The SPECjbb2011 benchmark is failing on OEL with a SIGSEGV in java.util.concurrent.ConcurrentHashMap.get or java.util.concurrent.ConcurrentHashMap.put when G1 is enabled but Biased Locking (or UseOptoBiasInlining) is disabled:


#
# A fatal error has been detected by the Java Runtime Environment:
#
#  SIGSEGV (0xb) at pc=0x00007fdc6453cc4f, pid=29870, tid=140555764786960
#
# JRE version: 7.0-b147
# Java VM: Java HotSpot(TM) 64-Bit Server VM (21.0-b17-fastdebug mixed mode linux-amd64 compressed oops)
# Problematic frame:
# J  java.util.concurrent.ConcurrentHashMap.get(Ljava/lang/Object;)Ljava/lang/Object;
#
# Core dump written. Default location: /net/cairnapple.us.oracle.com/export/g1-coredump/core or core.29870
#
#
# If you would like to submit a bug report, please visit:
#   http://bugreport.sun.com/bugreport/crash.jsp
#
Current thread is 140555764786960
Dumping core ...


The SPECjbb2011 benchmark is attached as zip file.

The crash can be seen locally - on sc14ia03.us.oracle.com with the following script:
======================================================
#!/bin/ksh

#JHOME=/java/re/jdk/7/latest/binaries/linux-x64/fastdebug
#JHOME=/net/cairnapple.us.oracle.com/export/jdk7/linux-x64

VM_OPTIONS="-d64 -server -XX:+UseG1GC  -Xms26g -Xmx26g -Xmn8g  -XX:MaxGCPauseMillis=100 -XX:-UseBiasedLocking  -verbosegc"

#VM_OPTIONS="-d64 -server -XX:+UseG1GC  -Xms26g -Xmx26g -Xmn8g  -XX:MaxGCPauseMillis=100 -XX:-UseOptoBiasInlining -verbosegc -XX:+PrintNMethods"

echo "${JHOME}/bin/java ${VM_OPTIONS} -jar target/specjbb2011-composite.jar"

${JHOME}/bin/java ${VM_OPTIONS} -jar target/specjbb2011-composite.jar 2> err.out
====================

The original report was reported with TieredCompilation enabled but +PrintCompilation shows that the twoe methods in which the SEGV occurs have been compiled at tiered-level 4. When TieredStopAt=1 is set the test case passes.

The crash is seen with both product and fastdebug JVMs.

The crash is not seen on a similarly configured machin running solaris (sc14ia01.us.oracle.com).

The problem is also not seen with compressed oops disabled.

The crash was not seen with +PrintOptoAssembly but did occur with +PrintNMethods (without a disassebler). Using the information from PrintNMethods I was able to obtain the generated code java.util.concurrent.ConcurrentHashMap.get (attached as 29870_get.s).

Heap verification (before and after the GCs) does not detect a problem.

hs_err file is also attached.