8322859: Parallel: Move transform_stack_chunk

albertnetymk · pull[bot] · commit 8294241bac59 · 2025-01-15T16:17:40.000Z
Reviewed-by: stefank, tschatzl
diff --git a/src/hotspot/share/gc/parallel/psPromotionManager.inline.hpp b/src/hotspot/share/gc/parallel/psPromotionManager.inline.hpp
@@ -249,10 +249,6 @@ inline oop PSPromotionManager::copy_unmarked_to_survivor_space(oop o,
   // Copy obj
   Copy::aligned_disjoint_words(cast_from_oop<HeapWord*>(o), cast_from_oop<HeapWord*>(new_obj), new_obj_size);
 
-  // Parallel GC claims with a release - so other threads might access this object
-  // after claiming and they should see the "completed" object.
-  ContinuationGCSupport::transform_stack_chunk(new_obj);
-
   // Now we have to CAS in the header.
   // Because the forwarding is done with memory_order_relaxed there is no
   // ordering with the above copy.  Clients that get the forwardee must not
@@ -271,6 +267,8 @@ inline oop PSPromotionManager::copy_unmarked_to_survivor_space(oop o,
       assert(young_space()->contains(new_obj), "Attempt to push non-promoted obj");
     }
 
+    ContinuationGCSupport::transform_stack_chunk(new_obj);
+
     // Do the size comparison first with new_obj_size, which we
     // already have. Hopefully, only a few objects are larger than
     // _min_array_size_for_chunking, and most of them will be arrays.
