RFR: 8302462: [REDO] 8297487: G1 Remark: no need to keep alive oop constants of nmethods on stack
Albert Mingkun Yang
ayang at openjdk.org
Tue Feb 14 20:57:44 UTC 2023
On Tue, 14 Feb 2023 15:38:54 GMT, Richard Reingruber <rrich at openjdk.org> wrote:
>> This is a REDO of JDK-8297487 after the issues that cause the BACKOUT were resolved with [JDK-8300915](https://bugs.openjdk.org/browse/JDK-8300915)
>>
>> Testing:
>> * The reproducer `test/langtools:tier1` was executed without crash for 12h.
>> * Multiple iterations of jtreg tests tiers 1-4 on standard platforms and also on ppc64le.
>
> src/hotspot/share/gc/shared/barrierSet.cpp line 64:
>
>> 62: // the constant pool of nmethods as part of the SATB snapshot, and to deal with
>> 63: // compiled frames contained in continuation stack chunks allocated after
>> 64: // concurent mark start.
>
> I've adapted the comment after https://mail.openjdk.org/pipermail/hotspot-gc-dev/2023-January/041194.html
I feel the comment is at the wrong level of abstraction, compared with the surrounding code.
Maybe just `Provide nmethod entry barrier to support concurrent marking and code cache unloading.` at the top of the whole method.
The more elaborate "inside" about nmethod-entry-barrier probably makes more sense at `BarrierSetNMethod::nmethod_entry_barrier`.
-------------
PR: https://git.openjdk.org/jdk/pull/12561
More information about the hotspot-gc-dev
mailing list