RFR(S): 8050973: CMS/G1 GC: add missing Resource and Handle Mark
Lindenmaier, Goetz
goetz.lindenmaier at sap.com
Mon Jul 21 13:01:10 UTC 2014
Hi Thomas,
we put the marks there as other work functions are setup similarly, as, e.g.,
CMConcurrentMarkingTask::work(), CMRemarkTask::work().
So we propose to fist go with our change to fix the issue, it can
be refactored afterwards.
It's a minor resource leak.
We don't think it's a showstopper, so let's only move it to 8u40.
Best regards,
Martin and Goetz.
-----Original Message-----
From: Thomas Schatzl [mailto:thomas.schatzl at oracle.com]
Sent: Montag, 21. Juli 2014 10:55
To: Lindenmaier, Goetz
Cc: hotspot-dev at openjdk.java.net
Subject: Re: RFR(S): 8050973: CMS/G1 GC: add missing Resource and Handle Mark
Hi,
On Fri, 2014-07-18 at 12:47 +0000, Lindenmaier, Goetz wrote:
> Hi,
>
> This fixes two missing Resource and Handle marks.
>
> Please review and test this change. We please need a sponsor
> to push it.
> http://cr.openjdk.java.net/~goetz/webrevs/8050973-mark/webrev-01/
I think the resource/handle marks should be added before the work method
is called for safety.
So I would prefer if the Marks were added to at least in the two places
in GangWorker::loop() and YieldingFlexibleGangWorker::loop() where the
work method is called, if not everywhere where work() is called.
The latter might be too big a change for now.
> Should this be pushed to 8u20?
Can you elaborate the consequences of not applying the patch to 8u20?
Only showstoppers can be pushed to 8u20 at this time, so we need a good
reason.
It seems to be a problem with ParallelRefProcEnabled only, and we have
not seen crashes.
Thanks,
Thomas
More information about the hotspot-dev
mailing list