This is a reland of 11045926ca
The first version of this CL uncovered a TSAN failure when not disabling
RCS anymore. The problem was that a background thread was still updating
RCS counters, while the main thread already dumped those values during
tear down. This CL fixes this by dumping counters only after all
background threads are stopped.
Original change's description:
> [test] Avoid disabling of RCS during runtime
>
> Test was flaky since disabling of RCS at runtime is unsafe. Some code
> (e.g. TRACE_GC) is run only in case RCS is enabled and such code paths
> might also DCHECK that RCS is enabled.
> A background thread (sweeping in this case) could've already
> entered such a code path right before RCS is disabled. In this case the
> guard at the entry still saw that RCS is enabled but subsequent DCHECKs
> that ensure that RCS is enabled might fail.
>
> We could fix this by completing the sweeper tasks before disabling RCS
> but there might be other similar tasks which might be run at that point
> in the future (e.g. memory unmapper).
>
> Disabling the flag doesn't even seem to be needed for the initial
> regression test, so it seems simpler to just not disable RCS anymore.
>
> Bug: v8:12026
> Change-Id: I878f7dd9a7a4abb6a501f7a7651a1240ef2082a6
> Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3162043
> Reviewed-by: Michael Lippautz <mlippautz@chromium.org>
> Commit-Queue: Dominik Inführ <dinfuehr@chromium.org>
> Cr-Commit-Position: refs/heads/main@{#76845}
Bug: v8:12026, v8:12234
Change-Id: I5cb577051d188d17b02ca7da79a3f1d9aa646cae
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3162050
Commit-Queue: Dominik Inführ <dinfuehr@chromium.org>
Reviewed-by: Jakob Gruber <jgruber@chromium.org>
Reviewed-by: Michael Lippautz <mlippautz@chromium.org>
Cr-Commit-Position: refs/heads/main@{#76859}
b8146576d6