RFR: 8203028: Simplify reference processing in light of JDK-8175797
thomas.schatzl at oracle.com
Tue May 15 08:17:28 UTC 2018
On Mon, 2018-05-14 at 18:09 -0400, Kim Barrett wrote:
> > On May 14, 2018, at 8:45 AM, Thomas Schatzl <thomas.schatzl at oracle.
> > com> wrote:
> > Hi Kim,
> > some initial comments…
> Thanks for looking at this.
> > On Sun, 2018-05-13 at 21:32 -0400, Kim Barrett wrote:
> > > […]
> > > CR:
> > - please remove the unused complete_gc parameter from
> > ReferenceProcessor::process_phase2.
> I intentionally left it alone. Removing it completely starts getting
> messy when one starts walking up the callers and hits the parallel
> task layer (either drop the argument there instead, or have two
> different work functions with consequent fannout). And I expect this
> interface is going to be completely changed by JDK-8202845, so it
> didn't seem worth doing much here.
Oh, I would have simply kept the parameter in the ProcessTask layer.
You are right, we want to discuss this later though.
Thanks for all the other clarifications in the documentation.
> New webrev, rebased to include JDK-8201491:
> New webrevs with above changes:
> full: http://cr.openjdk.java.net/~kbarrett/8203028/open.02/
> incr: http://cr.openjdk.java.net/~kbarrett/8203028/open.02.inc/
maybe some really stupid question, but: do we really not need to
update the next field if we discovered an (inactive) Reference? (in
instanceRefKlass.inline.hpp, in oop_oop_iterate_discovery).
One situation that comes to my mind is that while a (Final-)Reference
is on the pending list it might still be moved (concurrent
discovery/Remark puts it on the pending list, some mixed gc might move
it later). While nobody seems to be touching that "next" reference any
more for FinalReferences after being marked that way, it seems bad
style to have a dangling oop (and probably messes up some heap
Another is situation is that the next field is used as a link in the
ReferenceQueue (for non-FinalReference), so the next field is always be
a valid oop (and fixed up), shouldn't it? (Since these References
referent is NULL, they won't be discovered, but the next field needs to
be iterated over?)
I do not think the next field can be the only source of a new
undiscovered object graph (it is either NULL or self-looped or part of
a ReferenceQueue) though.
More information about the hotspot-gc-dev