Prevent race between GC and interrupts from other threads.

src/heap.cc

Index: src/heap.cc
diff --git a/src/heap.cc b/src/heap.cc
index e0b312067fc5c38013c36d76da9f69beae64c4a4..a40d5f18ff609942d6ee5999ce9eb19340426121 100644
--- a/src/heap.cc
+++ b/src/heap.cc
@@ -6081,7 +6081,13 @@ void Heap::IterateWeakRoots(ObjectVisitor* v, VisitMode mode) {
 
 
 void Heap::IterateStrongRoots(ObjectVisitor* v, VisitMode mode) {
-  v->VisitPointers(&roots_[0], &roots_[kStrongRootListLength]);
+  { // Avoid race with interrupts that set the stack limit on the root list.
+    // Note that GC does not actually care about the value of the stack limit
+    // as long as it is a smi.  We may want to revisit this if this causes
+    // performance regressions by blocking the optimizing compiler thread.
+    ExecutionAccess access(isolate());

[Michael Starzinger, 2014/02/07 10:24:50]

As discussed offline: This seems very much like a redundant lock to me. I know
the motivation is to make TSAN happy, but I have issues with taking a lock in
IterateStringRoots, especially for a field that is _always_ a SMI.

One suggestion from my end would be to reorder the strong roots list in a way
that groups SMIs at the end so that the GC doesn't need to look at them at all.

+    v->VisitPointers(&roots_[0], &roots_[kStrongRootListLength]);
+  }
   v->Synchronize(VisitorSynchronization::kStrongRootList);
 
   v->VisitPointer(BitCast<Object**>(&hidden_string_));