heap: Resolve data race in InAtomicMarkingPause

Incremental marking goes in and out of AtomicPauseScope during marking.
Checking InAtomicPause while concurrent marking can race with the
incremental marker exiting AtomicPauseScope.
Checking InAtomicMarkingPause only if !IsInIncrementalMarking checks
the same condition overall but concurrent markers won't need to check
InAtomicMarkingPause because IsInIncrementalMarking will always be true
while concurrent marking is active.

Bug: 986235
Change-Id: I6d21843b25db553dfcd0ae0e95ec3c43eaa99d49
Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/1893851Reviewed-by: Michael Lippautz <mlippautz@chromium.org>
Commit-Queue: Omer Katz <omerkatz@chromium.org>
Cr-Commit-Position: refs/heads/master@{#711421}

third_party/blink/renderer/platform/heap/marking_visitor.h

@@ -97,7 +97,7 @@ ALWAYS_INLINE void MarkingVisitorCommon::AccountMarkedBytes(
 ALWAYS_INLINE bool MarkingVisitorCommon::MarkHeaderNoTracing(
     HeapObjectHeader* header) {
   DCHECK(header);
-  DCHECK(State()->InAtomicMarkingPause() || State()->IsIncrementalMarking());
+  DCHECK(State()->IsIncrementalMarking() || State()->InAtomicMarkingPause());
   // A GC should only mark the objects that belong in its heap.
   DCHECK_EQ(State(),
             PageFromObject(header->Payload())->Arena()->GetThreadState());