RFR(S): 8149834: gc/shared/gcTimer.cpp:88 assert(_is_concurrent_phase_active) failed: A concurrent phase is not active

sangheon sangheon.kim at oracle.com
Thu Mar 3 07:41:27 UTC 2016


Hi all,

Could I have a couple of reviews for this quick fix?

There is a race between VMThread and ConcurrentMarkThread for 
ConcurrentGCTimer.
At the end of abort, VMThread is trying to end concurrent timer and if 
ConcurrentMarkThread is trying to start or end concurrent phase, timer 
related asserts will be fired.
(Only ConcurrentMarkThread starts a concurrent phase)
We have 3 different cases but the root cause is same[1].

This proposal is introducing 3 phases of started, not started and 
stopping for concurrent phase status.
And the status is updated by cmpxchg.

However, I think more proper fix would be eliminating the race.
Currently G1CollectedHeap has ConcurrentGCTimer but it is mostly used 
from ConcurrentMarkThread.
So moving the timer and related routines to ConcurrentMark seems better.
I filed a new RFE for this[2].

Many thanks to Jon, Bengt(base patch as well) and StefanK for the 
discussion.

CR: https://bugs.openjdk.java.net/browse/JDK-8151085
Webrev: http://cr.openjdk.java.net/~sangheki/8149834/webrev.00
Testing: JPRT, local test with adding some sleep at vm code.

[1] Related bugs:
https://bugs.openjdk.java.net/browse/JDK-8145996
https://bugs.openjdk.java.net/browse/JDK-8150819

[2] RFE for proper fix:
https://bugs.openjdk.java.net/browse/JDK-8151085

Thanks,
Sangheon






More information about the hotspot-gc-dev mailing list