Example 3-20 Acquiring twelve 256 MB shared memory segments18
$ a.out -s 12
[ 0] beginning address of 256 MB shared memory segment is 0x0000000030000000
[ 0] ending address of 256 MB shared memory segment is 0x000000003fffffff
[ 1] beginning address of 256 MB shared memory segment is 0x0000000040000000
[ 1] ending address of 256 MB shared memory segment is 0x000000004fffffff
...
[ 9] beginning address of 256 MB shared memory segment is 0x00000000c0000000
[ 9] ending address of 256 MB shared memory segment is 0x00000000cfffffff
[10] beginning address of 256 MB shared memory segment is 0x00000000e0000000
[10] ending address of 256 MB shared memory segment is 0x00000000efffffff
[11] shmat() at 58 in grabshm.c failed with errno = 24: Too many open files
[11] shmdt() at 76 in grabshm.c failed with errno = 22: Invalid argument
Example 3-21 shows the output from this program, when we have attempted to
allocate twelve 1 MB shared memory segments. Because the 32-bit user process
model supports up to 11 shared memory segments, by default, if the
EXTSHM=ON environment variable is not specified, the last attempt to allocate
the 12th shared memory segment would fail, even if the total shared memory
size is just 11 MB.
Example 3-21 Acquiring twelve 1 MB shared memory segments
$ a.out -m 12
[ 0] beginning address of 1 MB shared memory segment is 0x0000000030000000
[ 0] ending address of 1 MB shared memory segment is 0x00000000300fffff
[ 1] beginning address of 1 MB shared memory segment is 0x0000000040000000
[ 1] ending address of 1 MB shared memory segment is 0x00000000400fffff
...
[ 9] beginning address of 1 MB shared memory segment is 0x00000000c0000000
[ 9] ending address of 1 MB shared memory segment is 0x00000000c00fffff
[10] beginning address of 1 MB shared memory segment is 0x00000000e0000000
[10] ending address of 1 MB shared memory segment is 0x00000000e00fffff
18 The value 24 of errno means EMFILE (too many open files) and the value 22 of errno means
EINVAL (invalid argument).
Chapter 3. Understanding user process models
153

[11] shmat() at 58 in grabshm.c failed with errno = 24: Too many open files
You may notice the system selected different segments for allocating just 1MB
size shared memory segments. For example, the 0x300FFFFF - 0x3FFFFFFF
address cannot be used to allocate another shared memory segment.
3.5.2 Order in the 32-bit very large memory model with DSA
If the program shown in Example 3-19 on page 151 is executed in the 32-bit very
large memory model with DSA, then segments 0xF - 0x3 would be sequentially
used; however, higher address segments cannot be used, depending on the
maxdata variable.
Example 3-22 shows the output from this program, when we have attempted to
allocate eleven 1 MB shared memory segments with
LDR_CNTRL=MAXDATA=0x80000000@DSA. Because the 0x3 segment is
always reserved for the process heap in the very large memory model, the 11th
attempt to allocate another shared memory segment failed.
Note: If segments 0x3 - 0xE were attached to the process address heap for
the process heap before allocating shared memory segments, this program
could have failed earlier.
Example 3-22 LDR_CNTRL=MAXDATA=0x80000000@DSA
$ LDR_CNTRL=MAXDATA=0x80000000@DSA a.out -m 11
[ 0] beginning address of 1 MB shared memory segment is 0x00000000e0000000
[ 0] ending address of 1 MB shared memory segment is 0x00000000e00fffff
[ 1] beginning address of 1 MB shared memory segment is 0x00000000c0000000
[ 1] ending address of 1 MB shared memory segment is 0x00000000c00fffff
...
[ 8] beginning address of 1 MB shared memory segment is 0x0000000050000000
[ 8] ending address of 1 MB shared memory segment is 0x00000000500fffff
[ 9] beginning address of 1 MB shared memory segment is 0x0000000040000000
[ 9] ending address of 1 MB shared memory segment is 0x00000000400fffff
[10] shmat() at 58 in grabshm.c failed with errno = 24: Too many open files
[10] shmdt() at 76 in grabshm.c failed with errno = 22: Invalid argument
Example 3-23 on page 155 shows the output from this program, when we have
attempted to allocate fourteen 1 MB shared memory segments with
LDR_CNTRL=MAXDATA=0@DSA. Because the program ran out of all the
available 13 segments, the 14th attempt to allocate another shared memory
segment failed.
154
Developing and Porting C and C++ Applications on AIX

Example 3-23 LDR_CNTRL=MAXDATA=0@DSA
$ LDR_CNTRL=MAXDATA=0@DSA a.out -m 14
[ 0] beginning address of 1 MB shared memory segment is 0x00000000f0000000
[ 0] ending address of 1 MB shared memory segment is 0x00000000f00fffff
[ 1] beginning address of 1 MB shared memory segment is 0x00000000e0000000
[ 1] ending address of 1 MB shared memory segment is 0x00000000e00fffff
...
[11] beginning address of 1 MB shared memory segment is 0x0000000040000000
[11] ending address of 1 MB shared memory segment is 0x00000000400fffff
[12] beginning address of 1 MB shared memory segment is 0x0000000030000000
[12] ending address of 1 MB shared memory segment is 0x00000000300fffff
[13] shmat() at 58 in grabshm.c failed with errno = 24: Too many open files
3.5.3 Extended mode shared memory segments
As explained in the previous sections, up to 11 shared memory segments are
available in the 32-bit default memory model and up to 13 shared memory
segments are available in the 32-bit very large memory model.
Although most 32-bit applications are satisfied with this limitation, some
applications require more shared memory segments. To address this
requirement, AIX supports another type of shared memory segments in the
32-bit user process called extended mode shared memory segments
(EXTSHM).19
The extended mode shared memory segment capability is a process basis
dynamic feature. To use the capability, simply define the following environment
variable before executing your applications:
EXTSHM=ON
Note:
The keyword must be upper case characters. EXTSHM=on is invalid.
Do not insert this line into the /etc/environment file on your system. Some
32-bit applications do not support this capability. Please consult with the
publications shipped with the software products installed on your system.
Once defined, a 32-bit process can have shared memory segments up to
131,072 on AIX Version 4.3.2 and later.
Example 3-24 on page 156 shows the output from the program, when we have
attempted to allocate 100 1 MB shared memory segments with EXTSHM=ON.
19 The extended mode shared memory segment capability has been supported since AIX Version
4.2.1.
Chapter 3. Understanding user process models
155

Example 3-24 Acquiring 100 shared memory segments with EXTSHM=ON
$ EXTSHM=ON a.out -m 100
[ 0] beginning address of 1 MB shared memory segment is 0x0000000030000000
[ 0] ending address of 1 MB shared memory segment is 0x00000000300fffff
[ 1] beginning address of 1 MB shared memory segment is 0x0000000030100000
[ 1] ending address of 1 MB shared memory segment is 0x00000000301fffff
...
[98] beginning address of 1 MB shared memory segment is 0x0000000036200000
[98] ending address of 1 MB shared memory segment is 0x00000000362fffff
[99] beginning address of 1 MB shared memory segment is 0x0000000036300000
[99] ending address of 1 MB shared memory segment is 0x00000000363fffff
Restrictions of extended shared memory segments
Although they seem very convenient, the extended shared memory segments
have some restrictions, which are described in the following:
The EXTSHM environment variable will be ignored and has no effect in the
64-bit user process model.
When attaching a segment larger than 268,431,360 bytes (256 MB - 4 KB),
the EXTSHM environment variable will be ignored, and the process attaches
the segment with a granularity of 256 MB.
When calling the mmap() routine, the EXTSHM environment variable will be
ignored, and the process attaches the segment with a granularity of 256 MB.
The SHM_SIZE parameter of shmctl() is not supported for segments created
with EXTSHM=ON.
No raw I/O is allowed for segments created with EXTSHM=ON.
3.5.4 Order in the 64-bit memory model
Example 3-25 shows the output from this program, when we have attempted to
allocate twelve 1 MB shared memory segments in the 64-bit user process model.
Each 1 MB shared memory segment is allocated into separate 256 MB
segments, starting from 0x0700_0000_0000_0000.
Example 3-25 Acquiring 12 shared memory segments in 64-bit mode
$ cc -q64 -DCALL_SHMDT grabshm.c
$ file a.out
a.out: 64-bit XCOFF executable or object module not stripped
$ a.out -m 12
[ 0] beginning address of 1 MB shared memory segment is 0x0700000000000000
[ 0] ending address of 1 MB shared memory segment is 0x07000000000fffff
[ 1] beginning address of 1 MB shared memory segment is 0x0700000010000000
[ 1] ending address of 1 MB shared memory segment is 0x07000000100fffff
156
Developing and Porting C and C++ Applications on AIX

...
[10] beginning address of 1 MB shared memory segment is 0x07000000a0000000
[10] ending address of 1 MB shared memory segment is 0x07000000a00fffff
[11] beginning address of 1 MB shared memory segment is 0x07000000b0000000
[11] ending address of 1 MB shared memory segment is 0x07000000b00fffff
Although it is technically possible that a 64-bit process can request shared
memory segments up to 131,072 on AIX Version 4.3.2 or later, it is your
responsibility to assure that your application will not consume all the available
system memory.
Note: The EXTSHM environment variable will be ignored and has no effect in
the 64-bit user process model.
3.6 Large page support
Historically, AIX supported 4 KB page size only. Starting with AIX 5L Version 5.1
plus 5100-02 Recommended Maintenance Level, AIX supports alternate page
size (called large pages), in addition to the traditional 4 KB page size. The large
page size on systems using the POWER4 processor is 16 MB, but the large page
size may be different size on future architectures. It is recommended to call the
sysconf() routine with the _SC_LARGE_PAGESIZE parameter in your
applications, in order to determine the supported large page size.
To verify you are using POWER4 processor systems, run the lscfg -vpl procX
command, where X is the instance number of processors. The following example
shows the output of this command executed on an AIX 5L Version 5.2 partition
on the pSeries 690:
# lsdev -Cc processor
proc0 Available 00-00 Processor
proc1 Available 00-01 Processor
# lscfg -vpl proc0
proc0 U1.18-P1-C1 Processor
Device Specific.(YL)........U1.18-P1-C1
PLATFORM SPECIFIC
Name: PowerPC,POWER4
Node: PowerPC,POWER4@0
Device Type: cpu
Physical Location: U1.18-P1-C1
Chapter 3. Understanding user process models
157

The following sections are excerpted from the technical white paper, AIX Support
for Large Pages, found at:
http://www.ibm.com/servers/aix/whitepapers/large_page.html
3.6.1 Large page support overview
Large page usage is primarily intended to provide performance improvements to
high performance computing (HPC) applications. Memory access intensive
applications that use large amounts of virtual memory may obtain performance
improvements by using large pages. The large page performance improvements
are attributable to reduced translation look-aside buffer (TLB) misses due to the
TLB being able to map a larger virtual memory range. Large pages also improve
memory prefetching by eliminating the need to restart prefetch operations on
4 KB boundaries.
The POWER4 large page architecture requires that all virtual pages in a 256 MB
segment be the same size. AIX uses this architecture to support a mixed mode
process model. Some segments in a process are backed with 4 KB pages and
16 MB pages back other segments. Applications may request that their heap
segments be backed with large pages. Applications may also request that shared
memory segments be backed with large pages. Other segments in a process are
backed with 4 KB pages.
AIX supports large page usage with both 32- and 64-bit applications. Both the
32- and 64-bit versions of the AIX kernel support large pages.
AIX maintains separate 4 KB and 16 MB size physical memory pools. The
customer specifies the amount of physical memory in the 16 MB memory pool
using the vmo command on AIX 5L Version 5.2.20 This amount of physical
memory is allocated to the 16 MB memory pool at boot time. The remaining
physical memory is used to back 4 KB virtual pages. The size of the 16 MB pool
is fixed at boot time and cannot be changed without rebooting the system.
On AIX Versions of 5.1 and 5.2, large pages are not paged and treated as pinned
memory. Therefore, an application’s data backed by large pages remains in
physical memory until the application completes.21 A security access control
prevents unauthorized applications from using large pages. This prevents
unauthorized applications from using large page physical memory and
preventing authorized users from using large pages for their applications.
20 On AIX 5L Version 5.1, use the vmtune command instead of vmo.
21 The implementation of large pages may be changed in the future versions of AIX. Do not depend
on large pages being pinned when developing your applications.
158
Developing and Porting C and C++ Applications on AIX

3.6.2 Large page application usage
Applications may use large pages in two ways. An application may request that
large pages back its data and heap segments. An application may also request
shared memory segments be backed by large pages.
Large page data/heap segments
An application may request that its initialized program data, uninitialized program
data (BSS), and heap segments be backed with large pages. There are two ways
to request large pages back an application’s data/heap segments:
The executable file can be marked to request large pages.
An environment variable can be set to request large pages.
A program’s large page data/heap use is established when the program is
exec()ed. A program cannot switch modes after it has begun executing. Large
page use is inherited by children processes on fork().
Marking an executable for large page use
The XCOFF header in an executable file contains a new flag to indicate that the
program wants to use large pages to back its data and heap segments. This flag
can be set when the application is linked by specifying the -blpdata option on the
ld command. The flag can also be set or cleared using the ldedit command.
The ldedit –blpdata filename command sets the large page data/heap flag in
the specified file. The ldedit –bnolpdata filename clears the large page flag.
The ldedit command may also be used to set an executable’s maxdata value. To
check if the flags are correctly set, see 3.2.5, “Checking large memory model
executables” on page 124.
Environment variables for large page use
An environment variable is provided to allow users to indicate they want an
application to use large pages for an application’s data and heap segments. The
environment variable takes precedence over the executable large page flag.
Large page usage is provided as options on the LDR_CNTRL environment
variable.
LDR_CNTRL=LARGE_PAGE_DATA=Y
Specifies that the exec()ed program should use large
pages for its data and heap segments. This is the same
as marking the executable to use large pages.
LDR_CNTRL=LARGE_PAGE_DATA=N
Specifies that the exec()ed program should not use large
pages for its data and heap segments. This overrides the
setting in a executable marked to use large pages.
Chapter 3. Understanding user process models
159

LDR_CNTRL=LARGE_PAGE_DATA=M
Specifies that the exec()ed program should use large
pages in a mandatory mode for its data and heap
segments.
You can separate multiple options on the LDR_CNTRL environment variables by
using an ‘@’ character. For example, the following LDR_CNTRL environment
variable setting requests large page usage along with the maxdata option:
LDR_CNTRL=MAXDATA=0x80000000@LARGE_PAGE_DATA=Y
Users are advised to be cautious in their use of the environment variable to
specify large page usage. Performance tests have shown there can be a
significant performance loss in environments where a number of shell scripts or
small, short running applications are invoked. One example saw a shell script’s
execution time increase over 10 times when the large page environment variable
was specified. Customers are advised to only set the large page environment
variable around specific applications that can benefit from large page usage.
Advisory and mandatory modes
An application can indicate that it wants to use large pages for data/heap
segments in either advisory or mandatory mode. In advisory mode, the
application will use large pages if possible. The conditions needed to use large
pages are:
The user ID is authorized to use large pages.
The system is running on a machine that has the POWER4 large page
architecture feature.
The customer defined a large page memory pool.
There are enough pages in the large page memory pool to back the entire
segment with large pages.
If all of these conditions are met, the application’s data/heap segments will be
backed with large pages. Otherwise, the application’s data/heap segments will be
backed with 4 KB pages.
In advisory mode, an application may have some of its heap segments backed by
large pages and some of them backed by 4 KB pages. 4 KB pages are used to
back segments when there are not enough large pages available to back the
segment. Executable programs marked to use large pages use large pages in
advisory mode.
In mandatory mode, the brk() or sbrk() system calls, which are internally called
from the malloc() subroutine, will fail if the application requests a heap segment
and there are not enough large pages to satisfy the request. Customers that use
the mandatory mode must monitor the size of the large page pool and ensure it
160
Developing and Porting C and C++ Applications on AIX

does not run out of large pages. Otherwise, their mandatory large page mode
applications may fail.
Large page data/heap segments fully backed
The POWER4 architecture requires all pages in a segment (256 MB) be backed
with the same size physical pages. AIX backs the entire 256 MB segment with
large pages when an application requests a large page heap segment. Even if
only a few bytes are needed in the new heap segment, the entire 256 MB
segment is backed. AIX does this to avoid terminating applications when they
want to grow a heap segment (such as when using malloc() or sbrk()) and there
are no large pages available to back the new space. This supports the advisory
mode of large page usage. It also eliminates the need for installations to closely
monitor the size of their large page physical memory pools.
Using large pages to back shared memory segments
AIX uses the POWER4 large page architecture feature to provide large page
backing for shared memory segments. Applications can request their shared
memory segments be backed with large pages by specifying both the
SHM_LGPAGE and SHM_PIN flags on the shmget() function.
The request to use large pages to back a shared segment is advisory. Large
pages will back a shared memory segment under the same conditions as
advisory mode large page data/heap usage. A shared segment is silently backed
with 4 KB pages if large pages are not available.
The physical memory to back large page shared memory and large page
data/heap segments comes from the large page physical memory pool.
Customers must size their large page physical memory pool to contain enough
large pages for both shared memory and data/heap large page usage.
3.6.3 Large page usage security capability
AIX provides a security mechanism to control the use of large page physical
memory by non-root users. The large page physical memory pool is a fixed size,
pinned memory system resource. The security mechanism prevents
unauthorized users from using the large page pool and thus preventing its use by
the intended users or applications.
Non-root users must have a CAP_BYPASS_RAC_VMM capability in order to use
large pages. A system administrator can grant this capability to a user by using
the chuser command. The following command grants the ability to use large
pages to user lpuserid:
chuser capabilities=CAP_BYPASS_RAC_VMM,CAP_PROPAGATE lpuserid
Chapter 3. Understanding user process models
161

Both large page data/heap and large page shared memory segments are
controlled by this capability.
3.6.4 Configuring system to use large pages
The customer must configure the system to use large pages. The customer must
specify the amount of physical memory to be used to back large pages. The
default is to not have any memory allocated to the large page physical memory
pool.
AIX 5L Version 5.2
The vmo command is used to configure the size of the large page physical
memory pool on AIX 5L Version 5.2.22 The following command will allocate 256
pages x 16 MB = 4 GB to the large page physical memory pool:
vmo -r -o lgpg_regions=256 -o lgpg_size=1677216 -o v_pinshm=1
Where:
-r
Updates the /etc/tunable/nextboot file so that the
modified tunable values will take effect after the next
system reboot.
-o lgpg_regions=256
Specifies the reserved memory blocks for large
pages.
-o lgpg_size=1677216
Specifies the large page size in bytes. The allowable
value is 16777216 (16 MB) on POWER4-based
systems.
-o v_pinshm=1
Allows pinning of shared memory segments.
You must run the bosboot command and reboot before the new size large page
memory pool takes effect.
AIX 5L Version 5.1
The vmtune command is used to configure the size of the large page physical
memory pool on AIX 5L Version 5.1.23 The following command will allocate 256
pages x 16 MB = 4 GB to the large page physical memory pool:
vmtune -g 16777216 –L 256
The -g option specifies the large page size in bytes. The allowable value is
16777216 (16 MB) on POWER4-based systems. The -L option is the number of
the -g sized blocks that are allocated to the large page physical memory pool.
22 The vmo command is included in the bos.perf.tune fileset.
23 The vmtune command is located in the /usr/samples/kernel directory.
162
Developing and Porting C and C++ Applications on AIX

You must run the bosboot command and reboot before the new size large page
memory pool takes effect.
If you want to use large pages for shared memory in your applications, the
application source codes must be modified to use the SHM_PIN shmget()
system call flag. The following vmtune command makes the necessary changes
in the kernel to support the SHM_PIN flag:
vmtune –S 1
Note: The vmtune command must be called after every system boot. To place
a permanent change into the system, insert the following lines into /etc/inittab:
vmtune -g 16777216 –L 256
vmtune –S 1
Considerations when determining large page pool size
Here are some things to consider when determining the size of the large page
physical memory pool:
Memory allocated to the large page physical memory pool is not available to
back 4 KB pages. Allocating too much physical memory to large pages will
degrade system performance to the point of not having enough memory to
back 4 KB pages. During system boot, AIX reserves enough physical memory
for 4 KB pages to ensure that the system will boot. However, system failures
may occur after booting if there is not enough physical memory to back 4 KB
pages.
The size of the large page physical memory pool is fixed at boot time and
remains the same for the entire boot. A reboot is required to change the size
of the large page memory pool.
Large pages are only used for applications that explicitly request them. There
is no need for a large page memory pool if your applications do not request
them.
Advisory mode large page applications will use large pages if there are large
pages available. If not, advisory mode large page applications will use 4 KB
pages. However, the inverse is not true. A 4 KB application will not use large
pages if the system runs low on 4 KB pages.
Mandatory mode large page applications will fail if the application requests a
large page and one is not available.
Chapter 3. Understanding user process models
163

3.6.5 Other system changes for large pages
The mprotect() function can not be used against a large page. It returns a -1
return code with an EINVAL errno if called to modify the protection attributes of a
large page.
Some debug malloc tools use mprotect() to diagnose memory management
problems. These tools will not work properly with large pages. Such applications
must use 4 KB pages.
Multi-threaded applications may use large pages for their data/heap segments.
However, when large pages are used, the libpthreads library does not place a
protected red zone page at the bottom of a pthread’s stack.
The sysconf(_SC_LARGE_PAGESIZE) function call will return the large page
size on systems that have large pages.
The vmgetinfo() function returns information about large page pools size and
other large page related information.
3.6.6 Large page usage considerations
Large page is a special purpose performance improvement feature. It is not
recommended for general use. Large page usage provides performance value to
a select set of applications. These are primarily long running memory access
intensive applications that use large amounts of virtual memory.
Not all applications benefit by using large pages. Some applications can be
severely degraded by the use of large pages. Applications that do a large number
of fork()s (such as shell scripts) are especially prone to performance degradation
when large pages are used. Tests have shown a tenfold increase in shell script
execution time when the LDR_CNTRL environment specifies the large page
usage variable. Consider marking specific executable files to use large pages
rather than using the LDR_CNTRL environment variable. This limits large page
usage to the specific applications that benefit from large page usage.
Consider the overall performance effect that large pages may have on your
system. While some specific applications may benefit from large page use, the
overall performance of your system may be degraded by large page usage due to
having reduced the amount of 4 KB page storage available in the system.
Consider using large pages when your system has sufficient physical memory
such that reducing the number of 4 KB pages does not significantly impact
overall system performance.
164
Developing and Porting C and C++ Applications on AIX

4
Managing the memory heap
Chapter 4.
The term heap, or memory heap, generally means a free memory pool, from
which a process can dynamically allocate chunks of memory. Although the
management of the memory heap is the most basic programming task on any
operating system environments, the mismanagement of the memory heap is a
quite common mistake when developing applications using the C and C++
languages.
AIX provides a useful feature, called malloc debug, to diagnose these misuse of
the memory heap without recompiling or modifying application source codes.
The first two sections in this chapter provide basic information about the malloc
subsystem on AIX, and the third section explains how to use the malloc debug
feature. The last section contains the use of library functions to transparently use
process heap or shared memory segments to satisfy memory requests.
Section 4.1, “Malloc subsystem” on page 166
Section 4.2, “Memory allocators” on page 169
Section 4.3, “Use of MALLOCDEBUG options” on page 182
Section 4.4, “Heap management using MEMDBG” on page 199
© Copyright IBM Corp. 2000, 2003. All rights reserved.
165

4.1 Malloc subsystem
A process can dynamically allocate chunks of memory from the process heap by
calling malloc subsystem subroutines. If the subroutine call succeeds, a process
will be given the requested amount of virtual memory pages from the operating
system, and those pages will be contiguously mapped in the process address
space.
On AIX, the mapping addresses of newly allocated virtual pages are varied,
depending on the user process model. For the detailed explanation about the
available address range of the process heap, see 3.2, “The 32-bit user process
model” on page 109 and 3.3, “The 64-bit user process model” on page 130.
The malloc subsystem performs the following fundamental memory operations:
Allocation
Allocates the specified size of virtual memory.
Deallocation
Deallocates (frees) the previous acquired memory space.
Reallocation
Reallocates (adjusts) the size of the previous acquired
memory space.
The malloc subsystem provides the following subroutines grouped by the
function categories:
Allocation
– malloc()
– calloc()
– alloca()
– valloc()
Deallocation
– free()
Reallocation
– realloc()
Other purposes
– mallopt()
– mallinfo()
– mallinfo_heap()
– disclaim()
Once virtual memory pages are allocated by calling either malloc(), calloc(), or
valloc() sub-routines, those routines internally call the system call sbrk() and
166
Developing and Porting C and C++ Applications on AIX

increase the break value, which defines the maximum address of the process
data segment (see Figure 3-8 on page 126).
If the process calls the free() sub-routine with the address of previously allocated
memory, the corresponding virtual memory pages are marked free and placed in
the list,1 with which the malloc subsystem manages free virtual pages. The
reference to free virtual pages should return a segmentation violation
(SIGSEGV) to the caller.
If the process again requests to allocate memory through the malloc subsystem
after that, the subsystem will try to allocate the requested memory size from the
list. If sufficient chunks of memory are available in the list, the subsystem returns
it. Otherwise, sbrk() is again internally called by the subsystem to increase the
break value.
4.1.1 malloc(), calloc(), valloc(), and alloca()
Use the malloc() or calloc() subroutines to request only as much space as you
actually need. Never request and then initialize a maximum-sized array when the
actual situation uses only a fraction of it.
When a process touches a new page to initialize the array elements, the process
effectively forces the VMM to steal a page of real memory from someone. Later,
this results in a page fault when the process that owned that page tries to access
it again. The difference between the malloc() and calloc() subroutines is not just
in the interface.
Because the calloc() subroutine zeroes the allocated storage, it touches every
page that is allocated, whereas the malloc() subroutine touches only the first
page. If you use the calloc() subroutine to allocate a large area and then use only
a small portion at the beginning, you place an unnecessary load on the system.
Not only do the pages have to be initialized, but if their real-memory frames are
reclaimed, the initialized and never-to-be-used pages must be written out to
paging space. This situation wastes both I/O and paging space.
The valloc() subroutine, found in many BSD systems, is supported as a
compatibility interface in the Berkeley compatibility library (libbsd.a). The valloc()
subroutine calls the malloc() subroutine and automatically page-aligns requests
that are greater than one page. The only difference between the valloc()
subroutine in the libbsd.a library and the one in the standard C library (described
above) is in the value returned when the size parameter is zero. The valloc()
subroutine has the same effect as malloc(), except that the allocated memory is
aligned to a multiple of the value returned by sysconf(_ SC_PAGESIZE).
1 Technically, it is an internal data structure, not a linked-list.
Chapter 4. Managing the memory heap
167

The alloca() subroutine allocates the number of bytes of space specified by the
Size parameter in the stack frame of the caller. This space is automatically freed
when the subroutine that called the alloca subroutine returns to its caller.
If alloca() is used in the code and compiled with the C++ compiler,
#pragma alloca would also have to be added before the usage of alloca() in the
code. Alternatively, the -ma option would have to be used while compiling the
code.
4.1.2 mallopt(), mallinfo, and mallinfo_heap()
The mallopt() and mallinfo() subroutines are provided for source-level
compatibility with the System V malloc subroutines. Nothing done with the
mallopt() subroutine affects how memory is allocated by the system, unless the
M_MXFAST option is used.
The mallinfo() subroutine can be used to obtain information about the heap
managed by the malloc() subroutine. Refer to the malloc.h file for details of the
mallinfo structure.
Note: When MALLOCTYPE is set to buckets and the memory request is
within the range of block sizes defined for the buckets, the memory request is
serviced but the heap statistics that are reported by mallinfo() are not updated.
See 4.2.3, “The default memory allocator with the malloc buckets extension”
on page 173 for a detailed explanation about malloc buckets.
The mallinfo_heap() subroutine provides information about a specific heap if
malloc multiheap is enabled (see 4.2.6, “Malloc multiheap” on page 181). The
mallinfo_heap() subroutine returns a structure that details the properties and
statistics of the heap specified by the user. Refer to the malloc.h file for details
about the mallinfo_heap structure.
Note: The mallinfo_heap() subroutine should not be used with the 3.1 memory
allocator explained in 4.2.1, “The 3.1 memory allocator” on page 171.
4.1.3 disclaim()
If a large structure is used early and then left untouched for the remainder of the
process life, it should be released. It is not sufficient to use the free() subroutine
to free the space that was allocated with the malloc() or calloc() subroutines on
AIX, unless the MALLOCDISCLAIM environment variable is defined. The free()
subroutine releases only the address range that the structure occupied from the
process address space. To release the real memory and paging space, use the
168
Developing and Porting C and C++ Applications on AIX

disclaim() subroutine to disclaim the space as well. The call to disclaim() should
be before the call to free().
When the disclaim() routine is called with the specific range of memory address,
it instructs VMM to relinquish physical in-memory pages, and disk blocks in the
paging space. The resultant address-range is logically 0 again, just like it was
when it was created, before any accesses were made to it.
This is a very useful optimization technique that some applications can use to get
performance gains. For example, assume that an application has just finished
using a buffer that is a small portion of a virtual memory address range, and it
needs to manipulate the other data in the same address range. In this case, the
application can call disclaim() specifying that memory range, which will be
replenished with 0-filled pages on subsequent accesses. The call to disclaim()
avoids the unnecessary page-in of staled data from the paging space.
MALLOCDISCLAIM
If the MALLOCDISCLAIM environment variable is set as shown in the following
example before a process start-up, all calls to the free() subroutine automatically
instruct the malloc subsystem to call the disclaim() subroutine internally:
MALLOCDISCLAIM=true
This is useful in circumstances where a process has a high paging-space usage,
but is not actually using the memory.
Note: It is not supported to insert MALLOCDISCLAIM=true in
/etc/environment.
For a detailed description of the malloc subsystem subroutines, please refer to
the “System Memory Allocation Using the malloc subsystem” section in AIX 5L
Version 5.2 General Programming Concepts: Writing and Debugging Programs.
4.2 Memory allocators
There are several internal mechanisms, called memory allocators, in the malloc
subsystem on AIX. Each memory allocator implements a different memory
allocation policy. The allocation policy refers to a set of data structures and
algorithms to represent the heap and to implement allocation, deallocation, and
reallocation.
Chapter 4. Managing the memory heap
169

The malloc subsystem on AIX supports the following memory allocators:
3.1 memory allocator
Default memory allocator
Default memory allocator with the malloc buckets extension
Debug memory allocator
It is also supported to implement the user-defined memory allocator on AIX (see
4.2.5, “User-defined malloc replacement” on page 178).
The programming interface to the malloc subsystem is the same regardless of
the selected memory allocator. The selection of memory allocators is made on a
per-process basis by setting the MALLOCTYPE environment value when a
program is executed. There are several environment variables that affect the
behavior of the selected memory allocator.
Table 4-1 explains how to specify the MALLOCTYPE environment variable in
order to select a memory allocator (related environment variables and sections
are also shown).
Table 4-1 Memory allocators and MALLOCTYPE
MALLOCTYPE=
Memory allocator
Related environment
Detailed explanation found in
variables
3.1
3.1 memory
MALLOCDEBUG
Section 4.2.1, “The 3.1 memory
allocator
allocator” on page 171
(null)1
Default memory
MALLOCDEBUG
Section 4.2.2, “The default
allocator2
MALLOCMULTIHEAP
memory allocator” on page 172
buckets
Default memory
MALLOCDEBUG
Section 4.2.3, “The default
allocator with the
MALLOCMULTIHEAP
memory allocator with the
malloc buckets
MALLOCBUCKETS
malloc buckets extension” on
extension
page 173
debug
Debug malloc
MALLOCDEBUG
Section 4.2.4, “The debug
allocator
malloc allocator” on page 176
user:archive_name2
User-defined
N/A
Section 4.2.5, “User-defined
memory allocator3
malloc replacement” on
page 178
1. The default memory allocator is selected by un-setting the MALLOCTYPE environment variable.
2. Where the archive_name specifies the library archive name that contains the user-defined malloc
subsystem replacement subroutines.
3. A user-defined memory allocator can also be specified in the program code, as explained in 4.2.5,
“User-defined malloc replacement” on page 178.
170
Developing and Porting C and C++ Applications on AIX

By default, the default memory allocator is always selected unless the
MALLOCTYPE environment variable is explicitly set. For example, to specify the
default memory allocator with the malloc extension, do the following on the Korn
shell command line:
$ export MALLOCTYPE=buckets
To reset to the default memory allocator, do the following:
$ export MALLOCTYPE=
Note: These memory allocators are mutually exclusive.
4.2.1 The 3.1 memory allocator
The 3.1 memory allocator is mainly provided to support applications that were
originally developed on AIX Version 3. Some of those applications may depend
on the behavior of the memory allocator on AIX Version 3 and may misbehave
when used with the other memory allocators. Use this memory allocator only
when it is absolutely required by a specific set of applications.
To select the 3.1 memory allocator, set the MALLOCTYPE environment variable
before executing programs on the Korn shell prompt:
$ export MALLOCTYPE=3.1
The 3.1 memory allocator maintains the process heap as a set of 28 hash
buckets, each of which points to a linked list. Hashing is a method of transforming
a search key into an address for the purpose of storing and retrieving items of
data. The method is designed to minimize the average search time. A bucket is
one or more fields in which the result of an operation is kept. Each linked list
contains blocks of a particular size. The index into the hash buckets indicates the
size of the blocks in the linked list. The size of the block is calculated using the
following formula:
size = 2 i + 4
where i identifies the bucket. This means that the blocks in the list anchored by
bucket zero are 20 + 4 = 16 bytes long. Therefore, given that a prefix is 8 bytes in
size, these blocks can satisfy requests for blocks between 0 and 8 bytes long.
To use the 3.1 memory allocator, keep the following points in mind:
The 3.1 memory allocator supports the 32-bit user process environment only.
It does not support the 64-bit user process environment.
Chapter 4. Managing the memory heap
171

The 3.1 memory allocator does not support the following environment
variables. If these are set, the 3.1 memory allocator simply ignores them:
– MALLOCMULTIHEAP
– MALLOCBUCKETS
The algorithm can use as much as twice the amount of memory actually
requested by the application. An extra page is required for buckets larger than
4096 bytes because objects of larger page are page-aligned. Because the
prefix immediately precedes the block, an entire page is required solely for
the prefix.
The 3.1 memory allocator supports the MALLOCDEBUG environment variable
with the following keywords:
log
verbose
trace
4.2.2 The default memory allocator
The default allocation policy maintains the free space in the heap as a binary
tree, in which nodes are sorted vertically by length and horizontally by address.
The data structure imposes no limitation on the number of block sizes supported
by the tree, allowing a wide range of potential block sizes. Tree-reorganization
techniques optimize access times for node location, insertion, and deletion, and
also protect against fragmentation.
To select the default memory allocator, the MALLOCTYPE environment variable
must be unset before executing programs. To confirm if it is unset, do following on
the Korn shell prompt:
$ echo $MALLOCTYPE
$
The echo command should return a blank line, as shown in the above example.
The default memory allocator supports the following:
Both the 32- and 64-bit user process environment
The MALLOCMULTIHEAP environment variable
The MALLOCDEBUG environment variable with the following keywords:
– log
– verbose
– arena_check
172
Developing and Porting C and C++ Applications on AIX

– trace
The default memory allocator does not support the MALLOCBUCKETS
environment variable. If it is set, the allocator simply ignores it.
4.2.3 The default memory allocator with the malloc buckets extension
The default memory allocator with the malloc buckets extension, or simply malloc
buckets, provides an optional buckets-based extension of the default allocator. It
is intended to improve malloc subsystem performance for applications that issue
large numbers of small allocation requests. When malloc buckets is enabled,
allocation requests that fall within a predefined range of block sizes are
processed by malloc buckets. All other requests are processed in the usual
manner by the default allocator.
To select malloc buckets, set the MALLOCTYPE environment variable before
executing programs on the Korn shell prompt:
$ export MALLOCTYPE=buckets
Additional user configuration can be done by explicitly setting the
MALLOCBUCKETS environment value (see “MALLOCBUCKETS” on page 174).
Bucket composition and sizing
A bucket consists of a block of memory that is subdivided into a predetermined
number of smaller blocks of uniform size, each of which is an allocatable unit of
memory. Each bucket is identified using a bucket number. The first bucket is
bucket 0, the second bucket is bucket 1, the third bucket is bucket 2, and so on.
The first bucket is the smallest, and each succeeding bucket is larger in size than
the preceding bucket, using a formula described later in this section. A maximum
of 128 buckets is available per heap.
The block size for each bucket is a multiple of a bucket-sizing factor. The
bucket-sizing factor equals the block size of the first bucket. Each block in the
second bucket is twice this size, each block in the third bucket is three times this
size, and so on. Therefore, a given bucket's block size is determined as follows:
block size = (bucket number + 1) * bucket sizing factor
For example, a bucket-sizing factor of 16 would result in a block size of 16 bytes
for the first bucket (bucket 0), 32 bytes for the second bucket (bucket 1), 48 bytes
for the third bucket (bucket 2), and so on.
The bucket-sizing factor must be a multiple of 8 for 32-bit implementations and a
multiple of 16 for 64-bit implementations in order to guarantee that addresses
returned from malloc subsystem functions are properly aligned for all data types.
Chapter 4. Managing the memory heap
173

The bucket size for a given bucket is determined as follows:
bucket size = number of blocks per bucket *
(malloc overhead + ((bucket number + 1) * bucket sizing factor))
The preceding formula can be used to determine the actual number of bytes
required for each bucket. In this formula, malloc overhead refers to the size of an
internal malloc construct that is required for each block in the bucket. This
internal construct is 8 bytes long for 32-bit applications and 16 bytes long for
64-bit applications. It is not part of the allocatable space available to the user, but
is part of the total size of each bucket.
The default memory allocator with the malloc buckets extension supports the
following:
Both the 32- and 64-bit user process environment
The MALLOCMULTIHEAP environment variable
The MALLOCDEBUG environment variable with the following keywords:
– log
– verbose
– arena_check
– trace
MALLOCBUCKETS
The number of blocks per bucket, number of buckets, and bucket-sizing factor are
all set with the MALLOCBUCKETS environment variable. The syntax of the
variable is as follows (multiple keywords can be separated by a comma):
MALLOCBUCKETS=[[ number_of_buckets:N | bucket_sizing_factor:N |
blocks_per_bucket:N | bucket_statistics:[stdout|stderr|path_name]],...]
Where:
number_of_buckets:N
This option can be used to specify the number of buckets
available per heap, where N is the number of buckets. The
value specified for N will apply to all available heaps.
The default value for number_of_buckets is 16. The
minimum value allowed is 1. The maximum value allowed
is 128.
bucket_sizing_factor:N
This option can be used to specify the bucket-sizing
factor, where N is the bucket-sizing factor in bytes. The
value specified for bucket_sizing_factor must be a multiple
of 8 for 32-bit implementations and a multiple of 16 for
174
Developing and Porting C and C++ Applications on AIX

64-bit implementations.
The default value for bucket_sizing_factor is 32 for 32-bit
implementations and 64 for 64-bit implementations.
blocks_per_bucket:N
This option can be used to specify the number of blocks
initially contained in each bucket, where N is the number
of blocks. This value is applied to all of the buckets. The
value of N is also used to determine how many blocks to
add when a bucket is automatically enlarged because all
of its blocks have been allocated. The default value for
blocks_per_bucket is 1024.
bucket_statistics:[stdout|stderr|path_name]
The bucket_statistics option will cause the malloc
subsystem to output a statistical summary for malloc
buckets upon normal termination of each process that
calls the malloc subsystem while malloc buckets is
enabled. This summary shows buckets-configuration
information and the number of allocation requests
processed for each bucket. If multiple heaps have been
enabled by way of malloc multiheap, the number of
allocation requests shown for each bucket will be the sum
of all allocation requests processed for that bucket for all
heaps.
The buckets statistical summary will be written to one of
the following output destinations, as specified with the
bucket_statistics option.
stdout
Standard output
stderr
Standard error
path_name
A user-specified path name
If a user-specified path name is provided, statistical output
will be appended to the existing contents of the file (if
any).
Standard output should not be used as the output
destination for a process whose output is piped as input
into another process.
The buckets_statistics option is disabled by default.
Chapter 4. Managing the memory heap
175

Note:
1. One additional allocation request will always be shown in the first bucket for
the atexit() subroutine that prints the statistical summary.
2. For multi-threaded processes, additional allocation requests will be shown
for some of the buckets due to malloc subsystem calls issued by the
Pthread library.
If the MALLOCBUCKETS environment variable is not set, then the default values
shown in Table 4-2 are assumed.
Table 4-2 Default configuration values for malloc buckets
Configuration option
Default value (32-bit)
Default value (64-bit)
Number of buckets per
16
16
heap
Bucket sizing factor
32 bytes
64 bytes
Allocation range
1 to 512 bytes (inclusive)
1 to 1024 bytes (inclusive)
Number of blocks initially
1024
1024
contained in each bucket
Bucket statistical summary
disabled
disabled
4.2.4 The debug malloc allocator
Debugging applications that are mismanaging memory allocated via the malloc
subsystem can be difficult and tedious. Most often, the problem is that data is
written past the end of an allocated buffer. Since this has no immediate
consequence, problems do not become apparent until much later when the
space that was overwritten (usually belonging to another allocation) is used and
no longer contains the data originally stored there.
The AIX malloc subsystem includes the debug memory allocator2 to allow users
to identify memory overwrites, over-reads, duplicate frees, and reuse of freed
memory allocated by malloc(). This memory allocator is sometimes referred to as
debug malloc.
Memory problems detected by the debug memory allocator result in an abort() or
a segmentation violation (SIGSEGV). In most cases, when an error is detected
the application stops immediately and a core file is produced.
2 The debug memory allocator has been supported on AIX since Version 4.3.3.
176
Developing and Porting C and C++ Applications on AIX

To select the debug memory allocator, set the MALLOCTYPE environment
variable before executing programs on the Korn shell prompt:
$ export MALLOCTYPE=debug
Additional user configuration can be done by explicitly setting the
MALLOCDEBUG environment value as follows:
MALLOCDEBUG=<options...>
where options is a comma-separated list of one or more predefined configuration
options:
align:N
validate_ptrs
postfree_checking
allow_overreading
override_signal_handling
record_allocations
report_allocations
More than one option can be specified (and in any order) as long as options are
comma-separated, as shown in the following example:
MALLOCDEBUG=align:0,validate_ptrs,report_allocations
Each configuration option should only be specified once when setting
MALLOCDEBUG. If a configuration option is specified more than once per
setting, only the final instance will apply. For a further detailed explanation about
these options, see 4.3.1, “MALLOCDEBUG with the debug memory allocator” on
page 182.
Note: The debug memory allocator does not support the several
MALLOCDEBUG options explained in 4.3.2, “MALLOCDEBUG with memory
allocators other than debug” on page 190.
The debug memory allocator is not appropriate for full-time, constant, or
system-wide use. Although it is designed for minimal performance impact upon
the application being debugged, it may have significant negative impact upon
overall system throughput if it is used widely throughout a system. In particular,
setting MALLOCTYPE=debug in the /etc/environment file (to enable debug
malloc for the entire system) is unsupported, and will likely cause significant
system problems, such as excessive use of paging space. The debug memory
allocator should only be used to debug single applications or small groups of
applications at the same time.
Chapter 4. Managing the memory heap
177

In addition, please note that the debug memory allocator is not appropriate for
use in some debugging situations. Because the debug memory allocator places
each individual memory allocation on a separate page, programs that issue
many small allocation requests will see their memory usage increase
dramatically. These programs may encounter new failures as memory allocation
requests are denied due to a lack of memory or paging space. These failures are
not necessarily errors in the program being debugged, and they are not errors in
the debug memory allocator.
4.2.5 User-defined malloc replacement
If none of memory allocators provided by AIX did not satisfy the specific memory
management requirement for your applications, you can implement your own
memory allocator, called a user-defined memory allocator, on AIX. This
functionality is referred to as a user-defined malloc replacement.
Once the user-defined memory allocator has been prepared, all the calls to the
malloc subsystem from user applications, such as malloc(), free(), and so on, are
transparent; no modification is necessary.
In the C and C++ program languages linkage mechanism, it is always possible to
override system subroutines (external linkage symbols) by user-defined functions
(internal linkage symbols). For example, the sample source code shown in
Example 4-1 defines a fake version of malloc(), which simply returns the address
of global character array (as long as the program requires only one memory area
up to size of BUFSIZ, this fake version of malloc() is sufficient for this program). If
compiled and executed, the program prints the following output:
$ cc fake_malloc.c
$ a.out
p = 1234567890
$ echo $?
0
Example 4-1 fake_malloc.c
#include <stdio.h>
#include <stdlib.h>
#include <errno.h>
char global_buf[BUFSIZ];
/* fake malloc() routine. */
void *malloc(size_t sz)
{
/* return the address of global character array. */
return((void *)global_buf);
}
178
Developing and Porting C and C++ Applications on AIX

int
main(int argc, char *argv[])
{
char *p;
if ((p = (char *)malloc(BUFSIZ)) == (char *)NULL) {
perror("malloc() failed.\n");
exit(1);
}
/* write some data to the address pointed by p. */
strcpy(p, "1234567890");
/* print the written data. */
printf("p = %s\n", p);
exit(0);
}
It must be clear that this overriding of external symbol names is different from the
user-defined malloc replacement. If programmed appropriately, the user-defined
memory allocator will be used for all the malloc subsystem calls, not only from
user codes, but also from functions within shared libraries (many subroutines
within shared libraries provided by AIX call the malloc subsystem internally).
To implement user-defined memory allocators, the following requirements must
be satisfied:
A user defined memory memory allocator must provide both the 32- and
64-bit object modules. Both modules must be placed in a library archive file
and the 32-bit shared object must be named mem32.o and the 64-bit shared
object must be named mem64.o.
A user-defined memory memory allocator should be thread-safe. This is the
programmer’s responsibility; there are no automatic checks to verify it.
A user-defined memory memory allocator must implement the functions with
those names that start with double underscore characters, as shown in
Table 4-3.
Table 4-3 User-defined replacement subroutines
Function name, proto-type declaration
Description
void *__malloc__(size_t)
A user-defined replacement of malloc().
void __free__(void *)
A user-defined replacement of free().
void *__realloc__(void *, size_t)
A user-defined replacement of realloc().
void *__calloc__(size_t, size_t)
A user-defined replacement of calloc().
Chapter 4. Managing the memory heap
179

Function name, proto-type declaration
Description
int __mallopt__(int, int)
A user-defined replacement of mallopt().
struct mallinfo __mallinfo__()
A user-defined replacement of mallinfo().
void __malloc_once__()
Will be called once before any other
user-defined malloc entry point is called.
void __malloc_init__(void)
Called by the Pthread initialization routine
to initialize the user-defined memory
allocator in the multi-threaded
programming environment.1
void __malloc_prefork_lock__(void)
Called by Pthread when the fork()
subroutine is called.1
void __malloc_postfork_unlock__(void)
Called by Pthread when the fork()
subroutine is called.1
1. These functions are mandatory in the multi-threaded environment.
The shared objects (mem32.o for 32-bit and mem64.o for 64-bit) must export
the following symbols:
– __malloc__
– __free__
– __realloc__
– __calloc__
– __mallinfo__
– __mallopt__
– __malloc_init__
– __malloc_prefork_lock__
– __malloc_postfork_unlock__
The shared objects can optionally export __malloc_once__.
To select the user-defined memory allocator, set the MALLOCTYPE environment
variable before executing programs on the Korn shell prompt:
$ export MALLOCTYPE=user:archive_name
Where the archive_name specifies the library archive name that contains the
user-defined malloc subsystem replacement subroutines.
180
Developing and Porting C and C++ Applications on AIX

A user-defined memory allocator can also be specified in the program code by
declaring the global symbol _malloc_user_defined_name, as shown in the
following example:
char *_malloc_user_defined_name=”archive_name”;
If both the MALLOCTYPE environment variable and the global symbol are used
to specify the archive_name, the name specified by MALLOCTYPE will override
the one specified by the global symbol.
Note: User-defined memory allocators written in C++ are not supported,
because the C++ standard library libC.a depends on the standard malloc
subsystem provided by the C standard library libc.a.
4.2.6 Malloc multiheap
Historically, the malloc subsystem was designed for the non-threaded
programming environment. Therefore, there was an single memory pool, or
memory heap, per-process basis. After the evolution of the multi-threaded
programming environment, it was realized that the single heap does not satisfy
the memory allocation requests from multi-threaded applications, because a
single malloc() call from a user thread can lock the entire malloc subsystem and
the other user threads would be starving; thus, malloc() calls within a process
would be serialized.
By providing multiple heaps, malloc multiheap efficiently supports the memory
allocation requests from multi-threaded applications; thus, the malloc multiheap
has a finer locking mechanism than the single heap malloc subsystem. The
potential performance enhancement is particularly likely for multi-threaded C++
programs, because these may make use of the malloc subsystem whenever a
constructor or destructor is called.
Beginning with Version 5.1, the malloc multiheap is enabled by default in the
malloc subsystem on AIX. Therefore, it does not require any user settings,
though it can be tuned using the MALLOCMULTIHEAP environment variable. To
set the MALLOCMULTIHEAP environment variable, use the following syntax:
MALLOCMULTIHEAP=[[heaps:N],[considersize],...]
Where:
heaps:N
The heaps:N option can be used to change the maximum
number of heaps to any value from 1 through 32, where N
is the number of heaps. If n is set to a value outside the
given range, the default value of 32 is used.
Chapter 4. Managing the memory heap
181

considersize
By default, malloc multiheap selects the next available
heap. If the considersize option is specified, malloc
multiheap will use an alternate heap-selection algorithm
that tries to select an available heap that has enough free
space to handle the request. This may minimize the
working set size of the process by reducing the number of
sbrk subroutine calls. However, because of the additional
processing required, the considersize heap-selection
algorithm is somewhat slower than the default heap
selection algorithm.
Multiple keywords are separated by a comma.
Note: The malloc multiheap is enabled internally on AIX beginning with
Version 5.1 with the default value of 32. Therefore, echo $MALLOCMULTIHEAP
prints a blank line by default.
4.3 Use of MALLOCDEBUG options
This section explains the use of the MALLOCDEBUG environment variable by
providing the following sections:
Section 4.3.1, “MALLOCDEBUG with the debug memory allocator” on
page 182
Section 4.3.2, “MALLOCDEBUG with memory allocators other than debug”
on page 190
4.3.1 MALLOCDEBUG with the debug memory allocator
The debug memory allocator supports the following options specified by the
MALLOCDEBUG environment variable:
“align:N” on page 183
“validate_ptrs” on page 185
“postfree_checking” on page 186
“allow_overreading” on page 188
“override_signal_handling” on page 188
“record_allocations” on page 189
“report_allocations” on page 189
182
Developing and Porting C and C++ Applications on AIX

Note: The debug memory allocator does not support the MALLOCDEBUG
options explained in 4.3.2, “MALLOCDEBUG with memory allocators other
than debug” on page 190.
align:N
By default, malloc() returns a pointer aligned on a 2-word3 boundary in the 32-bit
and 4-word boundary in the 64-bit user process environment. The align:N option
can be used to change the default alignment, where N is the number of bytes to
be aligned and can be any power of 2 between 0 and 4096 inclusive (for
example, 0, 1, 2, 4, …). The values 0 and 1 are treated as the same, that is, there
is no alignment, so any memory accesses outside the allocated area will cause
an abort().
The following formula can be used to calculate how many bytes of over-reads or
over-writes the debug memory allocator will allow for a given allocation request
when MALLOCDEBUG=align:N and size is the number of bytes to be allocated:
((((size / N) + 1) * N) - size) % N
The example program shown in Example 4-2 demonstrates the effect of the
align:N option. This program allocates a character string array size of 10 bytes,
then prompts user to input some data, which will be stored in the previously
allocated array.
Example 4-2 debug_malloc_align.c
#include <stdio.h>
#include <stdlib.h>
#include <errno.h>
#define MAX_SIZE 10
#define EXIT_CODE -1
int main(int argc, char *argv[])
{
char *ptr = (char *)NULL;
char str[BUFSIZ];
if ((ptr = (char *)malloc(MAX_SIZE)) == (char *)NULL) {
perror("malloc() failed.\n");
exit(EXIT_CODE);
}
printf("Enter the value for ptr: ");
gets(str);
strcpy(ptr, str);
3 The term word means an implementation dependent unit of memory. On AIX, a word is 32 bits (4
bytes).
Chapter 4. Managing the memory heap
183

printf("ptr points at : %p\n", ptr);
printf("The value stored in ptr is : %s\n", ptr);
free(ptr);
}
Before executing the program, set the following environment variables to enable
the align:N option from the command prompt:
$ export MALLOCTYPE=debug
$ export MALLOCDEBUG=align:2
Applying the above mentioned formula for align:2, the number of bytes of
over-read or over-write allowed is:
align:2
((((10/2) + 1) * 2) - 10) % 2 = 0
Therefore, the debug memory allocator will not allow any over-reads or
over-writes. If executed, the program would print the following output (the
character string 12345678901 is user input):
$ a.out
Enter the value for ptr: 12345678901
Segmentation fault(coredump)
The program is terminated by a segmentation fault, because the length of ptr is
12 (11 printable characters plus the NULL-termination character ‘0x0’).
If align:4 or align:8 is specified, the allowed over-read or over-write memory byte
region size would be 2 or 6 bytes, as shown in the following calculation:
align:4
((((10/4) + 1) * 4) - 10) % 4 = 2
align:8
((((10/8) + 1) * 8) - 10) % 8 = 6
For example, if align:4 is specified, the same program prints the following output
(no segmentation fault occurs):
$ export MALLOCTYPE=debug
$ export MALLOCDEBUG=align:4
$ a.out
Enter the value for ptr: 12345678901
ptr points at : 20001ff4
The value stored in ptr is : 12345678901
The following points should be considered while setting the align:N option:
For allocated space to be word aligned, specify align:N with a value of 4.
If the align:N option is not explicitly set, it defaults to 8.
184
Developing and Porting C and C++ Applications on AIX

validate_ptrs
By default, free() does not validate its input pointer to ensure that it actually
references memory previously allocated by malloc(). If the parameter passed to
free() is a NULL value, free() will return to the caller without taking any action. If
the parameter is invalid, the results will be undefined. A core dump may or may
not occur in this case, depending upon the value of the invalid parameter.
Specifying the validate_ptrs option will cause free() to perform extensive
validation on its input parameter. If the parameter is found to be invalid (that is, it
does not reference memory previously allocated by a call to malloc() or realloc()),
debug malloc will print an error message stating why it is invalid. The abort()
function is then called to terminate the process and produce a core file.
The example program shown in Example 4-3 demonstrates the effect of the
validate_ptrs option. This is slightly modified from Example 4-2 on page 183 and
calls the free() subroutine twice at the end of the program. Though the second
call of free() is an error, it does not abort the execution of the program in normal
situations.
Example 4-3 debug_malloc_vptr.c
/*This program is a slightly modified version of debug_mallo_align.c. We are
just trying to free the ptr memory even after it is freed by the first free()
call*/
#include <stdio.h>
#include <stdlib.h>
#include <errno.h>
#define MAX_SIZE 10
#define EXIT_CODE -1
int main(int argc, char*argv[])
{
char *ptr = NULL;
char str[BUFSIZ];
if ((ptr = (char *)malloc(MAX_SIZE)) == (char *)NULL) {
perror("malloc() failed.\n");
exit(EXIT_CODE);
}
printf("Enter the value for ptr: ");
gets(str);
strcpy(ptr, str);
printf("ptr points at : %p\n", ptr);
printf("The value stored in ptr is : %s\n", ptr);
free(ptr);
free(ptr); /* This is invalid call. ptr is already freed. */
}
Chapter 4. Managing the memory heap
185

Before executing the program, set the following environment variables to enable
the validate_ptr option from the command prompt:
$ export MALLOCTYPE=debug
$ export MALLOCDEBUG=validate_ptrs
If executed, the program would print the following output (the character string
1234567890 is user input):
$ a.out
Enter the value for ptr: 1234567890
ptr points at : 20001ff0
The value stored in ptr is : 1234567890
Debug Malloc: Buffer (0x20001ff0) has already been free'd.
IOT/Abort trap(coredump)
As highlighted in the output, the debug memory allocator has detected the invalid
second free() call.
postfree_checking
By default, the malloc subsystem allows the calling program to access memory
that has previously been freed. This should result in an error in the calling
program. If the postfree_checking option is specified, any attempt to access
memory after it is freed will cause the debug memory allocator to report the error
and abort the program; then a core file will be produced.
Note: Specifying the postfree_checking option automatically enables the
validate_ptrs option.
If the same program shown in Example 4-3 on page 185 is executed with the
postfree_checking by setting the following environment variables:
$ export MALLOCTYPE=debug
$ export MALLOCDEBUG=postfree_checking
then it will result in a segmentation fault, though the reason for that is not clearly
reported in the following output:
$ a.out
Enter the value for ptr: 1234567890
ptr points at : 20001ff0
The value stored in ptr is : 1234567890
Segmentation fault(coredump)
The postfree_checking option identifies the access (if any) to the memory after it
is freed, but the validate_ptrs option does not. The example program shown
Example 4-4 on page 187 illustrates the difference between the
186
Developing and Porting C and C++ Applications on AIX

postfree_checking and validate_ptrs options. This program is trying to access the
ptr memory after it is freed by the free() call.
Example 4-4 debug_malloc_pfc.c
#include <stdio.h>
#include <stdlib.h>
#include <errono.h>
#define MAX_SIZE 10
#define EXIT_CODE -1
int main(int argc, char *argv[])
{
char *ptr = (char *)NULL;
char str[BUFSIZ];
if ((ptr = (char *)malloc(MAX_SIZE)) == (char *)NULL) {
perror("malloc() failed.\n");
exit(EXIT_CODE);
}
printf("Enter the value for ptr: ");
gets(str);
strcpy(ptr, str);
printf("ptr points at : %p\n", ptr);
printf("The value stored in ptr is : %s\n", ptr);
free(ptr);
/* Wrong. trying to access memory after it is freed. */
printf("The value stored in ptr is : %s\n", ptr);
}
Before executing the program, set the following environment variables to enable
the validate_ptr option from the command prompt:
$ export MALLOCTYPE=debug
$ export MALLOCDEBUG=validate_ptrs
If executed, the program would print the following output (the highlighted
character string 12345 is a user input):
$ a.out
Enter the value for ptr: 12345
ptr points at : 20001ff0
The value stored in ptr is : 12345
Apparently the debug memory allocator with the validate_ptr option did not detect
the error. The reason is that the validate_ptrs option checks for the validity of ptr
only when it is passed to a free() function call.
Chapter 4. Managing the memory heap
187

If the same program is executed with the validate_ptrs option, the program would
print the following output (the highlighted character string 12345 is a user input):
$ export MALLOCTYPE=debug
$ export MALLOCDEBUG=postfree_checking
$ a.out
Enter the value for ptr: 12345
ptr points at : 20001ff0
The value stored in ptr is : 12345
Segmentation fault(coredump)
With the postfree_checking option set, the debug memory allocator identifies the
erroneous access to the memory after it is freed.
allow_overreading
By default, the debug memory allocator will respond with a segmentation
violation and if the program attempts to read past the end of allocated memory.
The allow_overreading option instructs the debug memory allocator to ignore
over-reads of this nature so that other types of errors, which may be considered
more serious, can be detected first.
override_signal_handling
The debug memory allocator reports errors in one of two ways:
Memory access errors (such as trying to read or write past the end of
allocated memory) will cause a segmentation violation (SIGSEGV), resulting
in a core dump.
For other types of errors (such as trying to free space that was already freed),
the debug memory allocator will print an error message, then call abort(),
which will send a SIGIOT signal to terminate the current process.
If the calling program is blocking or catching the SIGSEGV and/or the SIGIOT
signals, the debug memory allocator will be prevented from reporting errors. The
override_signal_handling option provides a means of addressing this situation
without recording and rebuilding the application.
If the override_signal_handling option is specified, the debug memory allocator
will perform the following actions upon each call to one of the memory allocation
routines (malloc(), free(), realloc(), or calloc()):
1. Disables any existing signal handlers set up by the application.
2. Sets the action for both SIGIOT and SIGSEGV to the default (SIG_DFL).
3. Unblocks both SIGIOT and SIGSEGV.
188
Developing and Porting C and C++ Applications on AIX

When using the override_signal_handling option, keep in mind the following:
If an application signal handler modifies the action for SIGSEGV between
memory allocation routine calls and then attempts an invalid memory access,
the debug memory allocator will be unable to report the error (the application
will not exit and no core file will be produced).
The override_signal_handling option may be ineffective in a multi-threaded
application environment because the debug memory allocator uses
sigprocmask() and many multi-threaded processes use pthread_sigmask().
If a user thread calls sigwait() without including SIGSEGV and SIGIOT in the
signal set and the debug memory allocator subsequently detects an error, the
user thread will hang because the allocator can only generate SIGSEGV or
SIGIOT.
record_allocations
The record_allocations option instructs the debug memory allocator to create an
allocation record for each malloc() request. Each record contains the following
information:
The original address returned to the caller from malloc().
Up to six function trace backs starting from the call to malloc().
Each allocation record will be retained until the memory associated with it is
freed.
report_allocations
The report_allocations option instructs the debug memory allocator to report all
active allocation records at application exit. An active allocation record will be
listed for any memory allocation that was not freed prior to application exit.
Note: Specifying the report_allocations option automatically enables the
record_allocations option.
To demonstrate how the report_allocations works, we have slightly modified the
example program shown in Example 4-3 on page 185 by removing two free()
lines (highlighted in the example).
Before executing the program, set the following environment variables to enable
the validate_ptr option from the command prompt:
$ export MALLOCTYPE=debug
$ export MALLOCDEBUG=report_allocations
Chapter 4. Managing the memory heap
189

If executed, the program would print the following output (the highlighted
character string 12345678901 is a user input):
$ a.out
Enter the value for ptr: 12345678901
ptr points at : 20003ff0
The value stored in ptr is : 12345678901
Current allocation report:
Allocation #1: 0x20003FF0
Allocation size: 0xA
Allocation traceback:
0x100001B4 __start
0x1000035C main
0xD01D7104 malloc
Allocation #2: 0x20001FF0
Allocation size: 0x10
Allocation traceback:
0x1000035C main
0xD01D7104 malloc
0xD01D6C28 init_malloc
0xD01D6120 check_environment
0xD022BA10 malloc_debug_start
0xD01E42D4 atexit
Total allocations: 2.
The output contains the allocation report (Allocation #1) for the un-freed memory
in the program (the memory address printed for Allocation #1 is same as the
address location of ptr: 0x20003FF0). Because this memory has not been freed,
the debug memory allocator detects it and then prints the allocation report.
Note: One allocation record will always be listed for the atexit() handler that
prints the allocation records, as shown in the previous output (Allocation #2).
4.3.2 MALLOCDEBUG with memory allocators other than debug
The options shown in Table 4-4 on page 191, which can be specified by the
MALLOCDEBUG environment value, are supported by 3.1, default, and default
with malloc buckets extension memory allocators:
190
Developing and Porting C and C++ Applications on AIX

Table 4-4 MALLOCDEBUG options
Option
Feature name
Related section
verbose
Malloc report
“verbose” on page 191
arenaa
“arena” on page 192
trace
Malloc trace
“trace” on page 193
log
Malloc log
“log” on page 196
a. The arena option is not supported by the 3.1 memory allocator.
Multiple options can be specified by separating them using a comma as follows:
MALLOCDEBUG=option1,option2,...
verbose
The verbose option instructs memory allocators that information on errors that
occurred in the malloc subsystem will be reported and actions can be performed
if specified.
The verbose option is not enabled by default, but can be enabled and configured
prior to process startup by setting the MALLOCDEBUG environment variable as
follows:
MALLOCDEBUG=verbose
All errors caught in the malloc subsystem are output to standard error, along with
detailed information.
The verbose option allows the user to provide a function that the malloc
subsystem will call when it encounters an error. Before returning, the malloc
subsystem calls the user-provided function, if it is specified.
A global function pointer is available for use by the user. In the code, the following
function pointer should be set to the user’s function:
extern void (*malloc_err_function)(int, ...)
The following user-defined function must be implemented:
void malloc_err_function(int, ...)
For example, to use the user-defined function abort_sub, the following code must
be inserted into the user’s application:
malloc_err_function = &abort_sub;
Chapter 4. Managing the memory heap
191

A sample program shown in Example 4-5 illustrates the use of the verbose
option. When executed, the program prints the following error message:4
$ MALLOCDEBUG=verbose a.out
Malloc Report: Corruption in the Yorktown Malloc arena has been detected.
IOT/Abort trap(coredump)
The default memory allocator with the verbose option has detected the
corruption in this output.
Example 4-5 debug_malloc_verbose.c
#include <stdio.h>
#include <stdlib.h>
#include <errno.h>
#include <malloc.h>
int main(int argc, char *argv[])
{
char *ptr;
char buf[BUFSIZ];
if ((ptr = (char *)malloc(1200)) == (char *)NULL) {
sprintf(buf, "malloc() failed at %d in %s with errno = %d"
, __LINE__, __FILE__, errno);
perror(buf);
}
free(ptr);
memset(ptr - 8, (char)-1, 40);
free(ptr - 4096);
exit(0);
}
arena
The arena option instructs the malloc subsystem to check the structures that
contain the free blocks before every allocation request is processed. This option
will ensure that the arena is not corrupted. Also, the arena will also be checked
when an error occurs.
The checkarena option checks for NULL pointers in the free tree or pointers that
do not fall within a certain range. If an invalid pointer is encountered during the
descent of the tree, the program might perform a core dump depending on the
value of the invalid address.
4 Yorktown is an internal name used in the AIX development to refer to the default memory allocator.
192
Developing and Porting C and C++ Applications on AIX

The arena option is not enabled by default, but can be enabled and configured
prior to process startup by setting the MALLOCDEBUG environment variable as
follows:
MALLOCDEBUG=checkarena
A sample program shown in Example 4-6 illustrates the use of the arena option.
When executed, the program prints the following error message:
$ MALLOCDEBUG=verbose,arena a.out
Malloc Report: The address passed to free, 0x1ffff5f8, is outside the valid
range of addresses allocated by malloc (errno = 0).
IOT/Abort trap(coredump)
Example 4-6 debug_malloc_arena.c
#include <stdio.h>
#include <stdlib.h>
#include <errno.h>
#include <malloc.h>
int main(int argc, char *argv[])
{
char *ptr;
char buf[BUFSIZ];
if ((ptr = (char *)malloc(16)) == (char *)NULL) {
sprintf(buf, "malloc() failed at %d in %s with errno = %d"
, __LINE__, __FILE__, errno);
perror(buf);
exit(1);
}
free(ptr-4096);
exit(0);
}
trace
The trace option instructs the malloc subsystem to use the trace facility. Traces of
the malloc(), realloc(), and free() subroutines are recorded for use in problem
determination and performance analysis.
The trace option is not enabled by default, but can be enabled and configured
prior to process startup by setting the MALLOCDEBUG environment variable as
follows:
MALLOCDEBUG=trace
Chapter 4. Managing the memory heap
193

The trace option supports the following trace hook IDs:
HKWD_LIB_MALL_COMMON (hook ID: 60a)
When tracing is enabled for HKWD_LIB_MALL_COMMON, the input
parameters, as well as return values for each call to malloc(), realloc(), and
free() subroutines, are recorded in the trace subsystem. In addition to
providing trace information about the malloc subsystem, the trace option also
performs checks of its internal data structures. If these structures have been
corrupted, these checks will likely detect the corruption and provide temporal
data, which is useful in problem determination.
HKWD_LIB_MALL_INTERNAL (hook ID: 60b)
When tracing is enabled for HKWD_LIB_MALL_INTERNAL and corruption is
detected, information about the internal data structures are logged through
the trace subsystem.
To use the trace option, do the following:
1. Start the trace subsystem before executing the target application. If you use
the command line interface, type the following command as the root user:
# trace -j’60a,60b’ -a
If you use SMIT, run smit -C trace and select START Trace, and then select
the hook keywords 60a and 60b in the high-lighted field in Example 4-7.
Example 4-7 Start trace
START Trace
Type or select values in entry fields.
Press Enter AFTER making all desired changes.
[Entry Fields]
EVENT GROUPS to trace [] +
ADDITIONAL event IDs to trace [] +
Event Groups to EXCLUDE from trace [] +
Event IDs to EXCLUDE from trace [] +
Trace MODE [alternate] +
STOP when log file full? [no] +
LOG FILE [/var/adm/ras/trcfile]
SAVE PREVIOUS log file? [no] +
Omit PS/NM/LOCK HEADER to log file? [yes] +
Omit DATE-SYSTEM HEADER to log file? [no] +
Run in INTERACTIVE mode? [no] +
Trace BUFFER SIZE in bytes [131072] #
LOG FILE SIZE in bytes [1310720] #
Buffer Allocation [automatic] +
F1=Help F2=Refresh F3=Cancel F4=List
194
Developing and Porting C and C++ Applications on AIX

F5=Reset F6=Command F7=Edit F8=Image
F9=Shell F10=Exit Enter=Do
2. Run the target application. You should remember the process ID of the
application.
3. Stop the trace subsystem. If you use the command line interface, type the
following command as the root user:
# trcstop
If you use SMIT, run smit -C trace and select STOP Trace.
4. Generate a trace report. If you use the command line, type the following
command as the root user:
# trcrpt -O exec=y -O pid=y -O tid=y -O svc=y -O timestamp=1
If you use SMIT, run smit -C trace and select Generate a Trace Report.
Example 4-8 Trace output of the trace option
Fri Feb 28 14:29:20 2003
System: AIX 5.2 Node: murumuru
Machine: 000C91AD4C00
Internet Address: 0903046A 9.3.4.106
The system contains 2 cpus, of which 2 were traced.
Buffering: Kernel Heap
This is from a 32-bit kernel.
Tracing only these hooks, 60a,60b
/usr/bin/trace -j60a 60b -a
ID PROCESS NAME I SYSTEM CALL ELAPSED APPL SYSCALL KERNEL INTERRUPT
001 --1- 0.000000 TRACE ON channel 0
Fri Feb 28 14:29:20 2003
60A --1- 5.561069 HKWD_LIBC_MALL_COMMON
function=malloc() [Default Allocator]
size=000A
returnptr=20000728
60A --1- 5.561261 HKWD_LIBC_MALL_COMMON
function=free() [Default Allocator]
inptr=20000728
60A --1- 10.131345 HKWD_LIBC_MALL_COMMON
function=malloc() [Default Allocator]
size=0290
returnptr=20005878
Chapter 4. Managing the memory heap
195

log
The log option instructs the malloc subsystem to record information on the
number of active allocations of a given size and stack trace back of a user
program. This data can be used in problem determination and performance
analysis, if the user program is modified accordingly.
Data recorded with the log option
If the log option is enabled, the following data is recorded for each malloc or
realloc subroutine invocation:
The size of the allocation.
The stack trace back of the invocation. The depth of the trace back that is
recorded is a configurable option.
The number of currently active allocations that match the size and stack trace
back.
The data is stored into the following global structure:5
struct malloc_log *malloc_log_table;
#ifndef MALLOC_LOG_STACKDEPTH
#define MALLOC_LOG_STACKDEPTH 4
#endif
struct malloc_log {
size_t size;
size_t cnt;
uintptr_t callers [MALLOC_LOG_STACKDEPTH];
}
size_t malloc_log_size;
The size of the malloc_log structure can change. If the default call-stack depth is
greater than 4, the structure will have a larger size. The current size of the
malloc_log structure is stored in the globally exported malloc_log_size variable.
A user can define the MALLOC_LOG_STACKDEPTH macro to the stack depth
that was configured at process start time.
The malloc_log_table can be accessed in the following ways:
Using the get_malloc_log() sub-routine as follows:
#include <malloc.h>
size_t get_malloc_log (void *addr,void *buf,size_t bufsize);
This function copies the data from malloc_log_table into the provided buffer.
The data can then be accessed without modifying the malloc_log_table. The
data represents a snapshot of the malloc log data for that moment of time.
5 This is included in the /usr/include/malloc.h header file.
196
Developing and Porting C and C++ Applications on AIX

Using the get_malloc_log_live() sub-routine as follows:
#include <malloc.h>
struct malloc_log *get_malloc_log_live (void *addr);
The advantage of this method is that no data needs to be copied, therefore
performance suffers less. Disadvantages of this method are that the data
referenced is volatile and the data may not encompass the entire malloc
subsystem, depending on which malloc algorithm is being used.
To clear all existing data from the malloc log tables, use the reset_malloc_log()
subroutine as follows:
#include malloc.h
void reset_malloc_log(void *addr);
The sample program shown in Example 4-9 on page 198 demonstrates the use
of the get_malloc_log_live() sub-routine. After it is compiled, the program would
print the following output, if the log option is enabled:
$ cc get_malloc_live.c
$ export MALLOCTYPE=
$ export MALLOCDEBUG=log
$ a.out
i is 1217.
The size of the allocations is 8.
The number of matching allocations is 1.
i is 1241.
The size of the allocations is 16.
The number of matching allocations is 1.
i is 3293.
The size of the allocations is 8.
The number of matching allocations is 1.
Note: Do not set the MALLOCTYPE=log environment variable before invoking
the compiler, because it does not support this environment variable. If set, it
prints the following message:
cc: 1501-230 Internal compiler error; please contact your Service
Representative.
The program has a for loop to dump the content of log_ptr, if there are referenced
slots in the data structure referenced by log_ptr.
Chapter 4. Managing the memory heap
197

Example 4-9 get_malloc_log_live.c
#include <stdio.h>
#include <stdlib.h>
#include <errno.h>
#include <malloc.h>
int main(int argc, char *argv[])
{
char *ptr1 = (char *)NULL, *ptr2 = (char *)NULL, *ptr3 = (char *)NULL;
int i;
struct malloc_log *log_ptr;
ptr1 = (char *)malloc(8);
ptr2 = (char *)malloc(16);
ptr3 = (char *)malloc(8);
log_ptr = get_malloc_log_live(ptr1);
for (i = 0; i < DEFAULT_RECS_PER_HEAP; i++) {
if (log_ptr->cnt > 0) {
printf("i is %d.\n", i);
printf("The size of the allocations is %d.\n", log_ptr->size);
printf("The number of matching allocations is %d.\n"
, log_ptr->cnt);
}
log_ptr++;
}
exit(0);
}
Enabling malloc log
The log option is not enabled by default, but can be enabled and configured prior
to process startup by setting the MALLOCDEBUG environment variable as
follows:
MALLOCDEBUG=log
To enable the trace option with user-specified configuration options, set the
MALLOCDEBUG environment variable as follows:
MALLOCDEBUG=log:records_per_heap:stack_depth
Note: The records_per_heap and stack_depth parameters must be specified
in order. Leaving a value blank will result in setting that parameter to the
default value.
198
Developing and Porting C and C++ Applications on AIX

The predefined MALLOCDEBUG configuration options include the following:
records_per_heap
Used to specify the number of malloc log records that are
stored for each heap. This parameter affects the amount
of memory that is used by malloc log. The default value is
4096, the maximum value is 65535.
stack_depth
Used to specify the depth of the function-call stack that is
recorded for each allocation. This parameter affects the
amount of memory used by malloc log, as well as the size
of the malloc_log structure. The default value is 4, the
maximum is 32.
Limitations
The performance of all programs can degrade when the trace option is enabled,
due to the cost of storing data to memory. Memory usage will also increase.
4.4 Heap management using MEMDBG
The memdbg package provided by IBM C and C++ compiler products provides
several useful features, including the user-created heap (explained in this
section), to deal with managing the process heap. The user-created heap can be
used in your application source codes in order to replace the malloc subsystem
calls, such as malloc() and calloc(). It could be seen as a wrapper to the interface
to the malloc subsystem and shmat services.
If applications are modified to use the user-created heap, then they can use not
only the default process heap, but also the shared memory segments for memory
allocation requests.
The user-defined memory allocation functions are provided by the libhu.a library
and defined in the header file umalloc.h. You can either handle user-created
heaps or the default process heap with the following functions:
_ucalloc()
_umalloc()
_uheapmin()
As shown in Example 4-10 on page 200, the following filesets must be installed
on the AIX 5L Version 5.2 system in order to use these functions:
memdbg.adt
memdbg.aix50.adt
memdbg.msg.en_US
Chapter 4. Managing the memory heap
199

Example 4-10 /usr/lib/libhm.a and memdgb.* package
# ls -l /usr/lib/libhu.a
lrwxrwxrwx 1 bin bin 20 Jan 29 19:47 /usr/lib/libhu.a@ ->
/usr/vac/lib/libhu.a*
# lslpp -w /usr/lib/libhu.a
File Fileset Type
----------------------------------------------------------------------------
/usr/lib/libhu.a memdbg.adt Symlink
# lslpp -w /usr/vac/lib/libhu.a
File Fileset Type
----------------------------------------------------------------------------
/usr/vac/lib/libhu.a memdbg.aix50.adt Symlink
# lslpp -L memdbg.*
Fileset Level State Type Description (Uninstaller)
----------------------------------------------------------------------------
memdbg.adt 4.4.3.0 C F User Heap/Memory Debug Toolkit
memdbg.aix50.adt 4.4.3.0 C F User Heap/Memory Debug Toolkit
for AIX 5.0
memdbg.msg.en_US 4.4.3.0 C F User Heap/Memory Debug
Messages--U.S. English
Note: Although these sub-routines are convenient to use, the following points
must be understood before using these sub-routines:
Application source codes must be modified to call these routines.
These subroutines are not considered as a user-created malloc
replacement (see 4.2.5, “User-defined malloc replacement” on page 178).
Therefore, the user processes still require the default process heap for the
memory allocation requests made by functions in shared libraries, even if
you have specified to use shared memory segments for the user-created
heap.
More further information about these subroutines, please refer to Chapter 4,
“Using Memory Heaps”, in VisualAge C++ Professional for AIX Programming
Tasks and Library Reference, SC09-4963.
4.4.1 How to handle a user-created heap
To use a user-created heap, there are some basic tasks, as summarized in the
following:
1. Allocate memory chunks to be used for the user-created heap, depending on
the type of memory source:
– Call the system malloc() sub-routine to request memory from the process
heap.
200
Developing and Porting C and C++ Applications on AIX

– Call the shmget() and shmat() sub-routines to request from the shared
memory segments.
2. Call the _ucreate() sub-routine to create a handle to refer to the allocated
memory. The created handle will be used to refer to the allocated memory
pool in the subsequent _umalloc() subroutine calls.
3. Use the user-created heap.
– Call the _umalloc() subroutine with the handle instead of the system
malloc() subroutine.
– Call the _ucalloc() subroutine with the handle instead of the system
calloc() subroutine.
– Call the _ufree() subroutine with the handle instead of the system free()
subroutine.
Note: For any memory object allocated by _umalloc() or _ucalloc(), you
can find out from what heap it was allocated by calling the _mheap()
sub-routine.
4. Release the user heap memory before the process exit.
a. Deallocate the all acquired memory within the user-created heap by the
_ufree() function
b. Destroy the user-created heap with the _udestroy() function with the
handle.
c. Deallocate the memory chunks allocated in the first step depending on the
type of memory source:

Call the system free() sub-routine to free memory from the process
heap.

Call the shmdt() and shmctl() sub-routines to remove the allocated
shared memory segments from the process address space.
Managing the user-created heap size
Once a user-created heap is fully utilized, further allocation requests from that
heap will fail, unless you have added more memory chunks using the
_uaddmem() sub-routine. So make sure to allocate a block of memory large
enough to satisfy all the memory requests in your program.
Changing the default heap source
The use of _umalloc() and _ucalloc() does not conflict with the malloc subsystem
subroutines, such as malloc() and calloc(). If malloc() is called within a user
program, then the default process heap would be used to satisfy the memory
Chapter 4. Managing the memory heap
201

request, even if the program has allocated a user-created heap in the shared
memory segments.
However, if the _udefault() sub-routine with a handle parameter that points to the
shared memory segments is called within a user program, then subsequent calls
to malloc() and calloc() would start use of the user-created heap.
Note: The default heap source can be changed only when _udefault() is
called.
The default heap source changes only for the Pthread that has called _udefault()
within a process. You can use a different default heap for each thread of your
multi-threaded application program.
The _udefault() function call returns the current default heap, so that the value
can be saved and be used later to restore the default heap source. The default
process heap is specified by the _RUNTIME_HEAP6 macro.
Gathering statistical data about user-created heaps
The _ustats() sub-routine reports the following statistical data about the specified
user-created heap:
How much memory the heap holds (excluding memory used for overhead)?
How much memory is currently allocated from the heap?
What type of memory is in the heap?
The size of the largest contiguous piece of memory available from the heap?
4.4.2 A user-defined heap allocated from shared memory segments
An sample program shown in Example 4-12 on page 203 demonstrates how to
use a user-defined heap. The program allocate 512 MB memory in shared
memory segments, then calls the following sub-routines in turn:
_ucreate()
Creates a handle to refer to the user-created heap.
_umalloc()
Allocates memory size of 100000 bytes from the
user-created heap.
_ustats()
Prints the usage of the user-created heap.
_ufree()
Un-allocates the previously created memory object using
_umalloc() from the user-created heap.
_udestroy()
Destroys the user-create heap.
6 The _RUNTIME_HEAP macro is defined in the umalloc.h header file.
202
Developing and Porting C and C++ Applications on AIX

To compile this program, do the following:
$ cc user_heap.c -lhu
When executed, it prints the output shown in Example 4-11. It allocated a
memory chunk size of 100016 bytes (16 bytes are used for overhead to maintain
the statistics data) out of shared memory segments size of 512 MB.
Example 4-11 Output from _ustats()
$ a.out
----------------------------------------------------------
Heap information from _ustats():
..heapstat._provided = 536870384
..heapstat._used = 0
..heapstat._max_free = 536870384
----------------------------------------------------------
----------------------------------------------------------
Heap information from _ustats():
..heapstat._provided = 536870384
..heapstat._used = 100016
..heapstat._max_free = 536770368
----------------------------------------------------------
----------------------------------------------------------
Heap information from _ustats():
..heapstat._provided = 536870384
..heapstat._used = 0
..heapstat._max_free = 536870384
----------------------------------------------------------
Example 4-12 user_heap.c
#include <stdlib.h>
#include <stdio.h>
#include <errno.h>
#include <umalloc.h>
#include <sys/shm.h>
#define TWOSEGS 0x20000000
void heapstate(Heap_t usrheap)
{
_HEAPSTATS heapstat;
_ustats(usrheap, &heapstat);
printf("----------------------------------------------------------\n");
printf("Heap information from _ustats():\n");
printf("..heapstat._provided\t\t= %10u\n", heapstat._provided);
printf("..heapstat._used\t\t= %10u\n", heapstat._used);
printf("..heapstat._max_free\t\t= %10u\n", heapstat._max_free);
Chapter 4. Managing the memory heap
203

printf("----------------------------------------------------------\n");
return;
}
int main(void)
{
int shmid, rc;
char buf[BUFSIZ];
char *ptr1, *ptr2, *shmptr;
Heap_t myheap;
if ((shmid = shmget(IPC_PRIVATE, TWOSEGS, IPC_CREAT|S_IRUSR|S_IWUSR))
< 0) {
sprintf(buf, "shmget() failed at %d in %s with errno = %d"
, __LINE__, __FILE__, errno);
perror(buf);
exit(1);
}
if ((shmptr = shmat(shmid, (void *)0, 0)) == (void *) -1) {
sprintf(buf, "shmat() failed at %d in %s with errno = %d"
, __LINE__, __FILE__, errno);
perror(buf);
exit(1);
}
/* create a user-created heap. */
if ((myheap = _ucreate(shmptr, TWOSEGS, _BLOCK_CLEAN
, _HEAP_SHARED|_HEAP_REGULAR, NULL, NULL)) == (void *)NULL) {
sprintf(buf, "_ucreate() failed at %d in %s with errno = %d"
, __LINE__, __FILE__, errno);
perror(buf);
exit(1);
}
heapstate(myheap);
ptr1 = _umalloc(myheap, 100000); /*allocate from user heap. */
heapstate(myheap);
_ufree(ptr1);
heapstate(myheap);
/* destroy user heap. */
if (_udestroy(myheap, _FORCE)) {
sprintf(buf, "_destroy() failed at %d in %s with errno = %d"
, __LINE__, __FILE__, errno);
perror(buf);
exit(1);
}
if ((rc == shmdt(shmptr)) == -1) {
sprintf(buf, "shmdt() at %d in %s failed with errno = %d"
, __LINE__, __FILE__, errno);
204
Developing and Porting C and C++ Applications on AIX

perror(buf);
}
if (shmctl(shmid, IPC_RMID, 0) < 0) {
sprintf(buf, "shmctl() at %d in %s failed with errno = %d"
, __LINE__, __FILE__, errno);
perror(buf);
}
exit(0);
}
Chapter 4. Managing the memory heap
205

206
Developing and Porting C and C++ Applications on AIX

5
Creating DLPAR-aware
Chapter 5.
applications
This chapter provides a brief overview for understanding the new programming
paradigm available in a partitioned environment where resources can be
dynamically changed/triggered by DLPAR operations. We discuss the following
topics:
Section 5.1, “Dynamic logical partitioning overview” on page 208
Section 5.2, “The process flow of a DLPAR operation” on page 210
Section 5.3, “DLPAR-safe and DLPAR-aware applications” on page 214
Section 5.4, “Integrating a DLPAR operation into the application” on page 217
For further detailed information about the dynamic logical partitioning and the
new programming paradigm, please refer the following publications:
The Complete Partitioning Guide for IBM ^ pSeries Servers,
SG24-7039
IBM Hardware Management Console for pSeries Installation and Operations
Guide, SA38-0590
© Copyright IBM Corp. 2000, 2003. All rights reserved.
207

5.1 Dynamic logical partitioning overview
DLPAR, or dynamic logical partitioning, supports the following dynamic resource
change in a partition without requiring a partition reboot:
Resource addition
Resource removal
By achieving the resource changes sequentially in the following order on two
partitions in a system, the specified resource can be moved from a partition to
another partition:
1. Resource removal from a partition
2. Resource addition to another partition
This resource movement is implemented as single task on the IBM Hardware
Management Console for pSeries (hereafter referred to as HMC), although it is
actually composed of two separate tasks on two partitions internally.
A resource is either of the following types:
CPU
The granularity of a CPU resource of a DLPAR operation is one CPU. More
than one CPU can be specified as a resource of a DLPAR operation.
A partition must be assigned at least the minimum number of processors
specified in the partition profile, and it can be assigned up to the maximum
number of processors specified in the partition profile.
Therefore, you can dynamically add or remove processors for that partition
within the range of the minimum and maximum values.
Memory
The granularity of a memory resource of a DLPAR operation is 256 MB1.
Multiplies of 256 MB memory can be specified as a resource of a DLPAR
operation.
A partition must be assigned at least the minimum size of memory specified in
the partition profile, and it can be assigned up to the maximum size of
memory specified in the partition profile.
Therefore, you can dynamically add or remove memory for that partition
within the range of the minimum and maximum values.
1 This memory chunk is referred to as a logical memory block (LMB).
208
Developing and Porting C and C++ Applications on AIX

I/O resource
The granularity of an I/O resource of a DLPAR operation is a PCI slot with a
PCI adapter. Multiple I/O slots can be specified as a resource of a DLPAR
operation. If a PCI adapter has multiple ports, all the ports and devices
configured beneath the ports are treated as a resource.
For example, if a 10/100 4-Port Ethernet adapter (FC 4961) is selected, both
Ethernet devices (entX0) and interfaces (enX) configured on this adapter are
treated as a single resource.
A partition must be assigned all the adapters specified as required in the
partition profile, and it can be assigned adapters specified as desired in the
partition profile. Therefore, you can dynamically add or remove only adapters
specified as desired for that partition.
You cannot remove I/O slots listed as required; however, you can remove I/O
slots listed as desired, or those that were added as a result of a DLPAR
operation. In other words, a partition can currently contain an I/O slot that is
not listed as either desired or required in the active partition profile.
Note: A DLPAR operation can perform only one type of resource change. You
cannot add and remove memory to and from the same partition in a single
DLPAR operation. Also, you cannot move CPU and memory from a partition to
another partition in a single DLPAR operation.
Resources removed from a partition are marked free (free resources) and owned
by the global firmware of system; you can consider these resources as kept in
the “free resource pool.” Free resources can be added to any partition in a
system as long as the system has enough free resources.
It is imperative to understand that the DLPAR function is not solely provided by
AIX 5L Version 5.2, but it is supported by the integration of following components:
Hardware
A partitioning-capable pSeries server model is required.
Firmware
Depending on the models you have selected, a firmware update might be
required.
HMC
HMC software Release 3 Version 1 or later is required.
Operating system
AIX 5L Version 5.2 or later is required.
Chapter 5. Creating DLPAR-aware applications
209

If one of these components does not satisfy the requirement to support DLPAR,
the function is not available. For example, if a partition is installed with AIX 5L
Version 5.1, that partition does not support DLPAR, although the other partitions
installed with AIX 5L Version 5.2 support DLPAR.
5.2 The process flow of a DLPAR operation
A DLPAR operation initiated on the HMC is transferred to the target partition
through Resource Monitoring and Control (RMC). The request produces a
DLPAR event on the partition. After the event has completed, regardless of the
result from the event, a notification will be returned to the HMC in order to mark
the completion of the DLPAR operation. This means that a DLPAR operation is
considered as a single transactional unit; thus, only one DLPAR operation is
performed at a time.
210
Developing and Porting C and C++ Applications on AIX

A DLPAR operation is executed in the process flow, as illustrated in Figure 5-1.
B: DLPAR operation request via RMC from the HMC
Ethernet
C: DLPAR operation result from the partition
AIX 5L Version
RMC
5.2 Partition
IBM.DRM
GUI or command
HMC
drmgr
Platform-dependent
commands
A: Resource query and allocate
requests to the CSP before
the DLPAR operation over
Platform-dependent
Kernel
device driver
the serial line
D: Resource reclaim request
RTAS
to the CSP after the DLPAR
operation over the serial line
Global firmware / Hypervisor
CSP
Serial Line
Managed system
Figure 5-1 Process flow of a DLPAR operation2
The following steps explain the process flow of a DLPAR operation:
1. The system administrator initiates a DLPAR operation request on the HMC
using either the graphical user interface or command line interface.
2. The requested DLPAR operation is verified on the HMC with the current
resource assignment to the partition and free resources on the managed
system before being transferred to the target partition. In other words, the
HMC provides the policy that determines whether or not a DLPAR operation
request is actually performed on the managed system.
3. If the request is a resource addition, the HMC communicates with the global
firmware in order to allocate free resources to the target partition through the
service processor indicated as arrow A in Figure 5-1.
2 A managed system is a partitioning partitioning-capable pSeries server system managed by the HMC.
Chapter 5. Creating DLPAR-aware applications
211

If enough free resources exist on the system, the HMC assigns the requested
resource to the specified partition and updates the partition’s object to reflect
this addition, and then creates associations between the partition and the
resource to be added.
4. After the requested DLPAR operation has been verified on the HMC, it will be
transferred to the target partition using Resource Monitoring and Controlling
(RMC), which is an infrastructure implemented on both the HMC and AIX
partitions, as indicated as arrow B in Figure 5-1 on page 211. The RMC is
used to provide a secure and reliable connection channel between the HMC
and the partitions.
Note: The connection channel established by RMC only exists between
the HMC and the partition where the DLPAR operation is targeted to. There
are no connection paths required between partitions for DLPAR operation
purposes.
5. The request is delivered to the IBM.DRM resource manager running on the
partition, which is in charge of the dynamic logical partitioning function in the
RMC infrastructure in AIX. As shown in the following example, it is running as
the IBM.DRMd daemon process and is included in the devices.chrp.base.rte
fileset on AIX 5L Version 5.2 or later:
# lssrc -ls IBM.DRM
Subsystem : IBM.DRM
PID : 18758
Cluster Name : IW
Node Number : 1
Daemon start time : Wed Aug 21 16:44:12 CDT 2002
Information from malloc about memory use:
Total Space : 0x003502c0 (3474112)
Allocated Space: 0x0030b168 (3191144)
Unused Space : 0x00043e40 (278080)
Freeable Space : 0x00000000 (0)
Class Name(Id) : IBM.DRM(0x2b) Bound
# ps -ef | head -1 ; ps -ef | grep DRMd | grep -v grep
UID PID PPID C STIME TTY TIME CMD
root 18758 10444 0 Aug 21 - 0:22 /usr/sbin/rsct/bin/IBM.DRMd
# lslpp -w /usr/sbin/rsct/bin/IBM.DRMd
File Fileset Type
---------------------------------------------------------------------------
/usr/sbin/rsct/bin/IBM.DRMd
devices.chrp.base.rte File
212
Developing and Porting C and C++ Applications on AIX

Note: The absence of the IBM.DRM resource manager in the lssrc -a
output does not always mean that the partition has not been configured
appropriately for the dynamic logical partitioning. The resource manager
will be automatically configured and started by RMC after the first partition
reboot if the network configuration is correctly set up on the partition and
the HMC.
Resource managers are sub-systems used in the RMC infrastructure. For
further information about RMC and its sub-components, please refer to the
following publications:
A Practical Guide for Resource Monitoring and Control (RMC),
SG24-6615
IBM Reliable Scalable Cluster Technology for AIX 5L: Administration
Guide, SA22-7889
6. The IBM.DRM resource manager invokes the drmgr command, which is an
platform-independent command designed as the focal point of the dynamic
logical partitioning support on AIX.
As shown in the following example, the drmgr command is installed in the
/usr/sbin directory provided by the bos.rte.methods fileset:
# whence drmgr
/usr/sbin/drmgr
# lslpp -w /usr/sbin/drmgr
File Fileset Type

---------------------------------------------------------------------------
/usr/sbin/drmgr bos.rte.methods File
Note: The drmgr command should not be invoked by the system
administrator in order to directly perform resource changes in a partition. It
must be invoked in the context explained here to do so.
7. The drmgr command invokes several platform-dependent commands3
depending on the resource type (CPU, memory, or I/O resource) and request
(resource addition or removal) in order to instruct the kernel to process the
actual resource change with necessary information.
8. The kernel does many internal tasks.
9. After the DLPAR event has completed, regardless of the result, a notification
will be returned to the HMC in order to mark the completion of the DLPAR
3 On the current partitioning-capable pSeries server models, the platform-dependent commands are
included in the bos.chrp.base.rte fileset and installed in the /usr/lib/boot/bin directory.
Chapter 5. Creating DLPAR-aware applications
213

operation, indicated as arrow C in Figure 5-1 on page 211. The notification
also includes the exit code, standard out, and standard error from the drmgr
command. The system administrator who has initiated the DLPAR operation
will see the exit code and outputs on the HMC.
10.If the request is a resource removal, the HMC communicates with the global
firmware in order to reclaim resources to the free resource pool from the
source partition through the service processor, indicated as arrow D in
Figure 5-1 on page 211.
The HMC unassigns the resource from the partition and updates the
partition’s object to reflect this removal, and then removes associations
between the partition and the resource that was just removed.
A DLPAR operation can take a considerable amount of time, depending on the
availability and the capability to configure or deconfigure a specific resource.
5.3 DLPAR-safe and DLPAR-aware applications
The dynamic logical partitioning function on AIX 5L Version 5.2 is designed and
implemented to not impact the existing applications. In fact, most applications are
not affected by any DLPAR operations results. Therefore, those applications are
called DLPAR-safe applications.
There are two types of application classifications regarding DLPAR operations:
DLPAR-safe
Applications that do not fail as a result of DLPAR
operations. The application’s performance may suffer
when resources are removed, or it may not scale as
resources are added.
DLPAR-aware
Applications that incorporate DLPAR operations that allow
the application to adjust its use of the system resources
equal to the actual capacity of the system. DLPAR-aware
applications are always DLPAR-safe.
5.3.1 DLPAR-safe
Although, most applications are DLPAR-safe without requiring any modification,
there are certain instances where programs may not be inherently DLPAR-safe.
There are two cases where DLPAR operations may introduce undesirable effects
into the application:
Programs that are optimized for uni-processor may have problems when a
processor is added to the system resources.
214
Developing and Porting C and C++ Applications on AIX

On programs that are indexed by CPU numbers, the increased processor
number may cause the code to go down an unexpected code path during its
run-time checks.
In addition, applications that use uni-processor serialization techniques may
experience unexpected problems. In order to resolve these concerns, system
administrators and application developers need to be aware of how their
applications get the number of processors.
5.3.2 DLPAR-aware
DLPAR-aware applications adapt to system resource changes caused by DLPAR
operations. When these operations occur, the application will recognize the
resource change and accommodate accordingly.
Two techniques can be used to make applications DLPAR-aware:
The first method is to consistently poll for system resource changes. Polling is
not the recommended way to accommodate for DLPAR operations, but it is
valid for systems that do not need to be tightly integrated with DLPAR.
Because the resource changes may not be immediately discovered, an
application that uses polling may have limited performance. Polling is not
suitable for applications that deploy processor bindings, because they
represent hard dependencies.
Applications have other methods to react to the resource change caused by
DLPAR operations. See 5.4, “Integrating a DLPAR operation into the
application” on page 217.
Several applications should be made DLPAR-aware, because they need to scale
with the system resources. These types of applications can increase their
performance by becoming DLPAR-aware. Table 5-1 on page 216 lists some
examples of applications that should be made DLPAR-aware.
Note: These are only a few types of common applications affected by DLPAR
operations. The system administrator and application developer should be
sensitive to other types of programs that may need to scale with resource
changes.
Chapter 5. Creating DLPAR-aware applications
215

Table 5-1 Applications that should be DLPAR-aware
Application type
Reason
Database
The application needs to scale with the system. For example,
applications
the number of threads may need to scale with the number of
available processors, or the number of large pinned buffers
may need to scale with the available system memory.
License Managers
Licenses are distributed based on the number of available
processors or the memory capacity.
Workload Managers
Jobs are scheduled based on system resources, such as
available processors and memory.
Tools
Certain tools may report processor and memory statistics or
rely on available resources.
Applications should also be made DLPAR-aware, if they cause DLPAR requests
to fail.
The following application specific conditions can cause DLPAR remove requests
to fail:
Processor bindings
Large amounts of pinned memory
Processor bindings and attachments are obstacles to CPU removal, since they
represent work units that cannot be rescheduled on other processors.
DLPAR-aware programs remove these obstacles by unbinding, re-binding, or
terminating before the operating system attempts to remove the resource. AIX
provides several interfaces for binding and attaching to processors including
bindprocessor(), ra_exec(), ra_fork(), and ra_attachrset(). If the bindprocessor()
interface is used, then the target processor is always the Nth online processor.
The other interfaces target logical processors, so there is a direct correlation
between the processor being removed and the attachment, which is not
necessarily the case with bindprocessor(). It should be noted that
bindprocessor() bindings may also be resolved through application handlers that
may have already been provided for dynamic processor deallocation (for
example, CPU Guard).
Applications that pin large amounts of memory may cause DLPAR memory
removal failures, since they impede the system's capability to pin new memory
pages, which is a critical component of the DLPAR support for memory. AIX has
the capability to migrate pinned memory, but it must ensure that the target of the
migration is also pinned, so it must be able to pin memory on demand to satisfy
DLPAR memory remove requests. DLPAR-aware applications that pin lots of
memory should resize their buffers so that portions of them can be unpinned,
216
Developing and Porting C and C++ Applications on AIX

before the operating system attempts to remove memory. Applications pin
memory through the plock() and shmget(SHM_PIN) interfaces.
Applications that have processor attachments and pin memory are DLPAR-safe
and they should be made DLPAR-aware.
Applications are not impacted by the dynamic reconfiguration of I/O resources,
because access to the device is controlled through the device support layer,
which is not manipulated by DLPAR. The system administrator is expected to
unload device drivers before initiating DLPAR remove requests and, in a similar
vein, he is expected to manually configure them after performing DLPAR add
requests to make them available to applications. The DLPAR support for I/O
resources is therefore DLPAR-safe. For example, DLPAR I/O remove requests
will fail If a device driver that is associated with the I/O resource is loaded. The
device driver does not have to be open for the DLPAR remove operation to fail. It
simply has to be loaded.
5.4 Integrating a DLPAR operation into the application
The DLPAR operation can be integrated into the application using the following
two methods:
Script-based DLPAR event handling
If the application is externally controlled to use a specific number of threads or
to size its buffers, use this method. In order to facilitate this method, a new
command, drmgr, is provided. The drmgr command is the central focal point of
the dynamic logical partitioning function of AIX.
API-based DLPAR event handling
If the application is directly aware of the system configuration, and the
application source code is available, use this method.
Applications can monitor and respond to various DLPAR events, such as a
memory addition or processor removal, by utilizing these two methods. Although,
at the high-level, both methods share the same DLPAR events flow explained in
the following section, several key differences exist between these two methods.
One difference is that the script-based method externally reconfigures the
application once a DLPAR event takes place, while the API-based method can be
directly integrated into the application by registering a signal handler using the
dr_reconfig system call, so that the process can be notified with the
SIGRECONFIG signal when the DLPAR event occurs.
Chapter 5. Creating DLPAR-aware applications
217

Note: The DLPAR events of I/O resources are not notified applications in AIX
5L Version 5.2.
218
Developing and Porting C and C++ Applications on AIX

6
Programming hints and tips
Chapter 6.
In developing your application with C or C++, you may encounter problems while
compiling, linking, or executing for the first time.
This chapter starts off by offering some general programming tips that would
make an application perform better on AIX:
Section 6.1, “Programming recommendations” on page 220
The next sections continue to describe some of the more common errors that you
may encounter during the different phases of the development cycle:
Section 6.2, “Diagnosing compile-time errors” on page 227
Section 6.3, “Diagnosing link-time errors” on page 239
Section 6.4, “Diagnosing run-time errors” on page 244
All the options and #pragma directives referred to in this chapter are explained in
more details in the C for AIX Compiler Reference, SC09-4960 and VisualAge
C++ for AIX Compiler Reference, SC09-4959.
© Copyright IBM Corp. 2000, 2003. All rights reserved.
219

6.1 Programming recommendations
C and C++ are very powerful high level programming languages. Two
applications can be performing the same tasks, but written in completely different
ways using the many different features in the language. Programming style is
very personal, but there are programming practices that, when followed, will help
get the best results from the optimization techniques used by the compiler. This
section offers some tips and guidelines that will allow you to write programs that
exploit the C for AIX and VisualAge C++ for AIX compilers.
6.1.1 Variables and data structures
In C and C++, the default storage duration, scope, and linkage of variables and
objects depend on where they are declared. However, storage duration can be
explicitly overridden with storage class specifiers. Whenever possible, it is
recommended that you use local variables of the automatic storage class.
Several optimizations performed by the compiler rely on data flow analysis. For
example, whether a store into a variable is redundant and can be removed,
because a later store will invalidate the first store, before the variable is
referenced:
int func1()
{
int x;
x = 1;
func();
x = 2;
return x;
}
In this example, the first assignment, x = 1, can be safely removed, since the
second assignment will overwrite it, and the call to function func() will not access
variable x. Had variable x been declared as global, the compiler would not have
made the same assumption, since it is now possible for the function func() to be
referencing the global variable x.
Data flow analysis also helps in determining whether branches can be
eliminated. For example:
void func2()
{
int x;
x = 1;
func();
if (x == 1)
printf("true\n");
220
Developing and Porting C and C++ Applications on AIX

else
printf("false\n");
}
Again, since the function call to func() cannot possibly be updating the local
variable x, the compiler can safely remove the conditional statement, and
generate code for the true part of the statement only.
Using local, automatic storage variables is not always possible. If two functions
within the same source file need to share data, it is recommended that you use
variables with static storage. Static storage class variables and objects have
internal linkage, that is, the variable is accessible within the compilation unit only,
and not visible externally. The compiler in this case can use this information to
better optimize your program, since static variables appear as local within the
compilation unit.
If you must use external variables that are shared between more than one
compilation units, try to group the variables into structures to avoid excessive
memory access.
6.1.2 Functions
Without knowing what a function does, that is, if the function is not defined in the
current compilation unit, the compiler has to assume the worse, that calling the
function may have side effects. A side effect, according to the standard, is a
change in the state of the execution environment. Several operations can
produce a side effect:
Accessing a volatile object
Modifying an external object
Modifying a static object
Modifying a file
Calling a function that does any of the above
If a function requires input, it is recommended that you pass the input as
arguments, rather than having the function access the input from global
variables. If your function has no side effects, that is, it does not violate any of the
points above, you can use the #pragma isolated_call directive to specify that the
compiler may take a more aggressive approach in optimization. This generally
improves the execution performance. See C for AIX Compiler Reference,
SC09-4960 for more details on the directive. Functions that are marked with the
#pragma isolated_call directive are allowed to modify storage pointed to by
pointer arguments (that is, reference arguments).
Chapter 6. Programming hints and tips
221

If a function is only required in the current compilation unit, declaring it as static
will speed up calls to the function.
In C++, use virtual functions only when absolutely necessary. The location of
virtual functions are not known until execution time. The compiler generates an
indirect call to virtual functions via entries in a virtual function table, which gets
populated with the location of the virtual functions at load time. To reduce code
bloat, avoid declaring virtual functions inline, which tends to make the virtual
function table bigger and causes the virtual function defined in all compilation
units that use the class.
6.1.3 Pointers
The use of pointers causes uncertainty in data flow analysis, which can
sometimes inhibit optimization performed by the compiler. If a pointer is assigned
to the address of a variable, both the pointer and the variable can now be used to
change or reference the same memory location in an expression. For example:
void func()
{
int i = 55, a, b;
int *p = &i;
a = *p;
i = 66;
b = *p;
}
At first glance, it looks as if the expression *p does not need to be evaluated the
second time, and the variable a and b would have the same value. However,
since p and the address of i refer to the same memory location, the assignment i
= 66 invalidates the first *p evaluation and b = *p must be evaluated again.
When the compiler knows nothing about where a pointer can point to, it will have
to make the pessimistic assumption that the pointer can point to any variable.1
For example:
int rc;
void foo(int *p)
{
rc = 55;
*p = *p + 1;
1 In the standard conforming language levels, e.g. -qlanglvl=stdc89 and -qlanglvl=stdc99, or when
the -qalias=ansi option is in effect, a pointer can only point to variables of the same type. This is
referred to as type-based aliasing. The only exception is with character pointers and pointers to void;
they can point to any type. The use of -qalias=ansi to correct programming mistakes is not
recommended, as it inhibits optimization opportunities and degrades execution performance.
222
Developing and Porting C and C++ Applications on AIX

rc = 66;
}
In this case, the compiler has to assume that when the function foo is called,
pointer p can potentially point to the variable rc. The first assignment, rc = 55, is
relevant and cannot be safely removed, since *p on the following line can be
referring to rc.
If p is guaranteed not to be pointing to rc, you can use the #pragma disjoint
directive to inform the compiler:
#pragma disjoint(*p, rc)
Marking variables that do not share physical memory storage with the #pragma
disjoint directive allows the compiler to explore more optimization opportunities,
and the performance gain can be quite substantial with complex applications.
However, if the directive is used inappropriately on variables that may share
physical storage, the program may yield incorrect results.
6.1.4 Arithmetic operations
Multiplication in general is faster than division. If you need to perform many
divisions with the same divisor, it is recommended that you assign the reciprocal
of the divisor to a temporary variable, and change all your divisions to
multiplications with the temporary variable. For example, instead of:
double preTax(double total)
{
return total / 1.0825;
}
this will perform faster:
double preTax(double total)
{
return total * (1.0 / 1.0825);
}
The division (1.0 / 1.0825) will be evaluated once at compile time only due to
constant folding.
6.1.5 Selection and iteration statements
When using the if selection statement, order your conditional expressions
efficiently by putting the most decisive test first. For example:
struct {
char *name;
short len;
Chapter 6. Programming hints and tips
223

_Bool active;
} rec;
...
if (rec.active == true && rec.len == 9 && !memcmp(rec.name, “Elizabeth”, 9))
...
Also, order your case statements or if-else if conditions in a way that the most
likely occurring conditions appear first. For example:
typedef enum { VOWEL, CONSONANT, DIGIT, PUNCTUATION } _Type;
struct {
char ch;
_Type type;
} word;
...
switch (word.type) {
case VOWEL:
...
break;
case CONSONANT:
...
break;
case PUNCTUATION:
...
break;
case DIGIT:
...
break;
}
and:
if (!error) {
/* most likely condition first */
} else {
/* error condition that does not happen too often */
}
When using iteration statements such as for loop, do loop, and while loop, move
invariant expressions (expressions that do not change with loop iterations, nor
depend on the loop index of a for loop) out of the loop body.
6.1.6 Expression
The C and C++ compilers are able to recognize common sub-expressions, when
the sub-expression either:
Appears at the left end of the expression
Within parentheses
224
Developing and Porting C and C++ Applications on AIX

For example, the compiler recognizes the sub-expression a + b in the two
assignments:
x = a + b + c;
y = d * (a + b);
and evaluates the two sub-expressions only once. Essentially, it is logically
transforming the two lines into:
temp = a + b;
x = temp + c;
y = d * temp;
The use of the temp illustrates how the compiler can keep the result of the
sub-expression evaluation in a register, and simply reuse the register for both
assignments.
6.1.7 Memory usage
Avoid heap storage fragmentation with frequent allocations of small, temporary
objects using memory allocation functions such as malloc() and calloc().
while (list) {
char *temp = (char*)malloc(list->len+1);
strcpy(temp, list->element);
PrettyPrint(temp);
free(temp);
list = list->next;
}
In the example above, if the size of the longest element is known beforehand, you
can replace the call to malloc() and obtain storage from the stack instead by
using an array of characters:
char temp[MAX_ELEMENT_SIZE];
If the size is unknown, you can make use of the C99 feature, “Variable length
arrays” on page 7 to allocate storage dynamically for the array:
char temp[list->len+1];
In either case, the storage is obtained from the stack and not from the heap, and
it is freed automatically when the scope of the variable terminates.
6.1.8 Built-in functions
For performance reasons, a large number of the library functions are also
provided as compiler built-ins. A compiler built-in avoids any overhead associated
with function calls (for example, parameter passing, stack allocation and
Chapter 6. Programming hints and tips
225

adjustment, and so on) by expanding the function body at the call site. Various
hardware instruction built-ins are also provided to allow programmers direct
access to hardware instructions.
To use the built-in version of library functions, you need to include the appropriate
library header files. Including the proper header files also prevents parameter
type mismatch and ensures optimal performance.
6.1.9 Virtual functions
In general, when writing C++ code, you should try and avoid the use of virtual
functions. They are normally encoded as indirect function calls, which are slower
than direct function calls.
Usually, you should not declare virtual functions inline. In most cases, declaring
virtual functions inline does not produce any performance advantages. Consider
the following code sample:
class Base {
public:
virtual void foo() { /* do something. */ }
};
class Derived: public Base {
public:
virtual void foo() { /* do something else. */ }
};
int main(int argc, char *argv[])
{
Base* b = new Derived();
b->foo(); // not inlined.
}
In this example, b->foo() is not inlined because it is not known until run time
which version of foo() must be called. This is by far the most common case.
There are cases, however, where you might actually benefit from inlining virtual
functions, for example, if Base::foo() was a really hot function, it would get inlined
in the following code:
int main(int argc, char *argv[])
{
Base b;
b.foo();
}
226
Developing and Porting C and C++ Applications on AIX

If there is a non-inline virtual function in the class, the compiler generates the
virtual function table in the first file that provides an implementation for a virtual
function; however, if all virtual functions in a class are inline, the virtual table and
virtual function bodies will be replicated in each compilation unit that uses the
class. The disadvantage to this is that the virtual function table and function
bodies are created with internal linkage in each compilation unit. This means that
even if the -qfuncsect option is used, the linker cannot remove the duplicated
table and function bodies from the final executable. This can result in a very
bloated executable size.
6.2 Diagnosing compile-time errors
During compilation, the compiler emits diagnostic messages to the standard
error device file (the terminal by default) whenever it encounters programming
errors caused by invalid syntax or incorrect usage of features. There are various
compiler diagnostic aids that can help you in determining where the error
occurred.
6.2.1 Anatomy of a message
A compiler error message has the following format, which provides enough
information about the location and reason for the error:
file”, line line.column: 15cc-nnn (sev) msg
Where:
file
The name of the source file where the error occurred.
line
The line in the source file where the error occurred.
column
The column on the line where the error occurred.
cc
A two-digit code identifying the compiler component that
issued the diagnostic message:
00
Optimizer or code generator
01
Compiler services
05
C compiler
06
C compiler
40
C++ compiler
47
Munch utility
86
Interprocedural analysis (IPA)
nnn
The message number.
Chapter 6. Programming hints and tips
227

sev
The severity of the error.
I
Informational
W
Warning
E
Error
S
Severe error
U
Unrecoverable error
msg
Short message describing the error.
6.2.2 Useful options and compiler aids
The C and C++ compilers offer several options that help you detect and correct
programming errors in you program.
The -qsrcmsg option
By default, compiler diagnostic messages are issued in the format depicted in
6.2.1, “Anatomy of a message” on page 227. Sometimes it may not be apparent
what the problem is with the short explanation in the message text. The C
compiler -qsrcmsg option prints the source line and a finger line pointing to
where the compiler thinks the error is to the stderr file, giving a more precise
explanation by showing the actual source line containing the error. For example:
7 | char new[n];
.................a..
a - 1506-195 (S) Integral constant expression with a value greater than zero is
required.
The source line shown will be after macro expansion.
Compiler listing
You can use the -qsource option to ask for a compiler listing. A compiler listing
contains several sections that are useful in determining what has gone wrong in
a compilation. Diagnostic messages, if present, are written to the compiler listing
as well. See Example F-1 on page 490 for a sample compiler listing.
SOURCE SECTION
Shows the source code with line numbers. Lines containing macros will have
additional lines showing the macro expansion. By default, this section only
lists the main source file. Use the -qshowinc option to expand all header files
as well.
OPTIONS SECTION
Shows any non-default options that were in effect for the compilation. To list
all options in effect, specify the -qlistopt option.
228
Developing and Porting C and C++ Applications on AIX

FILE TABLE SECTION
Lists all the files used in the compilation. Each file is associated with a file
number, and it always begin with the main source file with file number 0. It
shows, for each file, from which file and line the file was included. If the
-qshowinc option is also in effect, each source line in the SOURCE SECTION
will have a file number to indicate which file the line came from.
COMPILATION EPILOGUE SECTION
Shows a summary of the diagnostic messages by severity level, the number
of lines read, and whether the compilation was successful or not.
ATTRIBUTE AND CROSS REFERENCE SECTION
The -qattr and -qxref options will cause this section to be produced.
Independently, they provide information on all identifiers used in the
compilation. This is where you will find pertinent information about variables
on their type, storage duration, scope, and where they are defined and
referenced.
OBJECT SECTION
The -qlist option will cause this section to be produced. It shows the pseudo
assembly code generated by the compiler. This section is invaluable for
diagnosing execution time problems, if you suspect the program is not
performing as expected due to code generation error.
The -qinfo option
The -qinfo option causes the compiler to produce additional informational
messages for possible programming errors. The extra diagnostic messages can
help you in debugging your program.
Messages related to the same problem area are grouped together, and each
group is controlled via a suboption. For instance, the -qinfo=ini suboption
diagnoses possible problems with variables that are not explicitly initialized but
should be. For a list of the groups (suboptions) supported, refer to the VisualAge
C++ for AIX Compiler Reference, SC09-4959.
One new info group introduced in the Version 6 C compiler is the -qinfo=c99
suboption. It diagnoses C code that may have different behavior between the
-qlanglvl=stdc89 and -qlanglvl=stdc99 language levels. For example:
$ cat test.c
#include <stdio.h>
int main()
{
printf("sizeof(2147483648) = %d\n", sizeof(2147483648));
return 0;
Chapter 6. Programming hints and tips
229

}
$ cc -qinfo=c99 -c test.c
"test.c", line 4.48: 1506-786 (I) Integral constant "2147483648" has an implied
type of unsigned long int under the C89 language level. It has an implied type
of long long int under C99.
The -qsuppress option
When you use diagnostic aids like the -qinfo option, you can sometimes get
numerous messages that you may not be interested in at the moment, clobbering
the terminal or listing file. Use the -qsuppress option to stop those messages
from being emitted by the compiler. You can suppress more than one message
by listing the message numbers in a colon separated list.
The -qflag option
In some instances, you may want to ignore all messages below a certain severity
level. For example, during a production build, warning messages that are
deemed to be harmless by development teams will only clobber log files and
cause unnecessary concerns from the build team. You can use the -qflag option
to stop diagnostic messages from being emitted to the terminal and the listing
file. Similarly, the -w compiler flag suppresses informational and warning
messages from being emitted, and is the same as specifying the -qflag=e:e
option. See 6.2.1, “Anatomy of a message” on page 227 for a list of single letter
severity codes that you can use with the -qflag option.
The -qhaltonmsg option
On the contrary, with the C++ compiler, if you want to stop the compilation when
a certain message is encountered, use the -qhaltonmsg option. The message is
issued with a higher severity level and compilation terminates.
The -qhalt and -qmaxerr options
To stop a compilation in general when a message of a certain severity level or
higher is encountered, use the -qhalt option. See 6.2.1, “Anatomy of a message”
on page 227 for a list of single letter severity codes that you can use with the
-qhalt option.
The -qmaxerr option allows you to stop the compilation when the specified
number of times messages of a certain severity level or higher is reached. This
option takes precedence over the -qhalt option.
6.2.3 Migrating from 32-bit to 64-bit
When migrating a 32-bit program to 64-bit, the data model differences (as
described briefly in 2.1, “32- and 64-bit development environments” on page 38)
230
Developing and Porting C and C++ Applications on AIX

may result in unexpected behavior at execution time. In 64-bit mode, the size of
pointers and long data type are now 8 bytes long, and can lead to several
conversion or truncation problems. The -qwarn64 option can be used to detect
these portability errors.
int and long types
In 32-bit mode, both int and long data types are 32 bits in size. Because of this
similarity, these types may have been used interchangeably. As shown in
Table 2-1 on page 39, the data type long is 64 bits in length in 64-bit mode. A
general guideline is to review the existing use of long data types throughout the
source code. If the values to be held in such variables, fields, and parameters will
fit in the range of [-231...231-1] or [0...232-1], then it is probably best to use int or
unsigned int instead. Also, review the use of the size_t type (used in many
subroutines), since it is typedef as unsigned long.
long to int truncation
Truncation problems can occur when converting between 64-bit and 32-bit data
objects. Since int and long are 32 bits in 32-bit mode, a mixed assignment or
conversion between these data types did not represent any problem. It does,
however, in 64-bit mode, as long is larger in size than int. When converting from
long to int, either implicitly or explicitly through a cast, truncation may now occur:
void foo(long l)
{
int i = l;
}
Without an explicit cast, the compiler is unable to determine whether the
narrowing assignment is intended. If the value l is always within the range
representable by an int, or if the truncation is intended by design, use a cast to
silence the -qwarn64 message that you will receive for this code.
Unexpected result due to conversion to long
Due to the difference in size for int and long in 64-bit mode, conversions to long
from other integral types may result in different execution behavior from 32-bit
mode in some boundary cases.
When a signed char, signed short, or signed int is converted to unsigned long,
sign extension may result in a different unsigned value in 64-bit mode. For
example:
#include <stdio.h>
void foo(int i)
{
unsigned long l = i;
printf("%lu (0x%lx)\n", l, l);
Chapter 6. Programming hints and tips
231

}
void main()
{
foo(-1);
}
This program will yield 4294967295 (0xffffffff) in 32-bit mode but
18446744073709551615 (0xffffffffffffffff) in 64-bit mode due to sign
extension.
When an unsigned int variable with values greater than INT_MAX is converted to
signed long, you will get different results between 32-bit and 64-bit mode. For
example:
#include <stdio.h>
#include <limits.h>
void foo(unsigned int i)
{
long l = i;
printf("%ld (0x%lx)\n", l, l);
}
void main()
{
foo(INT_MAX + 1);
}
In 32-bit mode, the value INT_MAX+1 will wrap around and yield -2147483648
(0x80000000). The same value can be represented in 64-bit mode by a 8-byte
signed long and will result in the correct value of 2147483648 (0x80000000).
When a signed long long variable with values greater than UINT_MAX or less
than 0 is converted to unsigned long, truncation will no longer occur in 64-bit
mode. For example:
#include <stdio.h>
#include <limits.h>
void foo(signed long long ll)
{
unsigned long l = ll;
printf("%lu (0x%lx)\n", l, l);
}
void main()
{
foo(-1);
foo(UINT_MA X+ 1ll);
}
232
Developing and Porting C and C++ Applications on AIX

This program will yield:
4294967295 (0xffffffff)
0 (0x0)
in 32-bit mode and in 64-bit mode:
18446744073709551615 (0xffffffffffffffff)
4294967296 (0x100000000)
When an unsigned long long variable with values greater than UINT_MAX is
converted to unsigned long, truncation will no longer occur in 64-bit mode:
#include <stdio.h>
#include <limits.h>
void foo(unsigned long long ll)
{
unsigned long l = ll;
printf("%ld (0x%lx)\n", l, l);
}
void main()
{
foo(UINT_MAX + 1ull);
}
The higher order word is truncated and will result in 0 (0x0) in 32-bit mode, but
yield the correct result of 4294967296 (0x100000000) without truncation in 64-bit
mode.
When a signed long long variable with values less than INT_MIN or greater than
INT_MAX is converted to signed long, truncation will no longer occur in 64-bit
mode. For example:
#include <stdio.h>
#include <limits.h>
void foo(signed long long ll)
{
signed long l = ll;
printf("%ld (0x%lx)\n", l, l);
}
void main()
{
foo(INT_MIN - 1ll);
foo(INT_MAX + 1ll);
}
This program will yield (in 32-bit):
2147483647 (0x7fffffff)
-2147483648 (0x80000000)
Chapter 6. Programming hints and tips
233

and in 64-bit mode:
-2147483649 (0xffffffff7fffffff)
2147483648 (0x80000000)
And finally, when an unsigned long long variable with values greater than
INT_MAX is converted to signed long, truncation will no longer occur in 64-bit
mode. For example:
#include <stdio.h>
#include <limits.h>
void foo(unsigned long long ll)
{
signed long l = ll;
printf("%ld (0x%lx)\n", l, l);
}
void main()
{
foo(INT_MAX + 1ull);
}
In 32-bit mode, the value INT_MAX+1ull will wrap around and yield -2147483648
(0x80000000). The same value can be represented in 64-bit mode by a 8-byte
signed long and will result in the correct value of 2147483648 (0x80000000).
Pointer assignment and arithmetic
When migrating a program from 32-bit environment to 64-bit environment, it is
crucial to avoid pointer corruption. Some of the possible problems are:
Assigning an int (32 bits) or a 32-bit hexadecimal constant to a pointer type
variable (64 bits) or casting a pointer to an int will yield an invalid address, and
will cause errors when the pointer is dereferenced. Also, the comparison of an
int to a pointer may cause unexpected results.
Pointers are converted to int or unsigned int with the expectation that the
pointer value will be preserved, as casting a pointer to an int will result in data
truncation.
Without proper function prototypes, functions that return pointers will return
truncated return values, as the functions are implicitly declared to return an int
that is just 32 bits, instead of the expected 64 bits of a pointer.
The code assumes that pointers and int are the same size in an arithmetic
context, as pointer arithmetic usually is a source of problems in migration. The
ISO C standard dictates that incrementing a pointer yields adding the size of
the data type to which it points to the pointer value. For example, if the
variable p is a pointer to long, then the operation (p+1) increments the value
of p by 4 bytes in 32-bit mode but by 8 bytes in 64-bit mode. Therefore, casts
234
Developing and Porting C and C++ Applications on AIX

between long* to int* are problematic because of the size differences of
pointer objects (32 bits versus 64 bits).
Incorrect pointer to int and int to pointer conversions
When a pointer is explicitly converted to an int, truncation of the high order word
occurs. When an int is explicitly converted to a pointer, the pointer may not be
correct and may cause invalid memory access if dereferenced. For example:
#include <stdio.h>
#include <stdlib.h>
void main()
{
int i, *p, *q;
p = (int*)malloc(sizeof(int));
i = (int)p;
q = (int*)i;
p[0] = 55;
printf("p = %p q = %p\n", p, q);
printf("p[0] = %d q[0] = %d\n", p[0], q[0]);
}
In 32-bit mode, the pointers p and q are pointing to the same memory location.
However, the pointer q is likely pointing to an invalid address in 64-bit mode, and
could result in a segmentation fault when q is dereferenced.
Integer constants
A loss of data can occur in some constant expressions because of lack of
precision. These types of problems are very hard to find and may have gone
unnoticed so far. You should therefore be very explicit about specifying the
type(s) in your constant expressions and use constant suffix {u,U,l,L,ll,LL} to
specify exactly its type. You might also use casts to specify the type of a constant
expression.
This is especially true when migrating to 64-bit, because integer constants may
have different types when compiled in 64-bit mode. The ISO C standard states
that the type of an integer constant, depending on its format and suffix, is the first
(that is, smallest) type in the corresponding list (see Table 6-1 on page 236) that
will hold the value. The quantity of leading zeros does not influence the type
selection.
Chapter 6. Programming hints and tips
235

Table 6-1 ISO C99 integer constant type selection
Suffix
Decimal constant
Octal or hexadecimal
constant
unsuffixed
int
int
long
unsigned int
long long
long
unsigned long
long long
unsigned long long
u or U
unsigned int
unsigned int
unsigned long
unsigned long
unsigned long long
unsigned long long
l or L
long
long
long long
unsigned long
long long
unsigned long long
Both u or U and l or L
unsigned long
unsigned long
unsigned long long
unsigned long long
ll or LL
long long
long long
unsigned long long
Both u or U and ll or LL
unsigned long long
unsigned long long
For example, a hexadecimal constant that could only be represented by an
unsigned long in 32-bit will now fit within a long in 64-bit. The change in type of
the constant in an expression may cause unexpected results.
Data alignment
Modern processor designs usually require data in memory to be aligned to their
natural boundaries in order to gain the best possible performance. The compiler
in most cases guarantees data objects to be properly aligned by inserting
padding bytes immediately before the misaligned data. Although the padding
bytes do not affect the integrity of the data, they can cause the layout and hence
the size of structures and unions to be different than expected.
Because the size of pointers and long is doubled in 64-bit mode, structures and
unions containing them as members will become bigger than in 32-bit mode. For
example, consider the structure in Example 6-1 on page 237.
236
Developing and Porting C and C++ Applications on AIX

Example 6-1 Structure alignment
#include <stdio.h>
#include <stddef.h>
void main()
{
struct T {
char c;
int *p;
short s;
} t;
printf("sizeof(t) = %d\n", sizeof(t));
printf("offsetof(t, c) = %d sizeof(c) = %d\n",
offsetof(struct T, c), sizeof(t.c));
printf("offsetof(t, p) = %d sizeof(p) = %d\n",
offsetof(struct T, p), sizeof(t.p));
printf("offsetof(t, s) = %d sizeof(s) = %d\n",
offsetof(struct T, s), sizeof(t.s));
}
When Example 6-1 is compiled and executed in 32-bit mode, the following result
indicates paddings have been inserted before the member p, and after the
member s:
sizeof(t) = 12
offsetof(t, c) = 0 sizeof(c) = 1
offsetof(t, p) = 4 sizeof(p) = 4
offsetof(t, s) = 8 sizeof(s) = 2
Three padding bytes are inserted before the member p to ensure p is aligned to
its natural 4-byte boundary. The alignment of the structure itself is the alignment
of its strictest member. In this example, it is 4-byte due to the same member p.
Therefore, two padding bytes are inserted at the end of the structure to make the
total size of the structure a multiple of 4-byte (see Figure 6-1). This is required so
that if you declare an array of this structure, each elements of the array will be
aligned properly.
padding
c padding
p
s
0 1 2 3 4 5 6 7 8 9 10 11
Figure 6-1 Structure padding in 32-bit mode
Chapter 6. Programming hints and tips
237

However, when Example 6-1 on page 237 is compiled and executed in 64-bit
mode, the size of the structure doubles, which is caused by more paddings
required to make the member p to fall on a natural alignment boundary of 8-byte
(see Figure 6-2):
sizeof(t) = 24
offsetof(t, c) = 0 sizeof(c) = 1
offsetof(t, p) = 8 sizeof(p) = 8
offsetof(t, s) = 16 sizeof(s) = 2
c
padding
p
s
padding
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23
Figure 6-2 Structure padding in 64-bit mode
And imagine if this structure is shared or exchanged among 32-bit and 64-bit
processes, the data fields (and paddding) of one environment will not match the
expectations of the other.
To eliminate the difference, and to allow the structure to be shared, you can
reorder the fields in the data structure to get the alignments in both 32-bit and
64-bit environments to match. However, this may not be possible in all cases. It
depends a great deal on the data types used in the structure, and the way in
which the structure as a whole is used (for example, whether the structure is
used as a member of another structure or array).
If you are unable to reorder the members of a structure, or if reordering alone
cannot provide correct alignment, you can introduce user-defined paddings to
cause the members of the structure to fall on their natural boundaries.
Depending on the data types involved, a conditional compile section may be
necessary. A conditional compile section will be required when the structure uses
data types that have different sizes in the 32-bit and 64-bit environments.
For example, if the structure layout of the Example 6-1 on page 237 is changed
to the following:
struct T {
char c;
short s;
#if !defined(__64BIT__)
char pad1[4];
#endif
int *p;
238
Developing and Porting C and C++ Applications on AIX

#if !defined(__64BIT__)
char pad2[4];
#endif
} t;
The structure will have the same size and member layout in both 32-bit and 64-bit
environments:
sizeof(t) = 16
offsetof(t, c) = 0 sizeof(c) = 1
offsetof(t, s) = 2 sizeof(s) = 2
offsetof(t, p) = 8 sizeof(p) = 4
Note that this output is from a 32-bit execution. The size of the member p,
sizeof(p), will be 8 in 64-bit mode. Figure 6-3 shows the member layout of the
structure with user-defined padding.
Important: This example is for illustrative purposes only. Sharing pointers
between 32-bit and 64-bit processes is not recommended and will likely yield
incorrect results.
padding
padding
p
64-bit
c
s
pad1
p
pad2
32-bit
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Figure 6-3 Structure with user-defined paddings in both 32-bit and 64-bit mode
When inserting paddings to structures, use an array of char to avoid any further
alignment requirement on the paddings themselves. The natural alignment of a
char is 1-byte, meaning it can reside anywhere in memory.
6.3 Diagnosing link-time errors
The C and C++ compiler drivers (see 1.7.1, “Default compiler drivers” on
page 30) handle object file linking by invoking the system link editor ld with the
appropriate options and flags. There is no need to use ld directly to link edit your
C and C++ object files. This section describes some of the more common
link-time errors that you may encounter during development.
Chapter 6. Programming hints and tips
239

6.3.1 Unresolved symbols
When linking your application with many libraries, particularly those supplied by a
third party product, such as a database, it is not unusual during the development
cycle to see a linker error warning about unresolved symbols.
The system link editor accepts input such as object files, shared object files,
archive object files, libraries, and import and export files, resolves external
symbol references, and creates a single executable object file that can be run.
When an external symbol cannot be resolved, the link would fail and an
executable is not generated.
Unresolved symbol errors can occur in many situations, the most common being
missing input files. For example, if you call a library function that is not in the
default C run-time library libc.a, you need to specify the archive library file in
which the symbol is found:
$ cat test.c
#include <stdio.h>
#include <math.h>
void main()
{
printf("%f\n", pow(2.0,3.0));
}
$ cc test.c
ld: 0711-317 ERROR: Undefined symbol: .pow
ld: 0711-345 Use the -bloadmap or -bnoquiet option to obtain more information.
$ cc test.c -lm
This is especially true when using functions provided by third-party libraries.
Aside from searching the product documentation, finding the library file in which
an unresolved symbol is defined can be quite a daunting task. You can include all
supplied libraries in the link command and let the link editor find out where the
symbol is, or you may use the nm command to try find it yourself.
The linker supports options that can be used to generate linker log files. These
log files can then be analyzed to determine the library or object file that
references the unresolved symbol. This can help in tracking interdependent or
redundant libraries being used in error.
The -bnoquiet option writes each binder sub command and its results to standard
output. It gives a better of picture of the unresolved symbol references if any
exists and also the list of symbols imported from the specified library modules.
The -bmap:filename option is used to generate an address map. Unresolved
symbols are listed at the top of the file, followed by imported symbols.
240
Developing and Porting C and C++ Applications on AIX

The -bloadmap:filename option is used to generate the linker log file. It includes
information on all of the arguments passed to the linker along with the shared
objects being read and the number of symbols being imported. If an unresolved
symbol is found, the log file produced by the -bloadmap option lists the object file
or shared object that references the symbol. In the case of using libraries
supplied by a third-party product, you can then search the other libraries supplied
by the product in an effort to determine which one defines the unresolved
symbol. This is particularly useful when dealing with database products that
supply many tens of libraries for use in application development.
If errors show up during linking or loading in a version controlled application
development environment, then try linking outside the version control system. If
errors happen only when linking or loading under version control, then check with
your version control system vendor for known linker/loader problems.
6.3.2 Duplicate symbols
Duplicate symbol errors usually indicate a programming error. It is invalid to have
multiple external function definitions of the same name. The link editor in this
case will use the first definition that it encounters, and the result may not be
desirable. Change the name of the function or use the static function.
The use of template functions in C++ may generate duplicate symbol errors at
link time. This happens when the template is implicitly instantiated in multiple
source files. For example:
// t.h
template <class T> class A {
public:
int f();
};
template <class T> int A<T>::f()
{
return 55;
}
// file1.C
#include "t.h"
int func();
int main()
{
A<int> a;
int obj1 = a.f() + func();
return obj1;
}
Chapter 6. Programming hints and tips
241

#include "t.h"
int func()
{
A<int> a;
int obj2 = a.f();
return obj2;
}
$ xlC -c file1.C file2.C
file1.C:
file2.C:
$ xlC file1.o file2.o
ld: 0711-224 WARNING: Duplicate symbol:
.A<int>::f()
ld: 0711-345 Use the -bloadmap or -bnoquiet option to obtain more information.
Using nm, you will note that the symbol A<int>::f() is defined in both object files:
$ nm -g file1.o
.A<int>::f()
T
100
.func()
U
-
.main
T
0
A<int>::f()
D
180
12
main
D
168
12
$ nm -g file2.o
.A<int>::f()
T
68
.func()
T
0
A<int>::f()
D
148
12
func()
D
136
12
In this case, since the duplicate symbol A<int>::f() has the same definition, it is
safe to suppress the link editor message with -bhalt:5:
$ xlC -bhalt:5 file1.C file2.C
file1.C:
file2.C:
The -bhalt link editor option specifies the maximum error level for the linking
process to continue, and suppresses any error below the maximum. The
recommended solution to fixing this error is to use the -qtemplateregistry
compiler option described in 10.5, “Template registry: The preferred method” on
page 387, and to take advantage of the improved template handling provided by
the VisualAge C++ for AIX Version 6 compiler. See the VisualAge C++ for AIX
Compiler Reference, SC09-4959 for details on the option.
242
Developing and Porting C and C++ Applications on AIX

6.3.3 Insufficient memory for the linker process
The linker can run out of memory when trying to link very large files. When this
happens, linking fails with the following error:
ld: 0711-101 fatal error, allocation of bytes fail in routine initsymtab_info.
There is not enough memory.
This may be because of low paging space or because of low resource limits for
the user invoking the command. The AIX linker offers a great deal more
functionality than traditional UNIX linkers, but it does require a reasonable
amount of virtual memory, particularly when linking large applications with many
libraries.
If this type of error is encountered, check the following and increase the space, if
necessary:
The available paging space on the machine.
The resource limits for the user invoking the linker using the ulimit command
(see 3.2.6, “Resource limits in 32-bit model” on page 125 for information on
how to use the command).
If the problem is not solved by these activities, you may consider running the
linker process in the 32-bit large memory model, as explained in 3.2.4, “Using the
large and very large memory model” on page 121.
6.3.4 The c++filt utility
The C++ programming language allows function and operator overloading. An
overloaded function is a function with the same name as a previously declared
function, but with a different parameter list. To distinguish overloaded function
names, the compiler uses name mangling to encode the names into unique
names so that the link editor will not mistake them as duplicate symbols. The
mangled name contains further information about the function and its
parameters.
The c++filt utility is provided with the VisualAge C++ for AIX compiler product to
convert these mangled names to their original source code names. With the
-bloadmap:filename link editor option, function names appearing in the loadmap
file, filename, are in mangled form. Use the c++filt utility to get the original
function names:
ld: 0711-228 WARNING: Duplicate symbols were found while resolving symbols.
The following duplicates were found:
Symbol Source-File(Object) OR Import-File{Shared-object}
------------------------- -------------------------------------------------
.f__1AXTi_Fv file1.C(file1.o)
Chapter 6. Programming hints and tips
243

** Duplicate ** file2.C(file2.o)
The c++filt command waits for input from the standard input. If you type in
mangled function names, it prints the original functions to the standard output:
$ c++filt
f__1AXTi_Fv
A<int>::f()
To quit the command, type ^D (Control-D).
6.4 Diagnosing run-time errors
Although your program may compile and link successfully, you may still
encounter unexpected results during execution. Programming errors that do not
violate the syntax of the language are not necessarily flagged by the compiler.
This section describes some of the most common errors, how to detect them,
and what you can do to rectify the problem.
6.4.1 Uninitialized variables
Unlike variables with static storage duration, according to the C and C++
programming language standards, an automatic (that is, stack allocated) storage
duration object is not implicitly initialized, and its initial value is indeterminate. In
other words, there is no guarantee that if an auto variable is used before it is set,
it will yield the same result in every execution of the program. For example, when
you first execute the following program, it may seem to be working correctly:
$ cat initauto.c
#include <stdio.h>
void func(int *p)
{
if (p == NULL)
printf("NULL pointer.\n");
else
*p = 0xdeadbeef;
}
void main()
{
int *ptr;
func(ptr);
}
$ cc initauto.c
$ a.out
NULL pointer.
244
Developing and Porting C and C++ Applications on AIX

However, there is still the potential that, given the right conditions, the program
will start to fail and yield unexpected results.
The C and C++ compilers provide the -qinitauto option, which causes the
compiler to generate extra code to initialize all automatic variables to a specified
initial value. Due to the extra code that the option generates, it reduces the
run-time performance of your program, and its use is only recommended for
debugging only. For example, compile and execute the above program with
-qinitauto=FE and you will receive a different result indicating a potential problem:
$ cc -qinitauto=FE initauto.c
$ a.out
Segmentation fault(coredump)
To find out where auto variables are used before being set, use the -qinfo=gen
compiler option:
$ cc -qinfo=gen initauto.c
"initauto.c", line 13.10: 1506-438 (I) The value of the variable "ptr" may be
used before being set.
6.4.2 Run-time checking
The -qcheck option inserts run-time checking code into the program executable.
The following checks are supported:
NULL pointer
Checks if the pointers used in any pointer referencing
have addresses greater than 512.
Array bounds
Checks for array indexing to be within the array bounds,
when the size of the array is known at compile time.
Divide by zero
Checks for integer divide by zero.
Depending on the suboptions specified, if a violation occurs, a run-time
SIGTRAP exception is raised. You can provide your own signal handler to handle
the exception:
$ cat check.c
#include <stdio.h>
#ifdef DEBUG
#include <signal.h>
#define SIGNAL(sig, handler) signal(sig, handler)
void trap_handler(int sig)
{
printf("SIGTRAP handled\n");
exit(-1);
}
#else
Chapter 6. Programming hints and tips
245

#define SIGNAL(sig, handler) ((void)0)
#endif
void func(int *p)
{
printf("p has address %p\n", p);
*p = 0xdeadbeef;
}
void main()
{
SIGNAL(SIGTRAP, trap_handler);
func(NULL);
}
$ cc -DDEBUG -qcheck check.c
$ a.out
p has address 0
SIGTRAP handled
Similar to the -qinitauto option, the -qcheck option reduces the run-time
performance of your program. It is recommended that you use it only for
debugging.
6.4.3 Unsignedness preservation in C
If you use the cc command, specify the -qlanglvl=extended option, or the
-qupconv option, and the C compiler will use unsignedness preservation in
integral promotion. This was the semantic used prior to the c89 standard, also
commonly referred to as K&R C.
Unsignedness preservation promotes unsigned integral types smaller than int,
that is, unsigned char and unsigned short, to unsigned int. On the contrary, c89
and c99 mandate value preservation, where they are promoted to int.
Unsignedness preservation may result in unexpected execution behavior. For
example:
$ cat upconv.c
#include <stdio.h>
void main()
{
unsigned char zero = 0;
if (-1 < zero)
printf("NOUPCONV: Value-preserving rules in effect \n");
else
printf("UPCONV: Unsignedness-preserving rules in effect \n");
}
$ cc upconv.c
$ a.out
246
Developing and Porting C and C++ Applications on AIX

UPCONV: Unsignedness-preserving rules in effect
$ cc -qnoupconv upconv.c
$ a.out
NOUPCONV: Value-preserving rules in effect
6.4.4 ANSI aliasing
The standard conforming language levels supported by the C and C++ compilers
enforce a type-based aliasing rule during optimization. Type-based aliasing,
sometimes referred to as ANSI-aliasing, restricts the lvalues2 that can be safely
used to access a data object. In essence, it mandates that a pointer can only
point to an object of the same type. In other words, any pointer cast followed by a
dereference violates this rule. The only exceptions are:
The sign and type qualifiers are not subject to type-based aliasing
A character pointer can point to any type.
For example, the following program gives different results when optimization is in
effect:
$ cat ansialias.c
#include <stdio.h>
unsigned int rc;
unsigned int function( float *ptr )
{
rc = 0;
*ptr = 1;
return rc;
}
void main()
{
unsigned int x = function((float *)&rc);
printf("x = %x rc = %x\n", x, rc);
}
$ c89 ansialias.c
$ a.out
x = 3f800000 rc = 3f800000
$ c89 -O ansialias.c
$ a.out
x = 0 rc = 3f800000
When optimization is turned on with -O, the compiler assumes that the pointer ptr
cannot be pointing to the external variable rc due to type-based aliasing. Hence,
it can return the value 0 to the caller because of the rc = 0 assignment.
2 An lvalue is an expression with an object type. It is a locator value representing an object.
Chapter 6. Programming hints and tips
247

If you use the cc compiler driver, or specify the -qalias=noansi option, the
compiler makes the worst case aliasing assumptions, and assumes that a pointer
of a given type can point to an external object or any object whose address is
already taken, regardless of type. This will correct the programming error in the
above example. However, it also greatly reduces optimization opportunities and
degrades run-time performance. It is therefore recommended that you change
you program to conform to type-based aliasing rule.
6.4.5 #pragma option_override
Unfortunately, it may not be easy to fix all programming errors in a complex
application, especially when the error only shows up when optimization is used.
In this case it may be worth while to turn off optimization for the function where
the programming error is, while the rest of the program still benefits from
optimization.
Use the #pragma option_override directive to specify alternate optimization
options for specific functions. In the example in 6.4.4, “ANSI aliasing” on
page 247, adding the directive:
#pragma option_override(function, “opt(level,0)”)
to the source will correct the programing error:
$ c89 -O ansialias.c
$ a.out
x = 3f800000 rc = 3f800000
The #pragma option_override directive is also useful in finding out, by a process
of elimination, which function is causing the problem. By selectively turning off
optimization for each function within the directive until the problem disappears, it
will let you identify where the programming error resides.
248
Developing and Porting C and C++ Applications on AIX

7
Debugging your applications
Chapter 7.
This chapter describes several methods for debugging C and C++ applications
by providing the following sections:
Section 7.1, “Working with core files” on page 250
Section 7.2, “Using the printf()-debug method” on page 258
Section 7.4, “Using dbx” on page 259
Section 7.5, “Debugging with the truss command” on page 264
Section 7.6, “Using the trace facility” on page 267
For further information about the debugging methods on AIX, please refer to AIX
5L Version 5.2 General Programming Concepts: Writing and Debugging
Programs.
© Copyright IBM Corp. 2000, 2003. All rights reserved.
249

7.1 Working with core files
A core file is generated when various errors occur in an application and the
process aborts. The reasons can be errors such as memory-address violations,
illegal instructions, bus errors, and user-generated quit signals. The core file
contains a memory image of the aborted process. For a detailed definition of the
contents of a core file, please refer to AIX 5L Version 5.2 Files Reference.
Even if the application does not stop, it might be useful to have a core file as an
image of the memory state of the process in a specific time. Core files can be
used for debugging and examination by yourself on your system or by remote
technical support specialists on their system. For the distribution of a core file,
there is a special way to collect all needed information.
7.1.1 Core file naming
Before AIX 5L Version 5.1, a core file was always stored in a file named core. If
the same or another application generated another core file before you renamed
the previous one, the original content was lost.
Beginning with AIX 5L Version 5.1, you can enable a unique naming of core files,
but be aware that the default behavior is to name the files core. You apply the
new enhancement by setting the environment variable CORE_NAMING to a
non-NULL value, for example:
CORE_NAMING=yes
After setting CORE_NAMING, you can disable this feature by setting the variable
to the NULL value. For example, if you are using the Korn shell, do the following:
export CORE_NAMING=
After setting CORE_NAMING, all new core will be stored in files of the format
core.pid.ddhhmmss, where:
pid Process
ID
dd
Day of the month
hh Hours
mm Minutes
ss Seconds
250
Developing and Porting C and C++ Applications on AIX

In the following example, two core files are generated by a process identified by
PID 30480 at different times:
$ ls -l core*
-rw-r--r-- 1 ausres01 itsores 8179 Jan 28 2003 core.30480.28232347
-rw-r--r-- 1 ausres01 itsores 8179 Jan 28 2003 core.30482.28232349
The time stamp used is in GMT1 and your time zone will not be used.
7.1.2 Creating core files with assert()
The assert() macro, provided by the assert.h header file, is a common way to
produce a core file, when you are sure where the application logic goes wrong.
For example, if you are sure the integer variable cnt should not be greater than
100 in the certain point of your code, you can insert the following line into your
source code:
#include <assert.h>
int func(void)
{
...
/* You are sure that cnt should not be greater than 100 in here. */
assert(cnt <= 100);
...
/* Application logic continues. */
}
If the variable cnt is greater than 100 in the highlighted line in the above example,
the assert() macro calls the abort() routine and generates a core file and the
process dies.
7.1.3 Creating core files with coredump()
If you have an application behaving unexpectedly, you can have a core file
without terminating the application process by adding the coredump()
sub-routine in your source code. The generated core file, which contains the
memory image of the process, can be used for debugging the problem with dbx.
In a multi-threaded process, only one thread at a time should attempt to call
coredump(). Subsequent calls to coredump() while a core dump (initiated by
another thread) is in progress will fail.
To use coredump(), you must compile your source code with the -bM:UR options;
otherwise, the routine will fail with an error code of ENOTSUP.
1 Greenwich Mean Time
Chapter 7. Debugging your applications
251

The syntax of coredump() is:
#include <core.h>
int coredump(struct coredumpinfo *coredumpinfop)
By specifying a file name and its string length in the coredumpinfo structure
parameter, you can specify any file name as the generating core file. For
example, if we compile the source file shown in Example 7-1 with cc -bM:UR
coredump.c, then run a.out, then the core file, mycore, will be generated as
shown in the following example:
$ cc -bM:UR coredump.c
$ ./a.out; echo $?
0
$ ls -l mycore
-rw-r--r-- 1 ausres01 itsores
8183 Feb 05 10:55 mycore
Example 7-1 coredump.c
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <errno.h>
#include <core.h>
int
main(int argc, char *argv[])
{
int rc;
struct coredumpinfo cdinfo;
cdinfo.name = "mycore";
cdinfo.length = strlen("mycore");
if ((rc = coredump(&cdinfo)) == -1) {
perror("coredump()");
}
exit(0);
}
For more information about coredump(), please refer to AIX 5L Version 5.2
Technical Reference: Base Operating System and Extensions.
252
Developing and Porting C and C++ Applications on AIX

7.1.4 Including shared memory information in the core file
On AIX, by default, the detailed information about shared memory segments and
thread stacks are not collected when generating core files. If you need to have
core files that contain detailed information (called full-core), there are two ways
to do so.
System wide full-core setting
Use the chdev command to change the full core attribute setting system wide.
The default value is not to take full core dump, as shown in the following example:
# lsattr -El sys0 -a fullcore
fullcore false Enable full CORE dump True
To change the setting, issue the chdev -l sys0 -a fullcore=true command as
the root user.
Add a signal handler with the SA_FULLDUMP flag
You can modify your application source code by adding a signal handler with the
SA_FULLDUMP flag set for the signal that will cause the core file. To register a
signal handler, use the sigaction().
Note: The full-core setting is required to debug multi-threaded applications
with most debugging methods.
7.1.5 Gathering core files
All associated information of a core file can be packed and archived in a pax file,
which can be stored on disk or tape or sent to another system for investigation. At
the time of the writing of this redbook, the Distributed Debugger only supports
debugging of core files on the machine that created them.
The snapcore command
Use the snapcore command to gather the core file, its program binary executable
file, and its dependent library files. The syntax of the command is:
snapcore core_filename [program_filename]
Specify the full path names for core file and program file name. Without defining
the program name, snapcore will read the program name out of the core file and
search it in the directories defined by PATH.
The command will produce a compressed pax file in the /tmp/snapcore directory
by default. Use the -d directory option to specify an alternative directory. In the
Chapter 7. Debugging your applications
253

following example, a core file generated by prog1 and its dependent library files
are archived in a compressed pax file, /tmp/snapcore/core.31442.pax.Z:
$ snapcore core.31374.29164438 prog1
Core file "core.31374.29164438" created by "prog1"
pass1() in progress ....
Calculating space required .
Total space required is 6578 kbytes ..
Checking for available space ...
Available space is 119148 kbytes
pass1 complete.
pass2() in progress ....
Collecting fileset information .
Collecting error report of CORE_DUMP errors ..
Creating readme file ..
Creating archive file ...
Compressing archive file ....
pass2 completed.
Snapcore completed successfully. Archive created in /tmp/snapcore.
$ ls -l /tmp/snapcore
total 5960
-rw-r--r-- 1 ausres01 itsores 3049565 Jan 29 10:52 snapcore_31442.pax.Z
You can check what files are gathered in the archive file, as shown in the
following example:
$ uncompress -c /tmp/snapcore/snapcore_31442.pax.Z | pax
core.31374.29164438
README
lslpp.out
errpt.out
prog1
./usr/lib/libc.a
./usr/lib/libcrypt.a
./usr/ccs/lib/libc.a
For more information about the snapcore command refer to AIX 5L Version 5.2
Reference Documentation: Commands Reference.
The check_core command
You can determinate the program that caused the core and the dependent
libraries for an existing core file by using the check_core2 command.
2 The check_core command is included in the bos.rte.serv_aid fileset.
254
Developing and Porting C and C++ Applications on AIX

The following example shows that the core file, core.31374.29164438, was
generated by the program file, prog1, and its dependent libraries:
$ /usr/lib/ras/check_core core.31374.29164438
/usr/lib/libc.a
/usr/lib/libcrypt.a
prog1
7.1.6 AIX error log entry
Each core dump creates a new entry in the AIX error log. It can be useful for
identifying an application that dumps. Use the following command for examining
all entries caused by an core dump:
errpt -aJ CORE_DUMP
Use the -s mmddhhmmyy option to filter error log entries starting after the given
time (mm = month, dd = day, hh = hours, mm = minutes, yy = year).
If you use the -A option instead of -a, you will see a more condensed output.
Example 7-2 shows that the program prog1 identified with PID 31242 generated
a core file because of the signal 11 (SIGSEGV) delivery.
Example 7-2 AIX error log: CORE_DUMP
$ errpt -A -J CORE_DUMP -s 0129130003
-----------------------------------------------------------------------
LABEL: CORE_DUMP
Date/Time: Wed Jan 29 13:20:35 CST
Type: PERM
Resource Name: SYSPROC
Description
SOFTWARE PROGRAM ABNORMALLY TERMINATED
Detail Data
SIGNAL NUMBER
11
USER'S PROCESS ID:
31242
FILE SYSTEM SERIAL NUMBER
13
INODE NUMBER
2048
PROGRAM NAME
prog1
ADDITIONAL INFORMATION
main 10C
main F8
__start 8C
Chapter 7. Debugging your applications
255

7.1.7 Lightweight core file support
Besides the standard core file format, you can use the lightweight core file
format, which complies with the Parallel Tools Consortium Lightweight Core File
Format. If a multi-threaded program crashes or hangs, it is nearly impossible to
find out how far the execution got. Dumping out the memory information of many
processors requires time and disk space. The resulting information in a standard
core file is of little use for this case. The lightweight core file format is a platform
independent format with a snapshot of the current location of an application. The
lightweight core file is a high level, symbolic collection of the program state. It is
an ASCII file, readable by humans and analysis programs. Running prog2 from
Example 7-3 on page 257, we receive the lightweight core file named lwcore:
$ more lwcore
+++PARALLEL TOOLS CONSORTIUM LIGHTWEIGHT COREFILE FORMAT version 1.0
+++LCB 1.0 Wed Feb 12 08:19:07 2003 Generated by IBM AIX 5.1
#
+++ID Node 0 Process 30820 Thread 1
***FAULT "SIGSEGV - Segmentation violation"
+++STACK
main : 0x0000004c
---STACK
---ID Node 0 Process 30820 Thread 1
---LCB
+++PARALLEL TOOLS CONSORTIUM LIGHTWEIGHT COREFILE FORMAT version 1.0
+++LCB 1.0 Wed Feb 12 08:31:09 2003 Generated by IBM AIX 5.1
#
+++ID Node 0 Process 37548 Thread 1
***FAULT "SIGSEGV - Segmentation violation"
+++STACK
main : 0x0000004c
---STACK
---ID Node 0 Process 37548 Thread 1
---LCB
Creating a lightweight core file with install_lwcf_handler()
The install_lwcf_handler() subroutine provides a lightweight core file instead of a
standard core file when an application crashes. It is part of the PTools Library
(libptools_ptr.a). There are two ways to create the lightweight core file:
Call the install_lwcf_handler() subroutine directly in your application to
register a signal handler (Example 7-3 on page 257).
Use the linker option -binitfini:install_lwcf_handler, so that the function will be
called by starting the program automatically. In this way, you do not have to
change your application code.
256
Developing and Porting C and C++ Applications on AIX

The default file name for the lightweight core file is lw_core. Use the
LIGHTWEIGHT_CORE environment variable to change it to your desired file
name, or set it to STDERR to redirect the lightweight core file content to the
standard error.
Example 7-3 prog2.c with install_lwcf_handler() subroutine
#include <stdlib.h>
#include <stdio.h>
void install_lwcf_handler (void);
main(int argc, char *argv[])
{
int i,j;
char s[10];
install_lwcf_handler();
printf("I will start counting!\n");
while (i >= j) {
s[j] = j;
j--;
i++;
}
printf("My result is %s\n",s);
exit(0);
}
$ cc prog2.c -o prog2 -lptools_ptr
$ export LIGHTWEIGHT_CORE="lwcore";
$ prog2
I will start counting!
$
Please refer to AIX 5L Version 5.2 Technical Reference: Base Operating System
and Extensions for more information.
Creating a lightweight core file with mt_trce()
The mt_trce() subroutine provides a lightweight core file with the trace back
information of all threads allocated in the process space. Threads, except for the
calling thread, are suspended during the execution of the mt_trce() subroutine.
Refer to the AIX 5L Version 5.2 Technical Reference: Base Operating System
and Extensions for a complete overview of this topic.
Chapter 7. Debugging your applications
257

7.2 Using the printf()-debug method
The printf()-debug method is easy to understand and is still the most common
method used to debug applications. To use the method, define the following
macro in your codes:
#if !defined(DEBUG)
#define DBGMSG(MSGSTR) ((void)0)
#else
#define DBGMSG(MSGSTR) {\
fprintf(stderr,"In %s() at line %d in file %s: %s"\
, __FUNCTION__, __LINE__, __FILE__, (MSGSTR));\
}
#endif
then call this macro where you would like to print messages, for example:
int main(int argc, char *argv[])
{
DBGMSG("Hi there!\n");
exit(0);
}
If you compile the source code with the -DDEBUG compiler option, then it will
print a debug message very similar to the following line:
In main() at line 16 in file printf-debug.c: Hi there!
Although it is a very commonly used method in the early development cycle, it
has the following disadvantages:
You may need to recompile your source code every time you want to insert
new printf()-debug lines.
Debugging multi-threaded applications with the printf()-debug method is not
trivial.
Because it may cause many calls to printf(), the application performance get
degraded in most cases.
If you insert the DBGMSG macro too many times, it can be hard to trace the
huge quantity of debug messages.
Function-like macros introduced by C99 are very useful when defining the debug
macro (see “Function-like macros with variable arguments” on page 9).
258
Developing and Porting C and C++ Applications on AIX

7.3 Preparing your application for debugging
To fully exploit any debugger, you need to compile your application with the -g
compiler option and without optimization. Otherwise, the debugger cannot
resolve referenced symbols in your application, and you cannot refer to variables
or functions by their name while debugging.
If you run the strip command to strip symbol information from your application
program, the debugger cannot resolve referenced symbols in your application
either.
The latest C and C++ compilers on AIX support the following useful options:
-qfullpath
This puts the full path of the source files into the debug
information. This can be very useful if you have a very
large source tree, since the debugger does not need to
ask for the location of the source file.
-qlinedebug
This eliminates the variable descriptions from the debug
information. This is recommended if the code is
optimized, since the variable monitoring in optimized code
will generally give incorrect results. This option
significantly reduces the size of the debug information,
especially in C++ programs.
-qdbxextra
This include debug information for all symbols in your
program, regardless of whether they are referenced.
Normally, only debug information for referenced symbols
is included. This option will significantly increase the size
of the resulting executable.
When debugging information is created in an application, the size of the
executable may grow dramatically. A C++ executable may grow in size by a factor
of 10 or more when debug information is generated. Intelligent use of
-qlinedebug and turning off debug information in uninteresting areas of the code
may help in this case.
7.4 Using dbx
The symbolic debugger, dbx, is the most commonly used debugger on UNIX
operating systems, including AIX. The command provides many useful functions
to debug application problems, including:
Examines object and core files
Provides a controlled environment for running a program
Chapter 7. Debugging your applications
259

Sets break points
Traces program flow
Supervises symbolic variables
Although the usage of dbx is well explained in Chapter 3, “Debugging Programs”,
in AIX 5L Version 5.2 General Programming Concepts: Writing and Debugging
Programs, we describe the basic use of dbx and show useful customization
examples.
Note: dbx does not support lightweight core file.
7.4.1 Starting a dbx session
If an application aborted and generated a core file, run dbx as follows:
dbx ObjectFile CoreFile
If you do not specify the core file name, then the core file in the current directory
is assumed to be the correct file. After starting a dbx session, it prints the
sub-command prompt (dbx). The following example shows that the core file,
core.30330.30144523, generated by prog1, is core dumped at line number 13 in
main() because of the segmentation fault:
$ dbx prog1 core.30330.30144523
Type 'help' for help.
reading symbolic information ...
[using memory image in core.30330.30144523]
Segmentation fault in main at line 13
13 s[j] = j;
To exit from a dbx session, type quit on the dbx sub-command prompt:
(dbx) quit
Attaching to a running process
If you need to debug a running process, for example, your application seems to
run into an endless loop, you can attach a dbx session to the process by
specifying the process ID (PID) with the -a option, as shown in the following
example:
$ dbx -a 25952
Waiting to attach to process 25952 ...
Successfully attached to prog6.
Type 'help' for help.
reading symbolic information ...
stopped in main at line 15
260
Developing and Porting C and C++ Applications on AIX

15 while (i >= j) {
(dbx) quit
[2] + Killed prog6 &
Note: If you type the quit sub-command on the sub-command prompt, the
attached process will be killed as well as the dbx session. To exit from the dbx
session without terminating the attached process, use the detach
sub-command.
7.4.2 Customizing a dbx session
The dbx command reads in an initial configuration file, .dbxinit, in the following
order:
1. The current directory
2. User’s home directory
If the .dbxinit file is found in the current directory, the command does not read the
one in the user’s home directory. Or, you can explicitly specify the file name using
the -c option.
By defining aliases of frequently used sub-commands or macros, you can
customize your dbx session.
After starting the dbx session, you can also instruct dbx to dynamically read the
configuration file by using the source sub-command as follows:
(dbx) source /home/ausres01/.dbxinit
The following settings are useful to customize your dbx session.
Displaying more readable output of structure
Using the print sub-command, you can print the value of a variable in a dbx
session. If you have to print complex C or C++ structures or unions, it could be
difficult to read and interpret the resulting output. Use the following
sub-command to display more readable output.
Set $pretty=”on”
Use the set $pretty=”on” sub-command to make your output more readable.
Every value is print on its own line, with its scope value indented, as shown in the
following example:
(dbx) set $pretty="off"
(dbx) print names[0]
(effort = (low = (0, 0), name1 = "Fabian", name2 = "Julian", name3 = "Manuel",
high = (
(a = (0), b = 0)
Chapter 7. Debugging your applications
261

(a = (0), b = 0)
)), range = (0))
(dbx) set $pretty="on"
(dbx) print names[0]
{
effort = {
low[0] = 0
low[1] = 0
name1 = "Fabian"
name2 = "Julian"
name3 = "Manuel"
high[0] = {
a[0] = 0
b = 0
}
high[1] = {
a[0] = 0
b = 0
}
}
range[0] = 0
}
Set $pretty=”verbose”
You can also use the set $pretty=”verbose” sub-command for other printing
styles, with each value on its own line and with qualified names:
(dbx) set $pretty="verbose"
(dbx) print names[0]
effort.low[0] = 0
effort.low[1] = 0
effort.name1 = "Fabian"
effort.name2 = "Julian"
effort.name3 = "Manuel"
effort.high[0].a[0] = 0
effort.high[0].b = 0
effort.high[1].a[0] = 0
effort.high[1].b = 0
range[0] = 0
(dbx)
Command line editor mode
You can set the dbx’s sub-command line editor mode to either vi or emacs by
using the sub-command set edit [vi/emacs] or set -o [vi/emacs].
262
Developing and Porting C and C++ Applications on AIX

7.4.3 Working with breakpoints: The stop subcommand
The stop subcommand halts the application program when certain conditions are
fulfilled. The program is stopped when:
The Condition is true when the if Condition flag is used.
The Procedure is called if the in Procedure flag is used.
The Variable is changed if the Variable parameter is specified.
The SourceLine line number is reached if the at SourceLine flag is used.
7.4.4 Redirection of library location in object files with the -p flag
If you examine a core file, dbx has to resolve all references for libraries and
shared objects loaded by the application when the core file is generated. The
information is kept in the loader section of the core file. All file names except the
main executable module are treated as absolute path names. The dbx command
uses this table, not the LIBPATH environment variable.
If you bring the core file to another system to examine it, the libraries can be
missing or in another location. Starting with AIX 5L Version 5.2, the -p flag of dbx
allows you to map the old library names to the new ones. You do not need to alter
the core file.
In the following example, the ./libone.so library (listed as entry 5) is used, which is
located in the current directory:
(dbx) map
Entry 1:
Object name: prog8
Text origin: 0x10000000
...
Entry 5:
Object name: ./libone.so
Text origin: 0xd00d7000
Text length: 0x2fd
Data origin: 0xf0323210
Data length: 0x34
File descriptor: 0x8
(dbx)
If the library is no longer at its original location, you will get an error upon the dbx
session start:
$ dbx prog8 core
Type 'help' for help.
Chapter 7. Debugging your applications
263

reading symbolic information ...dbx: fatal error: 1283-012 cannot open
./libone.so
To solve this problem, use the -p option to map the missing library to its new
location, as shown in the following example:
$ dbx -p ./libone.so=./lib/libone.so prog8 core
Type 'help' for help.
[using memory image in core]
reading symbolic information ...
Segmentation fault in func8 at 0xd00d7158
0xd00d7158 (func8+0x30) 98640048 stb r3,0x48(r4)
(dbx) map
Entry 1:
Object name: prog8
...
Entry 5:
Object name: ./lib/libone.so
Text origin: 0xd00d7000
Text length: 0x2fd
Data origin: 0xf0323210
Data length: 0x34
File descriptor: 0x9
You can also put the mapping information in a file and call the -p flag as
-pfilename.
7.4.5 Using dbx with gcc
Starting with AIX 5L Version 5.2, dbx has been modified to support broad
compatibility with the GNU C compiler (gcc), so the command now can be used
to debug applications compiled with gcc.
To debug application compiled with dbx, you have to specify the -gxcoff option
when building an executable file as follows:
$ gcc -gxcoff prog.c -o prog
7.5 Debugging with the truss command
The truss command traces library and system calls and signal activity for a
process.3 It allows you to see all the system calls being made by a process, the
parameters passed by those calls, and any data or errors returned from those
3 The truss command has been supported on AIX, starting from Version 5.1. Several enhancements
have been made in AIX 5L Version 5.2.
264
Developing and Porting C and C++ Applications on AIX

calls. You can watch what your application is requesting from AIX and the results
of those requests on live.
To demonstrate how it works, we present an application, which is composed of
the four source files shown in Example 7-4.
Example 7-4 Source codes of prog_to_truss
/* main.c */
int main(int argc, char *argv[])
{
func9_1(1);
func9_3(2);
exit(0);
}
/* sample9a.c */
void func9_1(int i)
{
int offset;
offset = 10;
func9_2(i);
func9_3(i + offset);
}
/* sample9b.c */
void func9_2(int i)
{
func9_3(i);
}
/* sample9c.c */
#include <stdio.h>
#include <stdlib.h>
void func9_3(int i)
{
printf("I am here in func9_3 with %d !!\n", i);
}
We have built the application, prog_to_truss, as follows:
$ cc -G -o libs91.so sample9a.c
$ cc -G -o libs92.so sample9b.c
$ cc -G -o libs93.so sample9c.c
$ cc -brtl prog9.c -L. libs91.so libs92.so libs93.so -o prog_to_truss
The prog_to_truss application contains an object module, main.o, and references
to three run-time linking shared objects (see 2.5, “Run-time linking” on page 68).
Chapter 7. Debugging your applications
265

By default, truss does not trace library calls. In the following example, we have
specified the -t!__loadx option, which means that any calls of the system call
__loadx will be excluded from the output. Because the program calls three
functions in run-time linking shared objects, if we do not specify this option, the
output becomes very unclear for our demonstration:
$ truss -t!__loadx prog_to_truss
execve("./prog_to_truss", 0x2FF229FC, 0x2FF22A04) argc: 1
sbrk(0x00000000) = 0x2000050C
sbrk(0x00000004) = 0x2000050C
sbrk(0x00010010) = 0x20000510
kioctl(1, 22528, 0x00000000, 0x00000000) = 0
I am here in func9_3 with 1 !!
kwrite(1, 0xF01B5168, 24) = 24
I am here in func9_3 with 11 !!
kwrite(1, 0xF01B5168, 25) = 25
I am here in func9_3 with 2 !!
kwrite(1, 0xF01B5168, 24) = 24
kfcntl(1, F_GETFL, 0x00000000) = 67110914
kfcntl(2, F_GETFL, 0xF01B5168) = 67110914
_exit(0)
In the following example, we have specified the -u’*’ option to display the tracing
output of any library function calls:
$ truss -t!__loadx -u'*' prog_to_truss
execve("./prog_to_truss", 0x2FF229FC, 0x2FF22A04) argc: 1
sbrk(0x00000000) = 0x2000050C
sbrk(0x00000004) = 0x2000050C
sbrk(0x00010010) = 0x20000510
->librtl.a:func9_1(0x1)
->librtl.a:func9_3(0x1)
kioctl(1, 22528, 0x00000000, 0x00000000) = 0
I am here in func9_3 with 1 !!
kwrite(1, 0xF01B5168, 24) = 24
<-librtl.a:func9_3() = 18 0.000000
->librtl.a:func9_3(0xb)
I am here in func9_3 with 11 !!
kwrite(1, 0xF01B5168, 25) = 25
<-librtl.a:func9_3() = 19 0.000000
<-librtl.a:func9_1() = 19 0.000000
->librtl.a:func9_3(0x2)
I am here in func9_3 with 2 !!
kwrite(1, 0xF01B5168, 24) = 24
<-librtl.a:func9_3() = 18 0.000000
->librtl.a:exit(0x0)
kfcntl(1, F_GETFL, 0x00000000) = 67110914
kfcntl(2, F_GETFL, 0xF01B5168) = 67110914
_exit(0)
266
Developing and Porting C and C++ Applications on AIX

As shown in the high-lighted lines, func9_3() called three times with an integer
parameter, 1 (0x1), 11 (0xb), and 2 (0x2). Apparently, these values are passed by
the parent function, func9_2(), as shown in Example 7-4 on page 265.
For further information about the truss command and its enhancements, please
refer to the AIX 5L Differences Guide Version 5.2 Edition, SG24-5765 and the
AIX 5L Version 5.2 Reference Documentation: Commands Reference.
7.6 Using the trace facility
The trace facility is a flexible system monitoring service that supplies a stream of
events. It is up to you what information you extract. Trace provides detailed
information about system activity based on a collection of pre-defined system
events. You can add new trace hooks into your application in order to examine
the application behavior. By utilizing trace hooks, your application may be
debugged without having core files.
Please refer to the “Analyzing Performance with the Trace Facility” section of the
AIX 5L Version 5.2 Performance Management Guide for further information.
7.6.1 Introduction to trace
Events are the basics of every trace running. These events are compiled into
kernel or application code, but are only traced if tracing is active. You activate
tracing with the trace command or the trcstart() subroutine. Tracing is stopped
with the trcstop command or the trcstop() subroutine.
When tracing is active, you suspend or resume it with the trcoff and trcon
commands or with the trcoff() and trcon() subroutines.
During the tracing, the events are stored in the trace log file. The recorded events
can be selected and formatted with the trcrpt command to be a
human-readable report.
Note: Tracing may impact system performance, if many events are traced.
Trace hooks
The events monitored by trace are the trace hooks. Each hook is assigned to a
unique number. The trace hooks are defined in the /usr/include/sys/trchkid.h
header file.
You can define additional user trace hooks used for tracing your application. The
range for user defined hooks is between 0x010 and 0x0FF.
Chapter 7. Debugging your applications
267

Trace report
Once the trace events are stored in the log file, use the trcrpt command to
produce a human-readable report.
7.6.2 Tracing an application on the command line
This is the most common and useful way of tracing user defined trace hooks in
user applications.
Defining the trace hook in the application
Before starting trace, you have to modify your application code. The trace
interfaces in your application will be only active if the trace daemon is active and
is collecting data.
We use the example source code shown in Example 7-5 to demonstrate a simple
usage of trace hook in the application source code.
Example 7-5 Sample source code using trchook (prog11.c)
#include <stdlib.h>
#include <stdio.h>
int main(int argc, char *argv[])
{
unsigned int d1,d2,d3,d4,d5;
int a,b;
trchook(0x02010000,d1,d2,d3,d4,d5); /* trace hook 0x020 */
trchook(0x02110000,d1,d2,d3,d4,d5); /* trace hook 0x021 */
a = 13;
b = 3;
trchook(0x02220000,(a+b),d2,d3,d4,d5);/* trace hook 0x022 */
exit(0);
}
The trchook() subroutine has the following syntax:
void trchook(unsigned int HkWord, unsigned int d1
, unsigned int d2, unsigned int d3
, unsigned int d4, unsigned int d5);
where HkWord is an unsigned integer consisting of a hook ID (HkID), a hook type
(Hk_Type), and two bytes of data from the calling program (HkData). A hook ID is
a 12-bit value. For user programs, the hook ID may be a value from 0x010 to
0x0FF. Hook identifiers are defined in the /usr/include/sys/trchkid.h file. Hk_Type
268
Developing and Porting C and C++ Applications on AIX

is a 4-bit value that identifies the amount of trace data to be recorded (see
Figure 7-1).
HkWord
12 bit HkID
4 bit
2 Bytes HkData
Hk_Type
Value Records
1 Hookword
9 Hookword and a time stamp
2 Hookword and one data word
A Hookword, one data word and a time stamp
6 Hookwordamd up to five data words
E Hookword, up to five data words and a time stamp
Figure 7-1 Definition of HkWord
Defining trace templates
Prepare the trace report by naming the desired messages in the output by
defining your user trace hooks (in our example, hook id 020, 021, and 022).
The trace templates for the trcrpt command are stored in the /etc/trcfmt file.
Instead of using the default /etc/trcfmt file, we have made our own copy file,
mytrcfmt, so that we can refer to this file by using the -t filename option.
In our example, we have inserted the following definition entries in mytrcfmt:
020 1.0 L=APPL "USER EVENT - HKWD_USER2" O2.0 \n \
"Program runs over here"
021 1.0 L=INT "USER EVENT - HKWD_USER2" O2.0 \n \
"Simulatied Interrupt"
022 1.0 L=APPL "USER EVENT - HKWD_USER2" O2.0 \n \
"Program runs over here value:" U4
Please refer to AIX 5L Version 5.2 Files Reference for the complete definition of
trace templates.
Tracing the application
We have compiled the program as follows:
$ cc prog11.c -o prog11
Chapter 7. Debugging your applications
269

Then we start to trace it as follows:
$ trace
-> trcon
-> ! prog11
-> trcoff
-> q
When invoked, the trace command gives you the sub-command prompt “->”.
The trcon sub-command starts the collection of trace hooks. We started the
program, prog11, and stopped the collection with the trcoff sub-command.
Reporting output of the trace events
The following example shows the trace report of newly defined trace hooks:
# trcrpt -d 020,021,022 -t mytrcfmt
Mon Feb 3 17:31:25 2003
System: AIX lpar02 Node: 5
Machine: 0021768A4C00
Internet Address: 09030442 9.3.4.66
The system contains 2 cpus, of which 2 were traced.
Buffering: Kernel Heap
This is from a 32-bit kernel.
Tracing all hooks.
trace
ID ELAPSED_SEC DELTA_MSEC APPL SYSCALL KERNEL INTERRUPT
006 0.864997761 864.997761
TRACEBUFFER WRAPAROUND 0002
006 1.802352538 937.354777
TRACEBUFFER WRAPAROUND 0003
006 2.789985334 987.632796
TRACEBUFFER WRAPAROUND 0004
006 3.687272139 897.286805
TRACEBUFFER WRAPAROUND 0005
006 4.454988847 767.716708
TRACEBUFFER WRAPAROUND 0006
006 4.657498888 202.510041
TRACEBUFFER WRAPAROUND 0007
020 4.669032274*
USER EVENT - HKWD_USER2
Program runs over here
021 4.669032274*
USER EVENT - HKWD_USER2
Simulatied Interrupt
022 4.669032274*
USER EVENT - HKWD_USER2
Program runs over here value: 16
006 4.749999720 80.967446
TRACEBUFFER WRAPAROUND 0008
006 5.695142879 945.143159 TRACEBUFFER WRAPAROUND 0009
005 5.695142879* LOGFILE WRAPAROUND 0001
006 6.534979135 839.836256 TRACEBUFFER WRAPAROUND 000A
002 6.609241540 74.262405 TRACE OFF channel 0000 Mon Feb 3
17:31:33 2003
270
Developing and Porting C and C++ Applications on AIX

7.6.3 Tracing an application with subroutine calls
You can include the trace control, starting the trace, enable and resume data
collection, as well as trace stopping completely inside the application.
Defining the trace hook in the application
In the prog10.c sample, we use a predefined macro called TRCHKL1T (see
Example 7-6). The TCHKL1T macro found in /usr/include/sys/trcmacros.h is
defined to call the utrchook() subroutine, which is similar to the trchook()
subroutine used in our previous example, but used for non generic hook events.
The utrchook() subroutine is called by the macro with hook ID 0x010 and HType
2.
In Example 7-6, instead of using trcon() and trcoff() subroutines, we use ioctl()
calls for the trcon and trcoff functions. trcon() and trcoff() subroutines perform the
following tasks:
1. Opens the trace control device (/dev/systrctl).
2. Issues the appropriate ioctl() subroutine.
3. Closes the control device.
4. Returns to the calling program.
So, using the ioctl()-calls directly tunes the trace, because we do not need
additional I/O and system calls.
Example 7-6 Sample source code using TRCHKL1T (prog10.c)
#include <stdlib.h>
#include <stdio.h>
#include <sys/trcctl.h>
#include <sys/trcmacros.h>
#include <sys/trchkid.h>
#include <sys/trchkid.h>
char *ctl_file = "/dev/systrctl";
int ctlfd;
int i;
main(int argc, char *argv[])
{
printf("Configuring trace collection \n");
if (trcstart("-ad")) {
perror("trcstart");
exit(1);
}
printf("Opening the trace device \n");
Chapter 7. Debugging your applications
271

if ((ctlfd = open(ctl_file, 0)) < 0) {
perror(ctl_file);
exit(1);
}
printf("Turning trace on \n");
if (ioctl(ctlfd, TRCON, 0)) {
perror("TRCON");
exit(1);
}
for (i=1;i <6; i++) {
TRCHKL1T(HKWD_USER1, i);
}
printf("Turning trace off\n");
if (ioctl(ctlfd, TRCSTOP, 0)) {
perror("TRCOFF");
exit(1);
}
printf("Stopping the trace daemon \n");
if (trcstop(0)) {
perror("trcstop");
exit(1);
}
exit(0);
}
We have compiled the application as follows:
$ cc prog10.c -o prog10 -lrts
Tracing the application
When the application starts, it automatically starts and stops the trace:
$ prog10
Configuring trace collection
Opening the trace device
Turning trace on
Turning trace off
Stopping the trace daemon
Reporting the output of the trace events
After defining the template for hook ID 0x010, we can generate the trace report,
as shown in Example 7-7 on page 273.
272
Developing and Porting C and C++ Applications on AIX

Example 7-7 Trace report for the hook ID 010
$ trcrpt -d 010 -t mytrcfmt
Mon Feb 3 18:20:06 2003
System: AIX lpar02 Node: 5
Machine: 0021768A4C00
Internet Address: 09030442 9.3.4.66
The system contains 2 cpus, of which 2 were traced.
Buffering: Kernel Heap
This is from a 32-bit kernel.
Tracing all hooks.
/usr/sbin/trace -ad
ID ELAPSED_SEC DELTA_MSEC APPL SYSCALL KERNEL INTERRUPT
001 0.000000000 0.000000 TRACE ON channel 0
Mon Feb 3 18:20:06 2003
010 0.000004057 0.004057 USER EVENT - HKWD_USER1
The # of loop iterations =1
The elapsed time of the last loop =
010 0.000005344 0.001287 USER EVENT - HKWD_USER1
The # of loop iterations =2
The elapsed time of the last loop = [1 usec]
010 0.000006565 0.001221 USER EVENT - HKWD_USER1
The # of loop iterations =3
The elapsed time of the last loop = [1 usec]
010 0.000007785 0.001220 USER EVENT - HKWD_USER1
The # of loop iterations =4
The elapsed time of the last loop = [1 usec]
010 0.000009020 0.001235 USER EVENT - HKWD_USER1
The # of loop iterations =5
The elapsed time of the last loop = [1 usec]
002 0.000043623 0.034603
TRACE OFF channel 0000 Mon Feb 3 18:20:06
2003
For more information about the trace facility, please refer to the “Trace Facility”
section of AIX 5L Version 5.2 General Programming Concepts: Writing and
Debugging Programs and the “Analyzing Performance with the Trace Facility”
section of AIX 5L Version 5.2 Performance Management Guide. See the AIX 5L
Version 5.2 Files Reference for the trcfmt file format.
Chapter 7. Debugging your applications
273

274
Developing and Porting C and C++ Applications on AIX

8
Introduction to POSIX
Chapter 8.
threads
This chapter gives you an introduction to parallel programming on AIX using
POSIX threads by providing several example programs that explain the following
concepts:
Creating and joining POSIX threads
Mutexes (mutual exclusive locks)
Condition variables
Read-write locks
Thread-specific data
Thread scheduling
Environment variables
For further information about POSIX thread programming on AIX, please refer to
the following sections in AIX 5L Version 5.2 General Programming Concepts:
Writing and Debugging Programs:
“Parallel Programming”
“Programming on Multiprocessor Systems”
“Threads Programming Guidelines”
© Copyright IBM Corp. 2000, 2003. All rights reserved.
275

8.1 Overview of threads
A thread is a single, sequential flow of control within a process. Within each
thread, there is a single point of execution. On most UNIX operating systems,
including AIX, a thread possesses the following characteristics:
It has its own independent flow of control within a process.
It shares resources with other threads within a process.
It can voluntary terminate before the process termination, or all the threads
within a process terminate when the process terminates.
Most traditional programs execute as a process with a single thread. From a
programmer’s point of view, a thread can be considered as a procedure that can
concurrently and independently run from the main processing flow.
8.1.1 Relationship between a process and a user thread
A process is a running program instance, which contains the following resources:
Address space, including program text, shared library, global data, and so on.
File descriptors
Signal handlers
Environment (variable) and working directory
Set of identifiers (PID, GID, and so on)
Inter-process communication (IPC) facilities, such as message queues, pipes,
semaphores, and shared memory segments.
Multiple user threads can exist within a process and use these process
resources, yet are able to be scheduled by the operating system and run as
independent entities within a process. A user thread can posses an independent
flow of control and can be scheduled because it maintains its own resources:
Priority
Program counter, stack pointer, and other registers
Stack
Scheduling priorities
Signal mask
Thread ID
errno1
1 See “Use of errno in multi-threaded programming” on page 283.
276
Developing and Porting C and C++ Applications on AIX

Figure 8-1 illustrates the relationship between a process and two user threads.
User process address space
Program text
main()
User process resource
{
...
...
func1(a, b, c);
...
func2(d, e, f);
...
}
pthread_create()
User thread 1
func1()
a, b, c
func2()
Thread (1) resource
User thread 2
d, e, f
Thread (2) resource
Process heap
Shared library
text and data
Figure 8-1 Two user threads in a process
A process can have multiple user threads, all of which share the resources within
a process and all of which execute within the same address space. At any given
point in time, there are multiple points of execution within a multi-threaded
process.
Because user threads within the same process share resources, changes made
by one user thread to a process resource will be transparently seen by all other
user threads. Also, because multiple user threads can read and write the same
memory locations within a process address space, the synchronization of data
must be explicitly taken among these threads.
The cost to create or manage user threads is relatively cheap compared with
processes in terms of CPU cycles. Also, the creation of a user thread requires a
very small amount of memory compared with a process creation. With careful
design and coding, the use of user threads gives programmers the ability to write
multi-threaded applications that run on both uni- and multi-processor systems,
taking advantage of the additional processors on SMP systems. Additionally,
multi-threaded applications can increase performance even in a uniprocessor
Chapter 8. Introduction to POSIX threads
277

environment when the application performs operations that are likely to block or
cause delays, such as file or socket I/O.
8.2 POSIX threads (Pthreads) on AIX
AIX supports the following standards:
The Single UNIX Specification Version 2
The specification includes the POSIX thread standard, known as the IEEE
POSIX 1003.1c standard.2 Thread implementations that conform to this
standard are referred to as POSIX threads or Pthreads.3
The Open Group UNIX 98 specification4
The specification defines many additional functions on the former UNIX 95
specification, as well as extended threads functions over POSIX threads,
based on industry input from major UNIX vendors.
On AIX, POSIX threads are defined as a set of C language programming types
and sub-routine calls, implemented with a header file (/usr/include/pthread.h) and
the POSIX thread library (/usr/lib/libpthreads.a).
AIX provides binary compatibility for existing multi-threaded applications that
were written for the draft of Version 7 of the POSIX threads standard. The
compatibility POSIX thread library, /usr/lib/libpthreads_compat.a, is only provided
for backward compatibility for those applications. The compatibility POSIX thread
library supports 32-bit applications only.
Both libraries are included in the bos.rte.libpthreads fileset, which is installed by
default, as shown in the following example:
# lslpp -w /usr/lib/libpthreads*.a
File Fileset Type
----------------------------------------------------------------------------
/usr/lib/libpthreads.a bos.rte.libpthreads Symlink
/usr/lib/libpthreads_compat.a bos.rte.libpthreads Symlink
2 AIX has been supporting the IEEE POSIX 1003.1c standard since Version 4.3. Later, it was
included in the Single UNIX Specification Version 2, which AIX has been supporting since Version
5.1.
3 Currently, AIX does not support the POSIX thread option real-time extension.
4 AIX has been supporting the UNIX 98 specification since Version 4.3.
278
Developing and Porting C and C++ Applications on AIX

8.2.1 Advantages of using Pthreads
The following are the advantages of using Pthreads:
The primary purpose of using Pthreads is to realize potential performance
gains. POSIX threads can be created and managed with much less operating
system overhead and system resources than creating and managing a
process.
Inter-thread communication is more efficient and, in many cases, easier to
use than inter-process communication.
Multi-threaded applications offer potential performance gains over
non-threaded applications in several other ways:
– Overlapping CPU work with I/O: For example, a program may have
sections where it is performing a long I/O operation. While one Pthread is
waiting for an I/O system call to complete, CPU intensive work can be
performed by other Pthreads.
– Priority/Real-time scheduling: High priority jobs can be scheduled to
supersede or interrupt lower priority jobs.
– Asynchronous event handling: Jobs that service events of indeterminate
frequency and duration can be interleaved. For example, a Web server
can both transfer data from previous requests and manage the arrival of
new requests.
POSIX threads provide the infrastructure to parallelize applications using
OpenMP, as explained in Chapter 9, “Program parallelization using OpenMP”
on page 335.
8.2.2 The POSIX threads API
The POSIX threads API is defined in the ANSI/IEEE POSIX 1003.1 standard. All
identifiers in the Pthreads library begin with pthread_. The POSIX threads API
contains over 60 subroutines. Brief descriptions about the POSIX threads
sub-routines supported on AIX are provided in Appendix D, “Subroutine
references for POSIX threads” on page 473.
Naming conventions
All the POSIX threads sub-routines are categorized into several groups
distinguished by the following function name prefixes:
pthread_
Threads and miscellaneous subroutines (see Table D-1
on page 474 and Table D-6 on page 481)
pthread_attr_
Thread attributes objects (see Table D-2 on page 476)
pthread_mutex_
Managing mutexes (see Table D-3 on page 477)
Chapter 8. Introduction to POSIX threads
279

pthread_mutexattr_ mutex attribute objects (see Table D-3 on page 477)
pthread_cond_
Condition variables (see Table D-3 on page 477)
pthread_condattr_
Condition attribute objects (see Table D-3 on page 477)
pthread_rwlock_
Read/Write lock objects (see Table D-5 on page 480)
pthread_rwlockattr_ Read/Write lock attributes (see Table D-5 on page 480)
pthread_key_
Thread-specific data keys (see Table D-4 on page 479)
8.2.3 Multi- and single-threaded processes
On AIX, the scheduling entity is a kernel thread. A single kernel thread can be
mapped to multiple user threads (Pthreads) within a process (detailed
information about the relationship between kernel threads and user threads is
provided in 8.7.1, “Thread models in AIX” on page 322).
In order to illustrate how multi-threaded applications run on AIX, we have
excerpted several lines from the ps -emo THREAD command output, as shown in
Example 8-1 on page 281. The -o THREAD option instructs the ps command to
display thread-level information; without this option, the command displays
process-level information only.
Each column in the output represents the following:
USER
The login name of the process owner.
PID
The process ID of the process.
PPID
The process ID of the parent process.
TID
The thread ID of the kernel thread.
S
The state of the process or kernel thread.
C
The CPU utilization of the process or kernel thread.
PRI
The priority of the process or kernel thread.
SC
The suspend count of the process or kernel thread.
WCHAN
The wait channel of the process or kernel thread.
FLAG
The flags of the process or kernel thread.
TTY
The controlling terminal of the process.
BND
The CPU to which the process or kernel thread is bound.
CMD
The command being executed by the process.
280
Developing and Porting C and C++ Applications on AIX

Example 8-1 ps -emo THREAD
# ps -emo THREAD
USER PID PPID TID ST CP PRI SC WCHAN F TT BND COMMAND
root 1 0 - A 0 60 1 - 200003 - - /etc/init
- - - 259 S 0 60 1 - 410410 - - -
... (some lines are omitted) ...
root 10432 13686 - A 0 60 11 f015ab98 240001 - - /usr/sbin/rpc.mountd
- - - 19881 S 0 60 1 f015ab98 c10400 - - -
- - - 23811 Z 0 60 1 - c00001 - - -
- - - 49449 Z 0 60 1 - c00001 - - -
- - - 53961 Z 0 60 1 - c00001 - - -
- - - 54215 Z 0 60 1 - c00001 - - -
- - - 55603 S 0 60 1 - 418400 - - -
- - - 60085 Z 0 60 1 - c00001 - - -
- - - 60705 Z 0 60 1 - c00001 - - -
- - - 62039 Z 0 60 1 - c00001 - - -
- - - 63481 Z 0 60 1 - c00001 - - -
- - - 63955 Z 0 60 1 - c00001 - - -
... (some lines are omitted) ...
root 16266 13686 - A 0 60 1 c0042100 240001 - - /usr/sbin/qdaemon
- - - 22197 S 0 60 1 c0042100 10400 - - -
... (rest of lines are omitted) ...
In this example, the rpc.mountd daemon process (PID 10432) is a multi-threaded
process, since it contains 11 lines underneath the highlighted line, while the
qdaemon daemon process (PID 16266) is a single-threaded process, since it has
only one line that represents the kernel thread for the process’s initial thread.
Two kernel threads (TID 19881 and 55603) of PID 10432 shown the status S,
which means the kernel thread is in the sleeping status. Other kernel threads did
not exist at the time the ps command was invoked, though TIDs were displayed.
the process could be verified by inspecting the /proc file system.5 The /proc file
system is a system interface to represent process information. The process
information for the process <PID> is shown as several files and sub-directories
under the /proc/<PID> directory.
As shown in Example 8-2 on page 282, there were only two sub-directories,
19881 and 55603, under the /proc/10432/lwp directory6 at the moment the ps
command was invoked. The sub-directory names, 19881 and 55603, were the
same TIDs for the PID 10432 in Example 8-1.
5 AIX has been supporting the /proc file system since AIX 5L Version 5.1.
6 LWP stands for light-weight process. In the /proc file system semantic, kernel threads are
represented as LWPs.
Chapter 8. Introduction to POSIX threads
281

Example 8-2 Inspecting the /proc file system
# ls /proc/10432
as ctl fd/ map psinfo status
cred cwd@ lwp/ object/ sigact sysent
# ls /proc/10432/lwp
19881/ 55603/
# ls -lR /proc/10432/lwp
total 0
dr-xr-xr-x 1 root system 0 Feb 21 18:41 19881/
dr-xr-xr-x 1 root system 0 Feb 21 18:41 55603/
/proc/10432/lwp/19881:
total 0
--w------- 1 root system 0 Feb 21 18:41 lwpctl
-r--r--r-- 1 root system 120 Feb 21 18:41 lwpsinfo
-r-------- 1 root system 1200 Feb 21 18:41 lwpstatus
/proc/10432/lwp/55603:
total 0
--w------- 1 root system 0 Feb 21 18:41 lwpctl
-r--r--r-- 1 root system 120 Feb 21 18:41 lwpsinfo
-r-------- 1 root system 1200 Feb 21 18:41 lwpstatus
For further information about the /proc file system, please refer to the AIX 5L
Version 5.2 Files Reference.
Although the rpc.mountd daemon process (PID 10432) was easily determined to
be a multi-threaded process in Example 8-1 on page 281, in order to determine if
the qdaemon daemon process (PID 16266) is non-threaded, use the ldd7
command against the executable file, as shown in Example 8-3. Because
qdaemon does not depend on the POSIX thread library, we are sure that it is not
a multi-threaded program.
Example 8-3 Inspecting an executable file using ldd
# ldd /usr/sbin/qdaemon
/usr/sbin/qdaemon needs:
/usr/lib/libc.a(shr.o)
/usr/lib/libqb.a(shr.o)
/unix
/usr/lib/libcrypt.a(shr.o)
Initial thread
When a process is created, one user thread is automatically created. This user
thread is called the initial thread. It ensures the compatibility between the
7 The ldd command is supported on AIX 5L Version 5.2 and later. On earlier versions of AIX, use the
dump -H command.
282
Developing and Porting C and C++ Applications on AIX

non-threaded processes with a unique implicit Pthread and the multi-threaded
processes. The initial thread has some special properties, not visible to the
programmer, that ensure binary compatibility between the non-threaded
processes and the multi-threaded operating system. It is also the initial thread
that executes the main function in multi-threaded programs.
Use of errno in multi-threaded programming
In the multi-process UNIX programming semantic, a system call or sub-routine
would set a non-zero value to the global variable errno in case of failure. This is
still true in multi-threaded programming, though there is a subtle difference.
Within a multi-threaded process, each Pthread has its own errno to avoid being
overwritten by the other Pthreads. On AIX, the Pthread-basis errno is
implemented as a macro to a function pointer to an internal function, as shown in
Example 8-4, which is excerpted from /usr/include/errno.h.8 In the multi-threaded
programming on AIX, the _THREAD_SAFE macro is always defined.
Example 8-4 Definition of errno
#if defined(_THREAD_SAFE) || defined(_THREAD_SAFE_ERRNO)
/*
* Per thread errno is provided by the threads provider. Both the extern int
* and the per thread value must be maintained by the threads library.
*/
extern int *_Errno( void );
#define errno (*_Errno())
#else
extern int errno;
#endif /* _THREAD_SAFE || _THREAD_SAFE_ERRNO */
Although the process level global symbol errno is still accessible from Pthreads
within a process, references to it is unreliable and useless in the multi-threaded
programming.
Therefore, all the multi-threaded applications that reference to errno must have
the following directive:
#include <errno.h>
and must not have the following declaration:
extern int errno;
8 In the POSIX thread standard, the Pthread-basis errno is defined as implementation-dependent.
Chapter 8. Introduction to POSIX threads
283

Note: Avoid directly referencing the internal function, because this is
implementation-dependent and may be subject to change in future versions of
AIX. Use the errno macro to assure the portability of your multi-threaded
applications.
8.3 Pthread management
This section explains some of the basic Pthread management routines used for
creating, joining, exiting, and detaching Pthreads.
8.3.1 Creating and terminating Pthreads
In order to create new Pthreads (besides the initial thread), the program must call
the pthread_create() sub-routine. Example 8-5 is our first multi-threaded sample.
Example 8-5 create_5threads.c
#include <pthread.h>
/* #A */
#include <stdio.h>
#include <stdlib.h>
#define NUM_OF_THREADS 6
#define EXIT_CODE -1
void *thr_func(void *id)
{
int j;
for (j = 0; j < 500000; j++) {
;
/* #B */
}
printf("Hello world from Pthread %d!\n", (int *)id);
pthread_exit(NULL);
/* #C */
}
int main (int argc, char *argv[])
{
int rc, i;
pthread_t tid[NUM_OF_THREADS];
for (i = 1; i < NUM_OF_THREADS; i++) {
printf("Creating Pthread: %d\n", i);
rc = pthread_create(&tid[i], NULL, thr_func, (void *)i);
/* #D */
if (rc != 0) {
fprintf(stderr
, "pthread_create() failed with rc = %d at %d in %s.\n"
284
Developing and Porting C and C++ Applications on AIX

, rc, __LINE__, __FILE__);
exit(EXIT_CODE);
}
}
pthread_exit(NULL);
}
This program creates five Pthreads, as illustrated in Figure 8-2.
Created Pthreads
Process creation
with pthread_create()
i = 1
i = 2
i = 3
i = 4
i = 5
executing
thr_func()
Initial thread
Process termination
Pthread termination with pthread_exit()
Figure 8-2 Five Pthreads created by pthread_create()
The highlighted lines with comments #A, #B, #C, and #D in Example 8-5 on
page 284 explain the following important programming manners when
developing multi-threaded applications:
Comment #A
To use the Pthread subroutines, the pthread.h header file must be included as
the first header file in the each source file using the Pthreads library. This is
because it defines some important macros that affect other system header
files. Having the pthread.h header file as the first included file ensures the
usage of thread-safe subroutines.
Chapter 8. Introduction to POSIX threads
285

Comment #B
A Pthread’s life-cycle can be very short; it may terminate just after it is
created. In this example, a delay loop commented as #B is necessary to
demonstrate that created Pthreads run in parallel. If this loop is removed or
the program is compiled with optimizing options, all the created Pthreads
could terminate before other Pthreads could be created.
The programmer must be aware of this volatile nature of Pthreads.
Comment #C
If a Pthread calls pthread_exit(), it terminates; this marks the end of Pthread’s
life-cycle. Thread resources for the Pthread will be freed after the termination.
However, other resources created by the terminating Pthread, such as file
descriptors and sockets, will not be freed.
Comment #D
Most Pthreads library sub-routines return an integer value, which is
interpreted as an error code. A value of zero indicates that the call was
successful. Other values are passed or retrieved through appropriately typed
arguments.
Also, the initial thread executing the main() function could have executed other
functions after creating the other Pthreads. Therefore, there were six
independent control flows in the process, as depicted in Figure 8-2 on page 285.
Compiling multi-threaded programs
To compile multi-threaded programs, use one of the compiler drivers with the _r
suffix (see Table 1-4 on page 30). To compile the program shown in Example 8-5
on page 284, we used cc_r, as shown in the following example:
$ cc_r create_5threads.c
Note: The POSIX thread library is automatically linked when programs are
compiled with _r compiler drivers. Therefore, the -lpthreads linker option is not
required.
The created executable file, a.out, is a 32-bit multi-threaded program, as shown
in the following:
$ file a.out
a.out: executable (RISC System/6000) or object module not stripped
$ ldd a.out
a.out needs:
/usr/lib/threads/libc.a(shr.o)
/usr/lib/libpthreads.a(shr_comm.o)
/usr/lib/libpthreads.a(shr_xpg5.o)
/unix
286
Developing and Porting C and C++ Applications on AIX

/usr/lib/libcrypt.a(shr.o)
Running multi-threaded programs
We ran this application on an AIX 5L Version 5.2 partition on the pSeries 690,
which is assigned two processors:
# lsdev -Cc processor
proc0 Available 00-00 Processor
proc1 Available 00-01 Processor
At the first invocation of this program, it produced the following output:
$ a.out
Creating Pthread: 1
Creating Pthread: 2
Creating Pthread: 3
Hello world from Pthread 1!
Hello world from Pthread 2!
Hello world from Pthread 3!
Creating Pthread: 4
Creating Pthread: 5
Hello world from Pthread 4!
Hello world from Pthread 5!
At the second invocation of this program, it produced the following output:
$ a.out
Creating Pthread: 1
Creating Pthread: 2
Creating Pthread: 3
Creating Pthread: 4
Creating Pthread: 5
Hello world from Pthread 2!
Hello world from Pthread 4!
Hello world from Pthread 5!
Hello world from Pthread 1!
Hello world from Pthread 3!
Apparently, the execution order of Pthreads is different between these two
outputs. In fact, it could be any order and you cannot predict which Pthread is
scheduled first in a multi-threaded process.
Chapter 8. Introduction to POSIX threads
287

Passing arguments to Pthreads
We discuss several aspects of the pthread_create() sub-routine in the following
sections:
Syntax
#include <pthread.h>
int pthread_create (thread, attr, start_routine (void), arg)
pthread_t
*thread;
const
pthread_attr_t *attr;
void
**start_routine (void);
void
*arg;
Parameters
thread
Points to where the thread ID will be stored.
attr
Specifies the thread attributes object to use in creating the
thread. If the value is NULL, the default attributes values
will be used.
start_routine
Points to the routine to be executed by the thread.
arg
Points to the single argument to be passed to the
start_routine routine.
Return values
If successful, the pthread_create function returns zero. Otherwise, an error
number is returned to indicate the error.
Error codes
The pthread_create function will fail if:
EAGAIN
If WLM9 is running, the limit on the number of threads in
the class may have been met.
EINVAL
The value specified by attr is invalid.
EPERM
The caller does not have the appropriate permission to
set the required scheduling parameters or scheduling
policy.
To pass multiple arguments the start_routine, create a structure that contains all
of the arguments, and then pass a pointer to the structure as the last parameter
of the pthread_create() routine.
9 Work Load Manager
288
Developing and Porting C and C++ Applications on AIX

For example, assuming that the following structure is defined:
struct thread_data {
int thread_id;
char *msg;
} thread_data_array[NUM_OF_THREADS];
then pass the structure to the pthread_create() routine as follows:
pthread_create(&tid[t], NULL, thr_func, (void *)&thread_data_array[i]);
8.3.2 Joining Pthreads
The pthread_join() subroutine will block the calling Pthread until the Pthread you
specify has terminated, and then, optionally, store the terminated Pthread’s
return value. Calling pthread_join() detaches the specified Pthread automatically.
When a Pthread is created, one of its attributes defines whether it is joinable or
detached. Detachable means that it cannot be joined. To explicitly create a
Pthread as joinable or detached, the second argument in the pthread_create()
routine is used. The pthread_detach() routine can be used to explicitly detach a
Pthread even though it was created as joinable. The following example
demonstrates how to wait for Pthread completions by using the pthread_join()
routine. The Pthreads in this example are explicitly created in a joinable state so
that they can be joined later:
/* File: pthread-join.c*/
#include <pthread.h>
#include <stdio.h>
#include <stdlib.h>
#define NUM_OF_THREADS 3
#define EXIT_CODE -1
void *Func_Join(void *t)
{
printf("The argument is %d\n", t);
pthread_exit((void *) 0);
}
int main(int argc, char *argv[])
{
pthread_t thread[NUM_OF_THREADS];
pthread_attr_t attr;
int rc, t, status;
/* Initialize and set thread detached attribute */
pthread_attr_init(&attr);
/*Creates a thread as Joinable*/
pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_JOINABLE);
Chapter 8. Introduction to POSIX threads
289

for (t = 0; t < NUM_OF_THREADS; t++) {
printf("Creating thread %d\n", (t+1));
rc = pthread_create(&thread[t], &attr, Func_Join, (void*)t);
if (rc) {
fprintf(stderr, "pthread_create() failed with rc = %d.\n", rc);
exit(EXIT_CODE);
}
}
/* Free attribute and wait for the other threads */
pthread_attr_destroy(&attr);
for (t = 0; t < NUM_OF_THREADS; t++) {
rc = pthread_join(thread[t], (void **)&status);
if (rc) {
fprintf(stderr, "pthread_join() failed with rc = %d.\n", rc);
exit(EXIT_CODE);
}
printf("Joining the thread %d \n",(t+1));
}
pthread_exit(NULL);
}
When executed, the above program printed the following output on our system:
Creating thread 1
Creating thread 2
The argument is 0
Creating thread 3
The argument is 1
Joining the thread 1
The argument is 2
Joining the thread 2
Joining the thread 3
As shown in the example above, to explicitly create a Pthread as joinable, the
following steps are followed:
1. Declare a Pthread attribute variable of the pthread_attr_t data type.
2. Initialize the attribute variable with the pthread_attr_init() routine.
3. Set the attribute detached status (PTHREAD_CREATE_JOINABLE) with the
pthread_attr_setdetachstate() routine (by default, the Pthread is joinable).
4. When done, free the library resources used by the attribute with the
pthread_attr_destroy() routine.
290
Developing and Porting C and C++ Applications on AIX

The following factors need to be considered when deciding whether a Pthread
has to be joined or not:
If a Pthread requires joining, consider (as shown above) creating it as
joinable. This provides portability, as not all implementations may create
Pthreads as joinable by default.
If you know in advance that a Pthread will never need to join with another
Pthread, consider creating it in a detached state. Some system resources
may be able to be freed.
8.3.3 Detaching a Pthread
We have seen how Pthreads can be joined using the pthread_join() function. In
fact, Pthreads that are in a joinable state must be joined by other Pthreads, or
else their memory resources will not be fully cleaned out. This is similar to what
happens with processes whose parents did not clean up after them (also called
orphan or zombie processes).
If we have a Pthread that we wish would exit whenever it wants without the need
to join it, we should put it in the detached state. This can be done either with an
thread attribute object with a PTHREAD_CREATE_DETACHED attached to the
pthread_create() function, or by using the pthread_detach() function.
The pthread_detach() function gets one parameter, of type pthread_t, that
denotes the Pthread we wish to put in the detached state. For example, we can
create a Pthread and immediately detach it with code similar to the following
example:
pthread_t a_thread;
/* store the thread's structure here. */
int rc;
/* return value for pthread functions. */
extern void* thread_loop(void*); /* declare the thread's main function. */
/* create a new thread. if succeeded, detach the newly created thread. */
if ((rc = pthread_create(&a_thread, NULL, thread_loop, NULL)) == 0) {
rc = pthread_detach(a_thread);
}
By default, the Pthread that is created using the pthread_create() routine is
joinable (the detach state is set to PTHREAD_CREATE_JOINABLE) and the
scope is set to the process level (PTHREAD_SCOPE_PROCESS), which means
that the user Pthread is not bound to a particular kernel thread. This is often
referred to as the M:N thread model, where M is the number of user threads and
N is the number of kernel threads.
Chapter 8. Introduction to POSIX threads
291

8.3.4 Thread stack
When a Pthread is created, its own thread stack is also created. All auto storage
class variables for the Pthread are allocated in its thread stack. When the
Pthread terminates, all data allocated in its thread stack will be unallocated.
On AIX, thread stacks (except for the initial thread’s stack) are created in the
process heap. The thread stack of the initial thread is created when the process
is created.10
By default, a thread stack size is 96 KB in the 32-bit user process model while it
is 192 KB in 64-bit. The thread stack size can be tuned through either
environment variables or through function calls, as explained in “Changing a
thread stack size” on page 295. At the end of a thread stack, there is a 4 KB
read/write protected page referred to as a guard page or red zone (see
“AIXTHREAD_GUARDPAGES” on page 331 for further information about the
guard page).
Figure 8-3 on page 293 illustrates how multiple thread stacks are allocated in the
process heap in the 32-bit default memory model. The default thread stack size
of 96 KB plus additional memory for the several necessary data structures will be
consumed from the process heap every time a new Pthread is created. If a
Pthread terminates, then the memory area for the terminated Pthread can be
used again if another Pthread is created.
10 For 32-bit process processes see 3.2.6, “Resource limits in 32-bit model” on page 125; for 64-bit
processes, see 3.3.8, “Resource limits in 64-bit mode” on page 136.
292
Developing and Porting C and C++ Applications on AIX

Process private segment (0x2)
Initialized data
0 MB (offset)
Un-initialized data
p
pThread#1
misc. data, page alignment
pThread#2
pThread attribute
pThread#3
s heas
thread stack (96KB)
Guard page (4KB)
pThread#N
Proce
misc. data, page alignment
Initial thread stack
256 MB (offset)
: Accessible virtual memory pages
: Access-prohibited virtual memory pages
Figure 8-3 Thread stacks (default 32-bit process model)
Because the creation of Pthreads consumes the process heap, depending on
number of Pthreads in a process, you may need to consider the large or very
large memory model,11 when developing 32-bit multi-threaded applications.
When developing 64-bit multi-threaded applications, the data resource limit value
only affects the maximum number of Pthreads within a process.12
To demonstrate how many Pthreads can be created in the 32-bit user process
model, we have prepared a short example program, which is shown in
Example 8-6 on page 294.
To compile the program, do the following:
$ cc_r -D_LARGE_THREADS max_threads.c
If the _LARGE_THREADS macro is defined, the PTHREAD_THREADS_MAX
compilation time symbolic constant macro, which defines the maximum number
of Pthreads per process, is set to 32767. If not defined,
PTHREAD_THREADS_MAX is set to 512, which is a sufficient number for most
multi-threaded applications, regardless of 32- or 64-bit applications.
11 See 3.2.2, “Large memory model” on page 116 and 3.2.3, “Very large memory model” on
page 117.
12 See 3.3.8, “Resource limits in 64-bit mode” on page 136.
Chapter 8. Introduction to POSIX threads
293

When executed, the program would print the number of Pthreads in the angle
brackets (highlighted in the example) within the process after failing to create
another Pthread, as shown in the following example:13
$ a.out
[ 1131]: pthread_create() failed with rc=11 at 21 in max_threads.c
$ for i in 1 2 3 4 5 6 7 8
do
echo "LDR_CNTRL=MAXDATA=0x${i}0000000"
LDR_CNTRL=MAXDATA=0x${i}0000000 a.out
done
LDR_CNTRL=MAXDATA=0x10000000
[ 2281]: pthread_create() failed with rc=11 at 21 in max_threads.c
LDR_CNTRL=MAXDATA=0x20000000
[ 4583]: pthread_create() failed with rc=11 at 21 in max_threads.c
LDR_CNTRL=MAXDATA=0x30000000
[ 6884]: pthread_create() failed with rc=11 at 21 in max_threads.c
LDR_CNTRL=MAXDATA=0x40000000
[ 9186]: pthread_create() failed with rc=11 at 21 in max_threads.c
LDR_CNTRL=MAXDATA=0x50000000
[11268]: pthread_create() failed with rc=11 at 21 in max_threads.c
LDR_CNTRL=MAXDATA=0x60000000
[13788]: pthread_create() failed with rc=11 at 21 in max_threads.c
LDR_CNTRL=MAXDATA=0x70000000
[15523]: pthread_create() failed with rc=11 at 21 in max_threads.c
LDR_CNTRL=MAXDATA=0x80000000
[18391]: pthread_create() failed with rc=11 at 21 in max_threads.c
Note: The numbers shown in this example should be considered theoretical
maximum values. If a process allocates memory objects from its heap, uses
many initialized or un-initialized data, or increases thread stack sizes, these
numbers could be much smaller ones.
Example 8-6 max_threads.c
#include <pthread.h>
#include <stdio.h>
#include <stdlib.h>
#include <errno.h>
void *thread_func(void *j)
{
sleep(60);
pthread_exit(0);
}
int main()
13 The return code 11 means EAGAIN (resource temporary unavailable).
294
Developing and Porting C and C++ Applications on AIX

{
int i, rc;
pthread_t tid[PTHREAD_THREADS_MAX];
for (i = 0; i < PTHREAD_THREADS_MAX; i++) {
if ((rc = pthread_create(&tid[i], NULL, thread_func, (void *)i)) != 0){
fprintf(stderr
, "[%5d]: pthread_create() failed with rc=%d at %d in %s\n"
, i, rc, __LINE__, __FILE__);
exit(1);
}
}
for (i = 0; i < PTHREAD_THREADS_MAX; i++) {
pthread_join(tid[i], NULL);
}
}
Changing a thread stack size
On AIX, the minimum thread stack size for a Pthread is 8 KB and the maximum
size is 256 MB. These values are defined by the compilation time symbolic
constants PTHREAD_STACK_MIN and PTHREAD_STACK_MAX, respectively.
Although the default stack sizes of 96 KB for 32-bit and 192 KB for 64-bit
applications suffice for most multi-threaded applications’ demand, the stack size
can be changed using the two methods described in the next two sections.
Setting the AIXTHREAD_STKSIZE environment value
If this environment value is set before the execution of a multi-threaded
application, then the default thread stack size of the application process will be
set to the value specified by the environment value (see “AIXTHREAD_STK” on
page 332). This is a process-basis setting.
Using the pthread_attr_setstacksize() sub-routine
By creating a thread attribute with the specified thread size before the creation of
a Pthread, the new Pthread will have the specified thread size. The following
pseudo-code explains how to use this sub-routine:
1. Create a thread attribute using pthread_attr_init().
2. Set the stack size in the thread attribute using pthread_attr_setstacksize().
3. Create a new Pthread using pthread_create() with the thread attribute.
This is a Pthread-basis setting and overrides the effect of the
AIXTHREAD_STKSIZE environment value.
Chapter 8. Introduction to POSIX threads
295

8.4 Data synchronization between Pthreads
Synchronization is a programming method that allows multiple Pthreads to
coordinate their data accesses, therefore avoiding the situation where one
Pthread can change a piece of data at the same time another one is reading or
writing the same piece of data. This situation is commonly called a race
condition.
Consider, for example, a single counter, X, that is incremented by two Pthreads,
A and B. If X is originally 1, then by the time Pthreads A and B increment the
counter, X should be 3. Both Pthreads are independent entities and have no
synchronization between them. If both Pthreads are executed concurrently on
two CPUs, or if the scheduling makes the Pthreads alternatively execute on each
instruction, the following steps may occur:
Pthread A executes the first instruction and puts X, which is 1, into the
Pthread A register. Then, Pthread B executes and puts X, which is 1, into the
Pthread B register.
Next, Pthread A executes the second instruction and increments the content
of its register to 2. Then, Pthread B increments its register to 2. Nothing is
moved to memory X, so memory X stays the same.
Last, Pthread A moves the content of its register, which is now 2, into memory
X. Then, Pthread B moves the content of its register, which is also 2, into
memory X, overwriting Pthread A's value.
Note that, in most cases, Pthread A and Pthread B will execute the three
instructions one after the other, and the result would be 3, as expected. Race
conditions are usually difficult to discover because they occur intermittently.
To avoid this race condition, each Pthread should lock the data before accessing
the counter and updating memory X. For example, if Pthread A takes a lock and
updates the counter, it leaves memory X with a value of 2. Once Pthread A
releases the lock, Pthread B takes the lock and updates the counter, taking 2 as
its initial value for X and incrementing it to 3, the expected result.
To write a program of any complexity using Pthreads, you will need to share data
between Pthreads, or cause various actions to be performed in some coherent
order across multiple Pthreads. To do this, you need to synchronize the activity of
the Pthreads. Data synchronization among Pthreads can be done using any of
the three types of locking primitive: mutexs, condition variables, and read-write
locks, as explained in the following sections:
Section 8.4.1, “Synchronizing Pthreads with mutexes” on page 297
Section 8.4.2, “Synchronizing Pthreads with condition variables” on page 303
Section 8.4.3, “Synchronizing Pthreads with read-write locks” on page 310
296
Developing and Porting C and C++ Applications on AIX

Although these sections provide enough information to understand the concept
of these locking primitives, data synchronization in the multi-threaded
programming environment can be a challenging task. For further information
about it, please refer to the following publications:
Programming with POSIX Threads, by Lewine
POSIX Programmer’s Guide: Writing Portable UNIX Programs, by Butenhof
8.4.1 Synchronizing Pthreads with mutexes
One of the basic problems when running several Pthreads that use the same
memory space is making sure that they do not interfere with each other. By this,
we refer to the problem of using a data structure from two different Pthreads.
The mutual exclusion lock (mutex) is the simplest synchronization primitive
provided by the Pthread library, and many of the other synchronization primitives
are built upon it.
It is based on the concept of a resource that only one person can use, in a period
of time, for example, a chair or a pencil. If one person sits in a chair, no one can
sit on it until the first person stands up. This kind of primitive is quite useful for
creating critical sections. A critical section is a portion of code that must run
atomically because they normally are handling resources, such as file
descriptors, I/O devices, or shared data. A critical section is a portion of code
delimited by the instructions that lock and unlock a mutex variable. Ensuring that
all Pthreads acting on the same resource or shared data obey this rule is a very
good practice to avoid trouble when programming with Pthreads. A mutex is a
lock that guarantees three things:
Atomicity
Locking a mutex is an atomic operation, meaning that the
operating system (or Pthreads library) assures you that if
you locked a mutex, no other Pthread succeeded in
locking this mutex at the same time.
Singularity
If a Pthread managed to lock a mutex, it is assured that no
other Pthread will be able to lock the Pthread until the
original Pthread releases the lock.
Non-busy wait
If a Pthread attempts to lock a Pthread that was locked by
a second Pthread, the first Pthread will be suspended
(and will not consume any CPU resources) until the lock is
freed by the second Pthread. At this time, the first Pthread
will wake up and continue execution, having the mutex
locked by it.
Very often, the action performed by a Pthread owning a mutex is the updating of
global variables. This is a safe way to ensure that when several Pthreads update
Chapter 8. Introduction to POSIX threads
297

the same variable, the final value is the same as what it would be if only one
Pthread performed the update. The variables being updated belong to a critical
section. A typical sequence of using mutex is as follows:
1. Create and initialize the mutex variable.
2. Several Pthreads attempt to lock the mutex.
3. Only one succeeds and that Pthread owns the mutex.
4. The owner Pthread performs some set of actions.
5. The owner unlocks the mutex.
6. Another Pthread acquires the mutex and repeats the process.
7. Finally, the mutex is destroyed.
Creating and initializing a mutex
In order to create a mutex, we first need to declare a variable of type
pthread_mutex_t, and then initialize it. The simplest way is by assigning it the
PTHREAD_MUTEX_INITIALIZER constant. For example, statically, it is declared
as follows:
pthread_mutex_t a_mutex = PTHREAD_MUTEX_INITIALIZER;
A mutex variable can also be dynamically initialized with the pthread_mutex_init()
routine. This method permits the setting of mutex object attributes.
The pthread_mutexattr_init() and pthread_mutexattr_destroy() routines are used
to create and destroy mutex attribute objects respectively.
Locking and unlocking a mutex
In order to lock a mutex, we may use the pthread_mutex_lock() function. This
function attempts to lock the mutex, or block the Pthread if the mutex is already
locked by another Pthread. In this case, when the mutex is unlocked by the first
Pthread, the function will return with the mutex locked by the other Pthreads.
Here is how to lock a mutex (assuming it was initialized earlier):
int rc = pthread_mutex_lock(&a_mutex);
if (rc) {
/* an error has occurred */
fprintf(stderr, "pthread_mutex_lock() failed with rc = %d.\n", rc);
pthread_exit(NULL);
}
/* mutex is now locked - do your stuff. */
...
298
Developing and Porting C and C++ Applications on AIX

After the Pthread did what it had to (change variables or data structures, handle
file, or whatever it intended to do), it should free the mutex, using the
pthread_mutex_unlock() function, as follows:
rc = pthread_mutex_unlock(&a_mutex);
if (rc) {
fprintf(stderr, "pthread_mutex_unlock() failed with rc = %d.\n", rc);
pthread_exit(NULL);
}
Destroying a mutex
After we finished using a mutex, we should destroy it. Finished using it means no
Pthread needs it at all. If only one Pthread is finished with the mutex, it should
leave it alive for the other Pthreads that might need to use it. Once all Pthreads
finished using it, the last one can destroy it using the pthread_mutex_destroy()
function:
rc = pthread_mutex_destroy(&a_mutex);
The pthread_mutex_destroy function destroys the mutex object referenced by
mutex; the mutex object becomes, in effect, uninitialized. An implementation may
cause pthread_mutex_destroy() to set the object referenced by mutex to an
invalid value. A destroyed mutex object can be re-initialized using
pthread_mutex_init(); the results of otherwise referencing the object after it has
been destroyed are undefined.
It is safe to destroy an initialized mutex that is unlocked. Attempting to destroy a
locked mutex results in undefined behavior.
An example of using mutexs
The example program shown in Example 8-7 on page 300 illustrates the use of
mutex variables in a multi-threaded program that increments a counter variable
declared globally. Each Pthread works on the same data. The initial thread waits
for all the other Pthreads to complete their executions, and then it prints the
resulting sum.
This program demonstrates how to use mutexes to synchronize the operation of
Pthreads. In this program, the main Pthread creates two Pthreads. Each Pthread
tries to access a global variable counter, increment its value, and print it. A mutex
is used to allow only a single Pthread to access the counter at any point of time.
Chapter 8. Introduction to POSIX threads
299

Example 8-7 pthread_mutex.c
#include <pthread.h>
#include <stdio.h>
#include <stdlib.h>
#define EXIT_CODE -1
pthread_t tid1, tid2; /* thread IDs. */
pthread_attr_t t_attr; /* thread attribute structure. */
pthread_mutex_t mut1; /* mutex. */
pthread_mutexattr_t attr;/* mutex attribute structure. */
int counter = 0;
/* global data that will be accessed by two threads.*/
/* first thead executes this function. */
void* mythread1(void* arg)
{
int rc;
while (counter < 10) {
/* block until the mutex lock is obtained, lock when you get it. */
if ((rc = pthread_mutex_lock(&mut1)) != 0) {
fprintf(stderr
, "pthread_mutex_lock() failed with rc = %d.\n", rc);
exit(EXIT_CODE);
}
if (counter < 10) {
counter++;
}
fprintf(stderr,"Thread 1 = %02d\n ", counter);
/* release or unlock the mutex */
if ((rc = pthread_mutex_unlock(&mut1)) != 0) {
fprintf(stderr
, "pthread_mutex_unlock() failed with rc = %d.\n", rc);
exit(EXIT_CODE);
}
sleep(1);
}
}
/* second thread executes this function */
void* mythread2(void* arg)
{
int rc;
while (counter < 10) {
/* block untill the mutex lock is obtained, lock when you get it. */
if ((rc = pthread_mutex_lock(&mut1)) != 0) {
fprintf(stderr
, "pthread_mutex_lock() failed with rc = %d.\n", rc);
exit(EXIT_CODE);
300
Developing and Porting C and C++ Applications on AIX

}
if (counter < 10) {
counter++;
}
printf("Thread 2 = %02d\n", counter);
/* release or unlock the mutex */
if ((rc = pthread_mutex_unlock(&mut1)) != 0) {
fprintf(stderr
, "pthread_mutex_unlock() failed with rc = %d.\n", rc);
exit(EXIT_CODE);
}
sleep(1);
}
}
/* main thread of execution. */
int main(int argc, char* argv[])
{
int rc, status;
/* create the mutex attribute structure. */
if ((rc = pthread_mutexattr_init(&attr)) != 0) {
fprintf(stderr, "pthread_mutexattr_init() failed with rc = %d.\n", rc);
exit(EXIT_CODE);
}
/* initialize the mutex to be used. */
if ((rc = pthread_mutex_init(&mut1, &attr)) != 0) {
fprintf(stderr, "pthread_mutex_init() failed with rc = %d.\n", rc);
exit(EXIT_CODE);
}
/* initialize the attribute structure for the threads to be created. */
if ((rc = pthread_attr_init(&t_attr)) != 0) {
fprintf(stderr, "pthread_attr_init() failed with rc = %d.\n", rc);
exit(EXIT_CODE);
}
/* create the first thread. */
if ((rc = pthread_create(&tid1, &t_attr, mythread1, 0)) != 0) {
fprintf(stderr, "pthread_create(#1) failed with rc = %d.\n", rc);
exit(EXIT_CODE);
}
/* create the second thread */
if ((rc = pthread_create(&tid2, &t_attr, mythread2, 0)) != 0) {
fprintf(stderr, "pthread_create(#2) failed with rc = %d.\n", rc);
exit(EXIT_CODE);
}
pthread_attr_destroy(&t_attr);
pthread_join(tid1, (void **)&status);
pthread_join(tid2, (void **)&status);
printf("\nSum is = %d\n", counter);
Chapter 8. Introduction to POSIX threads
301

pthread_mutex_destroy(&mut1);
pthread_exit(NULL);
}
The output of the above program is:
Thread 1 = 01
Thread 2 = 02
Thread 1 = 03
Thread 2 = 04
Thread 1 = 05
Thread 2 = 06
Thread 1 = 07
Thread 2 = 08
Thread 1 = 09
Thread 2 = 10
Sum is = 10
Please note that the program examines the value of the counter again after it
examines the value in the while condition (highlighted in Example 8-7 on
page 300). If it does not, the program results in a potential race condition, so that
the value of the counter could be 11 instead of 10.
Starvation and deadlock situations
We should remember that pthread_mutex_lock() might block for a undetermined
duration if the mutex is already locked. If it remains locked forever, it is said that
our Pthread is starved, that is, it tried to acquire a resource, but never got it. It is
up to the programmer to ensure that such starvation will not occur. The Pthread
library does not help us with that.
The Pthread library might, however, figure out a deadlock. A deadlock is a
situation in which a set of Pthreads are all waiting for resources taken by other
Pthreads, all in the same set. Naturally, if all Pthreads are blocked waiting for a
mutex, none of them will ever come back to life again. The Pthread library keeps
track of such situations, and thus would fail the last Pthread trying to call
pthread_mutex_lock() with an error of type EDEADLK. The programmer should
check for such a value, and take steps to solve the deadlock somehow.
The following code fragment will result in a deadlock:
pthread_mutex_t mutex_A;
pthread_mutex_t mutex_B;
int counter_A = 0;
int counter_B = 0;
302
Developing and Porting C and C++ Applications on AIX

/* func_1() will be executed by the Pthread #1. */
void func_1()
{
pthread_mutex_lock(&mutex_A);
counter_A++;
pthread_mutex_lock(&mutex_B);
counter_B++;
pthread_mutex_unlock(&mutex_B);
pthread_mutex_unlock(&mutex_A);
}
/* func_2() will be executed by the Pthread #2. */
void func_2()
{
pthread_mutex_lock(&mutex_B);
counter_B++;
pthread_mutex_lock(&mutex_A);
counter_A++;
pthread_mutex_unlock(&mutex_A);
pthread_mutex_unlock(&mutex_B);
}
In the above example, while mutex_A is already locked by Pthread #1 executing
func_1(), multex_B can be also locked by Pthread #2 executing func_B(). Both
Pthreads wait for a mutex until the other Pthread releases it. To avoid this
situation, both Pthreads must try to acquire to lock in the same order.
8.4.2 Synchronizing Pthreads with condition variables
Condition variables provide yet another way for Pthreads to synchronize. While
mutexes implement synchronization by controlling Pthread access to data,
condition variables allow Pthreads to synchronize based upon the actual value of
data. Without condition variables, the programmer would need to have Pthreads
continually polling (possibly in a critical section) to check if the condition is met.
This can be very resource consuming since the Pthread would be continuously
busy in this activity. A condition variable is a way to achieve the same goal
without polling.
What is a condition variable
A condition variable is a mechanism that allows Pthreads to wait (without wasting
CPU cycles) for some event to occur. Several Pthreads may wait on a condition
Chapter 8. Introduction to POSIX threads
303

variable, until some other Pthread signals this condition variable (thus sending a
notification). At this time, one of the Pthreads waiting on this condition variable
wakes up, and can act on the event. It is possible to also wake up all Pthreads
waiting on this condition variable by using a broadcast method on this variable.
Note that a condition variable itself does not provide any kind of locking. Thus, a
mutex is used along with the condition variable to provide the necessary locking
when accessing this condition variable.
Creating and initializing a condition variable
Condition variables must be declared with pthread_cond_t, and must be
initialized before they can be used. There are two ways to initialize a condition
variable.
Statically, when it is declared. For example,
pthread_cond_t a_cond_var = PTHREAD_COND_INITIALIZER;
To initialize the condition variable at run time, use the pthread_cond_init()
routine. This method permits setting condition variable object attributes. The
pthread_condattr_init() and pthread_condattr_destroy() routines are used to
create and destroy condition variable attribute objects.
Signalling a condition variable
In order to signal a condition variable, one should either use the
pthread_cond_signal() function (to wake up only one Pthread waiting on this
variable), or the pthread_cond_broadcast() function (to wake up all Pthreads
waiting on this variable). Here is an example of using a signal, assuming
a_cond_var is a properly initialized condition variable:
int rc = pthread_cond_signal(&a_cond_var);
Or by using the broadcast function:
int rc = pthread_cond_broadcast(&a_cond_var);
When either function returns, the return code rc is set to 0 on success, and to a
non-zero value on failure. In case of failure, the return code denotes the error that
occurred; EINVAL denotes that the given parameter is not a condition variable
and ENOMEM denotes that the system has run out of memory.
The success of a signaling operation does not mean any Pthread was awakened.
It might be that no Pthread was waiting on the condition variable, and thus the
signaling does nothing (that is, the signal is lost).
Note: The term signal, as used in this section, is a different concept from the
one used in the UNIX signal mechanism.
304
Developing and Porting C and C++ Applications on AIX

Waiting on a condition variable
If one Pthread signals the condition variable, other Pthreads would probably want
to wait for this signal. They may do so using one of two functions,
pthread_cond_wait() or pthread_cond_timedwait(). Each of these functions takes
a condition variable and a mutex (which should be locked before calling the wait
function), unlocks the mutex, and waits until the condition variable is signaled,
suspending the Pthread's execution. If this signaling causes the Pthread to
awake (see the discussion about pthread_cond_signal() in “Signalling a condition
variable” on page 304), the mutex is automatically locked again by the wait
function, and the wait function returns.
The only difference between these two functions is that
pthread_cond_timedwait() allows the programmer to specify a timeout for the
waiting, after which the function always returns with a proper error value
(ETIMEDOUT) to notify that condition variable was not signaled before the
timeout passed. The pthread_cond_wait() would wait indefinitely if it was never
signaled.
The code fragment shown in Example 8-8 depicts how to use
pthread_cond_wait(). We make the assumption that a_cond_var is a properly
initialized condition variable, and that request_mutex is a properly initialized
mutex.
Example 8-8 An example of using pthread_cond_wait()
/* first, lock the mutex. */
int rc;
if ((rc = pthread_mutex_lock(&a_mutex)) != 0) {
/* an error has occurred. */
fprintf(stderr, "pthread_mutex_lock() failed with rc = %d.\n", rc);
pthread_exit(NULL);
}
/* mutex is now locked - wait on the condition variable. */
/* During the execution of pthread_cond_wait, the mutex is unlocked. */
if ((rc = pthread_cond_wait(&a_cond_var, &a_mutex)) == 0) {
/*
* we were awakened due to the condition variable, which had been signaled.
* The mutex is now locked again by pthread_cond_wait().
* do your stuff....
*/
...
}
/* finally, unlock the mutex. */
pthread_mutex_unlock(&a_mutex);
The code fragment shown in Example 8-9 on page 306 shows how to use
pthread_cond_timedwait().
Chapter 8. Introduction to POSIX threads
305

Example 8-9 An example using pthread_cond_timedwait()
#include <sys/time.h>
/* struct timeval definition.
*/
#include <unistd.h>
/* declaration of gettimeofday().
*/
struct timeval now; /* time when we started waiting.
*/
struct timespec timeout; /* timeout value for the wait function.*/
int done; /* are we done waiting? */
/* first, lock the mutex. */
int rc = pthread_mutex_lock(&a_mutex);
if (rc) {
/* an error has occurred. */
fprintf(stderr, "pthread_mutex_lock() failed with rc = %d.\n", rc);
pthread_exit(NULL);
}
/* mutex is now locked. */
/* get current time. */
gettimeofday(&now);
/* prepare timeout value. */
/* t_sec member is represented in seconds, while tv_usec in microseconds. */
timeout.tv_sec = now.tv_sec + 5;
timeout.tv_nsec = now.tv_usec * 1000;
/* wait on the condition variable. */
/* we use a loop, since a UNIX signal might stop the wait before the timeout.*/
done = 0;
while (!done) {
/* remember that pthread_cond_timedwait() unlocks the mutex on entrance. */
rc = pthread_cond_timedwait(&a_cond_var, &a_mutex, &timeout);
switch(rc) {
case 0:
/*
* we were awakened due to the condition variable
* , which had been signaled.
* The mutex is now locked again by pthread_cond_wait().
* do your stuff....
*/
...
done = 0;
break;
case ETIMEDOUT: /* our time is up. */
done = 0;
break;
default: /* some error occurred (e.g. we got a UNIX signal). */
break; /* break this switch, but re-do the while loop.
*/
}
}
306
Developing and Porting C and C++ Applications on AIX

/* finally, unlock the mutex. */
pthread_mutex_unlock(&a_mutex);
It might be that a condition variable that has two or more Pthreads waiting on it is
signaled many times, and yet one of the Pthreads waiting on it never awakens.
This is because we are not guaranteed which of the waiting Pthreads is
awakened when the variable is signaled. It might be that the awakened Pthread
quickly comes back to waiting on the condition variables, and gets awakened
again when the variable is signaled again, and so on. The situation for the
un-awakened Pthread is called starvation. It is up to the programmer to make
sure that this situation does not occur if it implies bad behavior.
When the mutex is being broadcast using pthread_cond_broadcast(), this does
not mean all Pthreads are running together. Each of them tries to lock the mutex
again before returning from their wait function, and thus they will start running
one by one, each one locking the mutex, doing their work, and freeing the mutex
before the next Pthread gets its chance to run.
Destroying a condition variable
After we are done using a condition variable, we should destroy it to free any
system resources it might be using. This can be done using
pthread_cond_destroy(). In order for this to work, there should be no Pthreads
waiting on this condition variable. Here is how to use this function, again
assuming a_cond_var is a pre-initialized condition variable:
int rc;
if ((rc = pthread_cond_destroy(&a_cond_var)) == EBUSY) {
/* some Pthread is still waiting on this condition variable. */
/* handle this case here... */
}
What if some Pthread is still waiting on this variable? Depending on the case, it
might imply some flaw in the usage of this variable, or just lack of proper Pthread
cleanup code.
Using a condition variable
The example program shown in Example 8-10 on page 308 demonstrates the
use of several Pthread condition variable routines. The main routine creates
three Pthreads. Two Pthreads that execute the add_counter() function increment
a variable, named count, whenever the associated mutex is held. The third
Pthread that executes the monitor_counter() function waits until the count
variable reaches a specified value.
Chapter 8. Introduction to POSIX threads
307

Example 8-10 pthread_cond_variable.c
#include <pthread.h>
#include <stdio.h>
#define NUM_OF_THREADS 3
#define TOTAL_COUNT 5
#define MAX_COUNT 5
#define EXIT_CODE -1
int count = 0;
int thread_ids[3] = {0, 1, 2};
pthread_mutex_t a_mutex;
pthread_cond_t a_cond_var;
void *add_counter(void *arg)
{
unsigned int i, j;
double result = 0.0;
int *my_id = arg;
for (i = 0; i < TOTAL_COUNT; i++) {
pthread_mutex_lock(&a_mutex);
count++;
/*
* Check the value of count and signal waiting thread when condition is
* reached. Note that this occurs while mutex is locked.
*/
if (count == MAX_COUNT) {
pthread_cond_signal(&a_cond_var);
printf(
"add_counter(): thread %d, count = %d, threshold reached.\n"
, *my_id, count);
}
printf("add_counter(): thread %d, count = %d, unlocking mutex.\n"
, *my_id, count);
pthread_mutex_unlock(&a_mutex);
/* Do some work so threads can alternate on mutex lock. */
for (j = 0; j < 1000; j++)
result += j;
}
pthread_exit(NULL);
}
void *monitor_counter(void *arg)
{
int *my_id = arg;
308
Developing and Porting C and C++ Applications on AIX

printf("Starting monitor_counter(): thread %d\n", *my_id);
/*
* Lock mutex and wait for signal. Note that the pthread_cond_wait()
* routine will automatically and atomically unlock mutex while it waits.
* Also, note that if MAX_COUNT is reached before this routine is run by
* the waiting thread, the loop will be skipped to prevent
* pthread_cond_wait() from never returning.
*/
pthread_mutex_lock(&a_mutex);
while (count < MAX_COUNT) {
pthread_cond_wait(&a_cond_var, &a_mutex);
printf("monitor_counter(): thread %d Condition signal received.\n"
, *my_id);
}
pthread_mutex_unlock(&a_mutex);
pthread_exit(NULL);
}
int main (int argc, char *argv[])
{
int i, rc;
pthread_t threads[NUM_OF_THREADS];
pthread_attr_t attr;
/* Initialize mutex and condition variable objects. */
pthread_mutex_init(&a_mutex, NULL);
pthread_cond_init (&a_cond_var, NULL);
/*
* For portability, explicitly create threads in a joinable state
* so that they can be joined later.
*/
pthread_attr_init(&attr);
pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_JOINABLE);
if ((rc = pthread_create(&threads[0], &attr, add_counter
, (void *)&thread_ids[0])) != 0) {
fprintf(stderr, "pthread_create(#1) is failed with rc = %d.\n", rc);
exit(EXIT_CODE);
}
if ((rc = pthread_create(&threads[1], &attr, add_counter
, (void *)&thread_ids[1])) != 0) {
fprintf(stderr, "pthread_create(#2) is failed with rc = %d.\n", rc);
exit(EXIT_CODE);
}
if ((rc = pthread_create(&threads[2], &attr, monitor_counter
, (void *)&thread_ids[2])) != 0) {
fprintf(stderr, "pthread_create(#3) is failed with rc = %d.\n", rc);
Chapter 8. Introduction to POSIX threads
309

exit(EXIT_CODE);
}
/* Wait for all threads to complete */
for (i = 0; i < NUM_OF_THREADS; i++) {
if ((rc = pthread_join(threads[i], NULL)) != 0) {
fprintf(stderr, "pthread_join() is failed with rc = %d.\n", rc);
exit(EXIT_CODE);
}
}
printf ("Main(): Waited on %d threads. Done...\n", NUM_OF_THREADS);
/* Clean up and exit */
pthread_attr_destroy(&attr);
pthread_mutex_destroy(&a_mutex);
pthread_cond_destroy(&a_cond_var);
pthread_exit(NULL);
}
The output of this program is:
add_counter(): thread 0, count = 1, unlocking mutex.
add_counter(): thread 0, count = 2, unlocking mutex.
Starting monitor_counter(): thread 2
add_counter(): thread 0, count = 3, unlocking mutex.
add_counter(): thread 1, count = 4, unlocking mutex.
add_counter(): thread 0, count = 5, threshold reached.
add_counter(): thread 0, count = 5, unlocking mutex.
add_counter(): thread 1, count = 6, unlocking mutex.
add_counter(): thread 0, count = 7, unlocking mutex.
add_counter(): thread 1, count = 8, unlocking mutex.
monitor_counter(): thread 2 Condition signal received.
add_counter(): thread 1, count = 9, unlocking mutex.
add_counter(): thread 1, count = 10, unlocking mutex.
Main(): Waited on 3 threads. Done...
8.4.3 Synchronizing Pthreads with read-write locks
Another type of lock primitive, read-write locks allow a Pthread to exclusively lock
some shared data while updating that data, or allow any number of Pthreads to
have simultaneous read-only access to the data.
Unlike a mutex, a read-write lock distinguishes between reading data and writing
data. A mutex excludes all other Pthreads. A read-write lock allows other
Pthreads access to the data, providing no Pthread is modifying the data. Thus, a
read-write lock is less primitive than either a mutex-condition variable pair or a
semaphore.
310
Developing and Porting C and C++ Applications on AIX

Application developers should consider using a read-write lock rather than a
mutex to protect data that is frequently referenced but seldom modified. Most
Pthreads (readers) will be able to read the data without waiting and will only have
to block when some other Pthread (a writer) is in the process of modifying the
data. Conversely, a Pthread that wants to change the data is forced to wait until
there are no readers. This type of lock is often used to facilitate parallel access to
data on multiprocessor platforms or to avoid context switches on single
processor platforms where multiple Pthreads access the same data.
If a read-write lock becomes unlocked and there are multiple Pthreads waiting to
acquire the write lock, the implementation's scheduling policy determines which
Pthread shall acquire the read-write lock for writing. If there are multiple Pthreads
blocked on a read-write lock for both read locks and write locks, it is unspecified
whether the readers or a writer acquires the lock first. However, for performance
reasons, implementations often favor writers over readers to avoid potential
writer starvation.
A read-write lock object is an implementation-dependent opaque object of type
pthread_rwlock_t, as defined in pthread.h. There are two different sorts of locks
associated with a read-write lock: a read lock and a write lock.
The pthread_rwlockattr_init() function initializes a read-write lock attributes object
with the default value for all the attributes defined in the implementation. After a
read-write lock attributes object has been used to initialize one or more
read-write locks, changes to the read-write lock attributes object, including
destruction, do not affect previously initialized read-write locks.
The read-write lock attribute can be any of the following:
PTHREAD_PROCESS_SHARED
Any Pthread of any process that has access to the
memory where the read-write lock resides can manipulate
the read-write lock.
PTHREAD_PROCESS_PRIVATE
Only Pthreads created within the same process as the
Pthread that initialized the read-write lock can manipulate
the read-write lock. This is the default value.
The pthread_rwlockattr_setpshared() function is used to set the process-shared
attribute of an initialized read-write lock attributes object while the function
pthread_rwlockattr_getpshared() obtains the current value of the process-shared
attribute. A read-write lock attributes object is destroyed using the
pthread_rwlockattr_destroy() function.
Chapter 8. Introduction to POSIX threads
311

A Pthread creates a read-write lock using the pthread_rwlock_init() function. The
attributes of the read-write lock can be specified by the application developer;
otherwise, the default implementation-dependent read-write lock attributes are
used if the pointer to the read-write lock attributes object is NULL. In cases where
the default attributes are appropriate, the PTHREAD_RWLOCK_INITIALIZER
macro can be used to initialize statically allocated read-write locks.
A Pthread that wants to apply a read lock to the read-write lock can use either
pthread_rwlock_rdlock() or pthread_rwlock_tryrdlock(). If
pthread_rwlock_rdlock() is used, the Pthread acquires a read lock if a writer does
not hold the write lock and there are no writers blocked on the write lock. If a read
lock is not acquired, the calling Pthread blocks until it can acquire a lock.
However, if pthread_rwlock_tryrdlock() is used, the function returns immediately
with the error EBUSY if any Pthread holds a write lock or there are blocked
writers waiting for the write lock.
A Pthread that wants to apply a write lock to the read-write lock can use either of
two functions: pthread_rwlock_wrlock() or pthread_rwlock_trywrlock(). If
pthread_rwlock_wrlock() is used, the Pthread acquires the write lock if no other
reader or writer Pthreads holds the read-write lock. If the write lock is not
acquired, the Pthread blocks until it can acquire the write lock. However, if
pthread_rwlock_trywrlock() is used, the function returns immediately with the
error EBUSY if any Pthread is holding either a read or a write lock.
The pthread_rwlock_unlock() function is used to unlock a read-write lock object
held by the calling Pthread.
Supported read-write lock sub-routines are listed in Table D-5 on page 480.
For further information about the read-write lock programming on AIX, please
refer to the “Threads Programming Guidelines” section in AIX 5L Version 5.2
General Programming Concepts: Writing and Debugging Programs.
8.5 Thread-specific data
POSIX provides a mechanism that enables applications to maintain specified
data on a per-Pthread basis. The mechanism is motivated by the need of some
modules (that is, groups of related functions) to maintain selected data across
function invocations (that is, to maintain static data in the C programming
language). If such a module is being used by multiple Pthreads of a
multi-threaded process, then the module may need to maintain such data
separately for each calling Pthread, depending on the particular application.
312
Developing and Porting C and C++ Applications on AIX

Note: If a Pthread executes only one function during its life-cycle, it less likely
needs thread-specific data. Because each Pthread has its own thread stack,
all auto storage class variables in that function are allocated in the thread
stack, which would be automatically deallocated when the Pthread terminates
(see 8.3.4, “Thread stack” on page 292).
For example, consider a module consisting of the push, pop, and clear stack
functions. Suppose that the module declares the stack and stack pointer as static
data, instead of as function parameters. If the module is being used in a
multi-threaded process, in which each Pthread needs its own stack and stack
pointer, the module must have a mechanism for maintaining per-thread stacks
and stack pointers.
In the POSIX model, per-thread data is maintained through a key/value
mechanism, an approach designed for efficiency and ease of use. A key is an
opaque object of type pthread_key_t. The value of a key is thread-specific, that
is, each key that has been established for a multi-threaded process has a distinct
value for each Pthread of the process. For this reason, the thread-specific data of
a process is sometimes thought of as a matrix, with rows corresponding to keys
and columns to Pthreads, although implementations need not work this way. A
process can have up to PTHREAD_KEYS_MAX keys (or rows), where
PTHREAD_KEYS_MAX is defined at least at 128 in the POSIX thread
standard.14
Keys are of opaque data type so that operating system implementations can
have the freedom in setting them up to offer efficient access to thread-specific
data. Instead of holding thread-specific values directly, keys may hold means of
accessing thread-specific values. Conceptually, a key isolates a row of the
thread-specific data matrix, and then the key uses the Pthread ID of the calling
Pthread (the Pthread calling pthread_getspecific() or pthread_setspecific()) to
isolate an entry in the row, thus obtaining the desired key value.
Typically, the value associated with a given key for a given Pthread is a pointer to
memory dynamically allocated for the exclusive use of the given Pthread (for
example, per-thread stack and stack pointer). The scenario for establishment and
use of thread-specific data can be described as follows. A module that needs to
maintain static data on a per-thread basis creates a new thread-specific data key
as a part of its initialization routine. At initialization, all Pthreads of the process
are assigned null values for the new key. Then, upon each Pthread’s first
invocation of the module (which can be determined by checking for a null key
value), the module dynamically allocates memory for the exclusive use of the
14 Currently, AIX 5L Version 5.2 supports a maximum of 450 keys per process. However, it can be
increased in the future versions of AIX. Call the sysconf() routine with _SC_THREAD_KEYS_MAX to
determine the maximum number of keys.
Chapter 8. Introduction to POSIX threads
313

calling Pthread, and stores a pointer to the memory as the calling Pthread’s value
of the new key. Upon later invocations of the same module by the same Pthread,
the module can access the Pthread’s data through the new key (that is, the
Pthread’s value for the key). Other modules can independently create other
thread-specific data keys for other per-thread data for their own use.
An application process could maintain thread-specific data in other ways. For
example, it could use a hash function on a Pthread ID as a means of access to
an area of thread-specific data. Then the application process would have to
manage the use of sub-areas of the thread-specific data. The POSIX
thread-specific data interfaces, on the other hand, provide Pthreads more direct
access to sub-areas of their thread-specific data. Moreover, the sub-areas are
independent; different parts of the application process can use different
sub-areas without any need for process-wide cooperation.
8.5.1 Allocating thread-specific data
The pthread_key_create() function is used to allocate a new key. This key now
becomes valid for all Pthreads in our process. When a key is created, the value it
points to defaults to NULL. Later on, each Pthread may change its copy of the
value as it wishes. The following code fragment shows how to use this function:
/* rc is used to contain return values of pthread functions. */
int rc;
/* define a variable to hold the key, once created. */
pthread_key_t list_key;
/*
* cleanup_list() is a function that can clean up some data.
* it is specific to our program, not to thread-specific data.
*/
extern void* cleanup_list(void*);
/* create the key, supplying a function that'll be invoked when it's deleted.*/
rc = pthread_key_create(&list_key, cleanup_list);
Please note the following:
After pthread_key_create() returns, the variable list_key points to the newly
created key.
The function pointer, passed as a second parameter to pthread_key_create(),
will be automatically invoked by the Pthread library when our Pthread exits,
with a pointer to the key's value as its parameter. We may supply a NULL
pointer as the function pointer, and then no function will be invoked for key.
Note that the function will be invoked once in each Pthread, even though we
created this key only once in one Pthread.
If we created several keys, their associated destructor functions will be called
in an arbitrary order, regardless of the order of keys creation.
314
Developing and Porting C and C++ Applications on AIX

If the pthread_key_create() function succeeds, it returns 0. Otherwise, it
returns some error code.
There is a limit of PTHREAD_KEYS_MAX keys that may exist in our process
at any given time. An attempt to create a key after PTHREAD_KEYS_MAX
exits will cause a return value of EAGAIN from the pthread_key_create()
function.
8.5.2 Accessing thread-specific data
After we have created a key, we may access its value using two Pthread
functions: pthread_getspecific() and pthread_setspecific(). The first is used to get
the value of a given key, and the second is used to set the data of a given key. A
key's value is simply a void pointer (void *), so we can store in it anything that we
want. The following code fragment shows you how to use these functions,
assuming that a_key is a properly initialized variable of type pthread_key_t,
which contains a previously created key:
/* this variable will be used to store return codes of pthread functions. */
int rc;
/* define a variable into which we'll store some data. */
int* p_num = (int *)malloc(sizeof(int));
if (!p_num) {
perror("malloc() failed.\n");
exit(1);
}
/* initialize our variable to some value. */
*p_num = 4;
/*
* now let’s store this value in our thread-specific data key.
* note that we don't store p_num in our key.
* we store the value that p_num points to.
*/
rc = pthread_setspecific(a_key, (void *)p_num);
....
....
/*
* and somewhere later in our code, get the value of key a_key and print it.
*/
int* p_keyval = (int *)pthread_getspecific(a_key);
if (p_keyval != NULL) {
printf("value of 'a_key' is: %d\n", *p_keyval);
}
Chapter 8. Introduction to POSIX threads
315

Note that if we set the value of the key in one Pthread, and try to get it in another
Pthread, we will get a NULL, since this value is distinct for each Pthread.
There are two other cases where pthread_getspecific() might return NULL:
The key supplied as a parameter is invalid (for example, its key was not
created).
The value of this key is NULL. This means it either was not initialized, or was
set to NULL explicitly by a previous call to pthread_setspecific().
8.5.3 Deleting thread-specific data
The pthread_key_delete() function may be used to delete keys. It does not delete
memory associated with this key or call the destructor function in C++ defined
during the key's creation. Thus, we still need to do memory cleanup on our own if
we need to free this memory during run time. However, with global variables (and
thus also thread-specific data), we usually do not need to free this memory until
the thread terminates, in which case the pthread library will invoke our destructor
functions anyway.
Using this function is simple. Assuming list_key is a pthread_key_t variable
pointing to a properly created key, use this function like this:
int rc = pthread_key_delete(key);
The function will return 0 on success, or EINVAL if the supplied variable does not
point to a valid thread-specific data key.
An example of using thread-specific data
The example program shown in Example 8-11 illustrates the use of
thread-specific data routines. Each created Pthread maintains its own copy of the
data structure tsd_data, which is declared globally.
Example 8-11 pthread_tsd.c
#include <pthread.h>
#include <stdio.h>
#include <stdlib.h>
#define EXIT_CODE -1
/*
* Structure used as the value for thread-specific data key.
*/
typedef struct tsd_data {
pthread_t thread_id;
char *mesg;
} tsd_data_var;
316
Developing and Porting C and C++ Applications on AIX

pthread_key_t tsd_data_key; /* thread-specific data key */
pthread_once_t key_once = PTHREAD_ONCE_INIT;
/*
* init_routine() is an one-time initialization routine used with
* pthread_once().
*/
void init_routine(void)
{
int rc;
printf("Initializing key.\n");
if ((rc = pthread_key_create(&tsd_data_key, NULL)) != 0) {
fprintf(stderr, “pthread_create() failed with rc = %d.\n”, rc);
exit(EXIT_CODE);
}
}
/*
* tsd_routine() is a routine that uses pthread_once() to dynamically
* create a thread-specific data key.
*/
void *tsd_routine(void *arg)
{
tsd_data_var *value;
int rc;
if ((rc = pthread_once(&key_once, init_routine)) != 0) {
fprintf(stderr, “pthread_once() failed with rc = %d.\n”, rc);
exit(EXIT_CODE);
}
value = (tsd_data_var *)malloc(sizeof(tsd_data_var));
if (value == NULL) {
perror("Failed to allocate key value.");
exit(EXIT_CODE);
}
if ((rc = pthread_setspecific(tsd_data_key, value)) != 0) {
fprintf(stderr, “pthread_setspecific() failed with rc = %d.\n”, rc);
exit(EXIT_CODE);
}
printf("%s set thread-specific data value %p\n", arg, value);
value->thread_id = pthread_self();
value->mesg = (char *)arg;
value = (tsd_data_var *)pthread_getspecific(tsd_data_key);
printf("%s Starting...\n", value->mesg);
sleep(2);
value = (tsd_data_var *)pthread_getspecific(tsd_data_key);
printf("%s Done...\n", value->mesg);
Chapter 8. Introduction to POSIX threads
317

return NULL;
}
int main(int argc, char *argv[])
{
pthread_t thread1, thread2;
int rc;
if ((rc = pthread_create(&thread1, NULL, tsd_routine, "thread 1")) != 0) {
fprintf(stderr, “pthread_create(#1) failed with rc = %d.\n”, rc);
exit(EXIT_CODE);
}
if ((rc = pthread_create(&thread2, NULL, tsd_routine, "thread 2")) != 0) {
fprintf(stderr, “pthread_create(#2) failed with rc = %d.\n”, rc);
exit(EXIT_CODE);
}
pthread_exit (NULL);
}
The output of Example 8-11 on page 316 is:
Initializing key.
thread 1 set thread-specific data value 20209d58
thread 1 Starting...
thread 2 set thread-specific data value 20209d68
thread 2 Starting...
thread 1 Done...
thread 2 Done...
In Example 8-11 on page 316, the following two new Pthread routines are called:
pthread_once() and pthread_self().
We have called the pthread_once() routine to dynamically create a Pthread
specific data key. The pthread_once() routine calls an initialization routine
executed by one Pthread a single time. This routine allows you to create your
own initialization code that is guaranteed to be run only once, even if called
simultaneously by multiple Pthreads or multiple times in the same Pthread.
For example, a mutex or a thread-specific data key must be created exactly once.
Calling pthread_once() prevents the code that creates a mutex or thread-specific
data from being called by multiple Pthreads. Without this routine, the execution
must be serialized so that only one Pthread performs the initialization. Other
Pthreads that reach the same point in the code are delayed until the first Pthread
has finished the call to pthread_once().
This routine initializes the control record if it has not been initialized and then
determines if the client one-time initialization routine has executed once. If it has
not executed, this routine calls the initialization routine specified in init_routine().
318
Developing and Porting C and C++ Applications on AIX

If the client one-time initialization code has executed once, this routine returns.
The pthread_once_t data structure is a record that allows client initialization
operations to guarantee mutual exclusion of access to the initialization routine,
and that each initialization routine is executed exactly once.
This variable must be initialized using the pthread_once_init macro, as follows:
pthread_once_t key_once = PTHREAD_ONCE_INIT;
Similarly, we have called the pthread_self() routine to get the unique identifier of
the Pthread that is getting executed.
8.5.4 Thread-safe and reentrant functions
One very important point to take care of when building multi-threaded programs
is the resource handling. To avoid getting in trouble, be sure to call only
thread-safe and reentrant functions in multi-threaded applications. Reentrance
and thread-safety are separate concepts: A function can be either reentrant,
thread-safe, both, or neither.
Reentrant
A reentrant function does not hold static data over
successive calls, nor does it return a pointer to static data.
All data is provided by the caller of the function. A
reentrant function must not call non-reentrant functions.
Thread-safe
A thread-safe function protects shared resources from
concurrent access by locks. Thread-safety concerns only
the implementation of a function and does not affect its
external interface. The use of global data is
thread-unsafe. It should be maintained per-Pthread or
encapsulated so that its access can be serialized.
Reentrant and thread-safe libraries are useful in a wide range of parallel (and
asynchronous) programming environments, not just within Pthreads. Thus, it is a
good programming practice to always use and write reentrant and thread-safe
functions.
Some of the standard C subroutines are non-reentrant, such as the ctime() and
strtok() subroutines. The reentrant version of the subroutines typically have the
name of the original subroutine with a suffix _r (underscore r).
Chapter 8. Introduction to POSIX threads
319

When writing multi-threaded programs, the reentrant versions of subroutines
should be used instead of the original version. For example, the following code
fragment:
token[0] = strtok(string, separators);
i = 0;
do {
i++;
token[i] = strtok(NULL, separators);
} while (token[i] != NULL);
must be replaced by the following code fragment in the multi-threaded
programming:
char *pointer;
...
token[0] = strtok_r(string, separators, &pointer);
i = 0;
do {
i++;
token[i] = strtok_r(NULL, separators, &pointer);
} while (token[i] != NULL);
Thread-unsafe libraries may be used by only one Pthread in a program. The
uniqueness of the Pthread using the library must be ensured by the programmer;
otherwise, the program will have unexpected behavior or may even crash.
In order to determine which sub routines provided by the AIX base operating
system are reentrant or thread-safe, please refer to the “Threads Programming
Guidelines” section in AIX 5L Version 5.2 General Programming Concepts:
Writing and Debugging Programs.
8.6 Pthread cancellation
POSIX provides facilities for canceling Pthreads. The facilities enable a Pthread
to cancel another specified Pthread within the same process. The facilities are
aimed at allowing applications to cancel cooperating Pthreads.
A Pthread that is the target of cancellation must be cooperative in the following
sense. Each Pthread controls its own cancelability state and type. The
cancelability state can be enabled or disabled. For the enabled state, the type
can be asynchronous (cancellation requests accepted at any time) or deferred
(cancellation accepted only at designated cancellation points). By default,
Pthreads are initialized with state enabled and type deferred. An uncooperative
Pthread could disable cancellation, and thus thwart cancellation requests.
320
Developing and Porting C and C++ Applications on AIX

An application is expected to make cancellation graceful by specifying
cancellation cleanup handlers. The cleanup handlers perform actions such as
unlocking mutexes owned by the target Pthread (the Pthread being canceled),
signaling conditions that the target Pthread may have caused to become true,
releasing resources, and so on. These actions enable other Pthreads in the
application to make progress after the cancellation occurs.
The Pthread cancellation facilities specified in POSIX include the following
functions:
Canceling execution of a specified Pthread using the pthread_cancel()
routine. The cancellation occurs in accordance with the target Pthread’s
cancelability state and type. If and when cancellation does occur, cancellation
cleanup handlers are invoked in last-in-first-out (LIFO) order, followed by
thread-specific data destructor functions in unspecified order. Then the
Pthread is terminated.
Setting the cancelability state of the calling Pthread using the
pthread_setcancelstate() routine to either enabled or disabled.
Setting the cancelability type of the calling Pthread using
pthread_setcanceltype() to either asynchronous or deferred.
Creating a cancellation point using the pthread_testcancel() routine. To
designate a point in its execution as a cancellation point, a Pthread simply
makes a call to the pthread_testcancel() function at that point.
Establishing cancellation cleanup handlers using the pthread_cleanup_push()
and pthread_cleanup_pop() routines. The pthread_cleanup_push() function is
used to identify a specified routine as a cleanup handler. Conceptually, the
routine is pushed onto the top of the calling Pthread’s cancellation cleanup
stack. The pthread_cleanup_pop() function is used to remove and optionally
execute the cleanup handler at the top of the cancellation cleanup stack.
In order to determine which subroutines provided by the AIX base operating
system can be canceled (cancellation point(s) can occur when a Pthread is
executing that subroutine), please refer to the “Thread Programming Guidelines”
section in AIX 5L Version 5.2 General Programming Concepts: Writing and
Debugging Programs.
8.7 Pthread priority and scheduling
By default, each Pthread within a process has its own a dynamic or floating
priority on AIX. This mechanism ensures that all Pthreads on a system have an
opportunity to run and not be locked out by other Pthreads. As individual
Pthreads consume CPU time, they are charged. Pthreads that are not ready to
run, and that are not given a processor upon which they consume cycles, are not
Chapter 8. Introduction to POSIX threads
321

charged. Over time, CPU intensive Pthreads use more CPU cycles, are charged
more, and their priority degrades. Pthreads that run only briefly before being
preempted or relinquishing control of the processor tend to retain a fairly
constant priority. The net result is that I/O intensive tasks stay near a constant
priority, and computational tasks become lower in priority. The I/O intensive tasks
usually run briefly. The computational tasks run for most or all of their allotted
processor time. This allows all processes on the system to be responsive,
ensures that every task is treated fairly with respect to CPU utilization, and
prevents long-running tasks from hogging the system at the expense of others.
It is also possible to fix the priority of a Pthread at a specific value. This facility
allows processes with Pthreads that must run at a certain response level to
accomplish tasks in a predicable manner. While a normal process can fix the
priority of a Pthread up to a certain level, at high priority levels the root user
authority is required.
8.7.1 Thread models in AIX
Pthreads (user threads) are mapped to underlying kernel threads. A thread
model refers to the way this mapping is done. A light-weight process (LWP)
refers to a kernel thread.
There are three possible thread models, corresponding to three different ways to
map Pthreads to kernel threads (see Figure 8-4 on page 323).
The M:1 model15
Also known as the library thread scheduling, it refers to the case where
threads share a pool of available LWPs.
In the M:1 model, all user threads run on one VP and are linked to exactly one
LWP; the mapping is handled by a POSIX thread library scheduler. All user
threads programming facilities are completely handled by the library. This
model can be used on any system, especially on traditional single-threaded
systems.
The 1:1 model16
The 1:1 model is sometimes referred to as bound thread scheduling. In this
model, each user thread is bound to a VP and linked to exactly one kernel
thread (LWP). The VP is not necessarily bound to a real CPU (unless binding
to a processor was done). Each VP can be thought of as a virtual CPU
available for executing user code and system calls. A thread that is bound to a
VP is said to have system scope, because it is directly scheduled with all the
15 This was a thread model on AIX Version 3.2 when DCE is installed.
16 This was the thread model on AIX Version 4.1 and 4.2.
322
Developing and Porting C and C++ Applications on AIX

other kernel threads by the kernel scheduler. Most of the user threads
programming facilities are directly handled by the kernel threads.
The M:N model
In the M:N model (this is also known as multiplexed thread scheduling),
several user threads can share the same virtual processor or the same pool
of VPs. A thread that is not bound to a VP is said to be a local or process
scope, because it is not directly scheduled with all the other threads by the
kernel scheduler. The Pthreads library will handle the scheduling of user
threads to the VP and then the kernel will schedule the associated kernel
thread. This is the most efficient and most complex thread model; the user
threads programming facilities are shared between the threads library and the
kernel threads.
The mapping of user threads to kernel threads is done using virtual
processors. A virtual processor (VP) is a library entity that is usually implicit.
For a user thread, the virtual processor behaves as a CPU for a kernel thread.
In the library, the virtual processor is a kernel thread or a structure bound to a
kernel thread, that is, user threads sit on top of virtual processors, which are
themselves on top of kernel threads.
Note: The M:N model is the default thread model on AIX, beginning with
Version 4.3.
M:1 thread model
1:1 thread model
M:N thread model
Process A
Process B
Process A
Process B
Process A
Process B
VP
VP
VP
VP
VP
VP
VP
VP
VP
VP
VP
VP
CPU
CPU
CPU
CPU
CPU
CPU
CPU
CPU
CPU
User threads
Kernel scheduler
Pthreads library
Kernel threads (running)
VP
Virtual Processors
Kernel threads (runnable)
User threads sleeping (No VP)
Figure 8-4 Thread models
Chapter 8. Introduction to POSIX threads
323

Comparing bound and multiplexed threads
Bound thread scheduling differs from multiplexed thread scheduling in the
following important ways:
A bound thread is permanently associated to its kernel thread, hence, it is
exempt from the intermediate level of scheduling provided by the POSIX
threads library used under the M:1 and M:N thread models.
A bound thread executes exactly when its associated kernel thread is
scheduled by the kernel scheduler.
A bound thread’s scheduling policy is related to the underlying kernel thread.
In AIX, only kernel threads with root authority can use a fixed-priority
scheduling policy.
Multiplexed thread scheduling has the following important characteristic: A
multiplexed thread is subject to two levels of scheduling. First, the thread is
assigned to a kernel thread and preempted by a POSIX thread library scheduler.
Second, the kernel scheduler assigns the LWPs to processors and then
preempts them.
8.7.2 Scheduling Pthreads
A kernel component, called scheduler, is used to share processor cycles among
running processes. The kernel scheduler coordinates ready-to-run tasks of
varying priorities, running at least 100 times every second to reevaluate tasks on
the run queue and allow processes to run based on the scheduling policy.
Starting with Version 4.3, AIX supports Pthread scheduling at the process level.
The user scheduler is provided by the Pthreads library and supports the M:N
thread model on AIX. The Pthreads library scheduler has a pool of one or more
virtual processors (VPs) upon which a user-space Pthread may run. You can
think of a VP as a kernel-space thread. This mapping of the user thread to kernel
thread provides more effective use of kernel resources for processes that create
many threads, not all of which will be running at any given moment. For example,
in a producer-consumer thread pair, we will often see only one task running while
the other waits to send or receive data. The Pthread library scheduler itself is
also created as a Pthread in a process and runs on its own VP.
The Pthreads library allows the programmer to control the execution scheduling
of the Pthreads. The control can be performed in two different ways:
By setting scheduling attributes when creating a Pthread.
By dynamically changing the scheduling attributes of a created Pthread.
324
Developing and Porting C and C++ Applications on AIX

A Pthread has three scheduling parameters:
Scope
The contention scope of a Pthread is defined by the
thread model used in the Pthreads library.
Policy
The scheduling policy of a Pthread defines how the
scheduler treats the Pthread once it gains control of the
CPU.
Priority
The scheduling priority of a Pthread defines the relative
importance of the work being done by each Pthread.
The scheduling parameters can be set before the Pthread's creation or during
the Pthread's execution. In general, controlling the scheduling parameters of
Pthreads is important only for Pthreads that are compute-intensive. Thus, the
Pthreads library provides default values that are sufficient for most cases.
Controlling the scheduling of a Pthread is often a complicated task. Because the
scheduler can handle all system or process-wide Pthreads, depending on the
scope context, the scheduling parameters of a Pthread can interact with those of
all other Pthreads in the process and in the other processes on the system.
Scheduling policies
AIX also provides support for a number of scheduling policies. Each policy has
certain characteristics that will benefit a specific workload, but not all policies are
suitable for every task. We can create an application that assigns different
policies to different Pthreads, tuning our application to make optimum use of the
system. AIX provides the following policies at the system, or kernel, level:
SCHED_FIFO
SCHED_RR
SCHED_OTHER
SCHED_FIFO2
SCHED_FIFO3
As long as a process is using process-level scope (that is, the M:N thread model
is in effect), the process may modify the scheduling policy of the Pthreads within
that process. If the process is using system-scope (the 1:1 model), then the root
user authority is required to modify the scheduling or priority aspects of the
Pthread. Root privileges are also required to modify the policy of a Pthread
outside the scope of the process, or to set the priority of a Pthread or process to
a fixed value.
Chapter 8. Introduction to POSIX threads
325

SCHED_FIFO
SCHED_FIFO is First-In/First-Out scheduling. A Pthread with fixed priority and
SCHED_FIFO scheduling will be allowed to run to completion unless it:
It yields the processor voluntarily.
It blocks because something it needs is not available.
A kernel thread with a higher priority is available to run.
Running Pthreads under a FIFO policy is discouraged because it prevents the
system from balancing resource utilization in a way that lets all tasks run.
SCHED_RR
SCHED_RR is round-robin scheduling. Pthreads using this policy will receive a
time slice (time slicing is a mechanism that ensures that every Pthread is allowed
time to execute by preempting running Pthreads at fixed intervals) of the CPU.
Only Pthreads with a fixed priority can take advantage of round-robin scheduling.
As each Pthread completes its allotted time on the processor, it is placed at the
end of the queue. The Pthread at the front of the queue is allowed to run. This
circular sharing of the processor ensures that systems that require a given
number of Pthreads to execute on a regular basis will not starve any task.
SCHED_OTHER
SCHED_OTHER is the default scheduling policy used by AIX. Unlike
first-in/first-out and round-robin, the SCHED_OTHER policy is used to manage
Pthreads that have dynamic priority. In this policy the scheduler runs 100 times
per second, and at each interval charges the Pthread(s) running on a
processor(s) for its use of the CPU. Floating priorities are computed by using the
Pthread priority, the CPU usage, and various tunable kernel parameters.
By default, a Pthread is created with a priority of 40 plus its nice value. The
default nice value is 20 and can be changed within a relative range of plus or
minus 20 using the nice command. This results in a default priority for a newly
created Pthread of 60. If the Pthread has control of a processor when the
scheduler next runs, the Pthread is charged and its priority is recalculated. The
formula used to recalculate Pthread priority is:
recalculated_priority = P_nice + (C * R / 32)
Where:
P_nice
40 + NICE.
C (recent CPU usage)Old CPU Usage * D / 32 (default D is 16 and ranges from
0-32).
R / 32
The CPU penalty factor, where R is a kernel tunable and
ranges from 0-32.
326
Developing and Porting C and C++ Applications on AIX

The values for D and R are tunable. D defaults to 16 and may range from 0 to 32;
R also defaults to 16 and ranges in value from 0 to 32.
SCHED_FIFO2
SCHED_FIFO2 is similar to SCHED_FIFO, but allows a Pthread that has slept
for a short period of time to be added to the front of the run queue when it
awakened. The time period is a kernel parameter that can be modified. This
affinity limit is adjusted using the schedtune command:
/usr/samples/kernel/schedtune -a N
Where N is the number of clock ticks.
SCHED_FIFO3
SCHED_FIFO3 will always place an awakened Pthread at the front of the queue,
regardless of how long it was asleep. While this is similar to FIFO2, we should
remember that when a FIFO3 Pthread is on a run queue, the parameters of the
queue are modified to prevent a FIFO2 Pthread from being put at the head of the
queue.
Setting Pthread scheduling policies
Scheduling policies are inherited from the process that created the Pthread. If a
different scheduling policy is required, the pthread_attr_setinheritsched() routine
can be used to set the thread attribute prior to Pthread creation. Again, this
applies to Pthreads in a process that uses the M:N model:
pthread_attr_setinheritsched(pthread_attr_t *thread_attr, int value)
Where:
thread_attr
Thread attribute structure or object
Value
Either PTHREAD_INHERIT_SCHED or
PTHREAD_EXPLICIT_SCHED
If the Pthread is already created, the pthread_setschedparam() routine can be
used to modify a Pthread's scheduling policy. The scheduling change is not
realized until the next time the Pthread is put on the run queue. That is, if a
Pthread is currently running and a call is made to pthread_setschedparam(), the
change will not become effective until the Pthread has consumed its time slice:
pthread_setschedparam(pthread_t thread_id, int policy, *param)
Where:
thread_id
Thread ID to operate on.
policy
Scheduling policy, for example, SCHED_RR or
SCHED_FIFO.
param
This structure contains a policy field but is ignored.
Chapter 8. Introduction to POSIX threads
327

8.7.3 Scheduling limitations
Thread scope affects how the underlying kernel thread's scheduling will behave.
For system scope threads, there exists one kernel thread for every user thread,
that is, pthread. Modifying a system scope thread's scheduling policy modifies
the underlying kernel thread's scheduling policy; thus, system scope threads are
limited to SCHED_OTHER unless the process has root privileges. Also, in the
case of system scope threads, the other scheduling policies (FIFO and
round-robin) require a fixed priority and root is the only user who can modify a
thread's priority.
As stated earlier, process scope threads are threads that have M user threads to
N kernel threads. The scheduling policy for process scope threads is handled at
the library level and does not affect the underlying kernel thread(s). At the
process level, the scheduling policy affects how the ready-to-run Pthreads are
placed in the run queue. The scheduling policy of the underlying kernel thread
(which implements the virtual processor, or VP) is managed by the kernel.
8.8 Pthread specific environment variables in AIX
Tuning multi-threaded applications for optimal performance can be challenging
on any system. AIX provides a set of environmental variables that modify the
behavior of various threading functions. Let us look at some of these variables
one at a time.
AIXTHREAD_SCOPE
This environment variable controls the mapping of the application-level Pthreads
to entries in the OS scheduling queue. The default thread scope for AIX is
process, and this applies to the creation of Pthreads only when the attribute
argument to pthread_create() is NULL; otherwise, the contents of the attribute
structure determines the scope of the Pthread. This implies that a single process
may contain Pthreads of both scopes.
AIXTHREAD_SCOPE permits exploring application behavior by being able to
change the thread scope without having to modify the application. The default
thread scope can be set by setting the variable in the environment:
AIXTHREAD_SCOPE=[P|S]
Where:
P
For process scope
S
For system scope
328
Developing and Porting C and C++ Applications on AIX

The preferred setting for any particular application is dependent upon its
behavior. The performance difference between process and system scope, in
current levels of AIX 5L, has become almost negligible for certain applications.
AIXTHREAD_MNRATIO
This environment variable controls the scaling factor used within the Pthread
library when creating and terminating Pthreads. The default M:N ratio for current
levels of the operating system is 8:1, that is, eight Pthreads for every kernel
thread. This default ratio may not be suitable for all workloads, however, so the
user can modify the M:N ratio by setting and exporting the environmental variable
as follows:
AIXTHREAD_MNRATIO=M:N
Where:
M
Number of user threads
N
Number of kernel threads
AIXTHREAD_MNRATIO applies only to process-scope threads. It may be useful
for applications with a very large number of Pthreads.
SPINLOOPTIME
SPINLOOPTIME controls the number of times a Pthread will attempt to obtain a
mutex before blocking on the mutex availability. For example, sometimes it makes
more sense for a Pthread to keep trying just a bit longer to acquire a lock in
hopes of the other Pthread releasing the lock while the current Pthread is awake
and waiting. This can avoid the overhead of putting the Pthread to sleep,
managing the event that occurs when the lock is released, and waking the
Pthread up and setting it to run so that it can now acquire the lock.
On multiprocessor systems the default value for SPINLOOPTIME is 40. The spin
loop time can be modified by setting and exporting the environmental variable:
SPINLOOPTIME=N
Where:
N
Number of times to retry a busy lock before yielding to
another Pthread.
YIELDLOOPTIME
Similar to SPINLOOPTIME, YIELDLOOPTIME controls the number of times a
Pthread will yield itself before blocking on a mutex. In some cases, yielding the
processor will permit other Pthreads time to run, thus reducing the amount of
time the waiting Pthread will have to spin and/or wait for the mutex to be
unlocked. This allows work to be accomplished while one Pthread is waiting for a
Chapter 8. Introduction to POSIX threads
329

resource to be available. It also keeps the Pthread from going to sleep; instead,
the Pthread is put back on the run queue to try again to acquire the lock. Keeping
the Pthread from going to sleep can result in improved overall performance,
depending upon how much more time the Pthread needs to wait for the lock
acquisition.
The default for YIELDLOOPTIME is 0. If the Pthread spins on the lock and
cannot get it, it will be put to sleep. Increasing the value of this variable will
prevent the Pthread from going to sleep quite so quickly. We can modify the value
of yield loop time by setting and exporting the environmental variable:
AIXTHREAD_YIELDLOOPTIME=N
Where:
N
Number of times to yield to acquire a busy mutex or spin
lock
AIXTHREAD_MINKTHREADS
Sets the minimum number of kernel threads that should be used for a process.
The default is eight, which means that an average threaded process will have at
least eight kernel threads available as a resource for scheduling Pthreads. This
environment variable can be set as follows:
AIXTHREAD_MINKTHREADS=N
Where:
N
Number of kernel threads
AIXTHREAD_MUTEX_DEBUG
AIX is capable of maintaining, in running threaded processes, information
regarding mutexes; this information is then available for application debugging.
On current levels of AIX 5L, the default setting for this variable is OFF, but that
was not always the case. Since maintaining this information is not free, we may
observe unexplainable and undesirable performance characteristics. If we
suspect that our application may be impacted, set this variable in our
environment to ensure mutex debugging information is not being collected:
AIXTHREAD_MUTEX_DEBUG=[ON|OFF]
If you plan on debugging the Pthreads in your application, you will want to turn
this variable ON.
330
Developing and Porting C and C++ Applications on AIX

AIXTHREAD_COND_DEBUG
Similar to AIXTHREAD_MUTEX_DEBUG, this variable controls the collection of
debug information for condition variables. Its default setting is now OFF. For the
same conditions described above, we can set the variable in our environment:
AIXTHREAD_COND_DEBUG=[ON|OFF]
AIXTHREAD_RWLOCK_DEBUG
Causes the Pthreads library to maintain a list of all read-write locks that can be
viewed by debugging programs. Like the other two debugging variables above,
the default for this is now OFF. Set the variable in your environment to ON for
debugging the application.
AIXTHREAD_RWLOCK_DEBUG=[ON|OFF]
AIXTHREAD_GUARDPAGES
Each Pthread is created with its own stack; the stack size is controlled via the
thread attributes or via (for a default value) an environment variable. Since the
stack is allocated in the process heap, there is some potential for inadequately
sized stacks to allow a Pthread to cause a stack to grow beyond its maximum
size. To assist in detecting this behavior, and also to guard against errant
memory writes, the stack can be protected by setting pages at the top of the
stack as read-only. Any attempt to write onto these pages will result in the
application receiving a segmentation violation signal, which will occur
immediately. Debugging the application will allow us to investigate the conditions
at the time of the stack overflow and decide on an appropriate course of action
(increase the stack size, redesign part of the application, change where the data
is stored, and so on).
To enable and set the number of guard pages, we need to export the
environmental variable:
AIXTHREAD_GUARDPAGES=N
Where:
N
Number of 4 KB size pages
The default value is 0, which means no guard pages are created. If the
application specifies its own stack, or uses large pages17 for its process heap, no
guard pages are created.
AIXTHREAD_SLPRATIO
The “sleep ratio” value tells the system how many kernel threads should be held
in reserve for sleeping Pthreads. This tuning parameter allows greater
17 See 3.6, “Large page support” on page 157.
Chapter 8. Introduction to POSIX threads
331

management of kernel resources, as sleeping Pthreads do not require a
matching kernel thread. And since Pthreads are (usually) woken one at a time,
the scheduling and running of the Pthreads can be matched to existing resources
without needing to acquire additional kernel threads.
One of the control variables that impact the scheduling of Pthreads is created
with a process based contention scope. This environment variable will be set as
follows:
AIXTHREAD_SLPRATIO=k:p
Where:
k
The number of kernel threads that should be held in
reserve for p sleeping pthreads
Any positive integer value may be specified for p and k. If k > p, then the ratio is
treated as 1:1 (that is, you cannot specify more kernel threads than Pthreads).
The default sleep ratio is 1:12.
AIXTHREAD_STK
Each Pthread has its own stack; the default size (on current levels of AIX) is 96
KB for 32-bit applications and 192 KS for 64-bit applications. While we will find
this default adequate for many applications, it is also true that there are
exceptions. Instead of modifying code to adjust a parameter in the Pthread
attribute structure and then recompiling and rebuilding our application, we can
use this environment variable to specify stacks of up to 256MB in size. Keep in
mind, however, that Pthreads stacks are created in the process heap, and thus
impact the amount of space available for dynamically allocated memory.
We can modify the stack size for Pthreads created without specifying the stack
size programmatically by exporting the environment variable:
AIXTHREAD_STK=N
Where:
N
number of bytes
8.9 User API for Solaris threaded applications
The new user thread library provides for source compatibility with Solaris user
thread routines. This allows applications that are run on Solaris systems to be
recompiled without changing their application source code so that they can run
on AIX beginning with Version 5.2. The API for Solaris threaded applications for
332
Developing and Porting C and C++ Applications on AIX

AIX 5L Version 5.2 is designed to be compatible with Solaris Version 8 of the
thread library.
The Solaris user thread library does not alter the Pthread library, so compatibility
with POSIX and X/Open standards for Pthreads are maintained on AIX. The
Solaris user threads support is implemented on top of the Pthreads; therefore,
existing POSIX thread applications will not be affected by the Solaris user thread
support.
There is, however, no binary compatibility with applications compiled under
Solaris. All source code is required to be recompiled on AIX.
Note: Solaris user thread library function names have the prefix thr_, while
POSIX thread library function names have the prefix pthread_.
8.9.1 Application binary interface (ABI)
The design of the existing ABI of the pthread library is not altered with respect to:
Exported function names
Exported function signatures
Exported data structures
Exported data structures used in file formats
AIX LPP packaging
The filesets listed in Table 8-1 contain the AIX files needed for the user API for
Solaris threaded applications.
Table 8-1 Filesets for Solaris user thread library
File
Fileset
/usr/ccs/lib/libthread.a
bos.adt.lib
/usr/include/thread.h
bos.adt.include
/usr/include/synch.h
bos.adt.include
There are no user interfaces required for either the command line, SMIT, or
Web-based System Manager. All applications need to be recompiled.
Chapter 8. Introduction to POSIX threads
333

334
Developing and Porting C and C++ Applications on AIX

9
Program parallelization
Chapter 9.
using OpenMP
The latest versions of the IBM C and C++ compiler products for AIX support
OpenMP, which parallelizes your C and C++ applications. By inserting simple
OpenMP directives1 (start with “#pragma omp”) into your source code, your
application can be parallelized and efficiently performs on SMP systems.
This chapter explains how to use the OpenMP support of the IBM C and C++
compiler products for AIX.
The OpenMP support on AIX internally uses the POSIX threads explained in
Chapter 8, “Introduction to POSIX threads” on page 275. Although the
application programmer are not exposed to the POSIX threads when
programming C and C++ applications with OpenMP, several concepts in the
POSIX threads programming model are sometimes beneficial.
For further information about the OpenMP support provided by the compiler
products, please refer to C for AIX Compiler Reference, SC09-4960 and
VisualAge C++ for AIX Compiler Reference, SC09-4959.
1 In addition to the OpenMP directives, the IBM C and C++ compiler products for AIX also have been
supporting IBM SMP directives to parallelize your C and C++ applications. See Appendix E,
“Supported IBM SMP directives” on page 483 for more information about the IBM SMP directives.
© Copyright IBM Corp. 2000, 2003. All rights reserved.
335

9.1 Introduction to OpenMP
OpenMP is an industry specification describing a common set of APIs for
multi-platform SMP programming. OpenMP is a portable, scalable programming
model designed to provide SMP programmers with a simple and flexible interface
for developing parallel applications for platforms ranging from the desktop to the
super computer. It is comprised of three components, as shown in Table 9-1.
Table 9-1 OpenMP components
Components
Relevant section(s)
Compiler directives
From 9.3, “Classification of OpenMP directives” on
page 337 to 9.9, “Data-sharing attribute clauses” on
page 355
Run-time library
Section 9.10, “Run-time library functions” on page 363
functions
Environment variables
Section 9.11, “Environment variables” on page 373
For further information about OpenMP, please visit the following URL:
http://www.openmp.org
9.2 The OpenMP programming model
OpenMP uses the fork-join model of parallel execution. Although this fork-join
model can be useful for solving a variety of problems, it is somewhat tailored for
large array-based applications. OpenMP is intended to support programs that will
execute correctly both as parallel programs (multiple threads of execution and a
full OpenMP support library) and as sequential programs (directives will be
ignored and a stub library will be linked).
A program written with the OpenMP C/C++ API begins execution as a single
thread of execution called the master thread. The master thread executes in a
serial region until the first parallel construct is executed. In the OpenMP C/C++
API, the parallel directive constitutes a parallel construct. When a parallel
construct is executed, the master thread creates a team of threads, and the
master becomes master of the team. Each thread in the team executes the
statements in the dynamic extent of a parallel region, except for the work-sharing
constructs. Work-sharing constructs must be executed by all threads in the team
in the same order, and the statements within the associated structured block are
executed by one or more of the threads.
336
Developing and Porting C and C++ Applications on AIX

9.3 Classification of OpenMP directives
The IBM C and C++ compiler products for AIX support OpenMP pragma
directives that exploit shared memory parallelism.2 Directives are based on the
#pragma directive defined in the C and C++ standards. Compilers that support
the OpenMP C and C++ API will include a command line option that activates
and allows interpretation of all OpenMP compiler directives.
We categorized OpenMP pragmas into the four categories shown in Table 9-2.
Table 9-2 OpenMP directive categories
Category
Brief explanation and related sections
Parallel constructs
These pragmas enable the programmer to define the parallel
regions in which work is done by threads in parallel. Most of the
OpenMP directives either statically or dynamically bind to an
enclosing parallel region. See Section 9.4, “Parallel region
construct” on page 338 for more information.
Work-sharing
This category of pragmas enables the programmer to define
constructs
how work will be distributed across the threads in a parallel
region. See Section 9.5, “Work-sharing constructs” on
page 340 for more information.
Synchronization
This category enables the programmer to control
constructs
synchronization among threads. See Section 9.7,
“Synchronization constructs” on page 348 for more information.
Data sharing
This category of pragmas enables the programmer to define the
attributes
private/shared context of data within a parallel region. See
Section 9.9, “Data-sharing attribute clauses” on page 355 for
more information.
Note: Throughout this chapter, the term construct refers to the pragma line
and the following single code block. If a pragma line does not have any
following code blocks, then we use the term directive.
9.3.1 The OpenMP directive format
The syntax of an OpenMP directive is specified as follows:
#pragma omp directive-name [clause[ [,] clause]...] new-line
2 VisualAge C++ for AIX supports OpenMP since Version 6, while C for AIX has been supporting it
since Version 5.
Chapter 9. Program parallelization using OpenMP
337

Each directive starts with “#pragma omp”, to reduce the potential for conflict with
other (non-OpenMP or vendor extensions to OpenMP) pragma directives with
the same names. The remainder of the directive follows the conventions of the C
and C++ standards for compiler directives. In particular, white space can be used
before and after the #, and sometimes white space must be used to separate the
words in a directive. Preprocessing tokens following the #pragma omp are
subject to macro replacement.
Directives are case-sensitive. The order in which clauses appear in directives is
not significant. Clauses on directives may be repeated as needed, subject to the
restrictions listed in the description of each clause. If a variable-list appears in a
clause, it must specify only variables. Only one directive-name can be specified
per directive.
For example, the following directive is not allowed:
/* ERROR - multiple directive names not allowed */
#pragma omp parallel barrier
An OpenMP directive applies to at most one succeeding statement, which must
be a structured block. For those directives, which allow a structured block
following the directive name, are termed Constructs. Throughout this chapter, the
term Construct is used as appropriate.
9.4 Parallel region construct
The following directive defines a parallel region, which is a region of the program
that is to be executed by multiple threads in parallel. This is the fundamental
construct that starts parallel execution:
#pragma omp parallel [clause[ [, ]clause] ...] new-line
structured-block
The clause is one of the following:
if(scalar-expression)
private(variable-list)
firstprivate(variable-list)
default(shared | none)
shared(variable-list)
copyin(variable-list)
reduction(operator: variable-list)
num_threads(integer-expression)
For a detailed description of the clauses, see 9.9, “Data-sharing attribute
clauses” on page 355.
338
Developing and Porting C and C++ Applications on AIX

When a thread encounters a parallel construct, a team of threads is created if
one of the following cases is true:
No if clause is present.
The if expression evaluates to a nonzero value.
This thread becomes the master thread of the team, with a thread number of 0,
and all threads in the team, including the master thread, execute the region in
parallel. If the value of the if expression is zero, the region is serialized.
The number of threads in a parallel region is determined by the following factors,
in order of precedence:
1. Use of the omp_set_num_threads() library function.3
2. Setting of the OMP_NUM_THREADS environment variable.4
3. Implementation default.5
The sample code in Example 9-1 illustrates the parallel region execution. In this
program, every thread executes and prints the “Hello World.” enclosed in the
parallel region.
Example 9-1 omp_parallel.c
#include <stdio.h>
int main(int argc, char *argv[])
{
printf("Before forking a parallel region.\n");
/* Fork a team of threads giving them their own copies of variables. */
#pragma omp parallel
{
printf("Hello World.\n");
} /* All threads join master thread and terminate. */
printf("Exiting the program... Bye.\n");
}
To compile this code, run:
$ cc_r -qsmp=omp omp_parallel.c
3 See 9.10.1, “Execution environment functions” on page 364.
4 See 9.11.2, “OMP_NUM_THREADS” on page 374.
5 On AIX, it is equivalent to the number of equipped processors.
Chapter 9. Program parallelization using OpenMP
339

Note:
Use cc_r when compiling C programs with OpenMP directives. Use xlC_r
when compiling C++ programs with OpenMP directives.
The -qsmp=omp compiler option should be specified when compiling C
and C++ programs with OpenMP directives. If not specified, a stub library
(/usr/lib/libxlomp_ser.a), which does not create any threads, will be in-lined
instead of the actual OpenMP library (/usr/lib/libxlsmp.a).
Before running this program, set the environment variable
OMP_NUM_THREADS in order to set the number of threads that you want to
execute in parallel. For example,
$ export OMP_NUM_THREADS=3 a.out
The output of the program is as follows:
Before forking a parallel region
Hello World.
Hello World.
Hello World.
Exiting the program... Bye.
As instructed by the environment variable, the program forked three threads and
executed the parallel region construct.
9.5 Work-sharing constructs
A work-sharing construct distributes the execution of the associated statement
among the members of the team that encounter it. The work-sharing directives
do not launch new threads, and there is no implied barrier on entry to a
work-sharing construct.
OpenMP defines the three work-sharing constructs, for, sections, and single,
explained in the following sections:
Section 9.5.1, “for construct” on page 341
Section 9.5.2, “sections construct” on page 344
Section 9.5.3, “single construct” on page 345
A work-sharing construct must be enclosed dynamically within a parallel region
in order for the directive to execute in parallel. They must be executed by all
members of a team or none at all. Successive work-sharing constructs must be
executed in the same order by all members of a team.
340
Developing and Porting C and C++ Applications on AIX

9.5.1 for construct
The for directive specifies that the iterations of the loop immediately following it
must be executed in parallel by the team. This assumes a parallel region has
already been initiated; otherwise, it executes in serial on a single processor.
The syntax for the for construct is as follows:
#pragma omp for [clause[[,] clause] ... ] new-line
for-loop
The clause is one of the following:
private(variable-list)
firstprivate(variable-list)
lastprivate(variable-list)
reduction(operator: variable-list)
ordered
schedule(kind[, chunk_size])
nowait
Let us discuss a few of the clauses in brief. For a detailed description of the other
clauses, see 9.9, “Data-sharing attribute clauses” on page 355.
The ordered clause must be present when ordered directives bind to for
construct.
The schedule clause describes how iterations of the loop are divided among the
threads in the team. The schedule kind may be one of the following:
static
Loop iterations are divided into pieces of chunk_size and then
statically assigned to threads. If chunk is not specified, the
iterations are evenly (if possible) divided contiguously among the
threads.
dynamic
Loop iterations are divided into pieces of size chunk_size and
dynamically scheduled among the threads; when a thread
finishes one chunk, it is dynamically assigned another. The
default chunk_size is 1.
guided
The chunk_size is exponentially reduced with each dispatched
piece of the iteration space. The chunk_size specifies the
minimum number of iterations to dispatch each time. The default
chunk_size is 1.
runtime
The scheduling decision is deferred until run time by the
environment variable OMP_SCHEDULE. It is not allowed to
specify a chunk_size for this clause.
Chapter 9. Program parallelization using OpenMP
341

If the nowait clause is specified, then threads do not synchronize at the end of
the parallel loop. Threads proceed directly to the next statements after the loop.
The restrictions to the for directive are as follows:
The for loop must be a structured block, and, in addition, its execution must
not be terminated by a break statement.
The values of the loop control expressions of the for loop associated with a for
directive must be the same for all the threads in the team.
The for loop iteration variable must have a signed integer type.
Only a single schedule clause can appear on a for directive.
Only a single ordered clause can appear on a for directive.
Only a single nowait clause can appear on a for directive.
The value of the chunk_size expression must be the same for all threads in
the team.
The for directive requires that the for loop must have canonical form. In the
OpenMP specifications, the canonical form allows the number of loop iterations
to be computed on entry to the loop. For example, the following code fragment:
#pragma omp for reduction(+:overallsum)
for (n = 0; n != ARRSIZE; n++) {
overallsum += array[n];
}
would yield the following error when it is compiled:
" 1506-818 (S) Controlling expression of the for loop is not in the canonical
form."
because the loop test expression n != ARRSIZE, should be replaced by
n < ARRSIZE.
The example program shown in Example 9-2 on page 343 illustrates the use of
the for directive. In this example, the arrays a, b, and total are shared by all
threads. The variable i is private to each thread and each thread has its own copy
of it. The iterations of the for is distributed dynamically in chunk_size pieces. The
nowait clause ensures that threads will not synchronize upon completing their
individual pieces of work.
342
Developing and Porting C and C++ Applications on AIX

Example 9-2 omp_ws_for.c
#include <stdio.h>
#define CHUNKSIZE 5
#define N
10
int main(int argc, char *argv[])
{
int i, chunk_size;
float a[N], b[N], total[N];
/* Some initializations. */
for (i = 0; i < N; i++)
a[i] = b[i] = i * 1.0;
chunk_size = CHUNKSIZE;
#pragma omp parallel shared(a,b,total,chunk_size) private(i)
{
#pragma omp for schedule(dynamic,chunk_size) nowait
for (i=0; i < N; i++)
total[i] = a[i] + b[i];
} /* end of pragma omp parallel for. */
for (i = 0; i < N; i++)
printf(“The total value is = %f\n”, total[i]);
}
When executed, the program prints the output shown in Example 9-3.
Example 9-3 Output fromomp_ws_for.c
The total value is = 0.000000
The total value is = 2.000000
The total value is = 4.000000
The total value is = 6.000000
The total value is = 8.000000
The total value is = 10.000000
The total value is = 12.000000
The total value is = 14.000000
The total value is = 16.000000
The total value is = 18.000000
Chapter 9. Program parallelization using OpenMP
343

9.5.2 sections construct
The sections directive identifies a non-iterative work-sharing construct that
specifies a set of constructs that are to be divided among threads in a team.
Each section is executed once by a thread in the team. The syntax of the
sections directive is as follows:
#pragma omp sections [clause[[,] clause] ...] new-line
{
[#pragma omp section new-line]
structured-block
[#pragma omp section new-line
structured-block ]
...
}
The clause is one of the following:
private(variable-list)
firstprivate(variable-list)
lastprivate(variable-list)
reduction(operator: variable-list)
nowait
Each section is preceded by a section directive, although the section directive is
optional for the first section. The section directives must appear within the lexical
extent of the sections directive. There is an implicit barrier at the end of a
sections construct, unless a nowait is specified.
Restrictions to the sections directive are as follows:
A section directive must not appear outside the lexical extent of the sections
directive.
Only a single nowait clause can appear on a sections directive.
The example shown in Example 9-4 on page 345 illustrates the use of the
sections directive. The program is a slight modification of Example 9-2 on
page 343, which illustrated the use of the for directive. The first N/2 iterations of
the for loop will be distributed to the first thread, and the rest will be distributed to
the second thread. When each thread finishes its block of iterations, it proceeds
with whatever code comes next (nowait).
344
Developing and Porting C and C++ Applications on AIX

Example 9-4 omp_ws_section.c
#include <stdlib.h>
#define N 10
int main(int argc, char *argv[])
{
int i;
float a[N], b[N], total[N];
/* Some initializations. */
for (i = 0; i < N; i++)
a[i] = b[i] = i * 1.0;
#pragma omp parallel shared(a,b,total) private(i)
{
#pragma omp sections nowait
{
#pragma omp section
for (i = 0; i < N / 2; i++)
total[i] = a[i] + b[i];
#pragma omp section
for (i = N / 2; i < N; i++)
total[i] = a[i] + b[i];
} /* end of pragma omp sections nowait. */
} /* end of pragma omp parallel shared. */
for (i = 0; i < N; i++)
printf("The total value is = %f\n", total[i]);
}
When executed, the program prints the same output shown in Example 9-3 on
page 343.
9.5.3 single construct
The single directive identifies a construct that specifies that the associated
structured block is executed by only one thread in the team (not necessarily the
master thread). This may be useful when dealing with sections of code that are
not thread safe (such as I/O). The syntax of the single directive is as follows:
#pragma omp single [clause[[,] clause] ...] new-line
structured-block
The clause is one of the following:
private(variable-list)
firstprivate(variable-list)
copyprivate(variable-list)
Chapter 9. Program parallelization using OpenMP
345

nowait
Restrictions to the single directive are as follows:
Only a single nowait clause can appear on a single directive.
The copyprivate clause must not be used with the nowait clause.
single constructs cannot be nested within sections construct.
The following example program illustrates the use of a single construct:
/* File : omp_ws_single.c */
#include <stdio.h>
int main(int argc, char *argv[])
{
printf("Before forking a parallel region.\n");
/* Fork a team of threads giving them their own copies of variables. */
#pragma omp parallel
{
/* The printf() line is executed only once by any of the thread. */
#pragma omp single
printf("Before printing Hello World...\n");
/* The printf() line is executed by all the thread. */
printf("Hello World.\n");
/* The printf() line is executed only once by any of the thread. */
#pragma omp single
printf("Done.. Leaving.\n");
}
}
If executed, the program prints the following output:
Before forking a parallel region.
Before printing Hello World...
Hello World.
Done.. Leaving.
Hello World.
Hello World.
In the above output, the statements in the single construct were executed only
once by one of the threads from the team.
346
Developing and Porting C and C++ Applications on AIX

9.6 Combined parallel work-sharing constructs
Combined parallel work-sharing constructs are shortcuts for specifying a parallel
region that contains only one work-sharing construct. The semantics of these
directives are identical to that of explicitly specifying a parallel directive followed
by a single work-sharing construct.
The following sections describe the combined parallel work-sharing constructs:
Section 9.6.1, “parallel for construct” on page 347
Section 9.6.2, “parallel sections construct” on page 348
9.6.1 parallel for construct
The parallel for directive is a shortcut for a parallel region that contains only a
single for directive. The syntax of the parallel for directive is as follows:
#pragma omp parallel for [clause[[,] clause] ...] new-line
for-loop
This directive allows all the clauses of the parallel directive and the for directive,
except the nowait clause, with identical meanings and restrictions. The
semantics are identical to explicitly specifying a parallel directive immediately
followed by a for directive.
The following example illustrates the use of the parallel for construct. Iterations of
the for loop will be distributed in equal sized blocks to each thread in the team
(schedule static):
/* File : omp_parallel_for.c */
#include <stdlib.h>
#define N 10
#define CHUNKSIZE 5
int main(int argc, char *argv[])
{
int i, chunk_size;
float a[N], b[N], total[N];
/* Some initializations. */
for (i = 0; i < N; i++)
a[i] = b[i] = i * 1.0;
chunk_size = CHUNKSIZE;
#pragma omp parallel for \
shared(a,b,total,chunk_size) private(i) \
schedule(static,chunk_size)
for (i = 0; i < N; i++)
Chapter 9. Program parallelization using OpenMP
347

total[i] = a[i] + b[i];
for (i = 0; i < N; i++)
printf("Total value is = %f\n", total[i]);
}
When executed, the program prints the same output shown in Example 9-3 on
page 343.
9.6.2 parallel sections construct
The parallel sections directive provides a shortcut form for specifying a parallel
region containing only a single sections directive. The semantics are identical to
explicitly specifying a parallel directive immediately followed by a sections
directive. The syntax of the parallel sections directive is as follows:
#pragma omp parallel sections [clause[[,] clause] ...] new-line
{
[#pragma omp section new-line]
structured-block
[#pragma omp section new-line
structured-block ]
...
}
The clause can be one of the clauses accepted by the parallel and sections
directives, except the nowait clause.
9.7 Synchronization constructs
Synchronization refers to the time order in which threads access shared or global
variables. Because several parallel threads may need to access or update the
same shared variable at about the same time, explicit control of synchronization
is often required. A program with faulty synchronization may compile and run
without incident, but will produce incorrect results.
OpenMP provides a variety of synchronization constructs, explained in the
following sections, that control how the execution of each thread proceeds
relative to other team threads:
Section 9.7.1, “master construct” on page 349
Section 9.7.2, “critical construct” on page 349
Section 9.7.3, “barrier directive” on page 350
Section 9.7.4, “atomic construct” on page 352
348
Developing and Porting C and C++ Applications on AIX

Section 9.7.5, “flush directive” on page 352
Section 9.7.6, “ordered construct” on page 353
9.7.1 master construct
The master directive identifies a construct that specifies a structured block that is
executed by the master thread of the team. The syntax of the master directive is
as follows:
#pragma omp master new-line
structured-block
Other threads in the team do not execute the associated structured block. There
is no implied barrier either on entry to or exit from the master construct.
It is not allowed to branch into or out of master block.
9.7.2 critical construct
The critical directive identifies a construct that restricts execution of the
associated structured block to a single thread at a time. The syntax of the critical
directive is as follows:
#pragma omp critical [(name)] new-line
structured-block
An optional name may be used to identify the critical region. In the following
sample code, all threads in the team will attempt to execute in parallel; however,
because of the critical construct surrounding the increment of x, only one thread
will be able to read/increment/write x at any time:
#include <stdlib.h>
int main(int argc, char *argv[])
{
int x = 0;
#pragma omp parallel shared(x)
{
/* access allowed for one thread only at a time. */
#pragma omp critical
x = x + 1;
} /* end of pragma omp parallel shared. */
}
A small example program omp_reduction2.c, illustrating the use of the critical
directive, is shown in 9.9.6, “reduction clause” on page 358.
Chapter 9. Program parallelization using OpenMP
349

9.7.3 barrier directive
The barrier directive synchronizes all the threads in a team. When executed,
each thread in the team waits until all of the others have reached this point (see
Figure 9-1). In this figure, the thread 2 has not been reached at the barrier
statement; therefore, the other threads cannot proceed their execution flow.
Thread 0
Thread 1
Thread 2
Thread 3
Barrier
Time
Figure 9-1 Concept of barrier
The syntax of the barrier directive is as follows:
#pragma omp barrier new-line
After all threads in the team have executed the barrier, each thread in the team
begins executing the statements after the barrier directive in parallel.
In the following program, though the printf() statement is inside the parallel
construct, the value of x is printed only once after all the threads have finished
incrementing the variable. It is because the inclusion of the barrier directive has
made it possible:
/* File : omp_barrier.c */
#include <stdlib.h>
int main(int argc, char *argv[])
{
int x = 0;
#pragma omp parallel shared(x)
350
Developing and Porting C and C++ Applications on AIX

{
#pragma omp critical
x = x + 1;
/*
* The following block will be executed after all the threads finish
* their work.
*/
#pragma omp barrier
{
#pragma omp single
printf("The value of x is : %d\n", x);
} /* end of pragma omp barrier */
} /* end of pragma omp parallel shared. */
}
The output is printed as:
The value of x is : 3
If we remove the barrier construct from the code, we might not have achieved the
expected results. The output would have been any of the following lines:
The value of x is : 1
The value of x is : 2
The value of x is : 3
depending upon the thread that gets a chance to execute the single block.
Note: Because the barrier directive does not have a C language statement as
part of its syntax, there are some restrictions on its placement within a
program. The example below illustrates these restrictions:
/*
* ERROR - The barrier directive cannot be the immediate substatement of
* an if statement.
*/
if (x != 0)
#pragma omp barrier
your_statement_line;
...
/* OK - The barrier directive is enclosed in a compound statement. */
if (x != 0) {
#pragma omp barrier
your_statement_line;
}
...
Chapter 9. Program parallelization using OpenMP
351

9.7.4 atomic construct
The atomic directive ensures that a specific memory location is updated
atomically, rather than exposing it to the possibility of multiple, simultaneous
writing threads.
The syntax of the atomic directive is as follows:
#pragma omp atomic new-line
expression-stmt
The expression statement must have one of the following forms:
x binop= expr
x++
++x
x--
--x
Where:
x is an lvalue expression with scalar type.
expr is an expression with scalar type, and it does not reference the object
designated by x.
binop is not an overloaded operator and is one of +, *, -, /, &, ^, |, <<, or >>.
Only the load and store of the object designated by x are atomic; the evaluation
of expr is not atomic. To avoid race conditions, all updates of the location in
parallel should be protected with the atomic directive, except those that are
known to be free of race conditions.
9.7.5 flush directive
The flush directive identifies a synchronization point at which the implementation
must provide a consistent view of memory. Thread-visible variables are written
back to memory at this point. The syntax of the flush directive is as follows:
#pragma omp flush [(variable-list)] new-line
The optional variable list contains a list of named variables that will be flushed in
order to avoid flushing all variables. For pointers in the list, note that the pointer
itself is flushed, not the object it references to.
352
Developing and Porting C and C++ Applications on AIX

Note that because the flush directive does not have a C language statement as
part of its syntax, there are some restrictions on its placement within a program.
The example below illustrates these restrictions:
/*
* ERROR - The flush directive cannot be the immediate substatement of
* an if statement.
*/
if (x != 0)
#pragma omp flush (x)
your_statement_line;
...
/* OK - The flush directive is enclosed in a compound statement. */
if (x != 0) {
#pragma omp flush (x)
your_statement_line;
}
...
A variable specified in a flush directive must not have a reference type provided
by the C++ language. The flush directive only provides consistency between the
executing thread and global memory. To achieve a globally consistent view
across all threads, each thread must execute a flush operation.
9.7.6 ordered construct
The ordered directive specifies that iterations of the enclosed loop will be
executed in the same order as if they were executed on a serial processor. The
syntax of the ordered directive is as follows:
#pragma omp ordered new-line
structured-block
An ordered directive can only appear in the dynamic extent of the for or parallel
for directive. Only one thread is allowed in an ordered section at any time. It is
illegal to branch into or out of an ordered block. A loop which contains an ordered
directive must be a loop with an ordered clause. The following code fragment
illustrates the use of the ordered directive:
#pragma omp for ordered
for (i = 0; i
<n; i++) {
...
if (i <= 10) {
...
#pragma omp ordered
{ ... }
}
...
if (i > 10) {
Chapter 9. Program parallelization using OpenMP
353

...
#pragma omp ordered
{ ... }
}
...
}
9.8 Data environment: The threadprivate directive
The threadprivate directive is provided to make file-scope, namespace-scope, or
static block-scope variables local to a thread. The syntax of the threadprivate
directive is as follows:
#pragma omp threadprivate(variable-list) new-line
Each copy of a threadprivate variable is initialized once, at an unspecified point in
the program prior to the first reference to that copy, and in the usual manner (that
is, as the master copy would be initialized in a serial execution of the program).
As with any private variable, a thread must not reference another thread's copy of
a threadprivate object. During serial regions and master regions of the program,
references will be to the master thread's copy of the object.
After the first parallel region executes, the data in the threadprivate objects is
guaranteed to persist only if the dynamic threads mechanism has been disabled
and if the number of threads remains unchanged for all parallel regions.
The following code sample illustrates the use of the threadprivate variable:
/* File: omp_th_private.c */
#include <stdlib.h>
int a, b;
#pragma omp threadprivate(a)
int main(int argc, char *argv[])
{
/* First parallel region. */
#pragma omp parallel private(b)
a = b = 5;
/* Second parallel region. */
#pragma omp parallel
{
printf("The value of a is %d\n", a);
printf("The value of b is %d\n", b);
}
354
Developing and Porting C and C++ Applications on AIX

}
When executed, the program prints the following output:
The value of a is 5
The value of b is 0
The value of a is 5
The value of b is 0
The value of a is 5
The value of b is 0
threadprivate variables differ from private variables because they are able to
persist between different parallel sections of a code. That is the reason, in the
above output, the value of a is 5 and the value of b is zero or undefined.
9.9 Data-sharing attribute clauses
Several directives accept clauses that allow a user to control the sharing
attributes of variables for the duration of the region. Sharing attribute clauses
applies only to variables in the lexical extent of the directive on which the clause
appears. Not all of the following clauses are allowed on all directives. The list of
clauses that are valid on a particular directive are described with the directive.
The following sections describe the data-sharing attribute clauses:
Section 9.9.1, “private clause” on page 355
Section 9.9.2, “firstprivate clause” on page 356
Section 9.9.3, “lastprivate clause” on page 357
Section 9.9.4, “shared clause” on page 357
Section 9.9.5, “default clause” on page 358
Section 9.9.6, “reduction clause” on page 358
Section 9.9.7, “copyin clause” on page 362
Section 9.9.8, “copyprivate clause” on page 363
9.9.1 private clause
The private clause declares the variables in variable-list to be private to each
thread in a team. The syntax of the private clause is as follows:
private(variable-list)
Chapter 9. Program parallelization using OpenMP
355

The behavior of a variable specified in a private clause is as follows:
A new object with automatic storage duration is allocated for the construct. The
size and alignment of the new object are determined by the type of the variable.
This allocation occurs once for each thread in the team, and a default constructor
is invoked for a class object (if any) if necessary; otherwise, the initial value is
indeterminate. The original object referenced by the variable has an
indeterminate value upon entry to the construct, must not be modified within the
dynamic extent of the construct, and has an indeterminate value upon exit from
the construct.
The restrictions to the private clause are as follows:
A variable with a class type that is specified in a private clause must have an
accessible, unambiguous default constructor.
A variable specified in a private clause must not have a const-qualified type
unless it has a class type with a mutable member.
A variable specified in a private clause must not have an incomplete type or a
reference type.
Variables that appear in the reduction clause (explained in 9.9.6, “reduction
clause” on page 358) of a parallel directive cannot be specified in a private
clause on a work-sharing directive that binds to the parallel construct.
9.9.2 firstprivate clause
The firstprivate clause provides a superset of the functionality provided by the
private clause. The syntax of the firstprivate clause is as follows:
firstprivate(variable-list)
The firstprivate clause combines the behavior of the private clause with
automatic initialization of the variables in its list. The initialization or construction
happens as if it were done once per thread, prior to the thread’s execution of the
construct. For a firstprivate clause on a parallel construct, the initial value of the
new private object is the value of the original object that exists immediately prior
to the parallel construct for the thread that encounters it. For a firstprivate clause
on a work-sharing construct, the initial value of the new private object for each
thread that executes the work-sharing construct is the value of the original object
that exists prior to the point in time that the same thread encounters the
work-sharing construct. In addition, for C++ objects, the new private object for
each thread is copy constructed from the original object.
356
Developing and Porting C and C++ Applications on AIX

The restrictions to the firstprivate clause are as follows:
A variable specified in a firstprivate clause must not have an incomplete type
or a reference type.
A variable with a class type that is specified as firstprivate must have an
accessible, unambiguous copy constructor.
Variables that are private within a parallel region or that appear in the
reduction clause of a parallel directive cannot be specified in a firstprivate
clause on a work-sharing directive that binds to the parallel construct.
9.9.3 lastprivate clause
The lasprivate clause combines the behavior of the private clause with a copy
from the last loop iteration or section to the original variable object. The syntax of
the lastprivate clause is as follows:
lastprivate(variable-list)
The value copied back into the original variable object is obtained from the last
(sequentially) iteration or section of the enclosing construct.
For example, the team member which executes the final iteration for a for section,
or the team member which does the last section of a sections context performs
the copy with its own values.
The restrictions to the lastprivate clause are as follows:
All restrictions for private apply.
A variable with a class type that is specified as lastprivate must have an
accessible, unambiguous copy assignment operator.
Variables that are private within a parallel region or that appear in the
reduction clause of a parallel directive cannot be specified in a lastprivate
clause on a work-sharing directive that binds to the parallel construct.
9.9.4 shared clause
This clause shares variables that appear in the variable-list among all the threads
in a team. All threads within a team access the same storage area for shared
variables. The syntax of the shared clause is as follows:
shared(variable-list)
It is the programmer's responsibility to ensure that multiple threads properly
access shared variables (such as via critical sections).
Chapter 9. Program parallelization using OpenMP
357

9.9.5 default clause
The default clause allows the user to affect the data-sharing attributes of
variables. The syntax of the default clause is as follows:
default(shared | none)
Specifying default(shared) is equivalent to explicitly listing each currently visible
variable in a shared clause, unless it is threadprivate or const-qualified. In the
absence of an explicit default clause, the default behavior is the same as if
default(shared) were specified.
Specifying default(none) requires that at least one of the following must be true
for every reference to a variable in the lexical extent of the parallel construct:
The variable is explicitly listed in a data-sharing attribute clause of a construct
that contains the reference.
The variable is declared within the parallel construct.
The variable is threadprivate.
The variable has a const-qualified type.
Only a single default clause may be specified on a parallel directive.
9.9.6 reduction clause
This clause performs a reduction on the scalar variables that appear in
variable-list, with the operator op. The syntax of the reduction clause is as
follows:
reduction(op:variable-list)
A reduction is typically specified for a statement with one of the following forms:
x binop= expr
x = x op expr
x = expr op x (except for subtraction)
x++
++x
x--
--x
Where:
x is one of the reduction variables specified in the list.
expr is an expression with scalar type, and it does not reference the object
designated by x.
binop is not an overloaded operator and is one of +, *, -, &, ^, or |.
358
Developing and Porting C and C++ Applications on AIX

op is not an overloaded operator, but one of +, *, -, &, ^, |, &&, or ||.
The term reduction refers to repeated updating of a variable via +, -, *, or a binary
logical operator, with some other value. As will be apparent in the following, these
binary operators must be commutative; division is not permitted as a reduction
operator.
A reduction variable is a hybrid shared/private variable. It is initialized prior to the
parallel construct, such as a shared or global variable. Within the parallel
construct, it behaves as a private variable. But at the conclusion of the parallel
construct, the threads' private values are amalgamated per the designated
binary operation into the original shared variable.
Let us take the following example (Example 9-5). In this program, we are printing
Hello inside the nested for loop of a parallel for construct. We are also printing the
number of times the word Hello is printed. Let us look at the output of the
program.
Example 9-5 omp_reduction1.c
#include <stdio.h>
int main(int argc, char *argv[])
{
int i = 0, j = 0;
int result = 0;
#pragma omp parallel for private(i)
for (i = 0; i < 3; i++) {
for (j = i + 1; j < 4; j++) {
printf("Hello.\n");
result = result + 1;
}
}
printf("Number of times printed Hello = %d\n", result);
}
When executed, the program prints the following output:
Hello.
Hello.
Hello.
Hello.
Hello.
Hello.
Number of times printed Hello = 2
Chapter 9. Program parallelization using OpenMP
359

Alhough the word Hello is printed six times, the result variable is counted only
twice when it should have been six. Clearly, a synchronization problem exists in
the reading and writing of result. This problem can be resolved in three ways.
First, as we discussed earlier, the critical directive can be used to avoid the race
condition to update the result variable, as shown in Example 9-6.
Example 9-6 omp_reduction2.c
#include <stdlib.h>
int main(int argc, char *argv[])
{
int i = 0, j = 0;
int result = 0;
#pragma omp parallel for private(i)
for (i = 0;i < 3; i++) {
for (j = i+1; j <4; j++) {
printf("Hello.\n");
#pragma omp critical /* To avoid race condition. */
result = result + 1;
}
}
printf("Number of times printed Hello = %d\n", result);
}
When executed, the program prints the following output:
Hello.
Hello.
Hello.
Hello.
Hello.
Hello.
Number of times printed Hello = 6
Second is the use of the reduction clause. It supersedes the use of
synchronization constructs in these situations. It not only prevents the race
condition but also reduces the time of execution as it happens while using
synchronization constructs. In Example 9-6, we may treat result as a reduction
variable by modifying the parallel construct, as shown in Example 9-7 on
page 361.
360
Developing and Porting C and C++ Applications on AIX

Example 9-7 omp_reduction3.c
#include <stdio.h>
int main(int argc, char *argv[])
{
int i = 0, j = 0;
int result = 0;
/* reduction clause is used here for result. */
#pragma omp parallel for private(i) reduction(+:result)
for (i = 0; i < 3; i++) {
for (j = i + 1 ; j < 4; j++) {
printf("Hello.\n");
result = result + 1;
}
}
printf("Number of times printed Hello = %d\n",result);
}
When executed, the program prints the following output:
Hello.
Hello.
Hello.
Hello.
Hello.
Hello.
Number of times printed Hello = 6
For the third method of solving this problem using run-time library functions, see
9.10.2, “Lock functions” on page 368.
The restrictions to the reduction clause are as follows:
The type of the variables in the reduction clause must be valid for the
reduction operator, except that pointer types and reference types are never
permitted.
A variable that is specified in the reduction clause must not be const-qualified.
Variables that are private within a parallel region or that appear in the
reduction clause of a parallel directive cannot be specified in a reduction
clause on a work-sharing directive that binds to the parallel construct, as
shown in the following example.
#pragma omp parallel private(y)
{
/* ERROR - private variable y cannot be specified in a reduction clause. */
#pragma omp for reduction(+: y)
}
Chapter 9. Program parallelization using OpenMP
361

/* ERROR -
* variable x cannot be specified in both a shared and a reduction clause.
*/
#pragma omp parallel for shared(x) reduction(+: x)
9.9.7 copyin clause
The copyin clause provides a mechanism to assign the same value to
threadprivate variables for each thread in the team executing the parallel region.
For each variable specified in a copyin clause, the value of the variable in the
master thread of the team is copied, as if by assignment, to the threadprivate
copies at the beginning of the parallel region. The syntax of the copyin clause is
as follows:
copyin(variable-list)
The following code sample illustrates this:
/* File : omp_copyin.c */
#include <include.h>
int a, b;
#pragma omp threadprivate(a)
int main(int argc, char *argv[])
{
/* First parallel region. */
a = 10; /* a is initialized before the parallel region. */
/*
* copies the value of ‘a’ by assignment, if copyin is not used, ‘a’ is
* visible only to the master thread.
*/
#pragma omp parallel private(b) copyin(a)
b = 8;
/* Second parallel region. */
#pragma omp parallel
{
printf("The value of a is %d\n", a);
printf("The value of b is %d\n", b);
}
}
When executed, the program prints the following output:
The value of a is 10
The value of b is 0
The value of a is 10
The value of b is 0
The value of a is 10
The value of b is 0
362
Developing and Porting C and C++ Applications on AIX

The value of a is 10
The value of b is 0
The restrictions to the copyin clause are as follows:
A variable that is specified in the copyin clause must have an accessible,
unambiguous copy assignment operator (for C++).
A variable that is specified in the copyin clause must be a threadprivate
variable.
9.9.8 copyprivate clause
The copyprivate clause provides a mechanism to use a private variable to
broadcast a value from one member of a team to the other members. It is an
alternative to using a shared variable for the value when providing such a shared
variable would be difficult (for example, in a recursion requiring a different
variable at each level). The copyprivate clause can only appear on the single
directive. The syntax of the copyprivate clause is as follows:
copyprivate(variable-list)
Restrictions to the copyprivate clause are as follows:
A variable that is specified in the copyprivate clause must not appear in a
private or firstprivate clause for the same single directive.
If a single directive with a copyprivate clause is executed in the dynamic
extent of a parallel region, all variables specified in the copyprivate clause
must be private in the enclosing context.
A variable that is specified in the copyprivate clause must have an accessible
unambiguous copy assignment operator (for C++).
9.10 Run-time library functions
In addition to the directives described in previous sections, OpenMP provides a
set of run-time library functions. It includes run-time execution functions and lock
functions. The run-time functions allow an application to specify the mode in
which to run. An application developer may wish to maximize throughput
performance of the system, rather than time to completion. In such cases, the
developer may tell the system to dynamically set the number of processes used
to execute parallel regions. This can have a dramatic effect on the throughput
performance of a system with only a minimal impact on the time to completion for
a program.
Chapter 9. Program parallelization using OpenMP
363

The run-time functions also allow a developer to specify when to enable nested
parallelism. Enabling nested parallelism allows the system to act accordingly
when it encounters nested parallel constructs. On the other hand, by disabling
nested parallelism, a developer can write a parallel library that will perform in an
easily predictable fashion, whether executed dynamically from within or outside a
parallel region.
For further information about these library functions, please refer to the “Parallel
Processing Support” section of the VisualAge C++ for AIX Compiler Reference,
SC09-4959.
9.10.1 Execution environment functions
In this section, let us look at the following important OpenMP run-time functions
supported on AIX. The functions described in this section affect and monitor
threads, processors, and the parallel environment:
omp_set_num_threads
omp_get_num_threads
omp_get_max_threads
omp_get_thread_num
omp_get_num_procs
omp_set_dynamic
omp_get_dynamic
omp_in_parallel
omp_set_nested
omp_get_nested
An example program using these functions is provided in 9.10.3, “Example
usage of run-time library functions” on page 370.
omp_set_num_threads
The omp_set_num_threads() function sets the default number of threads to use
for subsequent parallel regions that do not specify a num_threads clause. The
syntax is as follows:
void omp_set_num_threads(int num_threads);
364
Developing and Porting C and C++ Applications on AIX

The dynamic threads mechanism modifies the effect of this routine.
Enabled: Specifies the maximum number of threads that can be used for any
parallel region.
Disabled: Specifies exact number of threads to use until next call to this
routine.
This routine can only be called from the serial portions of the code. This call has
precedence over the OMP_NUM_THREADS environment variable (see 9.11,
“Environment variables” on page 373).
omp_get_num_threads
The omp_get_num_threads() function returns the number of threads currently in
the team executing the parallel region from which it is called. The syntax is as
follows:
int omp_get_num_threads(void);
The num_threads clause, the omp_set_num_threads() function, and the
OMP_NUM_THREADS environment variable control the number of threads in a
team. If the number of threads has not been explicitly set by the user, the default
value is chosen. In AIX, it is the number of equipped processors on the system.
This function binds to the closest enclosing parallel directive. If called from a
serial portion of a program, or from a nested parallel region that is serialized, this
function returns 1.
omp_get_max_threads
It returns the maximum value that can be returned by calls to the
omp_get_num_threads() function. The syntax is as follows:
int omp_get_max_threads(void);
This function returns the maximum value, whether executing from a serial region
or from a parallel region. If a program uses omp_set_num_threads to change the
number of threads, subsequent calls to omp_get_max_threads() will return the
new value.
When the omp_set_dynamic() routine is set to TRUE, we can use
omp_get_max_threads() to allocate data structures that are maximally sized for
each thread.
Chapter 9. Program parallelization using OpenMP
365

omp_get_thread_num
The omp_get_thread_num() function returns the thread number, within its team,
of the thread executing the function. The thread number lies between 0 and
omp_get_num_threads() – 1, inclusive. The master thread of the team is thread
0. The syntax is as follows:
int omp_get_thread_num(void);
This function binds to the closest enclosing parallel directive. The function
returns zero when called from a serial region or from within a nested parallel
region that is serialized.
omp_get_num_procs
The omp_get_num_procs() function returns the number of processors that are
available to the program at the time the function is called. The syntax is as
follows:
int omp_get_num_procs(void);
omp_set_dynamic
The omp_set_dynamic() function enables or disables dynamic adjustment of the
number of threads available for execution of parallel regions. The syntax is as
follows:
void omp_set_dynamic(int dynamic_threads);
To obtain the best use of system resources, certain run-time environments
automatically adjust the number of threads that are used for executing sub-
sequent parallel regions. This adjustment is enabled only if the value of the scalar
logical expression to omp_set_dynamic is set to TRUE. If the value of the scalar
logical expression is set as FALSE, dynamic adjustment is disabled.
When dynamic adjustment is enabled, the number of threads specified by the
user becomes the maximum thread count. The number of threads remains fixed
throughout each parallel region and is reported by omp_get_num_threads(). A
call to omp_set_dynamic() has precedence over the OMP_DYNAMIC
environment variable.
The default for dynamic thread adjustment is implementation dependent. In AIX,
by default, dynamic adjustment is enabled. A user code that depends on a
specific number of threads for correct execution should explicitly disable dynamic
threads. Implementations are not required to provide the ability to dynamically
adjust the number of threads, but they are required to provide the interface in
order to support portability across platforms.
366
Developing and Porting C and C++ Applications on AIX

omp_get_dynamic
The omp_get_dynamic() function returns a nonzero value if dynamic adjustment
of threads is enabled, and returns 0 otherwise. The syntax is as follows:
int omp_get_dynamic(void);
This function returns 1 if dynamic thread adjustment is enabled; otherwise, it
returns 0. The function always returns 0 if dynamic adjustment of the number of
threads is not implemented.
omp_in_parallel
The omp_in_parallel() function returns a nonzero value if it is called within the
dynamic extent of a parallel region executing in parallel; otherwise, it returns 0.
The syntax is as follows:
int omp_in_parallel(void);
The omp_in_parallel() function determines whether a region is executing in
parallel. A parallel region that is serialized is not considered to be a region
executing in parallel.
omp_set_nested
The omp_set_nested() function enables or disables nested parallelism. The
syntax is as follows:
void omp_set_nested(int nested);
If the value of the scalar logical expression is FALSE, nested parallelism is
disabled, and nested parallel regions are serialized and executed by the current
thread. This is the default.
omp_get_nested
The omp_get_nested() function returns a nonzero value if nested parallelism is
enabled and 0 if it is disabled. The syntax is as follows:
int omp_get_nested(void);
If an implementation does not implement nested parallelism, this function always
returns 0.
Note: In the current implementation, nested parallel regions are always
serialized. As a result, omp_set_nested() does not have any effect, and
omp_get_nested() always returns 0 on AIX.
Chapter 9. Program parallelization using OpenMP
367

9.10.2 Lock functions
The functions described in this section manipulate locks used for
synchronization. For the following functions, the lock variable must have type
omp_lock_t. This variable must only be accessed through these functions. All
lock functions require an argument that has a pointer to omp_lock_t type.
omp_init_lock
omp_destroy_lock
omp_set_lock
omp_unset_lock
omp_test_lock
For the following functions, the lock variable must have type omp_nest_lock_t.
This variable must only be accessed through these functions. All nestable lock
functions require an argument that has a pointer to omp_nest_lock_t type.
omp_init_nest_lock
omp_destroy_nest_lock
omp_set_nest_lock
omp_unset_nest_lock
omp_test_nest_lock
An example program using these functions is provided in 9.10.3, “Example
usage of run-time library functions” on page 370.
omp_init_lock and omp_init_nest_lock
These functions provide the only means of initializing a lock. Each function
initializes the lock associated with the parameter lock for use in subsequent calls.
The syntax is as follows:
void omp_init_lock(omp_lock_t *lock);
void omp_init_nest_lock(omp_nest_lock_t *lock);
The initial state is unlocked (that is, no thread owns the lock). For a nestable lock,
the initial nesting count is zero. It is noncompliant to call either of these routines
with a lock variable that has already been initialized.
omp_destroy_lock and omp_destroy_nest_lock
These functions ensure that the pointed to lock variable lock is uninitialized. The
syntax is as follows:
void omp_destroy_lock(omp_lock_t *lock);
void omp_destroy_nest_lock(omp_nest_lock_t *lock);
368
Developing and Porting C and C++ Applications on AIX

It is noncompliant to call either of these routines with a lock variable that is
uninitialized or unlocked.
omp_set_lock and omp_set_nest_lock
Each of these functions blocks the thread executing the function until the
specified lock is available and then sets the lock. A simple lock is available if it is
unlocked. A nestable lock is available if it is unlocked or if it is already owned by
the thread executing the function. The syntax is as follows:
void omp_set_lock(omp_lock_t *lock);
void omp_set_nest_lock(omp_nest_lock_t *lock);
For a simple lock, the argument to the omp_set_lock() function must point to an
initialized lock variable. Ownership of the lock is granted to the thread executing
the function. For a nestable lock, the argument to the omp_set_nest_lock()
function must point to an initialized lock variable. The nesting count is
incremented, and the thread is granted, or retains, ownership of the lock.
omp_unset_lock and omp_unset_nest_lock
These functions provide the means of releasing ownership of a lock. The syntax
is as follows:
void omp_unset_lock(omp_lock_t *lock);
void omp_unset_nest_lock(omp_nest_lock_t *lock);
The argument to each of these functions must point to an initialized lock variable
owned by the thread executing the function. The behavior is undefined if the
thread does not own that lock.
For a simple lock, the omp_unset_lock() function releases the thread executing
the function from ownership of the lock. For a nestable lock, the
omp_unset_nest_lock() function decrements the nesting count, and releases the
thread executing the function from ownership of the lock if the resulting count is
zero.
omp_test_lock and omp_test_nest_lock
These functions attempt to set a lock but do not block execution of the thread.
The syntax is as follows:
int omp_test_lock(omp_lock_t *lock);
int omp_test_nest_lock(omp_nest_lock_t *lock);
The argument must point to an initialized lock variable. These functions attempt
to set a lock in the same manner as omp_set_lock() and omp_set_nest_lock(),
except that they do not block execution of the thread.
Chapter 9. Program parallelization using OpenMP
369

For a simple lock, the omp_test_lock() function returns a nonzero value if the lock
is successfully set; otherwise, it returns zero. For a nestable lock, the
omp_test_nest_lock() function returns the new nesting count if the lock is
successfully set; otherwise, it returns zero.
9.10.3 Example usage of run-time library functions
This section provides two example source codes in order to demonstrate the
usage of run-time library functions.
Note: Before calling any OpenMP run-time library functions, the omp.h header
file must be included in the source code.
Usage of run-time execution functions
The sample code shown in Example 9-8 illustrates the basic use of various
run-time routines that we have discussed so far. It displays the values returned
by the various run-time routines both in the serial and parallel region. For
simplicity, the single directive is used to just print the values returned by one
instance of a executing thread.
Example 9-8 omp_runtime.c
#include <stdio.h>
#include <omp.h>
int main(int argc, char *argv[])
{
printf("Before forking a parallel region.\n");
printf("---------------------------------\n");
printf("omp_get_num_threads returns %d\n", omp_get_num_threads());
printf("omp_get_max_threads return %d\n", omp_get_max_threads());
printf("omp_get_thread_num returns %d\n", omp_get_thread_num());
printf("omp_get_num_procs returns %d\n", omp_get_num_procs());
printf("omp_get_dynamic returns %d\n", omp_get_dynamic());
printf("omp_in_parallel returns %d\n", omp_in_parallel());
printf("omp_get_nested returns %d\n", omp_get_nested());
printf("\nAfter forking a parallel region.\n");
printf("---------------------------------\n");
omp_set_num_threads(6);/* set the number of threads at run time to 10. */
/* does not have any effect, just to illustrate that here. */
omp_set_nested(1);
#pragma omp parallel
{
#pragma omp single /* to print the values once. */
370
Developing and Porting C and C++ Applications on AIX

{
printf("omp_get_num_threads returns %d\n"
, omp_get_num_threads());
printf("omp_get_max_threads return %d\n", omp_get_max_threads());
printf("omp_get_thread_num returns %d\n", omp_get_thread_num());
printf("omp_get_num_procs returns %d\n", omp_get_num_procs());
printf("omp_get_dynamic returns %d\n", omp_get_dynamic());
printf("omp_in_parallel returns %d\n", omp_in_parallel());
printf("omp_get_nested returns %d\n", omp_get_nested());
}
} /* All threads join master thread and terminate. */
}
When executed, the program prints the following output:
Before forking a parallel region.
---------------------------------
omp_get_num_threads returns 1
omp_get_max_threads return 3
omp_get_thread_num returns 0
omp_get_num_procs returns 3
omp_get_dynamic returns 1
omp_in_parallel returns 0
omp_get_nested returns 0
After forking a parallel region.
---------------------------------
omp_get_num_threads returns 6
omp_get_max_threads return 6
omp_get_thread_num returns 4
omp_get_num_procs returns 3
omp_get_dynamic returns 1
omp_in_parallel returns 1
omp_get_nested returns 0
Usage of lock functions
The race condition problem that we discussed in 9.9.6, “reduction clause” on
page 358 can also be solved by using the lock functions discussed so far. In the
modified version of the program shown in Example 9-9 on page 372, we lock and
unlock the inner for loop section using the omp_set_lock() and omp_unset_lock()
functions respectively. This prevents the simultaneous entry of threads into the
for block; therefore, the variable result is incremented properly.
Chapter 9. Program parallelization using OpenMP
371

Example 9-9 omp_lock.c
#include <stdio.h>
#include <omp.h>
int main(int argc, char *argv[])
{
int i = 0, j = 0;
int result = 0;
omp_lock_t lock; /* lock variable to be initialized. */
omp_init_lock(&lock); /* Initializes the lock before using it. */
#pragma omp parallel for private(i)
for (i = 0; i < 3; i++) {
omp_set_lock(&lock); /* sets the lock here. */
for (j = i + 1; j < 4; j++) {
/*
* already locked, therefore it doesn't block the execution of the
* thread.
*/
if (!omp_test_lock(&lock)) {
printf("Hello.\n");
result = result + 1;
}
}
omp_unset_lock(&lock); /* releases the lock. */
}
printf("Number of times printed Hello = %d\n", result);
omp_destroy_lock(&lock); /*Finally, destroys the lock */
}
When executed, the program prints the following output:
Hello.
Hello.
Hello.
Hello.
Hello.
Hello.
Number of times printed Hello = 6
This is the same output produced by Example 9-7 on page 361.
372
Developing and Porting C and C++ Applications on AIX

9.11 Environment variables
This section explains the OpenMP C and C++ API environment variables that
control the execution of parallel code. The names of environment variables must
be uppercase. The values assigned to them are case insensitive and may have
leading and trailing white space. Modifications to the values after the program
has started are ignored. The environment variables are as follows:
OMP_SCHEDULE
Sets the run-time schedule type and chunk size.
OMP_NUM_THREADS
Sets the number of threads to use during execution.
OMP_DYNAMIC
Enables or disables dynamic adjustment of the
number of threads.
OMP_NESTED
Enables or disables nested parallelism.
For further information about these environment variables, please refer to the
“Parallel Processing Support” section in the VisualAge C++ for AIX Compiler
Reference, SC09-4959.
9.11.1 OMP_SCHEDULE
OMP_SCHEDULE applies only to for and parallel for directives that have the
schedule type run time. The schedule type and chunk size for all such loops can
be set at run time by setting this environment variable to any of the recognized
schedule types and to an optional chunk_size.
For for and parallel for directives that have a schedule type other than run time,
OMP_SCHEDULE is ignored. The default value for this environment variable is
implementation-defined. If the optional chunk_size is set, the value must be
positive. If chunk_size is not set, a value of 1 is assumed, except in the case of a
static schedule. For a static schedule, the default chunk size is set to the loop
iteration space divided by the number of threads applied to the loop.
The syntax of OMP_SCHEDULE is:
OMP_SCHEDULE=algorithm
where algorithm is one of the following:
dynamic[,N]
guided[,N]
runtime
static[,N]
If specified, the value of N must be an integer value of 1 or greater.
Chapter 9. Program parallelization using OpenMP
373

9.11.2 OMP_NUM_THREADS
The OMP_NUM_THREADS environment variable sets the default number of
threads to use during execution, unless that number is explicitly changed by
calling the omp_set_num_threads library routine or by an explicit num_threads
clause on a parallel directive.
Its effect depends upon whether dynamic adjustment of the number of threads is
enabled. If no value is specified for the OMP_NUM_THREADS environment
variable, or if the value specified is not a positive integer, or if the value is greater
than the maximum number of threads the system can support, the number of
threads to use is implementation-defined. On AIX, it depends on the number of
available processors.
The syntax of OMP_NUM_THREADS is:
OMP_NUM_THREADS=N
where N is the number of parallel threads requested, which must be a positive
integer. The number can be overridden during program execution by calling the
omp_set_num_threads() run-time library function.
9.11.3 OMP_DYNAMIC
The OMP_DYNAMIC environment variable enables or disables dynamic
adjustment of the number of threads available for execution of parallel regions.
The setting of this environment variable can be overridden by calling the
omp_set_dynamic() run-time library function.
The syntax of OMP_DYNAMIC is:
OMP_DYNAMIC=[TRUE | FALSE]
If set to TRUE, the number of threads that are used for executing parallel regions
may be adjusted by the run-time environment to best utilize system resources. If
set to FALSE, dynamic adjustment is disabled. The default condition is
implementation-defined.
9.11.4 OMP_NESTED
The OMP_NESTED environment variable enables or disables nested
parallelism. The setting of this environment variable can be overridden by calling
the omp_set_nested() run-time library function.
The syntax of OMP_NESTED is:
OMP_NESTED=[TRUE | FALSE]
374
Developing and Porting C and C++ Applications on AIX

If set to TRUE, nested parallelism is enabled; if it is set to FALSE, nested
parallelism is disabled. The default value is FALSE.
Chapter 9. Program parallelization using OpenMP
375

376
Developing and Porting C and C++ Applications on AIX

10
Dealing with C++ templates
Chapter 10.
This chapter explains how to deal with C++ templates on AIX for C++ application
programmers by providing the following sections:
Section 10.1, “What is a template” on page 378
Section 10.2, “AIX template implementations” on page 378
Section 10.3, “Simple code layout method” on page 381
Section 10.4, “Template instantiation file method” on page 383
Section 10.5, “Template registry: The preferred method” on page 387
Also, the ISO Standard C++ Library, including the Standard Template Library, is
explained in the following section:
Section 10.6, “Standard C++ Library and STL” on page 388
© Copyright IBM Corp. 2000, 2003. All rights reserved.
377

10.1 What is a template
A template defines a family of related classes or functions, where the parameters
in the declaration describe how they are specialized. Templates are generic
entities that can be instantiated to create specific user-defined types. The
compiler generates new classes or functions when you supply arguments for the
parameters. The class or function definition generated from a template with a set
of template parameters is called a specialization.
Templates are an area of the C++ language that provide a great deal of flexibility
for developers. The ISO C++ standard defines the language facilities and
features for templates. Unfortunately, the standard does not specify how a
compiler should implement templates. This means that there are sometimes
significant differences between the methods used to implement templates in
compiler products from different vendors.
Developers porting C++ code that uses templates to the AIX platform sometimes
have problems with the implementation model. The main problems experienced
are:
Long compile and link times
Linker warnings of duplicate symbols
Increase in code and executable size
Up until VisualAge C++ for AIX Version 6.0, all of the above problems are
generally caused by inefficient use of the compiler implementation of templates.
The Version 6 compiler provides new options that handle templates more
efficiently. The number of problems experienced will depend on the platform the
code is being ported from and the template implementation method used on that
platform. Sometimes, the problems can be fixed on AIX by simply adding a few
compiler options. In other instances, the code layout needs to be changed in
order to utilize the most efficient implementation method on AIX. In most of these
rare cases, the code changes are backwards compatible with the original
platform the code is being ported from. This is very important for developers who
maintain a single source tree that must compile correctly on multiple platforms.
10.2 AIX template implementations
The template mechanism provides a way of defining general container types,
such as list, vector, and stack, where the specific type of the elements is left as a
parameter. Two types of templates can be defined:
Class templates
Specify how individual classes can be constructed.
Function templates
Specify how individual functions can be constructed.
378
Developing and Porting C and C++ Applications on AIX

Regardless of the type of template being used, the code is essentially split into
three parts:
Template declaration
This is the part of the source code that declares the
template class or function. It does not necessarily
contain the definition of the class or function, although
it may optionally do so. For example, a template
declaration may describe a Stack template class, as
shown in Example 10-1.
Template definition
This portion of code is the definition of the template
function itself or the template class member functions.
For example, using the Stack class template, this
portion of code would define the member functions
used to manipulate the stack, as shown in
Example 10-2 on page 380.
Template instance
The template instance code is generated by the
compiler for each instance of the template. For
example, this would be the code to handle a specific
instance of the stack template class, such as a stack of
integer values.
The difference between the components is that the template declaration must be
visible in every compilation unit that uses the template. The template definition
and code for each instance of the template need only be visible once in each
group of compilation units that are linked together to make a single executable.
Example 10-1 Stack template declaration
template <class T> class stack
{
private:
T* v;
T* p;
int sz;
public:
stack(int);
~stack();
void push(T);
T pop();
};
Chapter 10. Dealing with C++ templates
379

Example 10-2 Stack template member function definition
template <class T> stack<T>::stack(int s)
{
v = p = new T[sz=s];
}
template <class T> stack<T>::~stack()
{
delete [] v;
}
template <class T> void stack<T>::push(T a)
{
*p++ = a;
}
template <class T> T stack<T>::pop()
{
T ret = *p;
p--;
return ret;
}
10.2.1 Generated function bodies
When you use class templates and function templates in your program, the
compiler automatically generates function bodies for all template functions that
are instantiated. The compiler follows four basic rules to determine when to
generate template functions. The compiler applies the rules in the following
order:
1. If a template declares a function to have internal linkage, the function must be
defined within the same compilation unit. The compiler generates the function
with internal linkage, and it is not shared with other compilation units. This is
the case if the template class has in-line member functions.
2. If a template function is referenced in a compilation unit, but it is not declared
to have internal linkage, the compiler looks for a definition of the function in
the same compilation unit. If a definition is found, the function is instantiated.
If the -qtemplateregistry option is in effect, and the function is already
instantiated in another compilation unit, the function is not instantiated in this
compilation unit.
3. If the -qtempinc option is in effect, the compiler creates a template
instantiation file when the template function is declared but not defined. The
template function is instantiated when the template instantiation file compiles.
380
Developing and Porting C and C++ Applications on AIX

4. If none of the preceding rules applies, the compiler does not generate the
definition of the template function. It must be instantiated in another
compilation unit.
10.3 Simple code layout method
The simplest method of using template code is to include both the declaration
and definition of the template in every compilation unit that uses instances of the
template. From a code layout point of view, this is very easy, since the template
declaration and definition can be kept in a single header file. Using the stack
example, the code in Example 10-1 on page 379 and Example 10-2 on page 380
would be combined into a single header file, for example, stack.h, which is then
included by every compilation unit that wishes to use the template. Alternatively,
the header file for a template declaration can include the source file that contains
the template definition. Using the stack template example, the header file,
stack.h, would #include the source file, stack.C.
There are a number of disadvantages to using this method. Some of them can be
overcome; others can not.
10.3.1 Disadvantages of the simple method
The first disadvantage is that using the header files can become complicated,
particularly when other header files need to declare an instance of the template.
In order to do this, they must #include the stack.h file, which potentially leads to
multiple #include’s of the file, resulting in multiple definitions of the member
functions. This problem can be fixed with the addition of preprocessor macros in
the header file to protect against multiple #include operations. For example:
#if !defined(stack_h)
#define stack_h
....
....declaration and definition of stack template
....
#endif
Using the macros shown above, the contents of the header file will only appear
once in the compilation unit, regardless of the number of times the file is
included. This resolves the problems of multiple definitions within a compilation
unit.
Chapter 10. Dealing with C++ templates
381

Template code bloat
The second disadvantage is that the code for each template instance will
potentially appear multiple times in the final executable, resulting in the problems
of large executable size and multiple symbol definition warnings from the linker.
As an example, consider an executable made up of two compilation units, main.C
and functions.C. If both compilation units include the stack.h header file, and
declare variables of the type stack<int>, then after the first stage of compilation,
both object files, main.o and functions.o, will contain the code for the member
functions of the stack<int> class. When the system linker parses the object files
to create the final executable, it cannot remove the duplicate symbols since, by
default, each compilation unit is treated as an atomic object by the linker. This
results in duplicate symbol linker warnings and a final executable that contains
redundant code.
The size of the final executable can be reduced by using the compiler option,
-qfuncsect, when compiling all of the source code modules. This option causes
the compiler to slightly change the format of the output object files. Instead of
creating an object file, which contains a single code section (CSECT) and must
be treated by the system linker as an atomic unit, the compiler creates an object
file where each function is contained in its own CSECT. This means that the
object files created are slightly larger than their default counterparts since they
contain extra format information in addition to the executable code. This option
does not remove the linker warnings, since at link time, there are still multiple
symbol definitions. The benefit of this option is that the linker can discard
multiple, identical function definitions by discarding the redundant CSECTs,
resulting in a smaller final executable. When the -qfuncsect option is not used,
the compiler cannot discard the redundant function definitions if there are other
symbols in the same CSECT that are required in the final executable.
Refer to 6.1.9, “Virtual functions” on page 226 for information on another
potential cause of C++ code bloat.
Template compile time
The use of the -qfuncsect option reduces the code size of the final executable. It
does not resolve the other disadvantage of using this method, that is, longer than
required compile times. The reason for this is that each compilation unit contains
the member functions for the templates that it instantiates. Using an extreme
example with the stack class, consider the situation where an application is built
from 50 source files, and each source file instantiates a stack<int> template. This
means the member functions for the class are generated and compiled 50 times,
yet the result of 49 of those compiles are discarded by the linker since they are
not needed. In the example used here, the code for the stack class is trivial, so in
absolute terms, the time saved would be minimal. In real life situations, where the
382
Developing and Porting C and C++ Applications on AIX

template code is complex, the time savings that can be made when compiling a
large application are considerable.
Because of the fact that not all of the disadvantages of the simple template
method can be overcome, it is only recommended for use when experimenting
with templates. An alternative method can be used, which solves all of the
problems of the simple method and scales very well for large applications.
10.4 Template instantiation file method
The template instantiation file method on AIX basically means letting the
compiler decide which template code to instantiate as a final step in the compile
and link process. This solves the long compile time disadvantage of the simple
template code layout method because the compiler need only compile each
template instance once.
This method requires that the declaration and definition of the template are kept
in separate files. This is because only the template declaration must be included
in every compilation unit that uses the template. If the definition of the template
were also in the header file, it would also be included in the source file, and thus
compiled, resulting in a situation similar to that in the simple method.
This template model can also benefit from the use of the -qfuncsect compiler
option, since it means the linker can discard code sections that are not
referenced in the final executable.
The template declaration should be left in the header file, as in the simple
template method. The definition of the template member functions needs to be in
a file with the same basename as the header file but with a .c (lower case C) file
name extension.
Note: By default, the file containing the template definition code must have the
same name as the template declaration header file, but with a file name
extension of .c (lowercase c), even though this extension normally indicates a
C language source file. It must also exist in the same directory as the template
declaration header file. If the template definition file is not in the same
directory, has a different basename, or has a different file name extension
(such as .C, .cxx, or .cpp, which are normally used for C++ source files), then
the compiler will not detect the presence of the template code to be used with
the template declaration header file.
Using the stack template example introduced earlier, the template declaration
shown in Example 10-1 on page 379 would be in the stack.h file, while the
Chapter 10. Dealing with C++ templates
383

template definition code shown in Example 10-2 on page 380 would be in the
stack.c file in the same directory. If the template definition code file was named
stack.cxx or stack_code.c, then the compiler will not associate the file with the
template declaration in the stack.h header file.
The name of the template definition file can be changed, if desired, using the
implementation pragma directive as follows:
#pragma implementation(string-literal)
where string-literal is the path name for the template definition file enclosed
in double quotes. For example, if the stack template definition code were to be
stored in the stack_code.cxx file, then the stack.h header file would have the
following directive:
#pragma implementation(“stack_code.cxx”)
Once the structure of the source code has been altered to conform to the
required layout, the templates can be used with this method.
10.4.1 The -qtempinc option
The -qtempinc option is used when compiling source code that instantiates
templates. When no directory is specified with the option, the compiler will create
a directory called tempinc in the current directory. For example:
$ xlC main.C -qtempinc
The user may optionally specify the name of a directory to use for storing the
information on the templates to be generated. This allows the same tempinc
directory to be used when creating an executable that consists of object files that
are compiled in different directories. For example:
$ xlC -c file1.C file2.C -qtempinc=../app1/templates
$ cd ../app1
$ xlC -o app1 main.C ../src/file1.o ../src/file2.o -qtempinc=./templates
The tempinc directory is used to store information about the templates that are
required to be generated. When invoked with the -qtempinc option, the compiler
collects information about template instantiations and stores the information in
the tempinc directory. As the last step of the compilation before linking, the
compiler generates the code for the required template instantiations. It then
compiles the code and includes it with the other object files and libraries that are
passed to the linker to create the final executable.
If the compiler detects a code layout structure that enables the tempinc method
to be used, it will automatically enable the -qtempinc option, even if it was not
specified on the command line. This causes the template instantiation
384
Developing and Porting C and C++ Applications on AIX

information to be stored in the tempinc directory. If you want to specify a different
directory, you should explicitly use the -qtempinc=dir_name option on the
command line. If you want to prevent the compiler from automatically generating
the template information, which may be the case when creating a shared object,
then use the -qnotempinc option. See 11.2, “Shared objects with templates” on
page 402 for more information on the use of the -qnotempinc option when
creating shared objects.
One important point to note about the -qtempinc option is that you should use the
same value when compiling all compilation units that will be linked together. In
other words, do not compile half of the application with -qtempinc, and the other
half with -qtempinc=dir_name. Only one tempinc directory can be specified on
the final C++ command line that is used to link the application, which means that
half of the template instance information will be missing. If more than one
tempinc option is specified on the command line, the last one encountered will
prevail.
10.4.2 Contents of the tempinc directory
The compiler generates a file in the tempinc directory for each template header
file that has templates instantiated. The file has the same name as the header
file, but with a .C (uppercase C) file name extension. The compiler generates the
file when it detects the first instantiation of a template that is declared in the
header file with the same name. Information on the subsequent instances of the
template is added to the file.
As the final step of the compilation before linking, the compiler compiles all of the
files in the tempinc directory and passes the object files to the linker along with
the user specified files.
The contents of a template information file are shown in Example 10-3.
Example 10-3 A sample template information file1
1: /*0965095125*/#include "/redbooks/examples/C++/stack.h"
2: /*0000000000*/#include "/redbooks/examples/C++/stack_code.cxx"
3: template stack<int>::stack(int);
4: template stack<int>::~stack();
5: template void stack<int>::push(int);
6: template int stack<int>::pop();
The code on line 1 includes the header file that declares the template. The
comment at the start of the line is a time stamp and is used by the compiler to
1 The line number at the beginning of each line is added intentionally for explanation. These numbers
and the colon characters “:” are not part of the file.
Chapter 10. Dealing with C++ templates
385

determine if the header file has changed, which would require the template
instance information file to be recompiled.
The code on line 2 includes the template implementation file that corresponds to
the header file in line 1. A time stamp consisting of all zeros indicates that the
compiler should ignore the time stamp. The file may include other header files
that define the classes that are used in template instantiations. For example, if
there was a user defined class Box, and the compiler detected an instantiation of
stack<Box>, then the header file that defines the class Box would be included in
the instance information file.
The subsequent lines in the example shown above cause the individual member
functions to be instantiated.
10.4.3 Forcing template instantiation
You can, if you wish, structure your program so that it does not use automatic
template instantiation. In order to do this, you must know which template classes
and functions need to be instantiated.
The #pragma define directive is used to force the instantiation of a template, even
if no reference is made to an instance of the generated template. For example:
#pragma define(stack<double>);
This, however, means that the template implementation file needs to be included
in the compilation units that have the #pragma define directives, which results in
the same disadvantages of the simple template method described in 10.3,
“Simple code layout method” on page 381.
An alternative to this is to manually emulate the process used by the compiler to
automatically create the appropriate template instances. Using the stack class as
an example, the following compilation unit shown in Example 10-4 could be used
to force the creation of the desired stack template classes, even though no
objects of those types are referenced in the source code.
Example 10-4 A sample usage of #pragma define2
1: #include "/redbooks/examples/C++/stack.h"
2: #include "/redbooks/examples/C++/stack_code.cxx"
3: #include “/redbooks/examples/C++/Box.h”
// definition of class Box
4: #pragma define(stack<int>);
5: #pragma define(stack<Box>);
6: #pragma define(stack<char>);
2 The line number at the beginning of each line is added intentionally for explanation. These numbers
and the colon characters “:” are not part of the file.
386
Developing and Porting C and C++ Applications on AIX

7: #pragma define(stack<short>);
This type of method will be useful when creating shared objects with the
makeC++SharedLib command. Users of the VisualAge C++ for AIX product should
use the -qmkshrobj option instead. See 11.2, “Shared objects with templates” on
page 402 for more information.
10.5 Template registry: The preferred method
New in VisualAge C++ for AIX Version 6.0 compiler is the template registry
method of template handling. The compiler decides which template code to
instantiate by referring to and updating a template registry file as compilation
occurs. This method is fast, efficient, and best of all, does not require
reorganization of your template code.
10.5.1 The -qtemplateregistry option
The -qtemplateregistry option maintains, in a registry file, records of all template
instantiations as they are encountered in the compilation unit, and ensures that
only one instantiation of each template is generated. By default, the compiler
writes to a file named templateregistry in the current directory. You may optionally
specify a different registry file name with the option. However, you must use the
same registry file for the entire program.
Unlike the template instantiation file method of template handling supported with
the -qtempinc option, the -qtemplateregistry option does not require the template
code to be structured in a certain way. The option is mutually exclusive with the
-qtempinc option, and you must not use both options at the same time. Before
using the -qtemplateregistry option, remove all instantiation files in the tempinc
directory. Also, any existing program that compiles successfully with
-qnotempinc, that is, when every reference is instantiated in every compilation
unit, will compile successfully with -qtemplateregistry. In other words, there is no
migration impact should you want to take advantage of this improved template
handling method.
The -qtemplateregistry option works by storing template instantiation information
in a registry as compilation occurs. The registry is read for each compilation, and
a check is done when an instantiation is required. If the instantiation has already
been seen, nothing will happen; otherwise the template will be instantiated in the
object file. In either case, a record is added for each template reference to the
registry to keep track of the information about use and instantiation for all
compilation units.
Chapter 10. Dealing with C++ templates
387

10.5.2 The -qtemplaterecompile option
If a change to a compilation unit removes a template instantiation, recompiling
only the changed source file will result in undefined symbols at link time, since
other compilation units may still require the template instantiation that is now
missing. This dependency information is maintained in the template registry file
in the -qtemplateregistry compiles. In this case, the -qtemplaterecompile option
will cause a recompilation of all source files that rely on the template.
The -qtemplaterecompile option is, by default, turned on with the
-qtemplateregistry option. It manages dependencies among compilation units,
and ensures that affected source files are recompiled automatically by consulting
the template registry file. It requires that object files originally generated with the
-qtemplateregistry option remain in the same directory. Should you need to
disable automatic recompilation for any reason, specify the
-qnotemplaterecompile option. This may be the case, for example, if your build
process moves the generated object files to a different directory.
10.6 Standard C++ Library and STL
With the removal of the IBM Open Class (IOC) Library in Version 6, the
VisualAge C++ for AIX compiler has standardized on the use of the Standard
C++ Library, including the Standard Template Library (STL).
When migrating from a previous version of the VisualAge C++ for AIX compiler,
refer to the IBM Open Class Library Transition Guide, SC09-4948 to determine
whether your application uses IOC, what version of IOC it is using, and the
general migration suggestions for application owners whose applications use the
IOC. It also contains migration suggestions for most classes included in the IBM
Open Class Library.
In general, where an overlap in functions exists between the IBM Open Class
Library and the Standard C++ Library, including the Standard Template Library,
the following is recommended:
Use the Standard Template Library (STL) containers, iterators, and algorithms
instead of the IOC collection classes.
Use the Standard C++ exception classes instead of the IOC exception
classes.
Use the Standard C++ string template classes instead of the IOC string
classes.
However, there are many classes in the IBM Open Class Library for which there
is no equivalent in the Standard C++ Library. The IBM Open Class Library
388
Developing and Porting C and C++ Applications on AIX

Transition Guide, SC09-4948 identifies some of the options available to
application owners to deal with this situation. The decision as to which option is
best depends on the version of the IBM Open Class Library you use and the
extent to which you use classes without an equivalent replacement in the
Standard C++ Library.
10.6.1 Standard Template Library
The motivation behind the design of the Standard Template Library, or STL, is to
support object-oriented programming with type generality. Implemented using
templates, STL provides a set of commonly used abstract data types, access
methods, and operation methods, allowing components to be easily created, and
operations performed with minimal lines of code.
STL is composed of three major components: containers, iterators, and
algorithms. The following sections explore each component in more detail.
Containers
A container is an abstract object data type. It is a template class that manages a
sequence or set of elements. Conceptually, it is an object containing other
objects, much like an ordinary array. The elements of a container can be of any
object type, and the allocation and deallocation of the elements are controlled by
the container methods. You can also perform operations such as insert, remove,
copy, or compare elements with operators provided by the container.
STL containers come in two different varieties: sequence containers and
associative containers. A sequence container is a linear block of objects, where
elements are referred to by their position in the container. STL provides the
following sequence containers:
Vector
A contiguous block of objects that allows random access
and fast insertions at the end.
List
Supports only sequential access, but has equally fast
insertions and deletions from anywhere within the list.
Deque
Allows random access and fast insertions and deletions
from either end of the queue.
Stack
Specialized container that supports stack operations only
(LIFO).
Queue
Specialized container that supports queue operations
only (FIFO).
Associative containers, on the other hand, do not store elements in any order,
and retrieval of elements are fast by using keys. Associative container template
Chapter 10. Dealing with C++ templates
389

classes are parameterized either on a key, or a key and a type. They can be
based on unique keys, where there is at most one element in the container for
each key. Alternatively, they can be based on equivalent keys, where there is
possibly multiple copies of the same key in the container. STL provides the
following associative containers:
Set
Supports unique keys and fast retrieval of keys
themselves.
Map
Supports unique keys and fast retrieval of values (of the
parameterized type) based on the keys.
Multiset
Supports equivalent keys and fast retrieval of keys
themselves.
Multimap
Supports equivalent keys and fast retrieval of values (of
the parameterized type) based on the keys.
To use a container, simply declare a variable with the desired container class. For
example, to create an empty vector of integers:
vector<int> v;
To add an integer n to the vector:
v.push_back(n);
To get the size of the vector:
int size = v.size();
See the VisualAge C++ Standard C++ Library Reference, SC09-4949 for more
information on the methods provided by the various container classes.
Iterators
Iterators are abstract pointers to containers, that allow you to work with the
various containers in much the same way as you would with normal C++
pointers. You can refer to a location in a container using an iterator, increment it
to get to the next position, dereference it to get the value of the element, and so
on. Two iterators can be compared to find the relative locations with respect to
each other, or subtracted to find the distance between the two locations. For
example:
vector<int>::iterator start = v.begin();
vector<int>::iterator end = v.end();
int size = end - start;
the above lines of code are equivalent to:
int size = v.size();
390
Developing and Porting C and C++ Applications on AIX

And to print the value of the first element of the vector:
cout << *start << endl;
The standard defines five categories of iterators, according to the type of
operations they perform:
Input
Provides data read access to a container in a sequential
manner.
Output
Provides data write access to a container in a sequential
manner.
Forward
Allows forward traversal (that is, increment only) of the
container in a sequential manner. This can replace an
input or output iterator.
Bidirectional
Allows traversal of the container in both directions (that is,
increment and decrement) in a sequential manner. This
can replace a forward iterator.
Random access
Allows random access of a container. This can replace a
bidirectional iterator.
Refer to the VisualAge C++ Standard C++ Library Reference, SC09-4949 for
more details.
Algorithms
Algorithms are template functions that are specialized to perform operations on
containers. They are parameterized by iterators instead of containers, in order to
support other user-defined types.
There are three main types of algorithmic operations. Non-modifying sequence
operations, as the name suggests, do not alter the sequence of the container in
any manner. It includes template functions such as find, count, equal, mismatch,
and search.
Mutating sequence operations, including functions such as copy, swap,
transform, replace, fill, generate, remove, unique, reverse, rotate,
random_shuffle, and partition, change the sequence of the container while the
operation is carried out.
Sorting and related operations may or may not change the sequence, and
include such functions as sort, nth_element, lower_bound, upper_bound,
equal_range, binary_search, merge, and so on. There are also set operations
functions like set_union, set_intersection, and set_difference that operate on
sorted containers.
Chapter 10. Dealing with C++ templates
391

For more information on all available algorithmic functions, refer to the VisualAge
C++ Standard C++ Library Reference, SC09-4949 for details.
10.6.2 A STL example
Putting it all together, the following is a simple example that takes an integer
number as input, and randomly generates 10 numbers between 0 to 100, using
the input as the seed value for the random generator function. The numbers are
stored in a vector container. They are then sorted and output to the terminal:
#include <algorithm>
#include <vector>
#include <iostream>
#include <stdlib.h>
using namespace std;
int main(int argc, char *argv[])
{
unsigned int seed;
if (argc >= 2 && (seed = atoi(argv[1])))
srand(seed);
vector<int> v;
for (int i = 0; i < 10; i++)
v.push_back(1+(int)(100.0*rand()/(RAND_MAX+1.0)));
sort(v.begin(), v.end());
for (int i = 0; i < v.size(); i++)
cout << v[i] << endl;
}
392
Developing and Porting C and C++ Applications on AIX

11
Creating shared objects
Chapter 11.
from C++ source codes
Creating shared objects from C++ source codes on AIX require some additional
knowledge in addition to the tasks required for creating shared objects from C
source codes.
This chapter explains how to create and manage shared objects from C++
source codes for C++ application programmers by providing the following
sections:
Section 11.1, “Creating shared objects from C++ source codes” on page 394
Section 11.2, “Shared objects with templates” on page 402
© Copyright IBM Corp. 2000, 2003. All rights reserved.
393

11.1 Creating shared objects from C++ source codes
The C++ language, although similar in some respects to the C language, offers
many additional facilities. One of these is known as function overloading, which
makes it possible to have multiple functions with the same name but different
parameter lists. This feature means it is not possible to use the function name
alone as a unique identifier in the symbol table of an object file. For this reason,
function names in C++ are mangled to produce the symbol name. The mangling
uses a code to indicate the number, type, and ordering of parameters to the
function.
The term mangle has a specific meaning in compiler products, especially in the
C++ compilers, because the language standard allows you to have multiple
instances of the same name function, but each function will be used with
specified parameter variable types. For example, the following two function
definitions can be mangled (this is a pseudo-mangling example):
int add(int, int)
double add(double, double)
to the following function names during the compilation process:
add_Fii(int, int)
add_Fdd(double, double)
Please note that the mangled names are referred to as symbols only after the
compilation; therefore, you do not have to understand what mangled names will
be produced after the compilation process (see 6.3.4, “The c++filt utility” on
page 243 for further information about the mangling process).
It is the name mangling feature of C++ that means the process of creating a
shared object, which includes object code created by the C++ compiler, is slightly
more complicated than when using code produced by the C compiler.
Although it would be possible to create import and export files manually, the
process is time consuming, since a unique symbol name is required for each
instance of an overloaded function.
394
Developing and Porting C and C++ Applications on AIX

11.1.1 Creating a C++ shared object
VisualAge C++ for AIX provides the makeC++SharedLib command, which
performs most of the dirty work behind the scenes. It is included in the
vacpp.cmd.tools fileset and installed automatically when the product is installed,
as shown in the following example:
# lslpp -w /usr/vacpp/bin/makeC++SharedLib
File Fileset Type
----------------------------------------------------------------------------
/usr/vacpp/bin/makeC++SharedLib vacpp.cmp.tools File
The process of creating a shared object from C++ source codes is very similar to
the process of creating a shared object from C source codes. See 2.8, “Creating
shared objects” on page 92 how to create shared objects from C program source
codes.
To create a shared object from C++ source codes, do the following:
1. Compile the C++ source code from which you wish to create to a shared
object. For example, the following example will produce two object modules,
cppsrc1.o and cppsrc2.o, in the current directory:
$ xlC -c cppsrc1.C cppsrc2.C
2. Run the makeC++SharedLib command1 to create the shared object with the
object file names. The following example shows how to create a shared
object, shr_cpp_12.o, from the object modules created in the previous step:
$ /usr/vacpp/bin/makeC++SharedLib -o shr_cpp_12.o cppsrc1.o cppsrc2.o
Use the -G option when you want to create a shared object enabled for use with
run-time linking, or one that uses the libname.so format. For example:
$ /usr/vac/bin/makeC++SharedLib -G -o shr_cpp_12.so cppsrc1.o cppsrc2.o
If the makeC++SharedLib command is used to build the C++ shared libraries and
export symbols, make sure that any system libraries required by the shared
object are always specified with the -l option (for example, -lX11) and not by
name (for example, /usr/lib/libX11.a) on the command line. This allows the
system libraries to be simply referenced by the shared object being created and
not go through the special C++ related processing.
11.1.2 Generating an export file
Another very useful option of the makeC++SharedLib command is the ability to
save the export file that is generated behind the scenes and normally discarded
1 If you add the directory path, /usr/vacpp/bin, into the PATH environment value, you do not have to
invoke the makeC++SharedLib command with the full pathname.
Chapter 11. Creating shared objects from C++ source codes
395

after use. If saved, this export file can then be used as an import file when
creating another shared object. The -e expfile option is used to save the export
file. Note that the export file produced does not have an object file name field (#!)
on the first line, so one will have to be manually added, if required.
11.1.3 The -qmkshrobj option
Starting from Version 5, VisualAge C++ for AIX provides a useful compiler option,
-qmkshrobj, which is used to instruct the compiler to create a shared object from
previously created object files and archive libraries. It provides similar
functionality to the makeC++SharedLib command and, in addition, makes it much
easier to create shared objects that use template functions. See 11.2, “Shared
objects with templates” on page 402 for more detailed information.
For example, to create a shared object shr1.o from the source1.o and source2.o
files, use the following command:
$ xlC -qmkshrobj -o shr1.o source1.o source2.o
The -G option can be also used in conjunction with the -qmkshrobj option to
create an object that uses the run-time linking shared object, as shown in the
following example:
$ xlC -G -qmkshrobj -o libshr1.so source1.o source2.o
To specify the priority of the shared object, which determines the initialization
order of the shared objects used in an application, append the priority number to
the -qmkshrobj option. For example, to create the shared object shr1.o, which
has an initialization priority of -100, use the following command:
xlC -qmkshrobj=-100 -o shr1.o source1.o source2.o
If none of the -bexpall, -bE:, -bexport:, or -bnoexpall options are specified, then
using the -qmkshrobj option will force the compiler to generate an exports file that
exports all symbols. This file can be saved for use as an import file when creating
other shared objects, if desired. This is done using the -qexpfile=filename option.
For example:
xlC -qmkshrobj -qexpfile=shr1.exp -o shr1.o source1.o source2.o
11.1.4 Mixing C and C++ object files
In addition to the mangling of symbol names, the C++ language differs from the C
language in the way function arguments are passed on the calling stack. The C
language pushes arguments onto the stack right to left, which means the
left-most argument is top-most on the stack. For various reasons, the C++
language uses right to left instead. This is termed linkage, and a complication
unit can have both the C and C++ linkages.
396
Developing and Porting C and C++ Applications on AIX

When mixing C and C++ code together, it is necessary to use a linkage block to
call a C routine from a C++ routine. This is to prevent the compiler from mangling
the name of the C routine, which would result in a symbol name that could not be
resolved. For example, to call the C function, foo(), from C++ code, the
declaration of foo must be in an external linkage block, as shown in the following
code fragment:
extern “C” {
void foo(void);
}
class1::class1(int a)
{
foo();
}
If the declaration of foo() was not contained in the extern “C” block, the C++
compiler would mangle the symbol name to foo__Fv.
When mixing C and C++ objects within a single shared object, either the
makeC++SharedLib command (which uses the C++ compiler) or the -qmkshrobj
option of the C++ compiler should be used to create the shared object. Do not
use the C compiler or the linker, since they may not produce the correct result, as
they are not aware of C++ constructors, destructors, templates, and other C++
language features.
11.1.5 Order of initialization
There are situations where the order of initialization of data objects within a
program is important to the correct operation of the application. A priority can be
assigned to an individual object file when it is compiled. This is done using the
-qpriority option. For example:
$ xlC -c zoo.C -qpriority=-50
The C++ compiler and the makeC++SharedLib command also support options that
can be used to indicate the relative order of initialization of shared objects. There
is a slight difference in the way the priority is specified when using each
command. When using the C++ compiler, the priority is specified as an additional
value with the -qmkshrobj option. For example:
$ xlC -qmkshrobj=-100 -o shr1.o source1.o
When using the makeC++SharedLib command, the priority is specified with the -p
option. For example:
$ makeC++SharedLib -p -100 -o shr1.o source1.o
Chapter 11. Creating shared objects from C++ source codes
397

Priority values can also be indicated within C++ code by inserting the priority
compiler directive as follows:
#pragma priority(value)
These values alter the order of initialization of data objects within the object
module.
Priority values
Priority values may be any number from -214782623 to 214783647. A priority
value of -214782623 is the highest priority. Data objects with this priority are
initialized first. A priority value of 214783647 is the lowest priority. Data objects
with this priority are initialized last. Priority values from -214783648 to
-214782624 are reserved for system use. If no priority is specified, the default
priority of 0 is used.
The explanation of priority values uses the example data objects and files shown
in Figure 11-1 on page 399.
398
Developing and Porting C and C++ Applications on AIX

myprogram.C
fish.o
fresh.C
salt.C
....................
..........
#pragma priority(-80)
main () {
........
#pragma priority(-200)
..........
.............
class trout A
class shark S
.......
..........
class Cage CAGE
#pragma priority(500)
#pragma priority(10)
..............
........
..........
class bass B
class tuna T
animals.o
house.C
farm.C
zoo.C
...........
...........
............
#pragma priority(20)
class horse H
class lion L
...........
...........
............
class dog D
#pragma priority(500)
#pragma priority(50)
#pragma priority(100)
...........
............
...........
class zebra Z
class cow W
class cat C
............
Figure 11-1 Illustration of objects in fish.o and animals.o
Chapter 11. Creating shared objects from C++ source codes
399

This example shows how to specify priorities when creating shared objects to
guarantee the order of initialization. The user should first of all determine the
order in which they want the objects to be initialized, both within each file and
between shared objects:
1. Develop an initialization order for the objects in house.C, farm.C, and zoo.C:
a. To ensure that the object lion L in zoo.C is initialized before any other
objects in either of the other two files in the shared object animals.o,
compile zoo.C using a -qpriority=nn option with nn less than zero so that
data objects have a priority number less than any other objects in farm.C
and house.C:
$ xlC -c -qpriority=-50 zoo.C
b. Compile the house.C and farm.C files without specifying the -qpriority=nn
option. This means the priority will default to zero. This means data objects
within the files retain the priority numbers specified by their #pragma
priority(nn) directives:
$ xlC -c house.C farm.C
c. Combine these three files into a shared library. Use the makeC++SharedLib
command to construct the shared object animals.o with a priority of 40:
$ makeC++SharedLib -o animals.o -p 40 house.o farm.o zoo.o
2. Develop an initialization order for the objects in fresh.C and salt.C, and use
the #pragma priority(value) directive to implement it:
a. Compile the fresh.C and salt.C files:
$ xlC -c fresh.C salt.C
b. To assure that all the objects in fresh.C and salt.C are initialized before any
other objects, including those in other shared objects and the main
application, use makeC++SharedLib to construct a shared object fish.o with
a priority of -100:
$ makeC++SharedLib -o fish.o -p -100 fresh.o salt.o
Because the shared object fish.o has a lower priority number (-100) than
animals.o (40), when the files are placed in an archive file with the ar
command, the objects are initialized first.
3. To create a library that contains the two shared objects, so that the objects
are initialized in the order you have specified, use the ar command. To
produce an archive file, libprio.a, enter the command:
$ ar rv libprio.a animals.o fish.o
Where libprio.a is the name of the archive file that will contain the shared
library files, and animals.o and fish.o are the two shared files created with
makeC++SharedLib.
400
Developing and Porting C and C++ Applications on AIX

4. Compile the main program, myprogram.C, that contains the function main to
produce an object file, myprogram.o. By not specifying a priority, this file is
compiled with a default priority of zero, and the objects in main have a priority
of zero:
$ xlC -c myprogram.C
5. Produce an executable file, animal_time, so that the objects are initialized in
the required order, and enter:
$ xlC -o animal_time main.o -lprio -L.
When the animal_time executable is run, the order of initialization of objects is
as shown in Table 11-1.
Table 11-1 Order of initialization of objects in prriolib.a
Object
Priority
Comment
value
fish.o
-100
All objects in fish.o are initialized first because they are in a
library prepared with makeC++SharedLib -p -100 (lowest
priority number; -p -100 specified for any files in this
compilation).
shark S
-100(-200)
Initialized first in fish.o because within file #pragma
priority(-200).
trout A
-100(-80)
#pragma priority(-80).
tuna T
-100(10)
#pragma priority(10).
bass B
-100(500)
#pragma priority(500).
myprog.o
0
File generated with no priority specifications; default is 0.
CAGE
0(0)
Object generated in main with no priority specifications;
default is 0.
animals.o
40
File generated with makeC++SharedLib with -p 40.
lion L
40(-50)
Initialized first in file animals.o compiled with -qpriority=-50.
horse H
40(0)
Follows with priority of 0 (since -qpriority=nn not specified at
compilation and no #pragma priority(nn) directive).
dog D
40(20)
Next priority number (specified by #pragma priority(20)).
zebra N
40(50)
Next priority number from #pragma priority(50).
cat C
40(100)
Next priority number from #pragma priority(100).
cow W
40(500)
Next priority number from #pragma priority(500).
Chapter 11. Creating shared objects from C++ source codes
401

11.2 Shared objects with templates
Templates are usually declared in a header file. Each time a template is used,
code is generated to instantiate the template with the desired parameters. Most
C++ compilers work with a template repository. No template code is generated at
compile time; the compiler just remembers where the template code came from.
Then, at link time, as the compiler/linker puts all parts together, it notices which
templates actually need to be generated. The code is then produced, compiled,
and linked into the application.
This becomes a problem when using templates with shared libraries, where no
actual linking takes place. Therefore, you must make sure that the template code
is generated when producing the shared library.
Therefore, one should keep track of compilation and inclusion of template
instantiations. This would mean that one has to manually keep track of all the
template instantiation and address them during the linking phase.
Starting from Version 5, the VisualAge C++ for AIX product supports a compiler
option, -qmkshrobj, to create a shared objects. This option, together with the
-qtempinc option, should be used in preference to the makeC++SharedLib
command when creating a shared object that uses templates. The advantage of
using these options instead of makeC++SharedLib is that the compiler will
automatically include and compile the template instantiations in the tempinc
directory.
11.2.1 Templates and makeC++SharedLib
The makeC++SharedLib command is supplied with all IBM C++ command line
compiler drivers for the AIX platform.2 The command is implemented as a shell
script that gathers the supplied input and then calls the linker to create the
shared object.
When creating a shared object that uses templates, the makeC++SharedLib
command needs to somehow find information on the templates that are to be
instantiated. Because the script calls the linker, and not the compiler, it does not
look at the contents of the tempinc directory. This means the method of creating
a shared object that uses templates relies on either using the simple template
method code layout, as described in 10.3, “Simple code layout method” on
page 381, or forcing templates to be instantiated, as described in 10.4.3,
“Forcing template instantiation” on page 386.
The best method to use will depend on the circumstances. Using the simple code
layout method means that all the required templates are generated automatically.
2 Type ls -l /usr/vacpp/bin/makeC++SharedLib* on the command line.
402
Developing and Porting C and C++ Applications on AIX

However, it also comes with the disadvantages of slower compile times and
larger code size. Forcing the templates to be instantiated is better from both the
code size and compile time aspect, but it does mean that the user needs to
maintain files that instantiate the required templates.
Suppose you want to create a shared object from the following two source files,
which use the preferred code layout method.
File source1.C contains the following code:
#include “stack.h”
stack<int> counter1;
void function1(int a)
{
counter1.push(a);
}
The file source2.C contains the following code:
include “stack.h”
stack<int> counter2;
void function2(int a)
{
counter2.push(a);
}
Using the makeC++SharedLib command, an attempt is made to create a shared
object as follows:
$ xlC -c source1.C source2.C
$ /usr/vacpp/bin/makeC++SharedLib -o shr1.o -p0 source1.o source2.o
ld: 0711-317 ERROR: Undefined symbol: .stack<int>::stack(int)
ld: 0711-317 ERROR: Undefined symbol: .stack<int>::~stack()
ld: 0711-317 ERROR: Undefined symbol: .stack<int>::push(int)
ld: 0711-345 Use the -bloadmap or -bnoquiet option to obtain more information.
The command failed, and based on the output, it is easy to see that the required
template functions have not been instantiated. At this point, note that because
the code uses the preferred code layout method, the compiler has, in fact,
automatically created the file tempinc/stack.C, which, if compiled, would supply
the required template definitions. You can, if you wish, copy this file and make it
an explicit compilation unit as part of your source code. In this case, that would
mean adding the following command to the sequence:
$ xlC -c tempinc/stack.C -o stack.o
The object, file stack.o, would then be passed to the makeC++SharedLib command
along with source1.o and source2.o.
Chapter 11. Creating shared objects from C++ source codes
403

11.2.2 Templates and -qmkshrobj
You should use the -qmkshrobj option instead of the makeC++SharedLib
command when creating a shared object. Because the option is a compiler
option, the compiler will automatically look in the tempinc directory (or the
directory specified with the -qtempinc=dirname option) for the automatically
generated template instance information. Using the same source files as
described in the previous section, the following commands can be used to create
the shared object:
$ xlC -c source1.C source2.C
$ xlC -qmkshrobj -o shr1.o source1.o source2.o
This time, the command works correctly, since the compiler looks in the tempinc
directory. Remember to use the same -qtempinc option (if any) that was used
when compiling the modules being used to create the shared object.
This option solves the problems associated with creating shared objects that use
template classes and functions. If you want to create a shared object that
contains pre-instantiated template classes and functions for use by other
developers, then you can create an additional compilation unit that explicitly
defines the required templates using the #pragma define directive.
404
Developing and Porting C and C++ Applications on AIX

12
Packaging your applications
Chapter 12.
If your applications are only executed on the systems where you are developing
and debugging, no software installation is required. However, applications are
most likely developed on development and test systems, then installed on
production systems.
The software installation tasks are commonly performed by customers or other
people who might not have enough knowledge about your applications. Even if
you have prepared the detailed software installation instruction for your
applications, there is no guarantee that those people follow your instruction; you
have no control where they try to install your applications.
By creating your application packages, the required conditions, such as
supported operating system software levels and adequate physical memory size
for your applications, will be enforced.
This chapter provides the following sections to explain how to package your
applications for AIX:
Section 12.1, “Understanding the AIX standard packaging” on page 406
Section 12.2, “Packaging applications using mkinstallp” on page 418
© Copyright IBM Corp. 2000, 2003. All rights reserved.
405

12.1 Understanding the AIX standard packaging
This section provides you with a basic understanding of the AIX standard
packaging. For further information about AIX standard packaging, please refer to
the AIX 5L Version 5.2 Installation Guide and Reference.
12.1.1 Filesets and package files
In AIX, the smallest installable unit is a fileset. A fileset logically groups files and
directories to be installed. A fileset also includes required control files and
optional installation customization files.
Several filesets can be packaged in a package file. A package file is an AIX
backup-format file; therefore, it is sometimes referred to as a bff-file.
Figure 12-1 illustrates the relationship among filesets, packages, and bundles.
Package A
Package B
Package C
Bundle 1
Fileset A1
Fileset C1
Bundle 2
Fileset A2
Fileset B1
Fileset C2
Fileset A3
Figure 12-1 Relationship among filesets, packages, and bundles
12.1.2 Bundles
In AIX, you have hundreds of filesets in the base operating system (BOS).
Therefore, to easily select many filesets, AIX offers simple facilities, called
bundles. In Figure 12-1, if you select bundle 1, then you specify installing the
filesets A1, B1, and C1. Please note that multiple bundles can include the same
filesets. For example, in Figure 12-1, both bundle 1 and 2 include the fileset B1.
A bundle is defined by a file installed in either the /usr/sys/inst.data/sys_bundles
or /usr/sys/inst.data/user_bundles directories. In the
/usr/sys/inst.data/sys_bundles directory, you can see the system defined1
bundles are already installed by default (see Example 12-1 on page 407).
406
Developing and Porting C and C++ Applications on AIX

Example 12-1 System defined bundles on AIX 5L Version 5.2
# cd /usr/sys/inst.data/sys_bundles; ls -l *.bnd
-rw-r--r-- 1 bin bin 896 Sep 13 09:53 AllDevicesKernels.bnd
-rw-r--r-- 1 bin bin 879 Sep 13 09:53 Alt_Disk_Install.bnd
-rw-r--r-- 1 bin bin 1391 Sep 13 09:53 App-Dev.bnd
-rw-r--r-- 1 bin bin 1051 Sep 13 09:53
CC_EVAL.DocServices.bnd
-rw-r--r-- 1 bin bin 1155 Sep 13 09:53 CC_EVAL.Graphics.bnd
-rw-r--r-- 1 bin bin 1176 Sep 13 11:03 CDE.bnd
-rw-r--r-- 1 bin bin 1026 Sep 13 09:53 DocServices.bnd
-rw-r--r-- 1 bin bin 1292 Sep 13 09:53 GNOME.bnd
-rw-r--r-- 1 bin bin 1517 Sep 13 09:53 Graphics.bnd
-rw-r--r-- 1 bin bin 879 Sep 13 09:53 HTTP_Server.bnd
-rw-r--r-- 1 bin bin 972 Sep 13 09:53 KDE.bnd
-rw-r--r-- 1 bin bin 829 Sep 13 09:53 Kerberos_5.bnd
-rw-r--r-- 1 bin bin 1201 Sep 13 09:53 Media-Defined.bnd
-rw-r--r-- 1 bin bin 1363 Sep 13 09:53 Netscape.bnd
-rw-r--r-- 1 bin bin 1214 Sep 13 11:03 Server.bnd
-rw-rw-r-- 1 root system 409 Jun 06 2002 devices.bnd
-rw-r--r-- 1 bin bin 2030 Sep 13 09:53 openssh_client.bnd
-rw-r--r-- 1 bin bin 2030 Sep 13 09:53 openssh_server.bnd
-rw-r--r-- 1 bin bin 1373 Sep 13 11:03 wsm_remote.bnd
In order to install filesets using a bundle, do the following:
1. Select the following SMIT menus (you can access it using the SMIT fast path
smit easy_install):
# smit
Software Installation and Maintenance
Install and Update Software
Install Software Bundle
2. Specify the installation device (typically /dev/cd0):
INPUT device / directory for software
[ ]
3. Select a bundle name on the panel, then press the Enter key; all the filesets
defined in the selected bundle file, shown in Example 12-1, will be installed.
Please note that you can only use bundles for installation purposes. AIX does not
offer simple methods to uninstall bundles or to confirm whether bundles are
correctly installed.
1 You can also create user defined bundles under the /usr/sys/inst.data/user_bundles directory.
Chapter 12. Packaging your applications
407

12.1.3 Managing filesets
The installed filesets can be classified under several states, as shown in
Table 12-1. The successfully installed filesets should be in either an applied or
committed state.
Table 12-1 Fileset state2
Status
Description
APPLIED
The specified fileset update is installed on the system. The
APPLIED state means that the fileset update can be rejected with
the installp command and the previous level of the fileset restored.
This state is only valid for fileset updates.
APPLYING*
An attempt was made to apply the specified fileset, but it did not
complete successfully, and cleanup was not performed.
BROKEN
The specified fileset or fileset update is broken and should be
reinstalled before being used.
COMMITTED
The specified fileset is installed on the system. The COMMITTED
state means that a commitment has been made to this level of the
software. A committed fileset update cannot be rejected, but a
committed fileset base level and its updates (regardless of state)
can be removed or deinstalled by the installp command.
OBSOLETE
The specified fileset was installed with an earlier version of the
operating system but has been replaced by a repackaged
(renamed) newer version. Some of the files that belonged to this
fileset have been replaced by versions from the repackaged fileset.
COMMITTING*
An attempt was made to commit the specified fileset, but it did not
complete successfully, and cleanup was not performed.
REJECTING*
An attempt was made to reject the specified fileset, but it did not
complete successfully, and cleanup was not performed.
In order to confirm the fileset status, you can use the lslpp -L command, as
shown in Example 12-2 on page 409. In this case, the fileset bos.rte.install is in
committed status (C).
2 Filesets with the state specified with an asterisk (*) are not shown on the lslpp -L command
output, since they are considered to be in a transient state.
408
Developing and Porting C and C++ Applications on AIX

Example 12-2 Listing a fileset status
# lslpp -L bos.rte.install
Fileset Level State Type Description (Uninstaller)
----------------------------------------------------------------------------
bos.rte.install 5.2.0.0 C F LPP Install Commands
State codes:
A -- Applied.
B -- Broken.
C -- Committed.
O -- Obsolete. (partially migrated to newer version)
? -- Inconsistent State...Run lppchk -v.
If a fileset status is broken, then you should deinstall and reinstall the fileset.3 If a
fileset status is obsolete, then the fileset may or may not be supported on your
system. If a fileset status is inconsistent, you should check it using the lppchk -v
command.
Figure 12-2 on page 410 illustrates the state diagram of the applied and
committed states. Once a base level fileset (fileset level 1.0.0.0) is installed, it is
always in the committed status. If you apply an update fileset (fileset level
1.0.1.0),4 the status is changed to the applied status.
If you commit the applied update fileset, then the updated fileset level is
persistent. In order to revert to the previous fileset level, you have to uninstall the
fileset and reinstall it. If you reject the applied update fileset, then the fileset level
is reverted to the last committed level.
3 If you install filesets over the network, you should verify the size and the checksum of the installing
package file on the target system.
4 An update fileset is sometimes referred to as a PTF (Program Temporary Fix).
Chapter 12. Packaging your applications
409

Fileset : foo.rte
Level : 1.0.0.0
Status : Committed
(Base level fileset)
Apply
Reject
Fileset : foo.rte
Uninstall and Reinstall
Level : 1.0.1.0
Status : Applied
(Update level fileset)
Fileset : foo.rte
Level : 1.0.1.0
Commit
Status : Committed
(Update level fileset)
Figure 12-2 State diagram between applied and committed state
This mechanism is used to precisely control the software levels on the running
system. When you encounter a software problem, you should investigate to solve
the problem. If the problem is caused by some defects, software vendors
supporting the software products might provide some software fixes.
APARs and fileset updates
In the IBM terminology, a software defect is uniquely identified by an identifier
called an APAR (authorized program analysis reports). An APAR can be, but
does not have to be, addressed by more than one fileset updates. If an APAR is
addressed by multiple fileset updates, then the software defect affects many files
included in multiple filesets. In Figure 12-3 on page 411,5 the APAR IZ98765
includes the fileset update of foo.rte with an update level 1.0.1.0. Therefore, in
order to fix the software defect identified by the APAR IZ98765, you have to apply
this single fileset update only. In order to fix the software defect identified by the
APAR IZ56789, you have to apply both fileset updates, for foo.rte and gnat.rte, as
shown in Figure 12-3 on page 411.
5 We use these two identifies, IZ98765 and IZ56789, for illustrative purposes only. They do not exist.
410
Developing and Porting C and C++ Applications on AIX

Fileset : foo.rte
APAR: IZ98765
Level : 1.0.1.0
(Update level fileset)
Fileset : gnat.rte
APAR: IZ56789
Level : 2.0.3.0
(Update level fileset)
Figure 12-3 Relationship between APARs and update fileset
You can confirm whether the specific APARs are applied or not using the instfix
command. In the following example, the APAR IY35444 is applied on the system:
# instfix -ivk IY35444
IY35444 Abstract: Add snapshot backup support to JFS2
Fileset bos.mp:5.2.0.1 is applied on the system.
Fileset bos.mp64:5.2.0.1 is applied on the system.
Fileset bos.up:5.2.0.1 is applied on the system.
All filesets for IY35444 were found.
If an APAR is not applied6 on the system, the instfix -ik APAR_ID command
shows the error message similar to either of the following examples:
All filesets for IZ98765 were not found.
or
There was no data for IZ56789 in the fix database.
Therefore, you do not have to remember which fileset update would fix the
specific software defect, as long as you know the corresponding APAR.
To find APARs for AIX, visit the following URL:
http://techsupport.services.ibm.com/server/support?view=pSeries
6 An APAR classified as a packaging APAR also shows this message. A packaging APAR is provided
to specify multiple APARs using one identifier, mainly for ordering and distribution purposes.
Chapter 12. Packaging your applications
411

Recommended maintenance level
Sometimes IBM ships a recommended maintenance level (also referred to as
RML), which includes a series of fileset updates. By using recommended
maintenance levels, you can easily track the latest level of all the filesets included
in AIX. The latest AIX installation media set usually includes the latest
recommended maintenance level in the Update CD.
To simply determine the latest recommended maintenance level applied on the
system, you can use the oslevel -r command as follows:
# oslevel -r
5200-01
The command output 5200-01 shows that RML 5200-01 is applied on that
system.
In order to determine what recommended maintenance levels are applied, you
can use the instfix command as follows:
# instfix -i | grep ML
All filesets for 5.0.0.0_AIX_ML were found.
All filesets for 5200-01_AIX_ML were found.
If some lines show the message All filesets for XXXX-YY_AIX_ML were not
found, then you can confirm which of the fileset updates included in the
recommended maintenance level are not applied on the system by running the
following command:
# instfix -ivk XXXX-YY_AIX_ML | grep not | grep :
To download the recommended maintenance level, visit the following URL:
http://techsupport.services.ibm.com/server/support?view=pSeries
Note: We strongly recommend that you purchase a software program support
contract for each AIX system, even if you understand the AIX software
packaging mechanism and can easily download the required fileset updates.
To purchase software program support, please contact your IBM sales
representative or the IBM Business Partner from which you purchased your
systems.
12.1.4 Viewing the TOC file (.toc)
Each AIX standard package file contains a table of contents (TOC). Before
installation, this information has to be retrieved from package files and placed in
the directory as a TOC file named .toc.
412
Developing and Porting C and C++ Applications on AIX

If you download some APARs or copy some packages from the install media to a
file system, you have to issue the inutoc command to create the .toc file. If you
copy additional packages into the /usr/sys/inst.images directory, you have to
manually rebuild the .toc file using the inutoc command. The following example
shows you how to create or update the .toc file in the /usr/sys/inst.images
directory:
# inutoc /usr/sys/inst.images
# cd /usr/sys/inst.images; ls -l
total 800
-rw-r--r-- 1 root system 552 Apr 10 15:55 .toc
-rw-r--r-- 1 root system 403456 Apr 10 15:55 U476599.bff
The created .toc file is a text file, so you can view the contents using a viewer
command, such as pg or more. Figure 12-4 shows an example entry of the .toc
file. A package has a block shown as A in Figure 12-4. A fileset has a block
shown as B in Figure 12-4. If a package contains multiple filesets, then you will
see multiple blocks of B in the package block.
Fileset name
Package type
Content
Package file name
Fileset level Bosboot flag
Fileset description
U476599.bff 4 R S bos {
bos.rte.shell 05.01.0000.0010 1 b U en_US Shells (bsh, ksh, csh)
[
%
/usr/bin 1176
/usr/lpp/SAVESPACE 1176
/usr/lib/objrepos 8
Required disk space
/etc/security 8
INSTWORK 56 24
%
%
B
A
APAR description
%
IY20636 1 csh coredump with malloc debug
IY20822 1 Process synchronisation problem while using pipes in csh
IY20823 1 POSIX/SUS Standards: ksh vi-mode R command
IY21276 1 Partial display of functions by typeset -f
IY21277 1 awk -F setting is ignored by getline
IY21779 1 nohup fails with bad file descriptor error with pipeline cmds.
]
}
Figure 12-4 Sample .toc file
Table 12-2 on page 414 explains entries shown in Figure 12-4.
Chapter 12. Packaging your applications
413

Table 12-2 Fields description of the .toc file
Field name
Description
Package file name
The file name of the package.
Package type
Indicates package type:
I (Installation): All the filesets contained in this package are
base filesets.
S (Single update): All the filesets contained in this package
are fileset updates.
Fileset name
The name of the fileset.
Fileset level
The fileset level represented by (V.R.M.F):
V: Version.
R: Release.
M: Maintenance level.
F: Fix level.
Bosboot flag
Indicates whether a bosboot is needed after installation:
N: Do not invoke bosboot.
b: Invoke bosboot.
Content
Indicates the parts included in the fileset or the fileset update:
B: usr and root part.
H: share part.
U: usr part only.
Fileset description
The description of the fileset.
Required disk space
Size required for each install target directory.
APAR descriptions
Information regarding the APARs contained in the fileset
update.
For further information about the format of the .toc file, please refer to AIX 5L
Version 5.2 General Programming Concepts: Writing and Debugging Programs.
Once a .toc file is created, you can list the table of contents using the installp
command, as shown in Example 12-3 on page 415.
414
Developing and Porting C and C++ Applications on AIX

Example 12-3 Listing the table of contents
# installp -ld /usr/sys/inst.images
Fileset Name Level I/U Q Content
====================================================================
bos.mp64 5.2.0.1 S b usr
# Base Operating System 64-bit Multiprocessor Runtime
bos.mp64 5.2.0.2 S b usr
# Base Operating System 64-bit Multiprocessor Runtime
12.1.5 Viewing package files
The AIX standard packaging uses the backup command to archive package files.
Therefore, you can un-archive it using the restore command. Example 12-4
shows how to view the contents of a package file.
Example 12-4 Viewing the contents of a package file
# restore -qTf bos.mp64.5.2.0.2.U
New volume on bos.mp64.5.2.0.2.U:
Cluster size is 51200 bytes (100 blocks).
The volume number is 1.
The backup date is: Mon Nov 18 11:24:17 CST 2002
Files are backed up by name.
The user is BUILD.
./
./lpp_name
./usr
./usr/lpp
./usr/lpp/bos.mp64/bos.mp64/5.2.0.2
./usr/lpp/bos.mp64/bos.mp64/5.2.0.2/liblpp.a
./usr/lib/boot/unix_64
The number of archived files is 7.
TOC information file (lpp_name)
A package always contains a file, named lpp_name, as its first archived file. This
file is the table of contents of this package file, as shown in Example 12-5 on
page 416.
Chapter 12. Packaging your applications
415

Example 12-5 Extracting the lpp_name file
# mp64.5.2.0.2.U ./lpp_name
x ./lpp_name
# ls -ld ./lpp_name
-r-xr-xr-x 1 root system 1823 Nov 18 11:24 ./lpp_name
# cat ./lpp_name
4 R S bos.mp64 {
bos.mp64 5.2.0.2 01 b U en_US Base Operating System 64-bit Multiprocessor
Runtime
[
%
/usr/lib/boot 19776
/usr/lpp/SAVESPACE 19776
/usr/lib/objrepos 8
/tmp 0 14656
INSTWORK 64 32
%
%
%
IY35438 3 dio read to an unpinned page
IY35432 3 Fixes for JFS2 on filesystems larger than 2TB
... many lines are skipped ...
]
}
Upon invoking the inutoc command, it parses the specified directory and
extracts the lpp_name file from each package file, then concatenates the
extracted information to the .toc file.
Installation control library file (liblpp.a)
A package also has to contain an installation control library file, named liblpp.a,
under the appropriate sub-directory. This file is an ar format archive file, which
contains the files, as shown in Example 12-6.
Example 12-6 Extracting the liblpp.a file
# restore -qxf bos.mp64.5.2.0.2.U ./usr/lpp/bos.mp64/bos.mp64/5.2.0.2/liblpp.a
x ./usr/lpp/bos.mp64/bos.mp64/5.2.0.2/liblpp.a
# cd ./usr/lpp/bos.mp64/bos.mp64/5.2.0.2
# ls -l liblpp.a
-r-xr-xr-x 1 root system 11382 Nov 18 11:24 liblpp.a
# file liblpp.a
liblpp.a: archive
# ar -t liblpp.a
productid
bos.mp64.copyright
bos.mp64.inventory
bos.mp64.size
416
Developing and Porting C and C++ Applications on AIX

bos.mp64.al
bos.mp64.fixdata
The <fileset_name>.al file contains the list of files in the fileset. Example 12-7
shows you the contents of the bos.mp64.al file, which contains one file that is
specified using the relative path name starting from the root directory.
Example 12-7 Contents of the bos.mp64.al file
./usr/lib/boot/unix_64
The <fileset_name>.inventory file contains the required information about files
within the fileset (see Example 12-8). Each file has entries explained in
Table 12-3.
Table 12-3 Definition of entries in <fileset_name>.inventory
Entry name
Description
owner
Specifies the file owner.
group
Specifies the file group.
mode
Specifies the permission bit of the file.
type
Specifies the file type.
links
Specifies a hard-link of the file (if available).
class
The logical group of the file.
size
Specifies the size of the file.
checksum
Specifies the result of the cksum command to the file.
These values are used to verify the contents of restored files from the package
file if those files are restored correctly.
Example 12-8 Contents of the bos.mp64.inventory file
/usr/lib/boot/unix_64:
owner = root
group = system
mode = 555
type = FILE
class = apply,inventory,bos.mp64
size = 10122374
checksum = "07429 9886 "
Chapter 12. Packaging your applications
417

For further information about the format of these files contained in the liblpp.a file,
please refer to AIX 5L Version 5.2 General Programming Concepts: Writing and
Debugging Programs.
12.2 Packaging applications using mkinstallp
Beginning with AIX 5L Version 5.2 plus 5200-01 Recommended Maintenance
Level, AIX provides a command, /usr/sbin/mkinstallp, to allow users to create
their own software packages for AIX. Packages created with mkinstallp are in
the AIX standard packaging format and can be installed or removed with SMIT or
the installp command.
12.2.1 mkinstallp
A new fileset, bos.adt.insttools, includes the following files:
/usr/sbin/mkinstallp
Front-end command
/usr/sbin/makebff.pl
Back-end command
/usr/lpp/bos/README.MKINSTALLP
Documentation
Note: The makebff.pl Perl script was originally written in the development
phase of the IBM Redbook Managing AIX Server Farms, SG24-6606.
Therefore, it is possible to package your applications on previous versions of
AIX; however, the script provided in the redbook, which is downloadable, might
be obsolete and is not supported by IBM.
Files to be packaged by mkinstallp must be in a directory structure so that the
location of the file relative to the root build directory is the same as the
destination of the file after installation.
For example, if /usr/your_app/bin/command1 is to be installed from a package,
command1 must be in the <build_root>/usr/your_app/bin directory when
mkinstallp is invoked as shown in Figure 12-5 on page 419.
418
Developing and Porting C and C++ Applications on AIX

<build_root>
.info/
tmp/
usr/your_app/
bin/
command1
lib/
command2
etc/
Figure 12-5 Directory structure for packaging
Once the contents of a package are in the correct directory structure, mkinstallp
prompts for basic package data from the command line interface. This data
includes the package name, requisites, descriptions of files to be packaged, and
so on. The mkinstallp command will then generate a template file based on
responses given by the user. Template files can be created or edited directly by
the user and passed to the mkinstallp command with the -T flag to prevent
command line prompting.
The mkinstallp command has the following command syntax:
mkinstallp [ -d build_root_dir ] [ -T template_file ]
Where:
-d build_root_dir
Specifies the build root directory containing the files to be
packaged. If omitted, the current working directory is
used.
-T template_file
Specifies the full path name of the template file to be
passed to mkinstallp. If omitted, mkinstallp will prompt
for package information and create a new template file
based on user responses.
12.2.2 Packaging examples
We demonstrate how to use the mkinstallp command in order to package a
sample program whose path name is /usr/redbooks/bin/vmgetinfo in this section.
Our vmgetinfo command is an executable file compiled from the C source file
shown in Example 3-17 on page 149.
Chapter 12. Packaging your applications
419

The sample application, vmgetinfo, would fail with errno 109 (ENOSYS), as
shown in the following example, if it is executed on AIX 5L Version 5.1, because it
calls the vmgetinfo() system call, which is supported from Version 5.2:
$ oslevel -r
5100-03
$ vmgetinfo
vmgetinfo() at 16 in my_application.c failed with errno = 109: Function not
implemented
Assuming that the build root directory is /tmp/packages, we have prepared the
install target files (the command itself and its source file), as shown in the
following directory structure:
# pwd
/tmp/packages
# ls -lFR .
total 8
drwxr-xr-x 3 root system 512 Feb 24 14:37 usr/
./usr:
total 8
drwxr-xr-x 4 root system 512 Feb 24 14:37 redbooks/
./usr/redbooks:
total 16
drwxr-xr-x 2 root system 512 Feb 24 14:39 bin/
drwxr-xr-x 2 root system 512 Feb 24 14:39 src/
./usr/redbooks/bin:
total 16
-r-xr-xr-x 1 root system 5898 Feb 24 14:39 vmgetinfo*
./usr/redbooks/src:
total 8
-rw-r--r-- 1 root system 922 Feb 24 14:39 vmgetinfo.c
Note: The owner, group, and permission modes must be carefully verified
before actually creating your packages.
Preparing a template file
We have prepared the template file shown in Example 12-9 on page 421 in order
to create a package, called vmgetinfo. In this example, the vmgetinfo package
consists of two filesets: vmgetinfo.rte and vmgetinfo.src.
As shown in the high-lighted lines in Example 12-9 on page 421, vmgetinfo.rte
requires the bos.rte.libc fileset to be installed with level 5.2.0.0 or higher. This
requisite condition would prevent users from installing this fileset on AIX 5L
Version 5.1 and earlier. Also, vmgetinfo.src requires that vmgetinfo.rte be
420
Developing and Porting C and C++ Applications on AIX

installed. Therefore, even if a user selects vmgetinfo.src only, vmgetinfo.rte will
be installed automatically.
For the complete definition of requisite condition keywords, refer to the
“Packaging Software for Installation” section in AIX 5L Version 5.2 General
Programming Concepts: Writing and Debugging Programs.
Example 12-9 A template file (redbook.tmplt)
Package Name: vmgetinfo
Package VRMF: 1.0.0.1
Update: N
Fileset
Fileset Name: vmgetinfo.rte
Fileset VRMF: 1.0.0.1
Fileset Description: vmgetinfo runtime
Bosboot required: N
License agreement acceptance required: N
Name of license agreement:
Include license files in this package: N
License file path:
Requisites: *prereq bos.rte.libc 5.2.0.0
Files
/usr/redbooks
/usr/redbooks/bin
/usr/redbooks/bin/vmgetinfo
EOFiles
EOFileset
Fileset
Fileset Name: vmgetinfo.src
Fileset VRMF: 1.0.0.0
Fileset Description: vmgetinfo source
Bosboot required: N
License agreement acceptance required: N
Name of license agreement:
Include license files in this package: N
License file path:
Requisites: *coreq vmgetinfo.rte 1.0.0.1
Files
/usr/redbooks/src
/usr/redbooks/src/vmgetinfo.c
EOFiles
EOFileset
The supported keywords in template files are shown in Table 12-4 on page 422.
Chapter 12. Packaging your applications
421

Table 12-4 Template file keywords
Keyword
Description
Package Name*
Name of the package.
Package VRMF*
Version, Release, Modification, and Fix
level of the package.
Update*
Is this an update package?
Fileset*
Start of a new fileset.
Fileset Name*
Name of the fileset.
Fileset VRMF*
VRMF of the fileset.
Fileset Description
Description of the fileset.
Bosboot required*
Is a bosboot required when installing this
fileset?
License agreement acceptance required
Is license agreement acceptance required
for this fileset?
Name of license agreement
Name of the license agreement.a
Include license files in this package
Are the license files included in this
package?
License file path
Path of the license file(s).b
Requisites
Requisite conditions (coreq, ifreq, instreq,
or prereq) for the fileset.c
Files*
Start of the files section.
/path/to/file
File path.d
EOFiles*
End of the files section.
EOFileset*
End of the fileset.
a. The Name of the license agreement is defined as LAR/path/to/license/agreement. The
%L tag can be used in place of a hardcoded path to represent the locale of the
machine that the package will be installed on. For example, if a package is installed in
the en_US locale, %L will be converted to en_US.
b. The License file path is defined as LAF/path/to/license/file. Multiple license files are
separated by semicolons.
c. Requisites are defined as *Type_keyword Name VRMF; Type_keyword is either coreq,
ifreq, instreq, or prereq. Multiple requisites are separated by semicolons.
422
Developing and Porting C and C++ Applications on AIX

d. The full path name for each file in the fileset must be listed in the Files section. Any
custom directories should also be listed in this section. For example, to package the
/usr/xyz/foo file, list both the /usr/xyz directory and the /usr/xyz/foo file in the Files
section. Each entity in the final package will have the same attributes
(owner/group/permissions) that it had at build time. The user must ensure that file
attributes in the build root directory are correct prior to running mkinstallp.
Note: Keywords with a * are required, and will cause mkinstallp to fail if left
blank or omitted in the template file.
Creating a package file
Once the template file is prepared, run the mkinstallp command to create a
package file. In the following example, the template file is redbook.tmplt and we
invoked the command in the build root directory, /tmp/packages:
# pwd
/tmp/packages
# ls -l
total 16
-rw-r--r-- 1 root system 837 Feb 24 15:01 redbook.tmplt
drwxr-xr-x 4 root system 512 Feb 24 15:00 usr/
# mkinstallp -T redbook.tmplt
Using '/tmp/packages' as the base package directory.
Cannot find '/tmp/packages/.info'. Attempting to create.
Using '/tmp/packages/.info' to store package control files.
Cleaning intermediate files from '/tmp/packages/.info'.
Using 'redbook.tmplt' as the template file.
vmgetinfo 1.0.0.1 I
processing vmgetinfo.rte
processing vmgetinfo.src
creating ./.info/liblpp.a
ar: Creating an archive file ./.info/liblpp.a.
creating ./tmp/vmgetinfo.1.0.0.1.bff
The command will generate the package file in the ./tmp directory underneath
the build root directory as shown in the following example:
# ls ./tmp
vmgetinfo.1.0.0.1.bff
# inutoc ./tmp
# ls ./tmp
.toc vmgetinfo.1.0.0.1.bff
# installp -ld ./tmp
Fileset Name Level I/U Q Content
====================================================================
vmgetinfo.rte 1.0.0.1 I N usr
# vmgetinfo runtime
Chapter 12. Packaging your applications
423

vmgetinfo.src 1.0.0.0 I N usr
# vmgetinfo source
12.2.3 Verification of packages
Once the package is created, it must be verified before the actual software
deployment phase on production systems.
Installation verification
The created package must be smoothly installed on the target system if it
satisfies the install requisite conditions. If the target system does not satisfy the
conditions, the package must be prevented from being installed.
In this example, the following tests must be done:
Installing only the vmgetinfo.rte fileset on AIX 5L Version 5.2 and later.
The vmgetinfo.src fileset should not be installed automatically.
Installing only the vmgetinfo.src fileset on AIX 5L Version 5.2 and later.
The vmgetinfo.src fileset should not be installed automatically, unless the
following conditions are met:
– The -g flag of installp, which instructs the installp command to
automatically install prerequisite filesets, is specified.
– The vmgetinfo.rte fileset has been already installed.
Installing the filesets on AIX 5L Version 5.1 and earlier.
The filesets must not be installed.
Verification of the installed files
Check if the installed files are placed, according to the fileset inventory
information, with the lppchk command. There are four flags of the lppchk
command for verifying the files:
-f
Fast check (file existence, file length)
-c
Checksum verification
-v
Fileset version consistency check
-l
File link verification
424
Developing and Porting C and C++ Applications on AIX

All four calls of lppchk with the different flags has to return with a zero exit code;
if not, there are some inconsistent files, as shown in the following example:
$ lppchk -f -m 3 vmgetinfo.rte; echo $?
lppchk: 0504-230 3 files have been checked.
0
$ lppchk -c -m 3 vmgetinfo.rte; echo $?
lppchk: 0504-230 3 files have been checked.
0
$ lppchk -l -m 3 vmgetinfo.rte; echo $?
0
$ lppchk -v -m 3 vmgetinfo.rte; echo $?
0
Verification of correct functionality
Use the lslpp command to show all installed files:
# lslpp -L vmgetinfo.rte
Fileset Level State Type Description (Uninstaller)
----------------------------------------------------------------------------
vmgetinfo.rte 1.0.0.1 C F vmgetinfo runtime
# lslpp -f vmgetinfo.rte
Fileset File
----------------------------------------------------------------------------
Path: /usr/lib/objrepos
vmgetinfo.rte 1.0.0.1
/usr/redbooks/bin/vmgetinfo
/usr/redbooks
/usr/redbooks/bin
After installing and validating of the installed package, test your installed
application to see if it works as designed. In this example, we have confirmed
that the command returned the same output shown in Example 3-16 on
page 148.
Uninstallation verification
As the last step of the verification process, check if the package can be
un-installed correctly by using the installp -u command, as shown in the
following example:
# installp -u vmgetinfo.rte
+-----------------------------------------------------------------------------+
Pre-deinstall Verification...
+-----------------------------------------------------------------------------+
Verifying selections...done
Verifying requisites...done
Results...
Chapter 12. Packaging your applications
425

SUCCESSES
---------
Filesets listed in this section passed pre-deinstall verification
and will be removed.
Selected Filesets
-----------------
vmgetinfo.rte 1.0.0.1 # vmgetinfo runtime
<< End of Success Section >>
FILESET STATISTICS
------------------
1 Selected to be deinstalled, of which:
1 Passed pre-deinstall verification
----
1 Total to be deinstalled
+-----------------------------------------------------------------------------+
Deinstalling Software...
+-----------------------------------------------------------------------------+
installp: DEINSTALLING software for:
vmgetinfo.rte 1.0.0.1
Finished processing all filesets. (Total time: 0 secs).
+-----------------------------------------------------------------------------+
Summaries:
+-----------------------------------------------------------------------------+
Installation Summary
--------------------
Name Level Part Event Result
-------------------------------------------------------------------------------
vmgetinfo.rte 1.0.0.1 USR DEINSTALL SUCCESS
After removing the fileset, the installation target directory (if it is unique to the
fileset) should be removed also. In this example, all files and sub-directories
under /usr/redbooks are removed.
Now this package is ready to be deployed on production systems.
12.2.4 Optional installation control executable files
The AIX standard packaging format allows you not only to package your
applications, but also to add optional files in order to control the installation
426
Developing and Porting C and C++ Applications on AIX

process. For example, if your application requires a configuration file that
depends on the host name where the application is installed, an optional
executable file, fileset.config, can be used to automatically create the
configuration file upon the fileset installation.
The optional installation control executable files shown in Table 12-5 are called
during the installation process. Unless otherwise noted, file names that end in _i
are used during installation processing only, and file names that end in _u are
used in file set update processing only. All files in Table 12-5 are optional and can
be either shell scripts or executable object modules. Each program should have a
return value of 0 (zero), unless the program is intended to cause the installation
or update to fail.
In order to include these optional installation control executable files, create the
files in the .info sub-directory underneath the build root directory before creating
the package.
Table 12-5 Optional installation control executable files
File name
Description
fileset.config
Modifies configuration near the end of the default installation or
fileset.config_u
update process. Fileset.config is used during installation
processing only.
fileset.odmdel
Updates ODM database information for the file set prior to adding
fileset.*.odmdel
new ODM entries for the file set. The odmdel file naming
conventions enables a file set to have multiple odmdel files.
fileset.pre_d
Indicates whether a file set may be removed. The program must
return a value of 0 (zero) if the file set may be removed. File sets
are removable by default. The program should generate error
messages indicating why the file set is not removable.
fileset.pre_i
Runs prior to restoring or saving the files from the apply list in the
fileset.pre_u
package, but after removing the files from a previously installed
version of the file set.
fileset.pre_rm
Runs during a file set installation prior to removing the files from a
previously installed version of the file set.
fileset.post_i
Runs after restoring the files from the apply list of the file set
fileset.post_u
installation or update.
fileset.unconfig
Undoes configuration processing performed in the installation or
fileset.unconfig_u
update. Fileset.unconfig is used during installation processing
only.
fileset.unodmadd
Deletes entries that were added to ODM databases during the
installation or update.
Chapter 12. Packaging your applications
427

File name
Description
fileset.unpost_i_0
Undoes processing that is performed following the restoration of
fileset.unpost_u
the files from the apply list in the installation or update.
fileset.unpre_i
Undoes processing performed prior to restoring the files from the
fileset.unpre_u
apply list in the installation or update.
For more information, please refer to the “Packaging Software for Installation”
section in the AIX 5L Version 5.2 General Programming Concepts: Writing and
Debugging Programs.
428
Developing and Porting C and C++ Applications on AIX

A
Previous versions of C and
Appendix A.
C++ compiler products
Over the years, IBM has offered a variety of compiler products to perform the
compilation of C and C++ programs on AIX. These C and C++ compilers have
evolved over time and can be tracked by their different version numbers. The C
and C++ compilers introduced new compiler features to take advantage of the
functionality included in new releases of the AIX operating system.
This appendix gives you background information on versions of C and C++
compilers prior to C for AIX Version 6.0 and VisualAge C++ for AIX Version 6.0,
by providing the following sections.
“Compiler product similarities” on page 430
“IBM C compilers” on page 431
“IBM C++ compilers” on page 435
© Copyright IBM Corp. 2000, 2003. All rights reserved.
429

Compiler product similarities
All of the IBM C and C++ compiler products for AIX Version 4 share some similar
characteristics, in particular, the way the products are installed on the system
and the configuration options available when using the products.
Multiple command line drivers
Each compiler product, with the exception of VisualAge C++ Professional for AIX
Version 4.0, has multiple command line driver interfaces available, each causing
a different set of default arguments to be used. For example, the C compiler
products provide commands, such as cc, xlc, c89, cc_r, and so on. These
commands are all links to a single compiler core, which uses a specific set of
options, depending on the name of the command used to invoke it.
In addition to the default invocation commands provided when the compiler is
installed, the system administrator can create new commands, which result in the
compiler being invoked with a customized set of default options. This feature is
controlled by the compiler configuration file, which lists the options to be used for
each invocation command. The exact name of the configuration file differs
between the compiler products, but generally has a name of the form
/etc/comp.cfg, where comp indicates the compiler product that uses the
configuration file.
Finding the compiler drivers
The earlier versions of the compiler products automatically created symbolic
links in /usr/bin for each invocation command supplied by the compiler. For
example, this means that if a user has the directory /usr/bin as part of their PATH
environment variable (which it is, by default), they need only type cc on the
command line to invoke the /usr/bin/cc command.
The later versions of the compiler products are designed to co-exist with earlier
versions, and, as a consequence, they do not automatically create the symbolic
links in /usr/bin when they are installed. This means that a user may have trouble
invoking the compiler on a system that only has a new version compiler product
installed. There are two solutions available in this instance:
When logged in as the root user, invoke the replaceCSET command supplied
with the compiler. This will create appropriate symbolic links in /usr/bin to the
compiler driver programs.
Alter the PATH environment variable to add the directory that contains the
compiler driver programs. For example:
PATH=/usr/vac/bin:$PATH; export PATH
430
Developing and Porting C and C++ Applications on AIX

The second solution should be used if two compilers are installed on a system,
since it allows each user to choose which version of the compiler they wish to
use. If the system only has one compiler installed, it makes sense to use the first
solution. If required, the root user can reverse the action of the replaceCSET
command by using the restoreCSET command, which is also supplied with the
compiler. The exact location of the replaceCSET and restoreCSET commands will
depend on the version of the compiler you are using. Note that the replaceCSET
and restoreCSET tools are removed in Version 6.0 of the C and C++ compiler
products.
Installation directory
The main components of the compiler product are installed on the system in the
/usr file system. The exact directory used depends on the compiler product.
Table A-1 on page 435 shows the location of this directory for C compiler
products. Table A-2 on page 439 shows the location of this directory for C++
compiler products.
IBM C compilers
This section describes the various IBM C compilers for AIX. The details provided
are limited to compatibility issues of the various compilers with the different
versions of the AIX operating system. This information can help in deciding which
version of a compiler product should be used in a given situation, based on the
version of AIX that will be used. The information can also help avoid problems
when upgrading a compiler product to a newer version.
AIX Version 3.2 included a bundled C compiler as part of the operating system.
This compiler was known as XL C Version 1.3. When AIX Version 4.1 was
introduced, the compiler product was unbundled from the operating system and
offered for sale as a separately orderable product. This appendix covers the C
and C++ compiler products prior to C for AIX Version 6.0.
IBM XL C Version 3
The IBM XL C Version 3 product was the first C compiler product from IBM for
AIX Version 4.1. The Version 3 in the name of the product specifies the version
number of the compiler product rather than the version of the AIX operating
system the compiler is compatible with.
Initially released as Version 3.1.0, this compiler evolved over time with the
addition of Program Temporary Fixes (PTFs) to become XL C Version 3.1.4,
Appendix A. Previous versions of C and C++ compiler products
431

which was supported on AIX Version 4.1 and AIX Version 4.2. This compiler was
not supported on AIX Version 3.2 and is not supported on AIX Version 4.3.
C programs written using C Set ++ Version 2 and XL C Version 1.3 on AIX
Version 3.2 are source compatible with XL C Version 3, with some exceptions in
detecting invalid programs or areas where results are undefined. These
exceptions are documented in the product README file.
The compiler product itself is installed in /usr/lpp/xlC, and symbolic links are
created in /usr/bin for the command line driver programs, for example, /usr/bin/cc
and /usr/bin/c89. The default PATH environment variable means that most users
need only type cc on the command line to invoke the /usr/bin/cc driver program,
which, in turn, is a symbolic link to the driver /usr/lpp/xlC/bin/xlC. The compiler
configuration file is /etc/xlC.cfg.
The XL C Version 3 compiler product uses the Net/LS licensing system to control
usage of the product.
This product has been withdrawn from marketing and is no longer available for
purchase. Support for this product has also been discontinued. Current users of
this product are encouraged to upgrade to the IBM C for AIX Version 6.0 compiler
product.
IBM C for AIX Version 4.1
A new version of the C compiler product was introduced with AIX Version 4.3.0.
The compiler product, IBM C for AIX Version 4.1, had a number of new features,
including new optimization routines for improved execution performance, new
inter-procedural analysis tools, precompiled headers for improved compiler
performance, improved memory management, and improved prototyping of
programs. This version of the compiler was supported on AIX Version 4.1.4, AIX
Version 4.2, and AIX Version 4.3.
C programs written using either Version 2 or 3 of IBM C Set ++ for AIX Version 3
of IBM XL C, or the XL C compiler component of AIX Version 3.2 are source
compatible with C for AIX Version 4.1, with some exceptions to detect invalid
programs or areas where results are undefined.
Two important configuration differences introduced with this version of the
compiler are:
1. The compiler product is now installed under /usr/vac rather than /usr/lpp/xlC.
2. The installation process does not create symbolic links to the driver programs
from the /usr/bin directory. This is because the C compiler has been designed
to co-exist on a system that already has the previous version of the C
compiler or a version of the C Set ++ compiler installed.
432
Developing and Porting C and C++ Applications on AIX

If the system does not have another version of the compiler installed, the
symbolic links in the /usr/bin directory can be created by invoking the
/usr/vac/bin/replaceCSET command, which, as the name implies, replaces the
symbolic links to the C Set driver programs.
The compiler product also includes the /usr/vac/bin/restoreCSET command,
which can be used to reverse the actions of the replaceCSET command.
Alternatively, if multiple versions of the compiler exist, or if the user does not want
to create symbolic links in /usr/bin, the setting of the PATH environment variable
can be used to determine which compiler product is used.
For example, setting the PATH environment variable as follows:
PATH=/usr/vac/bin:$PATH; export PATH
will result in the C for AIX Version 4.1 compiler being used when the cc command
is invoked.
The compiler configuration file is /etc/vac.cfg.
The C for AIX Version 4.1 compiler uses the License Use Management (LUM)
licensing system to control usage of the product. Refer to 1.4, “Activating the
compilers” on page 23 for information on configuring the licence system.
This product has been withdrawn from marketing and is no longer available for
purchase. Support for this product has also been discontinued. Current users of
this product are encouraged to upgrade to the IBM C for AIX Version 6.0 compiler
product.
IBM C for AIX Version 4.3
The IBM C for AIX Version 4.3 compiler product was introduced shortly after the
release of AIX Version 4.3.0 and the 64-bit hardware models of the RS/6000
family. This version of the compiler was similar to the IBM C for AIX Version 4.1
compiler, except that it added support for creating and debugging 64-bit
application binaries for use on the 64-bit hardware. This version of the compiler is
installed under /usr/vac, and uses the /etc/vac.cfg configuration file. If C for AIX
Version 4.1 is already installed, installing C for AIX Version 4.3 will overwrite and
upgrade the previous version.
The C for AIX Version 4.3 compiler uses the LUM licensing system to control
usage of the product. Refer to 1.4, “Activating the compilers” on page 23 for
information on configuring the licence system.
This product has been withdrawn from marketing and is no longer available for
purchase. Support for this product has also been discontinued. Current users of
Appendix A. Previous versions of C and C++ compiler products
433

this product are encouraged to upgrade to the IBM C for AIX Version 6 compiler
product.
IBM C for AIX Version 4.4
The IBM C for AIX Version 4.4 compiler product was an improved version of the
previously released C for AIX Version 4.3. The main enhancement was that this
compiler was designed to exploit the RS/6000 Symmetric Multi-Processing
(SMP) architecture. It supported automatic parallellization of a C program as well
as explicit parallellization through a set of directives that enabled the user to
parallelize selected sections of the application program. This version of the C
compiler was supported only by AIX Version 4.2 and 4.3.
C programs written using either Version 2 or Version 3 of IBM C Set ++ for AIX,
the XL C compiler component of AIX Version 3.2, or previous versions of the C
for AIX Version 4.x compilers, are source compatible with C for AIX Version 4.4,
with some exceptions to detect invalid programs or areas where results are
undefined.
The compiler is installed under /usr/vac, and uses the /etc/vac.cfg configuration
file. If a previous version of C for AIX 4.x is installed, installing C for AIX Version
4.4 will overwrite and upgrade the previous version.
The C for AIX Version 4.4 compiler uses the LUM licensing system to control
usage of the product. Refer to 1.4, “Activating the compilers” on page 23 for
information on configuring the licence system.
This product has been withdrawn from marketing and is no longer available for
purchase. Support for this product has also been discontinued. Current users of
this product are encouraged to upgrade to the IBM C for AIX Version 6.0 compiler
product.
IBM C for AIX Version 5.0
The C for AIX Version 5.0 compiler extends the existing symmetric
multi-processing (SMP) support available with C for AIX Version 4.4 by
supporting the OpenMP industry specification. OpenMP provides a model for
parallel programming that allows a program to be portable across shared
memory architectures from different vendors by using a common set of
application program interfaces. The compiler generates highly-optimized code for
all RS/6000 processors and can provide run-time address checking to detect
memory errors.
This compiler is supported only by IBM AIX Version 4.2.1 or later. Also, note that
64-bit applications will run only on AIX Version 4.3 and later when running on
434
Developing and Porting C and C++ Applications on AIX

64-bit hardware. C for AIX Version 5.0.2 adds support for AIX Version 5.1 and
Version 5.2.
C programs written using Version 3 of XL C or Version 4 of IBM C for AIX are
source compatible with IBM C for AIX Version 5.0. C programs written using
either Version 2 or 3 of IBM Set ++ for AIX or the XL C compiler component of
AIX Version 3.2 are source compatible with IBM C for AIX Version 5.0 with
exceptions to detect invalid programs or areas where results are undefined.
This version of the compiler is installed under /usr/vac and uses the /etc/vac.cfg
configuration file. If C for AIX Version 4.x is installed on a system, installing C for
AIX Version 5.0 will overwrite and upgrade the previous version.
The C for AIX Version 5.0 compiler uses the LUM licensing system to control
usage of the product. Refer to 1.4, “Activating the compilers” on page 23 for
information on configuring the licence system.
C compiler summary
Table A-1 summarizes the various versions of IBM C compiler products for AIX.
Table A-1 IBM C compilers for AIX
Compiler
Installation
Configuration
Supported
Licensing
Drivers in
directory
file
AIX levels
method
/usr/bin
XL C Version 3
/usr/lpp/xlC
/etc/xlC.cfg
4.1 and 4.2
Net/LS
Yes
C for AIX Version
/usr/vac
/etc/vac.cfg
4.1.4, 4.2, and
iFOR/LS
No
4.1
4.3
C for AIX Version
/usr/vac
/etc/vac.cfg
4.1.5, 4.2, and
LUM
No
4.3
4.3
C for AIX Version
/usr/vac
/etc/vac.cfg
4.2 and 4.3
LUM
No
4.4
C for AIX Version
/usr/vac
/etc/vac.cfg
4.2 and 4.3
LUM
No
5.0
IBM C++ compilers
This section describes the various IBM C++ Compilers for AIX. The details
provided here are, again, limited to compatibility issues of the various compilers
with the different versions of the AIX operating system. This information can help
decide which C++ Compiler product to use for a particular project, based on the
target version of AIX, and the nature of the C++ source code being compiled.
Appendix A. Previous versions of C and C++ compiler products
435

IBM C Set ++ for AIX Version 3
The IBM C Set ++ for AIX Version 3 product was the first C++ compiler product
from IBM for AIX Version 4.1. The Version 3 in the name of the product specifies
the version of the compiler product rather than the version of the AIX operating
system the compiler is compatible with. The C Set ++ for AIX Version 3 product
is, in effect, an extension of the C for AIX Version 3 compiler. An alternative view
is that the C for AIX Version 3 compiler is a subset of the C Set ++ for AIX
compiler.
This compiler was initially released as Version 3.1.0, and evolved over time with
the addition of Program Temporary Fixes (PTFs) to become C Set ++ for AIX
Version 3.1.4, which was supported on AIX Version 4.1 and AIX Version 4.2. It
was not supported on AIX Version 3.2 and is not supported on AIX Version 4.3.
C++ programs written using C Set ++ Version 2 on AIX Version 3.2 are source
compatible with C Set ++ for AIX Version 3, with some exceptions to detect
invalid programs or areas where results are undefined. These exceptions are
documented in the product README file.
The compiler product itself is installed in /usr/lpp/xlC, and symbolic links are
created in /usr/bin for the command line driver programs, for example,
/usr/bin/xlC. The default PATH environment variable means that most users need
only type xlC on the command line to invoke the /usr/bin/xlC driver program,
which, in turn, is a symbolic link to the driver /usr/lpp/xlC/bin/xlC. The compiler
configuration file is /etc/xlC.cfg.
The C Set ++ for AIX Version 3 compiler product uses the Net/LS licensing
system to control usage of the product.
This product has been withdrawn from marketing and is no longer available for
purchase. Support for this product has also been discontinued. Current users of
this product are encouraged to upgrade to IBM VisualAge C++ for AIX Version
6.0.
IBM C and C++ compilers Version 3.6
The official name of this product is IBM C and C++ compilers for AIX, OS/2®, and
Windows NT. The product is part of a family of related compilers, with versions
available for each of the mentioned platforms. The product is sometimes referred
to as C Set ++ Version 3.6. The AIX Version of the product is the follow on to C
Set ++ Version 3 for AIX.
The product offered a number of facilities to assist in the development of
cross-platform applications, where the same source code is used on multiple
platforms. The product includes a rich set of IBM class libraries, memory
436
Developing and Porting C and C++ Applications on AIX

management routines, graphical debuggers, and resource tools for creating and
compiling resources and converting between platform formats. The C compiler
component of the product can produce either 32-bit or 64-bit executable files
when used on AIX Version 4.3.
The product is supported on AIX Version 4.1.4, Version 4.2, and Version 4.3.
C++ programs written using Version 3 of C Set ++ for AIX and earlier are source
compatible with the C++ compiler of C Set ++ for AIX Version 3.6.
As with the other post Version 3.1 compiler products, the compiler command
drivers are not created in /usr/bin when the product is installed.
The installation directory is /usr/ibmcxx, and the configuration file is
/etc/ibmcxx.cfg.
The product uses the LUM license management system to control usage of the
product. Refer to 1.4, “Activating the compilers” on page 23 for information on
configuring the licence system.
This product has been withdrawn from marketing and is no longer available for
purchase. Support for this product has also been discontinued. Current users of
this product are encouraged to upgrade to IBM VisualAge C++ for AIX Version
6.0.
IBM VisualAge C++ Professional for AIX Version 4
VisualAge C++ Professional for AIX Version 4 is a powerful rapid application
development (RAD) tool for building C and C++ applications. This heterogeneous
RAD environment provides:
Tools, including a graphical debugger
Visual Builder and Data Access Builder
Incremental compiler and linker
A rich set of class libraries
Online help and a powerful full-text search engine
VisualAge C++ Professional for AIX provides a standards-compliant C++
compiler. Its incremental development environment and visual programming
tools improve programmer productivity.
This product features an incremental compiler and linker and, as such, is not
ideally suited for use when working with existing application code that uses
Makefiles or for a development environment that maintains a single source tree
for multiple platforms and uses Makefiles. For this reason, the product includes a
Appendix A. Previous versions of C and C++ compiler products
437

copy of the IBM C Set ++ Version 3.6 compiler for use in a batch compile
environment.
This compiler product runs on IBM AIX Version 4.1.5, Version 4.2, and Version
4.3 for RS/6000.
C++ programs written using Version 3.6 of IBM C and C++ compilers and earlier
are source compatible with the C++ compiler component of VisualAge C++
Professional for AIX Version 4.
Note: As described above, this version of VisualAge features an incremental
compiler. The implications of this for productivity and the code are impressive,
but if the application is moving from a batch environment, do spend time with
the application to adapt to the VisualAge products. For example, makefiles
cannot be processed directly by the incremental compiler.
But, once the migration is done, then the advantages of VisualAge products
are very impressive. This then would reduce the amount of time and memory
required to do each build as well as the time spent on rebuilding when some
changes are made to the source files.
This product has been withdrawn from marketing and is no longer available for
purchase. Support for this product has also been discontinued. Current users of
this product are encouraged to upgrade to IBM VisualAge C++ for AIX Version
6.0.
IBM VisualAge C++ Professional for AIX Version 5
VisualAge C++ Professional for AIX Version 5.0 features a fully incremental
compiler and a batch compiler. The Integrated Development Environment (IDE)
operates with the incremental compiler when used in the AIX Common Desktop
Environment (CDE). The batch compiler is run from the command line and is
suitable for use in a development environment that uses Makefiles. Both
compilers support the latest ANSI/ISO C++ language standard and the latest
version (Version 5) of the IBM Open Class library.
The main differences between Version 4 and Version 5 of this product are:
Version 5 supports multiple codestores in a single project.
Version 5 is a single product featuring both batch and incremental compilers.
The graphical interface of Version 5 has been redesigned with a host of helpful
features. Version 5 has improved optimization techniques and provides the
programmer with effective and efficient ways handling of C++ object code. Also,
438
Developing and Porting C and C++ Applications on AIX

this product allows the developer to carry out performance analysis to determine
the applications usage of system resources.
This product is supported on IBM AIX Version 4.2.1 and later versions for
RS/6000 hardware.
C++ programs written using Version 4 of IBM VisualAge C++ Professional for AIX
are source compatible with the VisualAge C++ Professional for AIX Version 5.
However, programs written using IBM C and C++ Compilers for AIX Version 3.6,
and earlier are not source compatible because the former compilers were based
on the ISO C++ Draft.
The C compiler component of VisualAge C++ Professional for AIX Version 5 is
provided by the IBM C for AIX Version 5 compiler.
C++ compiler summary
Table A-2 summarizes the various IBM C++ compiler products for AIX.
Table A-2 C++ compiler products
Compiler
Installation
Configuration
Supported
Licensing
Drivers in
directory
file
AIX levels
method
/usr/bin
C Set ++ for AIX
/usr/lpp/xlC
/etc/xlC.cfg
4.1, 4.2
Net/LS
Yes
Version 3
IBM C and C ++
/usr/ibmcxx
/etc/ibmcxx.cfg
4.1.4, 4.2, 4.3
LUM
No
compilers
Version 3.6
VisualAge C++
/usr/vacpp
/etc/vacpp.cfg
4.1.5, 4.2, 4.3
LUM
No
Professional for
AIX Version 4
VisualAge C++
/usr/vacpp
/etc/vacpp.cfg
4.2.1, 4.3
LUM
No
Professional for
/etc/vac.cfg
AIX Version 5
Appendix A. Previous versions of C and C++ compiler products
439

440
Developing and Porting C and C++ Applications on AIX

B
Useful information for
Appendix B.
linking and loading on AIX
This appendix gives you the back ground information about the linking and
loading processes on AIX by providing the following sections:
“A brief history of UNIX programming development” on page 442
“Historical view of linking and loading in AIX” on page 443
“Definitions” on page 443
© Copyright IBM Corp. 2000, 2003. All rights reserved.
441

A brief history of UNIX programming development
The traditional UNIX compilation technique historically involved static linking.
With this technique, multiple object files defining global symbols and containing
code and data were combined and written to an executable file with all the
references resolved. The executable was a single self contained file, and often
became quite large. However, since the name and location of all symbols were
resolved at link-time, a program could be executed by simply reading it into
memory and transferring control to its entry point. Static linking, which is still
used in certain circumstances, has the following drawbacks:
If any of the libraries used by the program are updated, the program needs to
be re-linked to take advantage of the updated libraries.
Disk space is wasted because every program on the system contains private
copies of every library functions that it needs.
System memory is wasted because every running process loads into memory
its own private copy of the same library functions.
To allow the libraries to be shared among different programs, the concept of
shared libraries evolved. When a program is linked with a shared library, the
shared library code is not included in the generated program executable file.
Instead, information is saved in the program that allows the library to be found
and loaded when the program is executed. Shared library code is loaded into
global system memory by the first program that needs it and is shared by all
programs that use it.
The advantages of shared libraries are:
Less disk space is used because the shared library code is not included in the
executable programs.
Less memory is used because the shared library code is only loaded once.
The time taken to start an application may be reduced because the shared
library code may already be in memory.
Performance may be improved because fewer page faults will be generated
when the shared library code is already in memory.
When a program is linked with shared libraries, the resulting executable contains
a list of symbols imported by it. The actual code is not included. Therefore, if one
of the shared libraries is updated, programs using that library do not need to be
re-linked, automatically picking up the current version of the library on its next
execution. This makes application service and support easier since new versions
of individual libraries containing patches and fixes can be shipped to customers
without having to rebuild and ship a whole new version of the application.
442
Developing and Porting C and C++ Applications on AIX

When a program uses imported symbols, each of those symbols is associated
with a specific shared library needed by the program. During program execution,
the symbols do not have to be searched, but can be found by simply looking up
their addresses in the specified shared library.
Historical view of linking and loading in AIX
The linker and system loader on AIX are designed so that modules are self
contained entities with well defined sets of imported and exported symbols.
Symbol resolution is performed at link-time, simplifying the work of the system
loader when a module is loaded. The system loader looks up symbols to relocate
references, but does not perform symbol resolution. As a result, a shared module
and its dependents can be pre-relocated in global memory. Once a set of
modules has been pre-relocated, a program using the modules can be loaded
more effectively since symbol lookup and relocation in the pre-relocated modules
does not have to be performed.
However, there has been several circumstances where the traditional AIX linking
mechanism did not satisfy the application programmers’ requirements. For
example, in many new applications, the location of some symbol definitions may
not be known at link-time.
To address this requirement, AIX has been supporting the run-time linking
method, which provide more flexibility in how and when symbol resolution takes
place (see 2.5, “Run-time linking” on page 68).
Definitions
The definitions shown in Table B-1 are very useful in understanding the technical
details of the linking and loading process on AIX.
Table B-1 Definitions of terms regarding linking and loading process on AIX
Term
Description
Object file
A generic term for a file containing executable code, data, relocation information, a symbol
table, and other information. Object files on AIX for POWER processor architecture are
defined by XCOFF (eXtended Common Object File Format). Please refer to the AIX 5L
Version 5.2 Files Reference for further information about the XCOFF format.
CSECT
The atomic unit of relocation as far as link-editing is concerned. A CSECT can contain code
or data. One CSECT can contain references to other CSECTs. These references are
described by relocation entries (RLDs) contained in a section of an object file.
Appendix B. Useful information for linking and loading on AIX
443

Term
Description
Module
The smallest, separately loadable and relocatable unit. A module is an object file containing
a loader section. Modules are loaded implicitly when they are dependents of another
loaded module. A module may have an entry point and may export a set of symbols.
Dependent
A module loaded automatically as part of the process of loading another module. A module
module
is loaded automatically if it is listed in the loader section of another module being loaded.
Executable
A module with an entry point. An executable may have exported symbols.
Linker
The application program that combines multiple object modules into an executable
program. On AIX, the ld command is the system linker. The ld command processes
command line arguments and generates a list of commands, which are piped to another
program called the binder (/usr/ccs/bin/bind). For the purpose of this document, the
terms linker and binder are used interchangeably.
System
A kernel subsystem that creates a process image from an executable file. The loader loads
loader
the executable file into memory, loads (or finds) the program's dependent modules, looks
up imported symbols, and relocates references.
Load-time
The time period during which the system loads a module and its dependents. This includes
the processing required to resolve symbols and relocate the modules. A module can be
loaded with the exec(), dlopen(), or load() system calls. The dlopen() and loadAndInit()
functions are higher level calls that ultimately call load().
Exec-time
Exec-time is load-time for the main program. A new program is loaded when the exec()
function is called.
Run time
Run time starts when the control enters main() and ends when the program exits.
Resolution
The act of associating a reference to a symbol, with the definition of that symbol. Symbol
resolution is performed by the linker at link-time and by the system loader at load-time.
Some special kinds of symbol resolution can occur during run time.
Archives
Archive files are composite objects that usually contain object files. On AIX, archives can
contain shared objects, import files, or other kinds of members. By convention, the name
of an archive usually ends with .a, but the magic number of a file (that is, the first few bytes
of the file) is used to determine whether a file is an archive or not.
Library
A library is a file that contain the definitions of multiple symbols.
Export and
Import files are ASCII files that list symbols to be resolved at load-time, and often times,
Import lists
their defining modules. This information is saved in the loader section of the generated
module, and is used by the system loader when the module is loaded. The import file can
be used in place of a corresponding shared object as an input file to the ld command.
Export files are ASCII files that list global symbols to be made available for another module
to import. The file formats of the import and export file are the same.
444
Developing and Porting C and C++ Applications on AIX

Term
Description
Strip
A system command that reduces the size of an XCOFF module by removing the linking and
debugging information. Object files that have been stripped cannot be used as input to the
linker. A module that has been stripped can still be loaded, since the module's loader
section contains the information needed to load the module.
Static
Executing the ld command so that shared objects are treated as ordinary (non shared)
linking
object files. A shared object that has been stripped cannot be linked statically.
load() call
The load() system call can be used to add a module into a running process, allowing a
program to expand its own capabilities. The unload() system call can be used to remove
object modules from an executing program, which were loaded with the load() routine.
loadbind()
A system call that allows deferred references to be resolved after a module has been
call
loaded.
Dynamic
The technique where imported symbols are looked up or resolved at load-time or run time.
binding
Run-time
The technique where imported symbols are resolved at load-time. When run-time linking is
linking
used, symbols may resolve to alternate definitions that were not available at link-time.
Dynamic
The addition of modules to a running (executing) process. Modules can be loaded with the
loading
load(), dlopen(), or loadAndInit() functions.
Lazy
The ability to defer loading of a dependent module until a function in the dependent module
loading
is first called by the process.
Rebinding
Associating an alternate definition with an imported symbol at run time. By default, the
linker identifies a specific dependent module with each imported symbol. When the
run-time linker is used, the imported symbol can be associated with a symbol in a different
module.
Appendix B. Useful information for linking and loading on AIX
445

446
Developing and Porting C and C++ Applications on AIX

C
Subroutine references for
Appendix C.
shmat and mmap services
This appendix contains the following subroutine references:
“References for shmat services” on page 448
“References for mmap services” on page 461
For detailed information about these routines, please refer to the AIX 5L Version
5.2 Technical Reference: Base Operating System and Extensions.
© Copyright IBM Corp. 2000, 2003. All rights reserved.
447

References for shmat services
This section includes the following subroutine references:
“The ftok() subroutine” on page 448
“The shmat() routine” on page 449
“The shmctl() subroutine” on page 454
“The shmget() routine” on page 457
“The shmdt() subroutine” on page 460
The ftok() subroutine
Generates a standard interprocess communication key.
Library
Standard C library (libc.a)
Syntax
#include <sys/types.h>
#include <sys/ipc.h>
key_t ftok(Path, ID)
char *Path;
int ID;
Description
The ftok() subroutine returns a key, based on the Path and ID parameters, to be
used to obtain interprocess communication identifiers. The ftok() subroutine
returns the same key for linked files if called with the same ID parameter.
Different keys are returned for the same file if different ID parameters are used.
All interprocess communication facilities require you to supply a key to the
msgget(), semget(), and shmget() subroutines in order to obtain interprocess
communication identifiers. The ftok() subroutine provides one method for
creating keys, but other methods are possible. For example, you can use the
project ID as the most significant byte of the key, and use the remaining portion
as a sequence number.
448
Developing and Porting C and C++ Applications on AIX

Attention:
If the Path parameter of the ftok() subroutine names a file that has been
removed while keys still refer to it, the ftok() subroutine returns an error. If
that file is then re-created, the ftok() subroutine will probably return a key
different from the original one.
Each installation should define standards for forming keys. If standards are
not adhered to, unrelated processes may interfere with each other's
operation.
The ftok() subroutine does not guarantee unique key generation. However,
the occurrence of key duplication is very rare and mostly for across file
systems.
Parameters
Path
Specifies the path name of an existing file that is
accessible to the process.
ID
Specifies a character that uniquely identifies a project.
Return values
When successful, the ftok() subroutine returns a key that can be passed to the
msgget(), semget(), or shmget() subroutines.
Error codes
The ftok() subroutine returns the value (key_t)-1 if one or more of the following
are true:
The file named by the Path parameter does not exist.
The file named by the Path parameter is not accessible to the process.
The ID parameter has a value of 0.
The shmat() routine
The shmat() subroutine attaches a shared memory segment or a mapped file to
the current process’s address space.
Library
Standard C library (libc.a)
Syntax
#include <sys/shm.h>
void *shmat(shm_mem_id, shm_mem_addr, shm_mem_flag)
Appendix C. Subroutine references for shmat and mmap services
449

int
shm_mem_id, shm_mem_flag;
const void *shared_mem_addr;
Description
The shmat() subroutine attaches the shared memory segment or mapped file
specified by the SharedMemoryID parameter (returned by the shmget()
subroutine), or file descriptor specified by the SharedMemoryID parameter
(returned by the openx() subroutine) to the address space of the calling process.
The following limits apply to shared memory:
The maximum shared-memory segment size is:
– 256 MB before AIX Version 4.3.1
– 2 GB for AIX Version 4.3.1 through AIX 5L Version 5.1
– 64 GB for 64-bit applications for AIX 5L Version 5.1 and later
The minimum shared-memory segment size is 1 byte.
The maximum number of shared memory IDs is 4096 for operating system
releases before AIX Version 4.3.2 and 131072 for AIX Version 4.3.2 and
following.
Note: The following applies to AIX Version 4.2.1 and later releases for 32-bit
processes only.
An extended shmat capability is available. If an environment variable
EXTSHM=ON is defined, then processes executing in that environment will be
able to create and attach more than eleven shared memory segments.
The segments can be of size, from 1 byte to 2 GB, although for segments larger
than 256 MB in size the environment variable EXTSHM=ON is ignored. The
process can attach these segments into the address space for the size of the
segment. Another segment could be attached at the end of the first one in the
same 256 MB segment region. The address at which a process can attach is at
page boundaries (a multiple of SHMLBA_EXTSHM bytes). For segments larger
than 256 MB in size, the address at which a process can attach is at 256 MB
boundaries, which is a multiple of SHMLBA bytes.
The segments can be of size from 1 byte to 256 MB. The process can attach
these segments into the address space for the size of the segment. Another
segment could be attached at the end of the first one in the same 256 MB
segment region. The address at which a process can attach will be at page
boundaries (a multiple of SHMLBA_EXTSHM bytes).
450
Developing and Porting C and C++ Applications on AIX

The maximum address space available for shared memory with or without the
environment variable and for memory mapping is 2.75 GB. An additional
segment register "0xE" is available so that the address space is from
0x30000000 to 0xE0000000. However, a 256 MB region starting from
0xD0000000 will be used by the shared libraries and is therefore unavailable for
shared memory regions or mmapped regions.
There are some restrictions on the use of the extended shmat feature. These
shared memory regions cannot be used as I/O buffers where the unpinning of the
buffer occurs in an interrupt handler. The restrictions on the use are the same as
that of mmap buffers.
The smaller region sizes are not supported for mapping files. Regardless of
whether EXTSHM=ON or not, mapping a file will consume at least 256 MB of
address space.
SHM_SIZE shmctl() is not supported for segments created with EXTSHM=ON.
A segment created with EXTSHM=ON can be attached by a process without
EXTSHM=ON. This will consume a 256 MB area of the address space
irrespective of the size of the shared memory region.
A segment created without EXTSHM=ON can be attached by a process with
EXTSHM=ON. This will consume a 256 MB area of the address space
irrespective of the size of the shared memory region.
The environment variable provides the option of executing an application either
with the additional functionality of attaching more than 11 segments when
EXTSHM=ON, or the higher-performance access to 11 or fewer segments when
the environment variable is not set.
Parameter
shm_mem_id
Specifies an identifier for the shared memory segment.
shm_mem_addr
Identifies the segment or file attached at the address
specified by the shm_mem_addr parameter, as follows:
If the shm_mem_addr parameter is not equal to 0, and
the SHM_RND flag is set in the shm_mem_flag
parameter, the segment or file is attached at the next
lower segment boundary. This address is given by
(shm_mem_addr - (shm_mem_addr modulo
SHMLBA_EXTSHM, if environment variable
EXTSHM=ON or SHMLBA if not). SHMLBA specifies the
low boundary address multiple of a segment.
Appendix C. Subroutine references for shmat and mmap services
451

If the shm_mem_addr parameter is not equal to 0 and the
SHM_RND flag is not set in the shm_mem_flag
parameter, the segment or file is attached at the address
given by the shm_mem_addr parameter. If this address
does not point to a SHMLBA_EXTSHM boundary if the
environment variable EXTSHM=ON or SHMLBA
boundary if not, the shmat() subroutine returns the value
-1 and sets the errno global variable to the EINVAL error
code. SHMLBA specifies the low boundary address
multiple of a segment.
shm_mem_flag
Specify several options. Its value is either 0 or is
constructed by logically ORing one or more of the values
in Table C-1.
Table C-1 Values for the third parameter of shmat()
Value
Description
SHM_MAP
Maps a file onto the address space instead of a shared memory
segment. The shm_mem_id parameter must specify an open file
descriptor in this case.
SHM_RDONLY
Specifies read-only mode instead of the default read-write mode.
SHM_RND
Rounds the address given by the shm_mem_addr parameter to the
next lower segment boundary, if necessary.
The shmat() subroutine makes a shared memory segment addressable by the
current process. The segment is attached for reading if the SHM_RDONLY flag is
set and the current process has read permission. If the SHM_RDONLY flag is not
set and the current process has both read and write permission, it is attached for
reading and writing.
If the SHM_MAP flag is set, file mapping takes place. In this case, the shmat
subroutine maps the file open on the file descriptor specified by the
SharedMemoryID onto a segment. The file must be a regular file. The segment is
then mapped into the address space of the process. A file of any size can be
mapped if there is enough space in the user address space.
When file mapping is requested, the SharedMemoryFlag parameter specifies
how the file should be mapped. If the SHM_RDONLY flag is set, the file is
mapped read-only. To map read-write, the file must have been opened for writing.
All processes that map the same file read-only or read-write map to the same
segment. This segment remains mapped until the last process mapping the file
closes it.
452
Developing and Porting C and C++ Applications on AIX

A mapped file opened with the O_DEFER update has a deferred update. That is,
changes to the shared segment do not affect the contents of the file resident in
the file system until an fsync subroutine is issued to the file descriptor for which
the mapping was requested. Setting the SHM_COPY flag changes the file to the
deferred state. The file remains in this state until all processes close it. The
SHM_COPY flag is provided only for compatibility with Version 2 of the operating
system. New programs should use the O_DEFER open flag.
A file descriptor can be used to map the corresponding file only once. To map a
file several times requires multiple file descriptors.
When a file is mapped onto a segment, the file is referenced by accessing the
segment. The memory paging system automatically takes care of the physical
I/O. References beyond the end of the file cause the file to be extended in
page-sized increments. The file cannot be extended beyond the next segment
boundary.
Return values
When successful, the segment start address of the attached shared memory
segment or mapped file is returned. Otherwise, the shared memory segment is
not attached, the errno global variable is set to indicate the error, and a value of
-1 is returned.
Error codes
The shmat() subroutine is unsuccessful and the shared memory segment or
mapped file is not attached if one or more of the following are true:
EACCES
The calling process is denied permission for the specified
operation.
EAGAIN
The file to be mapped has enforced locking enabled, and
the file is currently locked.
EBADF
A file descriptor to map does not refer to an open regular
file.
EEXIST
The file to be mapped has already been mapped.
EINVAL
The SHM_RDONLY and SHM_COPY flags are both set.
EINVAL
The shm_mem_id parameter is not a valid shared
memory identifier.
EINVAL
The shm_mem_addr parameter is not equal to 0, and the
value of (shm_mem_addr - (shm_mem_addr modulo
SHMLBA_EXTSHM if the environment variable
EXTSHM=ON or SHMLBA if not) points outside the
address space of the process.
Appendix C. Subroutine references for shmat and mmap services
453

EINVAL
The shm_mem_addr parameter is not equal to 0, the
SHM_RND flag is not set in the SharedMemoryFlag
parameter, and the shm_mem_addr parameter points to a
location outside of the address space of the process.
EMFILE
The number of shared memory segments attached to the
calling process exceeds the system-imposed limit.
ENOMEM
The available data space in memory is not large enough
to hold the shared memory segment. ENOMEM is always
returned if a 32-bit process tries to attach a shared
memory segment larger than 2 GB.
ENOMEM
The available data space in memory is not large enough
to hold the mapped file data structure.
ENOMEM
The requested address and length crosses a segment
boundary. This is not supported when the environment
variable EXTSHM=ON.
The shmctl() subroutine
Controls shared memory operations.
Library
Standard C library (libc.a)
Syntax
#include <sys/shm.h>
int shmctl (SharedMemoryID, Command, Buffer)
int SharedMemoryID, Command;
struct shmid_ds * Buffer;
Description
The shmctl() subroutine performs a variety of shared-memory control operations
as specified by the Command parameter.
The following limits apply to shared memory:
The maximum shared-memory segment size is:
– 256 MB before AIX Version 4.3.1
– 2 GB for AIX Version 4.3.1 through AIX 5L Version 5.1
– 64 GB for 64-bit applications for AIX 5L Version 5.1 and later
The minimum shared-memory segment size is 1 byte.
454
Developing and Porting C and C++ Applications on AIX

The maximum number of shared memory IDs is 4096 for operating system
releases before AIX Version 4.3.2 and 131072 for AIX Version 4.3.2 and
following.
Parameters
SharedMemoryID
Specifies an identifier returned by the shmget subroutine.
Buffer
Indicates a pointer to the shmid_ds structure. The
shmid_ds structure is defined in the sys/shm.h file.
Command
The commands listed in Table C-2 are available.
Table C-2 Values for the third parameter of shmctl()
Value
Description
IPC_STAT
Obtains status information about the shared memory segment
identified by the SharedMemoryID parameter. This information is
stored in the area pointed to by the Buffer parameter. The calling
process must have read permission to run this command.
IPC_ SET
Sets the user and group IDs of the owner as well as the access
permissions for the shared memory segment identified by the
SharedMemoryID parameter. This command sets the following fields:
shm_perm.uid /* owning user ID */
shm_perm.gid /* owning group ID */
shm_perm.mode /* permission bits only */
You must have an effective user ID equal to root or to the value of the
shm_perm.cuid or shm_perm.uid field in the shmid_ds data structure
identified by the SharedMemoryID parameter.
IPC_RMID
Removes the shared memory identifier specified by the
SharedMemoryID parameter from the system and erases the shared
memory segment and data structure associated with it. This command
is only executed by a process that has an effective user ID equal either
to that of superuser or to the value of the shm_perm.uid or
shm_perm.cuid field in the data structure identified by the
SharedMemoryID parameter.
Appendix C. Subroutine references for shmat and mmap services
455

Value
Description
SHM_SIZE
Sets the size of the shared memory segment to the value specified by
the shm_segsz field of the structure specified by the Buffer parameter.
This value can be larger or smaller than the current size. The limit is the
maximum shared-memory segment size. This command is only
executed by a process that has an effective user ID equal either to that
of a process with the appropriate privileges or to the value of the
shm_perm.uid or shm_perm.cuid field in the data structure identified by
the SharedMemoryID parameter. This command is not supported for
regions created with the environment variable EXTSHM=ON. This
results in a return value of -1 with errno set to EINVAL. Attempting to
use the SHM_SIZE on a shared memory region larger than 256 MB or
attempting to increase the size of a shared memory region larger than
256 MB results in a return value of -1 with errno set to EINVAL.
Return values
When completed successfully, the shmctl() subroutine returns a value of 0.
Otherwise, it returns a value of -1 and the errno global variable is set to indicate
the error.
Error codes
The shmctl() subroutine is unsuccessful if one or more of the following are true:
EACCES
The Command parameter is equal to the IPC_STAT value
and read permission is denied to the calling process.
EFAULT
The Buffer parameter points to a location outside the
allocated address space of the process.
EINVAL
The SharedMemoryID parameter is not a valid shared
memory identifier.
EINVAL
The Command parameter is not a valid command.
EINVAL
The Command parameter is equal to the SHM_SIZE
value and the value of the shm_segsz field of the
structure specified by the Buffer parameter is not valid.
EINVAL
The Command parameter is equal to the SHM_SIZE
value and the shared memory region was created with the
environment variable EXTSHM=ON.
ENOMEM
The Command parameter is equal to the SHM_SIZE
value, and the attempt to change the segment size is
unsuccessful because the system does not have enough
memory.
456
Developing and Porting C and C++ Applications on AIX

EOVERFLOW
The Command parameter is IPC_STAT and the size of the
shared memory region is greater than or equal to 4 GB.
This only happens with 32-bit programs.
EPERM
The Command parameter is equal to the IPC_RMID or
SHM_SIZE value, and the effective user ID of the calling
process is not equal to the value of the shm_perm.uid or
shm_perm.cuid field in the data structure identified by the
SharedMemoryID parameter. The effective user ID of the
calling process is not the root user ID.
The shmget() routine
The shmget() subroutine returns the shared memory identifier associated with
the specified key parameter.
Library
Standard C library (libc.a)
Syntax
#include <sys/shm.h>
int shmget(Key, Size, Shm_mem_flag)
key_t
Key;
size_t
Size;
int
Shm_mem_flag;
Description
The shmget() subroutine returns the shared memory identifier associated with
the specified Key parameter.
The following limits apply to shared memory:
The maximum shared-memory segment size is:
– 256 MB before AIX Version 4.3.1
– 2 GB for AIX Version 4.3.1 through AIX 5L Version 5.1
– 64 GB for 64-bit applications for AIX 5L Version 5.1 and later
The minimum shared-memory segment size is 1 byte.
The maximum number of shared memory IDs is 4096 for operating system
releases before AIX Version 4.3.2 and 131072 for AIX Version 4.3.2 and
following.
Appendix C. Subroutine references for shmat and mmap services
457

Parameters
Key
Specify either the IPC_PRIVATE value or an IPC key
constructed by ftok() routine.
Size
Specify the number of bytes of shared memory required.
Shm_mem_flag
Constructed by logically ORing one or more of the values
in Table C-3.
Table C-3 Values for the third parameter of shmget()
Value
Description
IPC_CREATE
Creates the data structure if it does not already exist.
IPC_EXCL
Cause the shmget() subroutine to be unsuccessful if the
IPC_CREATE flag is also set, and the data structure already exists.
SHM_LGPAGE
Attempts to create the region so it can be mapped through
hardware-supported, large-page mechanisms, if enabled. This is
purely advisory. For the system to consider this flag, it must be used
in conjunction with the SHM_PIN flag and enabled with the vmtune
command (-L to reserve memory for the region (which requires a
reboot) and -S to enable SHM_PIN). To successfully get
large-pages, the user requesting large-page shared memory must
have CAP_BYPASS_RAC_VMM capability. This has no effect on
shared memory regions created with the EXTSHM=ON environment
variable.
SHM_PIN
Attempts to pin the shared memory region if enabled. This is purely
advisory. For the system to consider this flag, the system must be
enabled with the vmtune command. This has no effect on shared
memory regions created with the EXTSHM=ON environment
variable.
S_IRUSR
Permits the process that owns the data structure to read it.
S_IWUSR
Permits the process that owns the data structure to modify it.
S_IRGRP
Permits the group associated with the data structure to read it.
S_IWGRP
Permits the group associated with the data structure to modify it.
S_IROTH
Permits others to read the data structure.
S_IWOTH
Permits others to modify the data structure.
Values that begin with S_I prefix are defined in the /usr/include/sys/mode.h file and are
a subset of the access permissions that apply to files.
458
Developing and Porting C and C++ Applications on AIX

A shared memory identifier, its associated data structure, and a shared memory
segment equal in number of bytes to the value of the Size parameter are created
for the Key parameter if one of the following is true:
The Key parameter is equal to the IPC_PRIVATE value.
The Key parameter does not already have a shared memory identifier
associated with it, and the IPC_CREAT flag is set in the SharedMemoryFlag
parameter.
Upon creation, the data structure associated with the new shared memory
identifier is initialized as follows:
The shm_perm.cuid and shm_perm.uid fields are set to the effective user ID
of the calling process.
The shm_perm.cgid and shm_perm.gid fields are set to the effective group ID
of the calling process.
The low-order 9 bits of the shm_perm.mode field are set to the low-order 9
bits of the SharedMemoryFlag parameter.
The shm_segsz field is set to the value of the Size parameter.
The shm_lpid, shm_nattch, shm_atime, and shm_dtime fields are set to 0.
The shm_ctime field is set to the current time.
Once created, a shared memory segment is deleted only when the system
reboots, by issuing the ipcrm command, or by using the following shmctl
subroutine:
if (shmctl(id, IPC_RMID, 0) == -1)
perror("error in closing segment"), exit (1);
Return values
Upon successful completion, a shared memory identifier is returned. Otherwise,
the shmget() subroutine returns a value of -1 and sets the errno global variable to
indicate the error.
Error codes
The shmget() subroutine is unsuccessful if one or more of the following are true:
EACCES
A shared memory identifier exists for the Key parameter,
but operation permission, as specified by the low-order 9
bits of the SharedMemoryFlag parameter, is not granted.
EEXIST
A shared memory identifier exists for the Key parameter,
and both the IPC_CREAT and IPC_EXCL flags are set in
the SharedMemoryFlag parameter.
Appendix C. Subroutine references for shmat and mmap services
459

EINVAL
A shared memory identifier does not exist and the Size
parameter is less than the system-imposed minimum or
greater than the system-imposed maximum.
EINVAL
A shared memory identifier exists for the Key parameter,
but the size of the segment associated with it is less than
the Size parameter, and the Size parameter is not equal
to 0.
ENOENT
A shared memory identifier does not exist for the Key
parameter, and the IPC_CREAT flag is not set in the
SharedMemoryFlag parameter.
ENOMEM
A shared memory identifier and associated shared
memory segment are to be created, but the amount of
available physical memory is not sufficient to meet the
request.
ENOSPC
A shared memory identifier will be created, but the
system-imposed maximum of shared memory identifiers
allowed will be exceeded.
The shmdt() subroutine
Detaches a shared memory segment.
Library
Standard C library (libc.a)
Syntax
#include <sys/shm.h>
int shmdt (SharedMemoryAddress)
const void * SharedMemoryAddress;
Description
The shmdt() subroutine detaches, from the data segment of the calling process,
the shared memory segment located at the address specified by the
SharedMemoryAddress parameter.
Mapped file segments are automatically detached when the mapped file is
closed. However, you can use the shmdt() subroutine to explicitly release the
segment register used to map a file. Shared memory segments must be explicitly
detached with the shmdt() subroutine.
If the file was mapped for writing, the shmdt() subroutine updates the mtime and
ctime time stamps.
460
Developing and Porting C and C++ Applications on AIX

The following limits apply to shared memory:
The maximum shared-memory segment size is:
– 256 MB before AIX Version 4.3.1
– 2 GB for AIX Version 4.3.1 through AIX 5L Version 5.1
– 64 GB for 64-bit applications for AIX 5L Version 5.1 and later
The minimum shared-memory segment size is 1 byte.
The maximum number of shared memory IDs is 4096 for operating system
releases before AIX Version 4.3.2 and 131072 for AIX Version 4.3.2 and
following.
Parameters
SharedMemoryAddress
Specifies the data segment start address of a
shared memory segment.
Return values
When successful, the shmdt subroutine returns a value of 0. Otherwise, the
shared memory segment at the address specified by the SharedMemoryAddress
parameter is not detached, a value of 1 is returned, and the errno global variable
is set to indicate the error.
Error codes
The shmdt() subroutine is unsuccessful if the following condition is true:
EINVAL
The value of the SharedMemoryAddress parameter is not the
data-segment start address of a shared memory segment.
References for mmap services
This section includes the following subroutine references:
“The mmap() subroutine” on page 461
“The mprotect() subroutine” on page 467
“The msync() subroutine” on page 469
“The munmap() subroutine” on page 471
The mmap() subroutine
Maps a file-system object into virtual memory.
Appendix C. Subroutine references for shmat and mmap services
461

Library
Standard C library (libc.a)
Syntax
#include <sys/types.h>
#include <sys/mman.h>
void *mmap(addr, len, prot, flags, filedes, offset)
void *
addr;
size_t
len;
int
prot, flags, filedes;
off_t
offset;
Description
The mmap() subroutine creates a new mapped file or anonymous memory
region by establishing a mapping between a process-address space and a
file-system object. Care needs to be taken when using the mmap() subroutine if
the program attempts to map itself. If the page containing executing instructions
is currently referenced as data through an mmap mapping, the program will
hang. Use the -H4096 binder option, and that will put the executable text on page
boundaries. Then reset the file that contains the executable material, and view
via an mmap mapping.
A region created by the mmap() subroutine cannot be used as the buffer for read
or write operations that involve a device. Similarly, an mmap region cannot be
used as the buffer for operations that require either a pin or xmattach operation
on the buffer.
Modifications to a file-system object are seen consistently, whether accessed
from a mapped file region or from the read or write subroutine.
Child processes inherit all mapped regions from the parent process when the
fork subroutine is called. The child process also inherits the same sharing and
protection attributes for these mapped regions. A successful call to any exec
subroutine will unmap all mapped regions created with the mmap() subroutine.
The mmap64() subroutine is identical to the mmap subroutine() except that the
starting offset for the file mapping is specified as a 64-bit value. This permits file
mappings which start beyond OFF_MAX.
In the large file enabled programming environment, mmap() is redefined to be
mmap64().
If the application has requested SPEC1170 compliant behavior, then the
st_atime field of the mapped file is marked for update upon successful
completion of the mmap() call.
462
Developing and Porting C and C++ Applications on AIX

If the application has requested SPEC1170 compliant behavior, then the
st_ctime and st_mtime fields of a file that is mapped with MAP_SHARED and
PROT_WRITE are marked for update at the next call to msync() subroutine or
munmap() subroutine if the file has been modified.
Note: A file-system object should not be simultaneously mapped using both
the mmap() and shmat() subroutines. Unexpected results may occur when
references are made beyond the end of the object.
Parameters
addr
Specifies the starting address of the memory region to be
mapped. When the MAP_FIXED flag is specified, this
address must be a multiple of the page size returned by
the sysconf() subroutine using the _SC_PAGE_SIZE
value. A region is never placed at address zero, or at an
address where it would overlap an existing region.
len
Specifies the length, in bytes, of the memory region to be
mapped. The system performs mapping operations over
whole page only. If the len parameter is not a multiple of
the page size, the system will include, in any mapping
operation, the address range between the end of the
region and the end of the page containing the end of the
region.
prot
Specifies the access permission for the mapped region.
The /usr/include/sys/mman.h file defines the following
access options:
PROT_READ
Region can be read.
PROT_WRITE
Region can be written.
PROT_EXEC
Region can be executed.
PROT_NONE
Region cannot be accessed.
The prot parameter can be the PROT_NONE flag, or any
combination of the other values logically ORed together. If
the PROT_NONE flag is not specified, access
permissions may be granted to the region in addition to
those explicitly requested. However, write access will not
be granted unless the PROT_WRITE flag is specified.
Appendix C. Subroutine references for shmat and mmap services
463

Note: The operating system generates a SIGSEGV signal if a program
attempts an access that exceeds the access permission given to a memory
region. For example, if the PROT_WRITE flag is not specified and a program
attempts a write access, a SIGSEGV signal results.
If the region is a mapped file that was mapped with the
MAP_SHARED flag, the mmap() subroutine grants read
or execute access permission only if the file descriptor
used to map the file was opened for reading. It grants
write access permission only if the file descriptor was
opened for writing.
If the region is a mapped file that was mapped with the
MAP_PRIVATE flag, the mmap() subroutine grants read,
write, or execute access permission only if the file
descriptor used to map the file was opened for reading. If
the region is an anonymous memory region, the mmap
subroutine grants all requested access permissions.
filedes
Specifies the file descriptor of the file-system object to be
mapped. If the MAP_ANONYMOUS flag is set, the filedes
parameter must be -1. After the successful completion of
the mmap() subroutine, the file specified by the filedes
parameter may be closed without executing the mapped
region or the contents of the mapped file. Each mapped
region creates a file reference, similar to an open file
descriptor, which prevents the file data from been
deallocated.
Note: The mmap() subroutine supports the mapping of regular files only. An
mmap() call that specifies a file descriptor for a special file fails, returning the
ENODEV error. An example of a file descriptor for a special file is one that
might be used for mapping either I/O or device memory.
offset
Specifies the file byte offset at which the mapping starts.
This offset must be a multiple of the page size returned by
the sysconf() routine using the _SC_PAGE_SIZE value.
flags
Specify attributes of the mapped region. Values for the
flags parameter are constructed by a bitwise-inclusive
ORing of values listed in Table C-4 on page 465, which
lists symbolic names defined in the
/usr/include/sys/mman.h file.
464
Developing and Porting C and C++ Applications on AIX

Table C-4 Values for the sixth parameter of mmap()
Value
Description
MAP_FILE
Specifies the creation of a new mapped file region by mapping
the file associated with the filedes file descriptor. The mapped
region can extend beyond the end of the file, both at the time
when the mmap() subroutine is called and while the mapping
persists. This situation could occur if a file with no contents was
created just before the call to the mmap() subroutine, or if a file
was later truncated. However, references to whole pages
following the end of the file result in the delivery of a SIGBUS
signal. Only one of the MAP_FILE and MAP_ANONYMOUS
flags must be specified with the mmap() subroutine.
MAP_ANONYMOUS
Specifies the creation of a new, anonymous memory region
that is initialized to all zeros. This memory region can be shared
only with the descendants of the current process. When using
this flag, the filedes parameter must be -1. Only one of the
MAP_FILE and MAP_ANONYMOUS flags must be specified
with the mmap() subroutine.
MAP_VARIABLE
Specifies that the system select an address for the new
memory region if the new memory region cannot be mapped at
the address specified by the addr parameter, or if the addr
parameter is NULL. Only one of the MAP_VARIABLE and
MAP_FIXED flags must be specified with the mmap()
subroutine.
MAP_FIXED
Specifies that the mapped region be placed exactly at the
address specified by the addr parameter. If the application has
requested SPEC1170 complaint behavior and the mmap()
request is successful, the mapping replaces any previous
mappings for the process’s pages in the specified range. If the
application has not requested SEPC170 compliant behavior
and a previous mapping exists in the range, then the request
fails. Only one of the MAP_VARIABLE and MAP_FIXED flags
must be specified with the mmap() subroutine.
Appendix C. Subroutine references for shmat and mmap services
465

Value
Description
MAP_SHARED
When the MAP_SHARED flag is set, modifications to the
mapped memory region will be visible to other processes that
have mapped the same region using this flag. If the region is a
mapped file region, modifications to the region will be written to
the file.
You can specify only one of the MAP_SHARED or
MAP_PRIVATE flags with the mmap() subroutine.
MAP_PRIVATE is the default setting when neither flag is
specified unless you request SPEC1170compliant behavior. In
this case, you must choose either MAP_SHARED or
MAP_PRIVATE.
MAP_PRIVATE
When the MAP_PRIVATE flag is set, modifications to the
mapped region by the calling process are not visible to other
processes that have mapped the same region. If the region is a
mapped file region, modifications to the region are not written
to the file.
If this flag is specified, the initial write reference to an object
page creates a private copy of that page and redirects the
mapping to the copy. Until then, modifications to the page by
processes that have mapped the same region with the
MAP_SHARED flag are visible.
You can specify only one of the MAP_SHARED or
MAP_PRIVATE flags with the mmap() subroutine.
MAP_PRIVATE is the default setting when neither flag is
specified unless you request SPEC1170compliant behavior. In
this case, you must choose either MAP_SHARED or
MAP_PRIVATE.
Return values
If successful, the mmap() subroutine returns the address at which the mapping
was placed. Otherwise, it returns -1 and sets the errno global variable to indicate
the error.
Error codes
Under the following conditions, the mmap subroutine fails and sets the errno
global variable to:
EACCES
The file referred to by the fildes parameter is not open for
read access, or the file is not open for write access and
the PROT_WRITE flag was specified for a
MAP_SHARED mapping operation, or the file to be
466
Developing and Porting C and C++ Applications on AIX

mapped has enforced locking enabled and the file is
currently locked.
EBADF
The fildes parameter is not a valid file descriptor, or the
MAP_ANONYMOUS flag was set and the fildes
parameter is not -1.
EFBIG
The mapping requested extends beyond the maximum file
size associated with fildes.
EINVAL
The flags or prot parameter is invalid, or the addr
parameter or off parameter is not a multiple of the page
size returned by the sysconf subroutine using the
_SC_PAGE_SIZE value for the Name parameter.
EINVAL
The application has requested SPEC1170 compliant
behavior and the value of flags is invalid (neither
MAP_PRIVATE nor MAP_SHARED is set).
EMFILE
The application has requested SPEC1170 compliant
behavior and the number of mapped regions would
exceed an implementation-dependent limit (per process
or per system).
ENODEV
The fildes parameter refers to an object that cannot be
mapped, such as a terminal.
ENOMEM
There is not enough address space to map len bytes or
the application has not requested Single UNIX
Specification Version 2 compliant behavior, and the
MAP_FIXED flag was set and part of the address-space
range (addr, addr+len) is already allocated.
ENXIO
The addresses specified by the range (off, off+len) are
invalid for the fildes parameter.
EOVERFLOW
The mapping requested extends beyond the offset
maximum for the file description associated with fildes.
The mprotect() subroutine
Modifies access protections for memory mapping.
Library
Standard C library (libc.a)
Appendix C. Subroutine references for shmat and mmap services
467

Syntax
#include <sys/types.h>
#include <sys/mman.h>
int mprotect ( addr, len, prot)
void *addr;
size_t len;
int prot;
Description
The mprotect() subroutine modifies the access protection of a mapped file region
or anonymous memory region created by the mmap() subroutine. The behavior
of this function is unspecified if the mapping was not established by a call to the
mmap() subroutine.
Parameters
addr
Specifies the address of the region to be modified. Must
be a multiple of the page size returned by the sysconf()
subroutine using the _SC_PAGE_SIZE value for the
Name parameter.
len
Specifies the length, in bytes, of the region to be modified.
If the len parameter is not a multiple of the page size
returned by the sysconf subroutine using the
_SC_PAGE_SIZE value for the Name parameter, the
length of the region will be rounded off to the next multiple
of the page size.
prot
Specifies the new access permissions for the mapped
region. Legitimate values for the prot parameter are the
same as those permitted for the mmap (mmap or
mmap64()) subroutine, as follows:
PROT_READ
Region can be read.
PROT_WRITE
Region can be written.
PROT_EXEC
Region can be executed.
PROT_NONE
Region cannot be accessed.
Return values
When successful, the mprotect() subroutine returns 0. Otherwise, it returns -1
and sets the errno global variable to indicate the error.
468
Developing and Porting C and C++ Applications on AIX

Note: The return value for mprotect() is 0 if it fails, because the region given
was not created by mmap() unless XPG 1170 behavior is requested by setting
the environment variable XPG_SUS_ENV to ON.
Error codes
Attention: If the mprotect() subroutine is unsuccessful because of a condition
other than that specified by the EINVAL error code, the access protection for
some pages in the (addr, addr + len) range may have been changed.
If the mprotect() subroutine is unsuccessful, the errno global variable may be set
to one of the following values:
EACCES
The prot parameter specifies a protection that conflicts
with the access permission set for the underlying file.
EINVAL
The prot parameter is not valid, or the addr parameter is
not a multiple of the page size as returned by the
sysconf() subroutine using the _SC_PAGE_SIZE value for
the Name parameter.
ENOMEM
The application has requested Single UNIX Specification
Version 2 compliant behavior, and addresses in the range
are invalid for the address space of the process or specify
one or more pages which are not mapped.
The msync() subroutine
Synchronizes a mapped file.
Library
Standard C library (libc.a).
Syntax
#include <sys/types.h>
#include <sys/mman.h>
int msync ( addr, len, flags)
void *addr;
size_t len;
int flags;
Appendix C. Subroutine references for shmat and mmap services
469

Description
The msync() subroutine controls the caching operations of a mapped file region.
Use the msync() subroutine to transfer modified pages in the region to the
underlying file storage device.
If the application has requested Single UNIX Specification Version 2 compliant
behavior then the st_ctime and st_mtime fields of the mapped file are marked for
update upon successful completion of the msync() subroutine call if the file has
been modified.
Parameters
addr
Specifies the address of the region to be synchronized.
Must be a multiple of the page size returned by the
sysconf() subroutine using the _SC_PAGE_SIZE value for
the Name parameter.
len
Specifies the length, in bytes, of the region to be
synchronized. If the len parameter is not a multiple of the
page size returned by the sysconf() subroutine using the
_SC_PAGE_SIZE value for the Name parameter, the
length of the region is rounded up to the next multiple of
the page size.
flags
Specifies one or more of the symbolic constants listed in
Table C-5 that determine the way caching operations are
performed.
Table C-5 The third parameter of msync()
Value
Description
MS_SYNC
Specifies synchronous cache flush. The msync subroutine does
not return until the system completes all I/O operations.
This flag is invalid when the MAP_PRIVATE flag is used with the
mmap() subroutine. MAP_PRIVATE is the default privacy setting.
When the MS_SYNC and MAP_PRIVATE flags both are used, the
msync() subroutine returns an errno value of EINVAL.
MS_ASYNC
Specifies an asynchronous cache flush. The msync() subroutine
returns after the system schedules all I/O operations.
This flag is invalid when the MAP_PRIVATE flag is used with the
mmap() subroutine. MAP_PRIVATE is the default privacy setting.
When the MS_SYNC and MAP_PRIVATE flags both are used, the
msync() subroutine returns an errno value of EINVAL.
MS_INVALIDATE
Specifies that the msync() subroutine invalidates all cached copies
of the pages. New copies of the pages must then be obtained from
the file system the next time they are referenced.
470
Developing and Porting C and C++ Applications on AIX

Return values
When successful, the msync() subroutine returns 0. Otherwise, it returns -1 and
sets the errno global variable to indicate the error.
Error codes
If the msync() subroutine is unsuccessful, the errno global variable is set to one
of the following values:
EIO
An I/O error occurred while reading from or writing to the
file system.
ENOMEM
The range specified by (addr, addr + len) is invalid for a
process' address space, or the range specifies one or
more unmapped pages.
EINVAL
The addr argument is not a multiple of the page size, as
returned by the sysconf() subroutine using the
_SC_PAGE_SIZE value for the Name parameter, or the
flags parameter is invalid. The address of the region is
within the process' inheritable address space.
The munmap() subroutine
Unmaps a mapped region.
Library
Standard C library (libc.a)
Syntax
#include <sys/types.h>
#include <sys/mman.h>
int munmap ( addr, len)
void *addr;
size_t len;
Description
The munmap() subroutine unmaps a mapped file region or anonymous memory
region. The munmap() subroutine unmaps regions created from calls to the
mmap() subroutine only.
If an address lies in a region that is unmapped by the munmap subroutine and
that region is not subsequently mapped again, any reference to that address will
result in the delivery of a SIGSEGV signal to the process.
Appendix C. Subroutine references for shmat and mmap services
471

Parameters
addr
Specifies the address of the region to be unmapped. Must
be a multiple of the page size returned by the sysconf()
subroutine using the _SC_PAGE_SIZE value for the
Name parameter.
len
Specifies the length, in bytes, of the region to be
unmapped. If the len parameter is not a multiple of the
page size returned by the sysconf() subroutine using the
_SC_PAGE_SIZE value for the Name parameter, the
length of the region is rounded up to the next multiple of
the page size.
Return values
When successful, the munmap() subroutine returns 0. Otherwise, it returns -1
and sets the errno global variable to indicate the error.
Error codes
If the munmap() subroutine is unsuccessful, the errno global variable is set to the
following value:
EINVAL
The addr parameter is not a multiple of the page size, as
returned by the sysconf() subroutine using the
_SC_PAGE_SIZE value for the Name parameter.
EINVAL
The application has requested Single UNIX Specification
Version 2 compliant behavior and the len argument is 0.
472
Developing and Porting C and C++ Applications on AIX

D
Subroutine references for
Appendix D.
POSIX threads
This appendix provides Pthread subroutine references supported on AIX 5L
Version 5.2 using the categories defined by the following sections:
“Subroutines defined in the POSIX thread standard” on page 474
“Subroutines defined in the UNIX 98 Specification” on page 479
“Extensions to POSIX thread” on page 481
In each category, Pthread subroutines are described by routine names and short
descriptions. For complete information about Pthread subroutines, please refer to
AIX 5L Version 5.2 Technical Reference: Base Operating System and
Extensions.
In addition to the Pthread subroutines (starting from pthread_), AIX provides a
set of subroutines starting from pthdb_, which are called as Pthread debug
library subroutines, in order to offer additional functions over the POSIX thread
standard. For further information about the Pthread debug library, please refer to
the “Parallel Programming“ section of AIX 5L Version 5.2 General Programming
Concepts: Writing and Debugging Programs.
© Copyright IBM Corp. 2000, 2003. All rights reserved.
473

Subroutines defined in the POSIX thread standard
We have categorized the Pthread subroutines defined in the POSIX thread
standard, which are supported on AIX, into the following:
Thread management sub-routines
See Table D-1.
Execution scheduling sub-routines
See Table D-2 on page 476.
Synchronization sub-routines
See Table D-3 on page 477.
Thread-specific data sub-routines
See Table D-4 on page 479.
Table D-1 Thread management sub-routines
Name
Description
pthread_attr_init()
Initializes a thread attributes object.
pthread_attr_destroy()
Destroys a thread attributes object.
pthread_attr_setdetachstate()
Sets the detachstate attribute of a thread attributes
object. This attribute determines if a thread created
with this thread attributes object is in a detached state
or not.
pthread_attr_getdetachstate()
Gets the detach state attribute from a thread attributes
object.
pthread_attr_setstackaddr()
Sets the value of the stackaddr attribute of a thread
attributes object. This attribute specifies the stack
address of a thread created with this attributes object.
pthread_attr_getstackaddr()
Gets the stackaddr attribute from a thread attributes
object.
pthread_attr_setstacksize()
Sets the value of the stacksize attribute of a thread
attributes object. This attribute specifies the minimum
stack size, in bytes, of a thread created with this
thread attributes object.
pthread_attr_getstacksize()
Gets the stacksize attribute from a thread attributes
object.
pthread_testcancel()
Creates a cancellation point in the calling thread.
pthread_setcancelstate()
Atomically sets the calling thread's cancelability state
to the indicated state and returns the previous
cancelability state at a specified location reference.
pthread_setcanceltype()
Atomically sets the calling thread's cancelability type
to the indicated type and returns the previous
cancelability type at a specified location reference.
474
Developing and Porting C and C++ Applications on AIX

Name
Description
pthread_create()
Creates a new thread and initializes its attributes
using the thread attributes object specified, or
standard values instead, if the NULL pointer is
specified. After thread creation, a thread attributes
object can be reused to create another thread, or
deleted.
pthread_exit()
Terminates the calling thread safely, and stores a
termination status for any thread that may join the
calling thread.
pthread_cancel()
Requests the cancellation of the specified thread. The
action depends on the cancelability of the target
thread.
pthread_kill()
Sends the specified signal to the specified thread. It
acts with threads like the kill subroutine with
single-threaded processes.
pthread_join()
Blocks the calling thread until the specified thread
terminates. If the specified thread is in a detached
state (non-joinable), an error is returned.
pthread_detach()
Used to indicate to the implementation that storage for
the specified thread can be reclaimed when that
thread terminates.
pthread_once()
Executes the specified routine exactly once in a
process. The first call to this subroutine by any thread
in the process executes the given routine, without
parameters. Any subsequent call will have no effect.
pthread_self()
Returns the calling thread's ID.
pthread_equal()
Compares the two specified thread IDs. Returns zero
if and only if the IDs are equal.
pthread_atfork()
Threads can fork processes. This routine registers
fork cleanup handlers. Three handlers can be
specified: prepare, parent, and child. The prepare
handler is called before the processing of the fork
subroutine commences. The parent handler is called
after the processing of the fork subroutine completes
in the parent process. The child handler is called after
the processing of the fork subroutine completes in the
child process.
Appendix D. Subroutine references for POSIX threads
475

Name
Description
pthread_cleanup_push()
Pushes the specified cancellation cleanup handler
routine onto the calling thread's cancellation cleanup
stack.
pthread_cleanup_pop()
Removes the routine at the top of the calling thread's
cancellation cleanup stack and optionally invokes it (if
execute is non-zero).
Table D-2 Execution scheduling sub-routines
Name
Description
pthread_attr_setschedparam()
Sets the value of the schedparam attribute of the
specified thread attributes object. The given
schedparam attribute specifies the scheduling
parameters of a thread created with this attributes
object.
pthread_attr_getschedparam()
Gets the value of the schedparam attribute of the
specified thread attributes object.
pthread_attr_setscope()
The contention scope can only be set before thread
creation by setting the contention-attribute of a
thread attributes object. The pthread_attr_setscope
subroutine sets the attribute to the specified value.
pthread_attr_getscope()
Gets the contention-scope attribute of the specified
thread attributes object.
pthread_attr_setinheritsched()
Sets the inheritsched attribute of the specified thread
attributes object to a given value.
pthread_attr_getinheritsched()
Gets the inheritsched attribute of the specified
thread attributes object.
pthread_attr_setschedpolicy()
Sets the schedpolicy attribute of the specified thread
attributes object.
pthread_attr_getschedpolicy()
Gets the schedpolicy attribute of the specified thread
attributes object.
476
Developing and Porting C and C++ Applications on AIX

Name
Description
pthread_setschedparam()
Dynamically sets the schedpolicy and schedparam
attributes of the specified thread. The given
schedpolicy attribute specifies the scheduling policy
of the thread. The given schedparam attribute
specifies the scheduling parameters. The
implementation of this subroutine is dependent on
the priority scheduling POSIX option. The priority
scheduling POSIX option is implemented in the
operating system. If the target thread has system
contention-scope, the process must have root
authority to set the scheduling policy to either
SCHED_FIFO or SCHED_RR.
pthread_getschedparam()
Returns the current schedpolicy and schedparam
attributes of the thread
thread. The schedpolicy attribute specifies the
scheduling policy of a thread.
Table D-3 Synchronization sub-routines
Name
Description
pthread_mutexattr_init()
Initializes a mutex attributes object with the default
value for all of the attributes defined by the
implementation.
pthread_mutexattr_destroy()
Destroys a mutex attributes object; the object
becomes, in effect, uninitialized.
pthread_mutexattr_setpshared()
Sets the process-shared attribute in a given initialized
attributes object.
pthread_mutexattr_getpshared()
Obtains the value of the process-shared attribute from
the given attributes object.
pthread_mutex_init()
Initializes the given mutex with attributes specified by
a given attributes object. If the attributes object is
NULL, the default mutex attributes are used.
pthread_mutex_destroy()
Destroys the specified mutex object; the mutex object
becomes, in effect, uninitialized.
pthread_mutex_lock()
The specified mutex object is locked by calling. If the
mutex is already locked, the calling thread blocks until
the mutex becomes available.
Appendix D. Subroutine references for POSIX threads
477

Name
Description
pthread_mutex_trylock()
Identical to pthread_mutex_lock(), except that if the
referenced mutex object is currently locked (by any
thread, including the current thread), the call returns
immediately.
pthread_mutex_unlock()
Releases the referenced mutex object. The manner in
which a mutex is released is dependent upon the
mutex's type attribute.
pthread_condattr_init()
Initializes a specified condition variable attributes
object with the default value for all of the attributes
defined by the implementation.
pthread_condattr_destroy()
Destroys a specified condition variable attributes
object; the object becomes, in effect, uninitialized.
pthread_condattr_setpshared()
Sets the value of the pshared attribute of the specified
condition attributes object. This attribute specifies the
process sharing of the condition variable created with
this attributes object.
pthread_condattr_getpshared()
Returns the value of the pshared attribute of the
specified condition attribute object. This attribute
specifies the process sharing of the condition variable
created with this attributes object.
pthread_cond_init()
Initializes the given condition variable with attributes
given by a condition attributes object. If that object is
NULL, the default condition variable attributes are
used.
pthread_cond_destroy()
Destroys the given condition variable; the object
becomes, in effect, uninitialized.
pthread_cond_wait()
Blocks on a condition variable. Must be called with a
specified mutex locked by the calling thread or
undefined behavior will result.
pthread_cond_timedwait()
Same as pthread_cond_wait(), except that an error is
returned if the specified absolute time passes (that is,
system time equals or exceeds the specified absolute
time) before the specified condition is signaled or
broadcasted, or if the absolute time specified has
already been passed at the time of the call.
pthread_cond_signal()
Unblocks one or more threads blocked on the
specified condition.
478
Developing and Porting C and C++ Applications on AIX

Name
Description
pthread_cond_broadcast()
Unblocks all the blocked threads on the specified
condition.
Table D-4 Thread-specific data sub-routines
Name
Description
pthread_key_create()
Creates a thread-specific data key. The key is shared among
all threads within the process, but each thread has specific
data associated with the key. The thread-specific data is a
void pointer, initially set to NULL. An optional destructor
routine can be specified. It will be called for each thread when
it is terminated and detached, after the call to the cleanup
routines, if the specific value is not NULL.
pthread_key_delete()
Deletes the given thread-specific data key previously created
with the pthread_key_create() subroutine. The application
must ensure that no thread-specific data is associated with
the key.
pthread_setspecific()
Associates a thread-specific value with a key obtained
through a previous call to pthread_key_create(). Different
threads may bind different values to the same key.
pthread_getspecific()
Returns the value currently bound to the specified key on
behalf of the calling thread.
Note: Currently, AIX does not support the following sub-routines. Although the
symbols are provided in the Pthread library, but calls to these routines always
return with ENOSYS:
pthread_mutexattr_setprioceiling()
pthread_mutexattr_getprioceiling()
pthread_mutexattr_setprotocol()
pthread_mutexattr_getprotocol()
pthread_mutex_setprioceiling()
pthread_mutex_getprioceiling()
Subroutines defined in the UNIX 98 Specification
We have categorized the Pthread subroutines defined in the UNIX 98
Specification, which are supported on AIX 5L Version 5.2, into the following:
Read-write lock sub-routines
See Table D-5 on page 480.
Additional POSIX threads sub-routines
See Table D-6 on page 481.
Appendix D. Subroutine references for POSIX threads
479

Table D-5 Read-write lock sub-routines
Name
Description
pthread_rwlockattr_init()
Initializes the read-write specified lock with the
attributes referenced by the given read-write lock
attribute object. If that object is NULL, the default
read-write lock attributes are used.
pthread_rwlockattr_destroy()
Destroys the specified read-write lock attribute object
and releases any resources used by the lock.
pthread_rwlockattr_setpshared()
Sets the process-shared attribute in the given
initialized read-write lock attributes object.
pthread_rwlockattr_getpshared()
Obtains the value of the process-shared attribute
from the given initialized read-write lock attributes
object.
pthread_rwlock_init()
Initializes the specified read-write lock with the
attributes from a given read-write lock attributes
object. If that object is NULL, the default read-write
lock attributes are used.
pthread_rwlock_destroy()
Destroys the specified read-write lock object and
releases any resources used by the lock.
pthread_rwlock_rdlock()
Applies a read lock to the given read-write lock. The
calling thread acquires the read lock if a writer does
not hold the lock and there are no writers blocked on
the lock.
pthread_rwlock_tryrdlock()
Applies a read lock as in the pthread_rwlock_rdlock()
function with the exception that the function fails if any
thread holds a write lock on the specified read-write
lock or there are writers that sblocked the lock.
pthread_rwlock_wrlock()
Applies a write lock to the given read-write lock. The
calling thread acquires the write lock if no other
thread (reader or writer) holds the read-write lock.
Otherwise, the thread blocks (that is, does not return
from the pthread_rwlock_wrlock() call) until it can
acquire the lock.
pthread_rwlock_trywrlock()
Applies a write lock like the pthread_rwlock_wrlock()
function, with the exception that the function fails if
any thread currently holds the specified read-write
lock (for reading or writing).
pthread_rwlock_unlock()
Releases a lock held on the specified read-write lock
object.
480
Developing and Porting C and C++ Applications on AIX

Table D-6 Additional POSIX threads sub-routines defined in UNIX 98
Name
Description
pthread_setconcurrency()
Allows an application to inform the threads
implementation of its desired concurrency level. The
actual level of concurrency provided by the
implementation as a result of this function call is
unspecified.
pthread_getconcurrency()
Returns the value set by a previous call to the
pthread_setconcurrency() function.
pthread_attr_setguardsize()
Sets the guardsize attribute in a thread attribute
object. The guardsize attribute controls the size of the
guard area for the created thread's stack. The
guardsize attribute provides protection against
overflow of the thread’s stack pointer.
pthread_attr_getguardsize()
Gets the guardsize attribute of a thread attributes
object.
pthread_mutexattr_settype()
Sets the mutex type attribute of a mutex attributes
object to a given type.
pthread_mutexattr_gettype()
Gets the type attribute of a given mutex attributes
object.
pthread_suspend()
Suspends execution of specified thread.
pthread_continue()
Resumes execution of specified thread.
Extensions to POSIX thread
The POSIX thread standard leaves certain aspects of implementation defined,
unspecified, or even undefined. Therefore, most UNIX operating system vendors
provide several extensions in order to complement those areas in the POSIX
thread standard.
On AIX, those extension subroutines are starting from pthread_ and ending in
_np to signify that a library routine is non-portable and should not be used in
code that will be ported to other UNIX-based systems (see Table D-7 on
page 482).
Appendix D. Subroutine references for POSIX threads
481

Table D-7 Non-portable thread routines in AIX
Routine
Routine (continued)
pthread_atfork_unregister_np()
pthread_getthrds_np()
pthread_attr_getsuspendstate_np()
pthread_getunique_np()
pthread_attr_setstacksize_np()
pthread_join_np()
pthread_attr_setsuspendstate_np()
pthread_lock_global_np()
pthread_cleanup_information_np()
pthread_mutexattr_getkind_np()
pthread_cleanup_pop_np()
pthread_mutexattr_setkind_np()
pthread_cleanup_push_np()
pthread_set_mutexattr_default_np()
pthread_clear_exit_np()
pthread_setcancelstate_np()
pthread_continue_np()
pthread_signal_to_cancel_np()
pthread_continue_others_np()
pthread_suspend_np()
pthread_delay_np()
pthread_suspend_others_np()
pthread_get_expiration_np()
pthread_test_exit_np()
pthread_getrusage_np()
pthread_unlock_global_np()
Relevant information about these non-portable subroutines could be find in the
following publications:
AIX Version 4.3 Differences Guide, SG24-2014
AIX 5L Differences Guide Version 5.2 Edition, SG24-5765
482
Developing and Porting C and C++ Applications on AIX

E
Supported IBM SMP
Appendix E.
directives
This appendix explains IBM SMP directives supported by the following compiler
products:
C for AIX Version 6.0
VisualAge C++ for AIX Version 6.0
When developing new applications, it is recommended that you use OpenMP
directives explained in Chapter 8, “Introduction to POSIX threads” on page 275.
For further information about IBM directives, please refer to:
C for AIX Compiler Reference, SC09-4960
VisualAge C++ for AIX Compiler Reference, SC09-4959
© Copyright IBM Corp. 2000, 2003. All rights reserved.
483

IBM SMP directives
IBM SMP directives for parallelization are based on the possibility of parallelizing
countable loops. A loop is considered countable when the following rules can be
applied:
There is no branching into or outside of the loop.
The incremental expression (incr_expr) is not within a critical section.
Table E-1 shows the C language control flow statements and the regular
expressions that define when they can be treated as countable loops.
Table E-1 Regular expressions for countable loops
C control flow statement keywords
Regular expression
for
for ([iv]; exit_cond; incr_expr)
statement
for ([iv]; exit_cond; [expr] {
[declaration_list]
[statement_list]
incr_expr;
[statement_list]
}
while
while (exit_cond) {
[declaration_list]
[statement_list]
incr_expr;
[statement_list]
}
do
do {
[declaration_list]
[statement_list]
incr_expr;
[statement_list]
} while (exit_cond)
Where:
exit_cond iv
<=
ub
iv < ub
iv >= ub
iv > ub
incr_expr
++iv
484
Developing and Porting C and C++ Applications on AIX

iv++
--iv
i--
iv += incr
iv -= incr
iv = iv + incr
iv = incr + iv
iv = iv - incr
iv
Iteration variable. The iteration variable is a signed integer
that has either automatic or register storage class, does
not have its address taken, and is not modified anywhere
in the loop except in incr_expr.
incr
Loop invariant signed integer expression. The value of the
expression is known at compile-time and is not 0. incr
cannot reference extern or static variables, pointers or
pointer expressions, function calls, or variables that have
their address taken.
ub
Loop invariant signed integer expression. ub cannot
reference extern or static variables, pointers or pointer
expressions, function calls, or variables that have their
address taken.
In general, a countable loop is automatically parallelized only if all of the following
conditions are met:
The order in which loop iterations start or end does not affect the results of
the program.
The loop does not contain I/O operations.
Floating point reductions inside the loop are not affected by round-off error,
unless the -qnostrict option is in effect.
The -qnostrict_induction compiler option is in effect.
The -qsmp compiler option is in effect without its omp sub option. The
compiler must be invoked using a thread-safe compiler mode.
The IBM SMP directives syntax
When using IBM SMP directives for explicitly defining parallel portions of code,
use the following syntax:
#pragma ibm pragma_name_and_args
Appendix E. Supported IBM SMP directives
485

<countable for|while|do loop>
Pragma directives must appear immediately before the section of code to which
they apply. For most parallel processing pragma directives, this section of code
must be a countable loop, and the compiler will report an error if one is not found.
More than one parallel processing pragma directive can be applied to a
countable loop. For example:
#pragma ibm independent_loop
#pragma ibm independent_calls
#pragma ibm schedule(static,5)
<countable for|while|do loop>
Some pragma directives are mutually-exclusive. If mutually-exclusive pragmas
are specified for the same loop, the last pragma specified applies to the loop. In
the example below, the parallel_loop pragma directive is applied to the loop, and
the sequential_loop pragma directive is ignored:
#pragma ibm sequential_loop
#pragma ibm parallel_loop
Other pragmas, if specified repeatedly for a given loop, have an additive effect.
For example:
#pragma ibm permutation (a,b)
#pragma ibm permutation (c)
is equivalent to:
#pragma ibm permutation (a,b,c)
Table E-2 shows all IBM pragma directives supported by the latest compilers.
Table E-2 Supported IBM pragma directives
Pragma
Description
#pragma ibm critical
Instructs the compiler that the statement or
statement block immediately following this
pragma is a critical section.
#pragma ibm independent_calls
Asserts that specified function calls within the
chosen loop have no loop-carried
dependencies.
#pragma ibm independent_loop
Asserts that iterations of the chosen loop are
independent, and that the loop can therefore
be parallelized.
#pragma ibm iterations
Specifies the approximate number of loop
iterations for the chosen loop.
486
Developing and Porting C and C++ Applications on AIX

Pragma
Description
#pragma ibm parallel_loop
Explicitly instructs the compiler to parallelize
the chosen loop.
#pragma ibm permutation
Asserts that specified arrays in the chosen
loop contain no repeated values.
#pragma ibm schedule
Specifies scheduling algorithms for parallel
loop execution.
#pragma ibm sequential_loop
Explicitly instructs the compiler to execute the
chosen loop sequentially.
Appendix E. Supported IBM SMP directives
487

488
Developing and Porting C and C++ Applications on AIX

F
Sample compiler listing
Appendix F.
This appendix includes a sample compiler listing, as explained in 6.2,
“Diagnosing compile-time errors” on page 227.
© Copyright IBM Corp. 2000, 2003. All rights reserved.
489

Compiler listing
The following is a sample compiler listing showing the various sections described
in “Compiler listing” on page 228.
Example: F-1 Sample compiler listing
C for AIX Compiler Version 6.0.0.2 --- hello.c 02/11/03 13:31:29 (C)
>>>>> SOURCE SECTION <<<<<
1 | #include <stdio.h>
2 |
3 | void main() {
4 | printf(“hello\n”);
5 | }
>>>>> OPTIONS SECTION <<<<<
C for AIX Compiler Version 6.0.0.2 ---
*** Command Line Invocation ***
*** Options In Effect ***
NOA NOAE NOALLOCA NOBROWSE
NOCOMPACT NOCPLUSCMT NODBCS NODBXEXTRA
NODIGRAPH DOLLAR NOEXTCHK NOFDPR
NOFULLPATH NOFUNCSECT NOG NOGRAPHICS
NOHEAPDEBUG NOIDIRFIRST NOIGNERRNO NOINLGLUE
NOLARGEPAGE NOLIBANSI NOLINEDEBUG LINEDIR
LIST LISTOPT LONGLONG NOMACPSTR
NOMAKEDEP NOMBCS NOOFFSET NOP
NOPASCAL NOPDF1 NOPDF2 NOPHSINFO
PRINT NOPROTO NOREPORT NORO
NOROCONST NOSMALLSTACK SOURCE NOSTATSYM
STDINC STRICT STRICT_INDUCTION NOSYNTAXONLY
NOTHREADED NOTOCMERGE UNWIND UPCONV
NOWARN64 NOXCALL NOXCOFF NOXPH2
OPTIMIZE=0
INLINE THRESHOLD=20
AGGRCOPY=NOOVERLAP
ALIAS=NOANSI:NOTYPEPTR:NOALLPTRS:NOADDRTAKEN
ALIGN=POWER
ATTR
BITFIELDS=UNSIGNED
CHARS=UNSIGNED
DATAIMPORTED
ENUM=INT
FLAG=I:I
490
Developing and Porting C and C++ Applications on AIX

FLOAT=NOHSFLT:NORNDSNG:NOHSSNGLE:MAF:NORSQRT:NORRM:FOLD:NOSPNANS:NOFLTINT:NOEMU
LATE
FLTTRAP=NOOV:NOUND:NOZERO:NOINV:NOINEX:NOEN:NOIMP
NOGENPROTO
HALT=S
NOHOT
INFO=NOCLS:NOCMP:NOCND:NOCNS:NOCNV:NOCPY:NODCL:NOEFF:NOENU:NOEXT:NOGEN:NOGNR:NO
GOT:NOINI:NOLAN:NOOBS:NOORD:NOPAR:NOPOR:NOPPC:NOPPT:NOPRO:NOREA:NORET:NOTRD:NOT
RU:TRX:NOUND:NOUNI:NOUSE:NOVFT:NOPRIVATE:NOREDUCTION:NOC99
LANGLVL=EXTENDED:NOUCS
LONGDOUBLE=128
NOMAXERR
MAXMEM=8192
OS=AIX
PROCUNKNOWN
SHOWINC=NOSYS:NOUSR
NOSMP
SPILL=512
TBTABLE=DEFAULT
TUNE=DEFAULT
UNROLL=AUTO
XREF
YN (ROUND NEAR)
REACHABLE=setjmp
REACHABLE=_setjmp
REACHABLE=sigsetjmp
REACHABLE=_sigsetjmp
>>>>> ATTRIBUTE AND CROSS REFERENCE SECTION <<<<<
fhandle struct tag
4-528.8$ 4-528.16{ 4-530.1} 4-531.16
fid struct tag
4-521.8$ 4-521.12{ 4-524.1} 4-525.16
fileid struct tag
4-500.8$ 4-500.15{ 4-505.1} 4-535.16
label_t struct tag
8-48.16$ 8-51.16 8-49.1{ 8-67.1}
main extern function returning void
0-3.6Y
printf extern function returning int
Appendix F. Sample compiler listing
491

1-263.17X 0-4.9Z
unique_id struct tag
4-545.8$ 4-545.18{ 4-550.1} 4-551.16
>>>>> FILE TABLE SECTION <<<<<
FILE CREATION FROM
FILE NO FILENAME DATE TIME FILE LINE
0 hello.c 02/11/03 13:23:32
1 /usr/include/stdio.h 09/13/02 10:34:34 0 1
2 /usr/include/standards.h 09/13/02 10:27:52 1 43
3 /usr/include/va_list.h 09/13/02 10:27:45 1 185
4 /usr/include/sys/types.h 09/13/02 10:27:48 1 399
5 /usr/include/sys/inttypes.h 09/13/02 10:27:52 4 55
6 /usr/include/stdint.h 09/13/02 10:27:52 5 62
7 /usr/include/standards.h 09/13/02 10:27:52 6 28
8 /usr/include/sys/m_types.h 09/13/02 10:52:54 4 393
9 /usr/include/sys/vm_types.h 09/13/02 10:53:54 8 40
10 /usr/include/va_list.h 09/13/02 10:27:45 1 432
11 /usr/include/sys/limits.h 09/13/02 10:28:31 1 466
12 /usr/include/float.h 09/13/02 10:51:20 11 263
>>>>> COMPILATION EPILOGUE SECTION <<<<<
C for AIX Summary of Diagnosed Conditions
TOTAL UNRECOVERABLE SEVERE ERROR WARNING INFORMATIONAL
(U) (S) (E) (W) (I)
0 0 0 0 0 0
Source records read......................................... 3591
1501-008 Compilation successful for file hello.c. Object file created.
>>>>> OBJECT SECTION, NO OPTIMIZATION <<<<<
GPR’s set/used: ss-s ssss ssss s--- ---- ---- ---- ---s
FPR’s set/used: ssss ssss ssss ss-- ---- ---- ---- ----
CCR’s set/used: ss-- -sss
| 000000 PDEF main
3| PROC
0| 000000 mfspr 7C0802A6 1 LFLR gr0=lr
492
Developing and Porting C and C++ Applications on AIX

0| 000004 stw 93E1FFFC 0 ST4A #stack(gr1,-4)=gr31
0| 000008 stw 90010008 2 ST4A #stack(gr1,8)=gr0
0| 00000C stwu 9421FFC0 0 ST4U gr1,#stack(gr1,-64)=gr1
3| 000010 lwz 83E20004 1 L4A gr31=.$STATIC(gr2,0)
4| 000014 addi 387F0008 2 AI gr3=gr31,8
4| 000018 bl 4BFFFFE9 0 CALL
gr3=printf,1,gr3,printf”,gr1,cr[01567]”,gr0”,gr4”-gr12”,fp0”-fp13”
4| 00001C ori 60000000 1
5| CL.1:
5| 000020 lwz 80010048 1 L4A gr0=#stack(gr1,72)
5| 000024 mtspr 7C0803A6 2 LLR lr=gr0
5| 000028 addi 38210040 1 AI gr1=gr1,64
5| 00002C lwz 83E1FFFC 0 L4A gr31=#stack(gr1,-4)
5| 000030 bclr 4E800020 2 BA lr
| Tag Table
| 000034 00000000 00002041 80010001 00000034 00046D61 696E
| Instruction count 13
| Straight-line exec time 13
Appendix F. Sample compiler listing
493

494
Developing and Porting C and C++ Applications on AIX

Abbreviations and acronyms
ANSI
American National Standard
FTSS
Field Technical Support
Institute
Specialist
APAR
Authorized Problem Analysis
GB
Gigabyte
Report
GID
Group ID
API
Application Programming
GMT
Greenwich Mean Time
Interface
GNU
GNU is Not UNIX
ASCII
American National Standard
Code for Information
GPR
General Purpose Register
Interchange
GUI
Graphical User Interface
BOS
Base Operating System
HMC
IBM Hardware Management
BSD
Berkeley Software Distribution
Console for pSeries
BSS
Block Started by Symbol
HPC
High-Performance Computing
CD
Compact Disk
HTML
Hyper-Text Markup Language
CDE
Common Desktop
HTTP
Hyper-Text Transfer Protocol
Environment
I/O
Input/Output
CD-ROM
CD-Read Only Media
IBM
International Business
CDT
Central Daylight Time
Machines Corporation
CPU
Central Processing Unit
IDE
Integrated Development
Environment
CST
Central Standard Time
IEEE
Institute of Electrical and
DCE
Distributed Computer
Electronics Engineers
Environment
ILP32
integer/long/pointer 32-bit
DLL
Dynamic Link Library
IOC
IBM Open Class Library
DLPAR
Dynamic Logical Partitioning
IPA
Interprocedural Analysis
DSA
Dynamic Segment Allocation
IPC
Inter-Process Communication
DVD
Digital Versatile Disk
ISBN
International Standard Book
DVD-R
DVD-Recordable
Number
DVD-RAM
DVD-Random Access Media
ISO
International Organization for
DVD-ROM
DVD-Read Only Media
Standardization
EB
Exabyte
ISV
Independent Software Vendor
EXTSHM
Extended Mode Shared
ITSO
International Technical
Memory Segments
Support Organization
FC
Feature Code
JFS
Journaled File System
FIFO
First-in First-out
JFS2
Enhanced Journaled File
System
© Copyright IBM Corp. 2000, 2003. All rights reserved.
495

K&R
The C programming language
SMP
Symmetric Multiple Processor
standard defined by Brian W.
STL
Standard Template Library
Kernighan and Dennis M.
Ritchie
TB
Terabyte
KB
Kilobyte
TID
Thread ID
LIFO
Last-In First-Out
TLB
Translation Look-Aside Buffer
LP64
Long/Pointer 64-bit
TOC
Table of Contents
LPP
License Program Product
TSD
Thread-Specific Data
LUM
License Use Management
TTY
Teletypewriter
LWP
Light-Weight Process
UCS
Universal Coded Character
Set
MB
Megabyte
UID
User ID
MP
Multiprocessor
UP
Uniprocessor
NFS
Network File System
URL
Universal Resource Locator
NIS
Network Information Service
VMM
Virtual Memory Manager
OS
Operating System
VP
Virtual Processor
PB
Petabyte
WLM
Work Load Manager
PCI
Peripheral Component
Interconnect
XCOFF
eXtended Common Object
File Format
PDF
Portable Document Format or
Profile-Directed Feedback
PID
Process ID
POSIX
Portable Operating System
Interface
POWER
Performance Optimization
with Enhanced RISC
PPID
Parent Process ID
PTF
Program Temporary Fix
RAD
Rapid Application
Development
RISC
Reduced Instruction Set
Computer
RMC
Resource Monitoring and
Control
RML
Recommended Maintenance
Level
RSCT
Reliable Scalable Cluster
Technology
SMIT
System Management
Interface Tool
496
Developing and Porting C and C++ Applications on AIX

Related publications
The publications listed in this section are considered particularly suitable for a
more detailed discussion of the topics covered in this redbook.
IBM Redbooks
For information on ordering these publications, see “How to get IBM Redbooks”
on page 500.
AIX 5L Differences Guide Version 5.2 Edition, SG24-5765
AIX 5L Porting Guide, SG24-6034
AIX Version 4.3 Differences Guide, SG24-2014
The Complete Partitioning Guide for IBM ^ pSeries Servers,
SG24-7039
IBM ^ pSeries 670 and pSeries 690 System Handbook, SG24-7040
Managing AIX Server Farms, SG24-6606
The POWER4 Processor