LI75707: Hang in ipa signal handler

September 2010 Update for XL Fortran Advanced Edition for Blue Gene/L, V11.1

APAR status

• Closed as program error.

Error description

• There appears to be a race condition that prevents the compiler
  from being interrupted with ctrl+C under some conditions.
  
  When this happens, it seems to be waiting on the ipa subprocess.
  
  
  
  With a sample Fortran program that has some modules, the
  following will reproduce the problem:
  
  1. Type the compile commands
  2. As soon as it outputs "End of Compilation XX", hit ctrl+C
  
  Normally, you would see this:
    Calling signal handler...
  and it would exit immediately.
  
  However, when the race condition is hit, it hangs right after
  printing the first message:
    Calling signal handler...
  
  and will remain there until you kill the ipa subprocess
  w/SIGTERM from another shell.
  
  Additional ^Cs will not have any effect.
  
  This hang does not happen always, depending on the timing you
  hit CTRL+C. But it always hangs in ipa process (add -V
  compiler option to view what process it is in).
  
  This causes inconvenience to users.
  

Local fix

• To kill the hanging process, launch another shell, and use ps -u
  to show the process you own.
  
  For example:
  
  awcheng@xlpfen15:~> ps -u awcheng
    PID TTY          TIME CMD
   2635 ?        00:00:00 sshd
   2636 pts/0    00:00:00 bash
   6342 pts/0    00:00:00 mpixlf77_r
   6357 pts/0    00:00:00 bgxlf_r
   6360 pts/0    00:00:00 ipa
   7413 ?        00:00:00 sshd
   7414 pts/2    00:00:00 bash
   7871 pts/2    00:00:00 ps
  
  awcheng@xlpfen15:~> kill 6360
  
  This way, the hang process in the original window will exit
  immediately.
  

Problem summary

• USER AFFECTED:
  Users who use ctrl+C to interrupt the IPA process during
  compilation.
  
  PROBLEM DESCRIPTION:
  The signal handler cannot call malloc/free on Linux because it
  may deadlock
  due to a lock held by malloc.
  

Problem conclusion

• The problem is fixed.
  The compiler now does not use malloc/free on the signal
  handler.
  It prints the interrupt message and call _exit().
  

Temporary fix

Comments

APAR Information

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

  LI74716

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

Fix information

• Fixed component name

  XL FORTRAN AE B

• Fixed component ID

  5799HJG00

Applicable component levels

• RB10 PSN LI74716

     UP06/09/13