--- a/hotspot/src/share/vm/gc/cms/concurrentMarkSweepGeneration.cpp	Tue May 03 07:23:23 2016 -0700
+++ b/hotspot/src/share/vm/gc/cms/concurrentMarkSweepGeneration.cpp	Thu May 05 08:51:27 2016 -0700
@@ -1019,7 +1019,7 @@
       return NULL;
     }
   }
-  assert(promoInfo->has_spooling_space(), "Control point invariant");
+  assert(!promoInfo->tracking() || promoInfo->has_spooling_space(), "Control point invariant");
   const size_t alloc_sz = CompactibleFreeListSpace::adjustObjectSize(word_sz);
   HeapWord* obj_ptr = ps->lab.alloc(alloc_sz);
   if (obj_ptr == NULL) {
@@ -1094,6 +1094,12 @@
   CMSParGCThreadState* ps = _par_gc_thread_states[thread_num];
   ParScanWithoutBarrierClosure* dummy_cl = NULL;
   ps->promo.promoted_oops_iterate_nv(dummy_cl);
+
+  // Because card-scanning has been completed, subsequent phases
+  // (e.g., reference processing) will not need to recognize which
+  // objects have been promoted during this GC. So, we can now disable
+  // promotion tracking.
+  ps->promo.stopTrackingPromotions();
 }
 
 bool ConcurrentMarkSweepGeneration::should_collect(bool   full,
@@ -2032,6 +2038,12 @@
   _capacity_at_prologue = capacity();
   _used_at_prologue = used();
 
+  // We enable promotion tracking so that card-scanning can recognize
+  // which objects have been promoted during this GC and skip them.
+  for (uint i = 0; i < ParallelGCThreads; i++) {
+    _par_gc_thread_states[i]->promo.startTrackingPromotions();
+  }
+
   // Delegate to CMScollector which knows how to coordinate between
   // this and any other CMS generations that it is responsible for
   // collecting.
@@ -2118,9 +2130,15 @@
 void ConcurrentMarkSweepGeneration::gc_epilogue(bool full) {
   collector()->gc_epilogue(full);
 
-  // Also reset promotion tracking in par gc thread states.
+  // When using ParNew, promotion tracking should have already been
+  // disabled. However, the prologue (which enables promotion
+  // tracking) and epilogue are called irrespective of the type of
+  // GC. So they will also be called before and after Full GCs, during
+  // which promotion tracking will not be explicitly disabled. So,
+  // it's safer to also disable it here too (to be symmetric with
+  // enabling it in the prologue).
   for (uint i = 0; i < ParallelGCThreads; i++) {
-    _par_gc_thread_states[i]->promo.stopTrackingPromotions(i);
+    _par_gc_thread_states[i]->promo.stopTrackingPromotions();
   }
 }
 
@@ -2431,9 +2449,6 @@
 void ConcurrentMarkSweepGeneration::save_marks() {
   // delegate to CMS space
   cmsSpace()->save_marks();
-  for (uint i = 0; i < ParallelGCThreads; i++) {
-    _par_gc_thread_states[i]->promo.startTrackingPromotions();
-  }
 }
 
 bool ConcurrentMarkSweepGeneration::no_allocs_since_save_marks() {

--- a/hotspot/src/share/vm/gc/cms/parNewGeneration.cpp	Tue May 03 07:23:23 2016 -0700
+++ b/hotspot/src/share/vm/gc/cms/parNewGeneration.cpp	Thu May 05 08:51:27 2016 -0700
@@ -476,7 +476,6 @@
 
     // Inform old gen that we're done.
     _old_gen.par_promote_alloc_done(i);
-    _old_gen.par_oop_since_save_marks_iterate_done(i);
   }
 
   if (UseConcMarkSweepGC) {
@@ -619,6 +618,16 @@
 
   // "evacuate followers".
   par_scan_state.evacuate_followers_closure().do_void();
+
+  // This will collapse this worker's promoted object list that's
+  // created during the main ParNew parallel phase of ParNew. This has
+  // to be called after all workers have finished promoting objects
+  // and scanning promoted objects. It should be safe calling it from
+  // here, given that we can only reach here after all thread have
+  // offered termination, i.e., after there is no more work to be
+  // done. It will also disable promotion tracking for the rest of
+  // this GC as it's not necessary to be on during reference processing.
+  _old_gen->par_oop_since_save_marks_iterate_done((int) worker_id);
 }
 
 ParNewGeneration::ParNewGeneration(ReservedSpace rs, size_t initial_byte_size)

--- a/hotspot/src/share/vm/gc/cms/promotionInfo.cpp	Tue May 03 07:23:23 2016 -0700
+++ b/hotspot/src/share/vm/gc/cms/promotionInfo.cpp	Thu May 05 08:51:27 2016 -0700
@@ -252,13 +252,15 @@
 }
 
 void PromotionInfo::startTrackingPromotions() {
+  assert(noPromotions(), "sanity");
   assert(_spoolHead == _spoolTail && _firstIndex == _nextIndex,
          "spooling inconsistency?");
   _firstIndex = _nextIndex = 1;
   _tracking = true;
 }
 
-void PromotionInfo::stopTrackingPromotions(uint worker_id) {
+void PromotionInfo::stopTrackingPromotions() {
+  assert(noPromotions(), "we should have torn down the lists by now");
   assert(_spoolHead == _spoolTail && _firstIndex == _nextIndex,
          "spooling inconsistency?");
   _firstIndex = _nextIndex = 1;

--- a/hotspot/src/share/vm/gc/cms/promotionInfo.hpp	Tue May 03 07:23:23 2016 -0700
+++ b/hotspot/src/share/vm/gc/cms/promotionInfo.hpp	Thu May 05 08:51:27 2016 -0700
@@ -144,7 +144,7 @@
     return _promoHead == NULL;
   }
   void startTrackingPromotions();
-  void stopTrackingPromotions(uint worker_id = 0);
+  void stopTrackingPromotions();
   bool tracking() const          { return _tracking;  }
   void track(PromotedObject* trackOop);      // keep track of a promoted oop
   // The following variant must be used when trackOop is not fully