diff --git a/src/heap.cpp b/src/heap.cpp
index 4c81b0d6ef..a70f176b3e 100644
--- a/src/heap.cpp
+++ b/src/heap.cpp
@@ -1709,6 +1709,11 @@ class MyHeap: public Heap {
     }
   }
 
+  virtual bool needsMark(void* p, unsigned offset) {
+    return needsMark(p) and targetNeedsMark
+      (mask(*(static_cast<void**>(p) + offset)));
+  }
+
   bool targetNeedsMark(void* target) {
     return target
       and not c.gen2.contains(target)
diff --git a/src/heap.h b/src/heap.h
index 8b4331ca1e..bf4dc5a360 100644
--- a/src/heap.h
+++ b/src/heap.h
@@ -62,6 +62,7 @@ class Heap: public Allocator {
   virtual void* allocateImmortal(Allocator* allocator, unsigned sizeInWords,
                                  bool objectMask, unsigned* totalInBytes) = 0;
   virtual bool needsMark(void* p) = 0;
+  virtual bool needsMark(void* p, unsigned offset) = 0;
   virtual void mark(void* p, unsigned offset, unsigned count) = 0;
   virtual void pad(void* p) = 0;
   virtual void* follow(void* p) = 0;
diff --git a/src/machine.h b/src/machine.h
index 4077745d61..cca8a3b564 100644
--- a/src/machine.h
+++ b/src/machine.h
@@ -1487,7 +1487,10 @@ mark(Thread* t, object o, unsigned offset, unsigned count)
 inline void
 mark(Thread* t, object o, unsigned offset)
 {
-  mark(t, o, offset, 1);
+  if (t->m->heap->needsMark(o, offset / BytesPerWord)) {
+    ACQUIRE_RAW(t, t->m->heapLock);
+    t->m->heap->mark(o, offset / BytesPerWord, 1);
+  }
 }
 
 inline void