IZ90631: PROCESS HANG DUE TO DEADLOCK BETWEEN SELPOLL AND KFCNTL APPLIES TO AIX 5300-11

Obtain the fix for this APAR.

APAR status

• Closed as program error.

Error description

• A process hang is possible if its one thread is calling
  poll() with POLLRDNORM events to check a file descriptor
  while its another thread is calling fcntl() with F_DUP2FD
  command on the same file descriptor.
  When process is hung, it has the following stack trace:
  
  pvthread+047800 STACK:
  [00066870]e_block_thread+000290 ()
  [000674C8]e_sleep_thread+0000E8 (??, ??, ??)
  [00565ABC]_getf_threaded+00011C (??, ??, ??)
  [005662C4]getft+0000C4 (??, ??, ??)
  [0058C128]fdtosth+000068 (??)
  [005F7C24]selpoll+000184 (??, ??, ??, ??, ??, ??, ??)
  [0062EB34]fp_poll+000274 (??, ??, ??, ??)
  [0062E3D8]_poll+000258 (??, ??, ??, ??)
  
  pvthread+046200 STACK:
  [00066870]e_block_thread+000290 ()
  [000674C8]e_sleep_thread+0000E8 (??, ??, ??)
  [004E207C]closex+00039C (??, ??)
  [006E7F1C]kfcntl+000FBC (??, ??, ??)
  

Local fix

• There are two workrounds to avoid
  such kind of process hang:
  
  1. Use POLLIN instead of POLLRDNORM.
  2. Avoid to poll and dup2 on same
     file descriptor at the same time.
  

Problem summary

• A process hang is possible if its one thread is calling
  poll() with POLLRDNORM events to check a file descriptor
  while its another thread is calling fcntl() with F_DUP2FD
  command on the same file descriptor.
  When process is hung, it has the following stack trace:
  
  pvthread+047800 STACK:
   00066870 e_block_thread+000290 ()
   000674C8 e_sleep_thread+0000E8 (??, ??, ??)
   00565ABC _getf_threaded+00011C (??, ??, ??)
   005662C4 getft+0000C4 (??, ??, ??)
   0058C128 fdtosth+000068 (??)
   005F7C24 selpoll+000184 (??, ??, ??, ??, ??, ??, ??)
   0062EB34 fp_poll+000274 (??, ??, ??, ??)
   0062E3D8 _poll+000258 (??, ??, ??, ??)
  
  pvthread+046200 STACK:
   00066870 e_block_thread+000290 ()
   000674C8 e_sleep_thread+0000E8 (??, ??, ??)
   004E207C closex+00039C (??, ??)
   006E7F1C kfcntl+000FBC (??, ??, ??)
  

Problem conclusion

• Avoid getft() calls
  

Temporary fix

Comments

• 5300-10 - use AIX APAR IZ90906
  5300-11 - use AIX APAR IZ90631
  5300-12 - use AIX APAR IZ90627
  6100-03 - use AIX APAR IZ94041
  6100-04 - use AIX APAR IZ93575
  6100-05 - use AIX APAR IZ82482
  6100-06 - use AIX APAR IZ93137
  6100-06 - use AIX APAR IZ89358
  6100-07 - use AIX APAR IZ88332
  7100-00 - use AIX APAR IZ89112
  

APAR Information

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

  IZ82482

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

Fix information

• Fixed component name

  AIX 5.3

• Fixed component ID

  5765G0300

Applicable component levels

• R530 PSY U841917

     UP11/02/22 I 1000