GitLab

Object size and mark bit reside in the same half word. Reading the size
during marking causes data races. Specifically, the size is read from
the write barrier and from Member's CheckPointer method.
Adding an atomic version to accessing the size. This version is used by
FindHeaderFromObject in the write barrier slow path and by CheckPointer
in member methods other than allocation (members during allocation will
not be traced yet so atomics are not needed there).
Note that since the size never changed during marking (added a DCHECK to
verify that incremental marking is off during size change), it is safe
to read the size relaxed.

Bug: 986235
Change-Id: I73019f9f9bfeb9562d4d1f29fc020a9b6f71866a
Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/1893202
Commit-Queue: Omer Katz <omerkatz@chromium.org>
Reviewed-by: Kentaro Hara <haraken@chromium.org>
Reviewed-by: Michael Lippautz <mlippautz@chromium.org>
Cr-Commit-Position: refs/heads/master@{#712551}