Does openjdk8u need pre-barrier for jweak?
Aleksey Shipilev
shade at redhat.com
Wed Aug 15 16:59:29 UTC 2018
Hi Liu,
Yes, I am committer in 8u. But this would require approval from current 8u maintainers. The
maintainers for jdk8u project would normally look at what GC devs said in patch review for
8u-specific patch. So, any patch to G1 in 8u should be RFR'ed on this list (hotspot-gc-dev@) first,
to check with G1 folks that is a proper way to go.
G1 and Shenandoah are definitely affected. I am not sure CMS has this problem: IIRC, remark is
supposed to find live-but-disconnected objects like these.
-Aleksey
On 08/15/2018 06:42 PM, Liu, Xin wrote:
> Agree. I will try to use the similar idea in G1. Aleksey, are you committer of JDK8u? I need to find a sponsor.
> IMHO, all concurrent GCs have this race condition, right? Shall we fix CMS?
>
> Thanks,
> --lx
>
> On 8/15/18, 12:50 AM, "Aleksey Shipilev" <shade at redhat.com> wrote:
>
> Hi Kim,
>
> On 08/15/2018 07:01 AM, Kim Barrett wrote:
> > However, I think there might be a much simpler solution that retains
> > at least some clearing of jweaks. Change the initial mark pause to
> > always scan and mark the JNI weak globals too, just as the normal JNI
> > globals are marked. Leave process_phaseJNI alone (i.e. don't use
> > Shenandoah's patch.)
>
> That is actually a much saner idea for Shenandoah as well, thanks! Should have thought about it at
> the time. We have moved the workaround block to root processor, which can also distinguish between
> root processing for concurrent mark and last-ditch STW Full GC:
> http://mail.openjdk.java.net/pipermail/shenandoah-dev/2018-August/007094.html
>
> I think/agree we can apply similar workaround to G1, and move on.
>
> -Aleksey
>
>
>
>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 833 bytes
Desc: OpenPGP digital signature
URL: <https://mail.openjdk.org/pipermail/hotspot-gc-dev/attachments/20180815/64b6a5cb/signature.asc>
More information about the hotspot-gc-dev
mailing list