IJ04249: JAVA JIT PPC - CRASH EXECUTING SOME JAVA.UTIL.CONCURRENT.ATOMIC.ATOMIC* ROUTINES

APAR status

• Closed as program error.

Error description

• Error Message: A SIGSEGV when executing
  AtomicInteger.getAndDecrement(), but the issue might effect the
  following Atomic functions as well.
  java_util_concurrent_atomic_AtomicBoolean_getAndSet
  java_util_concurrent_atomic_AtomicInteger_getAndAdd
  java_util_concurrent_atomic_AtomicInteger_getAndIncrement
  java_util_concurrent_atomic_AtomicInteger_getAndDecrement
  java_util_concurrent_atomic_AtomicInteger_getAndSet
  java_util_concurrent_atomic_AtomicInteger_addAndGet
  java_util_concurrent_atomic_AtomicInteger_incrementAndGet
  java_util_concurrent_atomic_AtomicInteger_decrementAndGet
  java_util_concurrent_atomic_AtomicIntegerArray_getAndAdd
  java_util_concurrent_atomic_AtomicIntegerArray_getAndIncrement
  java_util_concurrent_atomic_AtomicIntegerArray_getAndDecrement
  java_util_concurrent_atomic_AtomicIntegerArray_getAndSet
  java_util_concurrent_atomic_AtomicIntegerArray_addAndGet
  java_util_concurrent_atomic_AtomicIntegerArray_incrementAndGet
  java_util_concurrent_atomic_AtomicIntegerArray_decrementAndGet
  .
  Stack Trace: 0xfff94406b9c
  {com/ibm/io/async/ResultHandler.runEventProcessingLoop} JIT
  0xfff95fa1570 {com/ibm/io/async/ResultHandler$2.run} JIT
  0xfff93da6b18 {com/ibm/ws/util/ThreadPool$Worker.run} JIT
  The issue was seen when executing in the above stack trace, but
  the issue could occur in any java code that calls the
  java.util.concurrent.atomic.Atomic* routines listed in this
  APAR.
  .
  The issue can only happen on the PowerPC platform and only on
  Java8 SR5 through Java 8 SR5FP10.
  The problem will only occur when a level of contention exists
  that causes the atomic update to execute a 2nd attempt at
  performing the atomic update. If there is no contention then the
  1st atomic update will succeed and the issue will not occur.
  In addition, the issue can only occur when there the JIT is
  compiling a method that is complex enough to generate a
  significant amount of "register pressure" such that the JIT
  assigns register R0 to a specific purpose within the atomic
  update code.
  Given the above two restrictions, it is very unlikely that this
  issue will occur under normal conditions.
  

Local fix

• The problem can be avoided by excluding the compilation of the
  failing method using the following java command line argument:
  -Xjit:exclude={failing_method}
  Where "failing_method" is the full method name as seen in the
  javacore from the failure. Consider using (* after the method
  name to avoid having to provide the complete parameter and
  return type signatures, e.g.
  
  -Xjit:exclude={com/ibm/SomeClass.amethod(*}
  

Problem summary

• The JIT did not prevent the use of register R0 when implementing
  some set of Atomic functions that use PPC instructions with
  special handling for R0. This caused the JIT register assigner
  to incorrectly "spill" the register in the middle of a loop
  which results in a crash on the 2nd iteration of the loop.
  

Problem conclusion

• The JIT was modified such that the use of R0 is prevented in
  handling Atomic methods where instruction are generated that
  can't use R0.
  .
  This APAR will be fixed in the following Java Releases:
     8    SR5 FP11  (8.0.5.11)
  .
  Contact your IBM Product's Service Team for these Service
  Refreshes and Fix Packs.
  For those running stand-alone, information about the available
  Service Refreshes and Fix Packs can be found at:
             https://www.ibm.com/developerworks/java/jdk/
  

Temporary fix

Comments

APAR Information

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

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

Fix information

• Fixed component name

  JIT

• Fixed component ID

  620700124

Applicable component levels