IV45149: performance improvement in fail case of huge malloc() request APPLIES TO AIX 7100-03

Obtain the fix for this APAR.

APAR status

• Closed as program error.

Error description

• When a process trying to allocate huge amount of memory
  using the calls like malloc or calloc, it is possible
  that
  these calls may fail upon hitting the one of the
  resource limit like paging space, pinned memory etc ...
  In the failure case these calls may take longer time to
  return back to user with error number.
  During these longer periods before returning back to
  user,
  the thread stack of the process may look like:-
  
      (0)> stack 40210
      pvthread+9D1200 STACK:
      Use current context [F00000002FF47600] of cpu 126
      [005C0C30]shm_findsptrx+0000F0 (0000000000000000,
        F000000020000000,F00000002FF47270 [??])
      [005C0658]shm_removex+000018 (??, ??, ??)
      [005C0E28]shm_deletex+000048 (??, ??)
      [0063D5C4]_brk64+0003C4 (??, ??)
      [0063F288]_sbrk64+000068 (??)
      [0063D1B8]__sbrk+000238 (??)
      [00003850]ovlya_addr_sc_flih_main+000130 ()
      [90000000000F5DC]extend_brk+00005C (??, ??, ??)
      [9000000000A85A0]malloc_y+0003A0 (002386F26FC0FFFF,
       0000000000000000,09001000A0AB1530, 09001000A0A3FA80,
       3FFC000000000003,000000000001BA80,
         0000000000000000, 0000000000000000)
      [900000000010F48]malloc_common@AF102_86+000028 (??)
      [900000000010CD4]malloc+0000D4 (??)
      [9000000000103DC]calloc_common+00003C (??, ??)
      [90000000000EFE4]init_calloc+000044 (??, ??)
      [900000000010DA8]calloc+000028 (??, ??)
      [1000004F8]fopen_custom_buffer_size+000098 (??)
      [1000006C8]main+000088 (??, ??, ??)
      [1000002B0]__start+000070 ()
  

Local fix

• Set the data limit using ulimit to avoid performance
  degradation with malloc/calloc failures.
  Same can be achieved by setting MAXDATA with LDR_CNTRL
  process environment variable
  

Problem summary

• A process takes an abnormally large time to exit after a failed
  huge memory allocation. kdb would show the process spending
  most of its time in shm_findsptr64x().
  

Problem conclusion

• Improve the hinting algorithm to catch a corner case.
  

Temporary fix

Comments

• 6100-09 - use AIX APAR IV66299
  7100-03 - use AIX APAR IV45149
  

APAR Information

• APAR is sysrouted FROM one or more of the following:

• APAR is sysrouted TO one or more of the following:

  IV66299 IV66309

Fix information

• Fixed component name

  AIX V7.1

• Fixed component ID

  5765H4000

Applicable component levels

• R710 PSY U865837

     UP15/05/19 I 1000

PTF to Fileset Mapping

• U865837 bos.mp64 7.1.3.45