Bug ID: JDK-7026700 regression in 6u24-rev-b23: Crash in C2 compiler in PhaseIdealLoop::build_loop_late

Versions (Unresolved/Resolved/Fixed)

The Version table provides details related to the release that this issue/RFE will be addressed.

Unresolved : Release in which this issue/RFE will be addressed.
Resolved: Release in which this issue/RFE has been resolved.
Fixed : Release in which this issue/RFE has been fixed. The release containing this fix may be available for download as an Early Access Release or a General Availability Release.

To download the current JDK release, click here.

JDK 7	Other
7Fixed	hs21Fixed

C2 compiler crashes in PhaseIdealLoop::build_loop_late_post
This is reproducible easily and quickly after customer application launches.

V  [libjvm.so+0x6ea4d8] [0x6e988c]'void PhaseIdealLoop::build_loop_late_post(Node*)'+0xc4c
V  [libjvm.so+0x6e9838] [0x6e95c4]'void PhaseIdealLoop::build_loop_late(VectorSet&,Node_List&,Node_Stack&)'+0x274
V  [libjvm.so+0x6e785c] [0x6e7098]'void PhaseIdealLoop::build_and_optimize(bool,bool)'+0x7c4
V  [libjvm.so+0x1d93f4] [0x1d907c]'void Compile::Optimize()'+0x378
V  [libjvm.so+0x428b5c] [0x427db8]'Compile::Compile(ciEnv*,C2Compiler*,ciMethod*,int,bool,bool)'+0xda4
V  [libjvm.so+0x1d3cf4] [0x1d3c4c]'void C2Compiler::compile_method(ciEnv*,ciMethod*,int)'+0xa8
V  [libjvm.so+0x1d4bd8] [0x1d417c]'void CompileBroker::invoke_compiler_on_method(CompileTask*)'+0xa5c
V  [libjvm.so+0x247e34] [0x2475d4]'void CompileBroker::compiler_thread_loop()'+0x860
V  [libjvm.so+0x858590] [0x858540]'void JavaThread::thread_main_inner()'+0x50
V  [libjvm.so+0x77d890] [0x77d738]'java_start'+0x158

for details see the attached error log and corefile

WORK AROUND Also, remove -XX:-ReduceFieldZeroing
28-06-2011
EVALUATION No action required in JDK6 right now, workaround in place. Future JDK 6 updates will inherit the fix via hotspot express. Closing the 6-pool entry as will not fix, nothing required.
28-06-2011
EVALUATION http://hg.openjdk.java.net/jdk7/hotspot-rt/hotspot/rev/66b0e2371912
05-05-2011
EVALUATION http://hg.openjdk.java.net/jdk7/hotspot-comp/hotspot/rev/66b0e2371912
21-04-2011
EVALUATION Confirmed there's no problem when removing ReduceFieldZeroing. They had previously been asked to remove that as it was a workaround from a previous case, but hadn't done so until presented with this analysis!
19-04-2011
EVALUATION On 07/04/2011 20:45, Tom Rodriguez wrote: So I managed to reproduce it with replay and it's an EA bug. It seems to be connected to their use of the -XX:-ReduceFieldZeroing flag. If you don't turn that off the bug doesn't occur. I decided to run CTW with that flag and got a couple failures so we can reproduce it outside of replay. /java/re/jdk/1.7.0/latest/binaries/solaris-amd64/fastdebug/bin/java -server -XX:-ReduceFieldZeroing -Xss4m -verify -XX:-ShowMessageBoxOnError -XX:+CompileTheWorld -XX:CompileTheWorldStartAt=1090 -Xbootclasspath/p:/net/sqenfs-1.sfbay/export1/comp/vm/testbase/CompileTheWorld/jarfiles/Dynamo4.1.0.jar -XX:CompileTheWorldStopAt=109 crashes with the idom problem. I debugged it a bit this morning and I think that the problem is that we're not actually eliminating some loads from the eliminated object because the phi structures is complex. I notice that LoadNode::split_through_phi will sometimes give up if it doesn't appear profitable. Doesn't that code always have to do a split? Otherwise it can give up on eliminating a load. I modified it to always do the split by ignoring the wins count and skipping the early when cnt == 3 and that seems to fix the bug. As far as the customer goes if they drop the -XX:-ReduceFieldZeroing flag they should be fine. Why are they running with -XX:+ReduceFieldZeroing -XX:+ReduceInitialCardMarks -XX:-ReduceBulkZeroing. There were some bugs with those flags and CMS but they are fixed by now I think.
07-04-2011
WORK AROUND -XX:-DoEscapeAnalysis
11-03-2011

Relates :	JDK-6984348 - Fix typo in escape.cpp
Relates :	JDK-6873799 - enable escape analysis by default
Relates :	JDK-7045506 - assert(!can_reshape \|\| !new_phi) failed: for igvn new phi should be hooked