IZ80188: ON LINUX, JVM HANGS IN SIGNAL HANDLER WHEN MALLOC HEADER STRUCTU RE CORRUPTED

APAR status

• Closed as program error.

Error description

• Error Message: JVM freezes in signal handler when memory
  corruption reported in malloc header structure
  .
  Stack Trace: #0  0xffffe410 in __kernel_vsyscall ()
  #1  0xb7f276fe in __lll_mutex_lock_wait () from  /lib/libc.so.6
  #2  0xb7ec3d59 in _L_lock_113 () from /lib/libc.so.6
  #3  0xb7ebf312 in ptmalloc_lock_all () from /lib/libc.so.6
  #4  0xb7ee798f in fork () from /lib/libc.so.6
  #5  0xb7faec74 in fork () from /lib/libpthread.so.0
  #6  0xb7d8ba61 in j9dump_create () from
  /opt/WebSphere61/AppServer/java/jre/bin/libj9prt23.so
  #7  0xb7c23cd6 in doSystemDump () from
  /opt/WebSphere61/AppServer/java/jre/bin/libj9dmp23.so
  .
  

Local fix

• Use -Xrs option to disable all JVM signal handlers
  

Problem summary

• The problem is reported when a JVM signal handler is invoked as
  a consequence of memory corruption in the malloc header
  structure on LINUX. JVM signal handler uses the fork system
  function to process the system dump and fork on LINUX uses
  malloc for its functionality. So, when JVM signal handler is
  invoked for such memory corruption, the subsequent call to
  malloc (from fork) resulted in a hang.
  

Problem conclusion

• JVM has created the option -Xrs:sync which on Linux, z/OS and
  AIX to instruct the JVM to not register signal handlers for
  SIGILL/BUS/FPE/SEGV/TRAP/ABRT signals while preserving JVM
  signal handlers for other signals (like SIGQUIT, etc.). This
  feature enables the JVM not to register signal handlers for
  these selective signals & register signal handler for other
  signals. This way JVM avoids the problematic code path(involves
  fork) on such error conditions without disturbing signal actions
  of other signals. Specifically, while -Xrs disables the
  triggering of the -Xdump:event user event when SIGQUIT is
  injected to the process, -Xrs:sync does not. With -Xrs:sync,
  users can still generate javacore files by injecting SIGQUIT to
  the process.  Notes: 1. -Xrs:sync will reduce performance by
  approximately 2-4% depending on the application as an
  optimization needs to be disabled when it is used. 2. If
  -Xrs:sysc is used and any of the signals it disables are
  delivered to the process, the default OS action for the signal
  will occur. The default OS action is normally to generate a
  system core file, but users confirm this on their system.
  
  This defect will be fixed in:
  5.0.0 SR12
  6.0.0 SR9
  

Temporary fix

Comments

APAR Information

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

  IZ80185

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

Fix information

• Fixed component name

  J9 COMMON CODE

• Fixed component ID

  620700127

Applicable component levels

• R600 PSY

     UP