6700789: G1: Enable use of compressed oops with G1 heaps

Modifications to G1 so as to allow the use of compressed oops.

Reviewed-by: apetrusenko, coleenp, jmasa, kvn, never, phh, tonyp
diff --git a/hotspot/src/cpu/sparc/vm/assembler_sparc.cpp b/hotspot/src/cpu/sparc/vm/assembler_sparc.cpp
index f2d5878..b53cc4f 100644
--- a/hotspot/src/cpu/sparc/vm/assembler_sparc.cpp
+++ b/hotspot/src/cpu/sparc/vm/assembler_sparc.cpp
@@ -4208,6 +4208,7 @@
                   PtrQueue::byte_offset_of_active()),
          tmp);
   }
+
   // Check on whether to annul.
   br_on_reg_cond(rc_z, /*annul*/false, Assembler::pt, tmp, filtered);
   delayed() -> nop();
@@ -4215,13 +4216,13 @@
   // satb_log_barrier_work1(tmp, offset);
   if (index == noreg) {
     if (Assembler::is_simm13(offset)) {
-      ld_ptr(obj, offset, tmp);
+      load_heap_oop(obj, offset, tmp);
     } else {
       set(offset, tmp);
-      ld_ptr(obj, tmp, tmp);
+      load_heap_oop(obj, tmp, tmp);
     }
   } else {
-    ld_ptr(obj, index, tmp);
+    load_heap_oop(obj, index, tmp);
   }
 
   // satb_log_barrier_work2(obj, tmp, offset);
diff --git a/hotspot/src/cpu/x86/vm/assembler_x86.cpp b/hotspot/src/cpu/x86/vm/assembler_x86.cpp
index 174e542..4a67b42 100644
--- a/hotspot/src/cpu/x86/vm/assembler_x86.cpp
+++ b/hotspot/src/cpu/x86/vm/assembler_x86.cpp
@@ -6805,14 +6805,18 @@
   jcc(Assembler::equal, done);
 
   // if (x.f == NULL) goto done;
-  cmpptr(Address(obj, 0), NULL_WORD);
+#ifdef _LP64
+  load_heap_oop(tmp2, Address(obj, 0));
+#else
+  movptr(tmp2, Address(obj, 0));
+#endif
+  cmpptr(tmp2, (int32_t) NULL_WORD);
   jcc(Assembler::equal, done);
 
   // Can we store original value in the thread's buffer?
 
-  LP64_ONLY(movslq(tmp, index);)
-  movptr(tmp2, Address(obj, 0));
 #ifdef _LP64
+  movslq(tmp, index);
   cmpq(tmp, 0);
 #else
   cmpl(index, 0);
@@ -6834,8 +6838,7 @@
   if(tosca_live) push(rax);
   push(obj);
 #ifdef _LP64
-  movq(c_rarg0, Address(obj, 0));
-  call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::g1_wb_pre), c_rarg0, r15_thread);
+  call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::g1_wb_pre), tmp2, r15_thread);
 #else
   push(thread);
   call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::g1_wb_pre), tmp2, thread);
diff --git a/hotspot/src/cpu/x86/vm/methodHandles_x86.cpp b/hotspot/src/cpu/x86/vm/methodHandles_x86.cpp
index 24210bf..a45af4a 100644
--- a/hotspot/src/cpu/x86/vm/methodHandles_x86.cpp
+++ b/hotspot/src/cpu/x86/vm/methodHandles_x86.cpp
@@ -269,11 +269,11 @@
 
 #ifndef PRODUCT
 void trace_method_handle_stub(const char* adaptername,
-                              oop mh,
+                              oopDesc* mh,
                               intptr_t* entry_sp,
                               intptr_t* saved_sp) {
   // called as a leaf from native code: do not block the JVM!
-  printf("MH %s "PTR_FORMAT" "PTR_FORMAT" "INTX_FORMAT"\n", adaptername, mh, entry_sp, entry_sp - saved_sp);
+  printf("MH %s "PTR_FORMAT" "PTR_FORMAT" "INTX_FORMAT"\n", adaptername, (void*)mh, entry_sp, entry_sp - saved_sp);
 }
 #endif //PRODUCT
 
diff --git a/hotspot/src/cpu/x86/vm/stubGenerator_x86_32.cpp b/hotspot/src/cpu/x86/vm/stubGenerator_x86_32.cpp
index e38fc1d..3d6ca91 100644
--- a/hotspot/src/cpu/x86/vm/stubGenerator_x86_32.cpp
+++ b/hotspot/src/cpu/x86/vm/stubGenerator_x86_32.cpp
@@ -709,7 +709,7 @@
   //
   //  Input:
   //     start   -  starting address
-  //     end     -  element count
+  //     count   -  element count
   void  gen_write_ref_array_pre_barrier(Register start, Register count) {
     assert_different_registers(start, count);
     BarrierSet* bs = Universe::heap()->barrier_set();
@@ -757,7 +757,6 @@
           __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_post)));
           __ addptr(rsp, 2*wordSize);
           __ popa();
-
         }
         break;
 
diff --git a/hotspot/src/cpu/x86/vm/stubGenerator_x86_64.cpp b/hotspot/src/cpu/x86/vm/stubGenerator_x86_64.cpp
index ec322b5..6b07314 100644
--- a/hotspot/src/cpu/x86/vm/stubGenerator_x86_64.cpp
+++ b/hotspot/src/cpu/x86/vm/stubGenerator_x86_64.cpp
@@ -1207,9 +1207,9 @@
           __ pusha();                      // push registers (overkill)
           // must compute element count unless barrier set interface is changed (other platforms supply count)
           assert_different_registers(start, end, scratch);
-          __ lea(scratch, Address(end, wordSize));
-          __ subptr(scratch, start);
-          __ shrptr(scratch, LogBytesPerWord);
+          __ lea(scratch, Address(end, BytesPerHeapOop));
+          __ subptr(scratch, start);               // subtract start to get #bytes
+          __ shrptr(scratch, LogBytesPerHeapOop);  // convert to element count
           __ mov(c_rarg0, start);
           __ mov(c_rarg1, scratch);
           __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_post)));
@@ -1225,6 +1225,7 @@
           Label L_loop;
 
            __ shrptr(start, CardTableModRefBS::card_shift);
+           __ addptr(end, BytesPerHeapOop);
            __ shrptr(end, CardTableModRefBS::card_shift);
            __ subptr(end, start); // number of bytes to copy
 
@@ -2251,6 +2252,7 @@
     // and report their number to the caller.
     assert_different_registers(rax, r14_length, count, to, end_to, rcx);
     __ lea(end_to, to_element_addr);
+    __ addptr(end_to, -heapOopSize);      // make an inclusive end pointer
     gen_write_ref_array_post_barrier(to, end_to, rscratch1);
     __ movptr(rax, r14_length);           // original oops
     __ addptr(rax, count);                // K = (original - remaining) oops
@@ -2259,7 +2261,7 @@
 
     // Come here on success only.
     __ BIND(L_do_card_marks);
-    __ addptr(end_to, -wordSize);         // make an inclusive end pointer
+    __ addptr(end_to, -heapOopSize);         // make an inclusive end pointer
     gen_write_ref_array_post_barrier(to, end_to, rscratch1);
     __ xorptr(rax, rax);                  // return 0 on success
 
diff --git a/hotspot/src/share/vm/gc_implementation/g1/bufferingOopClosure.hpp b/hotspot/src/share/vm/gc_implementation/g1/bufferingOopClosure.hpp
index 1124e5d..b9283bd 100644
--- a/hotspot/src/share/vm/gc_implementation/g1/bufferingOopClosure.hpp
+++ b/hotspot/src/share/vm/gc_implementation/g1/bufferingOopClosure.hpp
@@ -42,35 +42,40 @@
     BufferLength = 1024
   };
 
-  oop          *_buffer[BufferLength];
-  oop         **_buffer_top;
-  oop         **_buffer_curr;
+  StarTask  _buffer[BufferLength];
+  StarTask* _buffer_top;
+  StarTask* _buffer_curr;
 
-  OopClosure  *_oc;
-  double       _closure_app_seconds;
+  OopClosure* _oc;
+  double      _closure_app_seconds;
 
   void process_buffer () {
-
     double start = os::elapsedTime();
-    for (oop **curr = _buffer; curr < _buffer_curr; ++curr) {
-      _oc->do_oop(*curr);
+    for (StarTask* curr = _buffer; curr < _buffer_curr; ++curr) {
+      if (curr->is_narrow()) {
+        assert(UseCompressedOops, "Error");
+        _oc->do_oop((narrowOop*)(*curr));
+      } else {
+        _oc->do_oop((oop*)(*curr));
+      }
     }
     _buffer_curr = _buffer;
     _closure_app_seconds += (os::elapsedTime() - start);
   }
 
-public:
-  virtual void do_oop(narrowOop* p) {
-    guarantee(false, "NYI");
-  }
-  virtual void do_oop(oop *p) {
+  template <class T> inline void do_oop_work(T* p) {
     if (_buffer_curr == _buffer_top) {
       process_buffer();
     }
-
-    *_buffer_curr = p;
+    StarTask new_ref(p);
+    *_buffer_curr = new_ref;
     ++_buffer_curr;
   }
+
+public:
+  virtual void do_oop(narrowOop* p) { do_oop_work(p); }
+  virtual void do_oop(oop* p)       { do_oop_work(p); }
+
   void done () {
     if (_buffer_curr > _buffer) {
       process_buffer();
@@ -88,18 +93,17 @@
 class BufferingOopsInGenClosure: public OopsInGenClosure {
   BufferingOopClosure _boc;
   OopsInGenClosure* _oc;
-public:
+ protected:
+  template <class T> inline void do_oop_work(T* p) {
+    assert(generation()->is_in_reserved((void*)p), "Must be in!");
+    _boc.do_oop(p);
+  }
+ public:
   BufferingOopsInGenClosure(OopsInGenClosure *oc) :
     _boc(oc), _oc(oc) {}
 
-  virtual void do_oop(narrowOop* p) {
-    guarantee(false, "NYI");
-  }
-
-  virtual void do_oop(oop* p) {
-    assert(generation()->is_in_reserved(p), "Must be in!");
-    _boc.do_oop(p);
-  }
+  virtual void do_oop(narrowOop* p) { do_oop_work(p); }
+  virtual void do_oop(oop* p)       { do_oop_work(p); }
 
   void done() {
     _boc.done();
@@ -130,14 +134,14 @@
     BufferLength = 1024
   };
 
-  oop                      *_buffer[BufferLength];
-  oop                     **_buffer_top;
-  oop                     **_buffer_curr;
+  StarTask     _buffer[BufferLength];
+  StarTask*    _buffer_top;
+  StarTask*    _buffer_curr;
 
-  HeapRegion               *_hr_buffer[BufferLength];
-  HeapRegion              **_hr_curr;
+  HeapRegion*  _hr_buffer[BufferLength];
+  HeapRegion** _hr_curr;
 
-  OopsInHeapRegionClosure  *_oc;
+  OopsInHeapRegionClosure*  _oc;
   double                    _closure_app_seconds;
 
   void process_buffer () {
@@ -146,15 +150,20 @@
            "the two lengths should be the same");
 
     double start = os::elapsedTime();
-    HeapRegion **hr_curr = _hr_buffer;
-    HeapRegion *hr_prev = NULL;
-    for (oop **curr = _buffer; curr < _buffer_curr; ++curr) {
-      HeapRegion *region = *hr_curr;
+    HeapRegion** hr_curr = _hr_buffer;
+    HeapRegion*  hr_prev = NULL;
+    for (StarTask* curr = _buffer; curr < _buffer_curr; ++curr) {
+      HeapRegion* region = *hr_curr;
       if (region != hr_prev) {
         _oc->set_region(region);
         hr_prev = region;
       }
-      _oc->do_oop(*curr);
+      if (curr->is_narrow()) {
+        assert(UseCompressedOops, "Error");
+        _oc->do_oop((narrowOop*)(*curr));
+      } else {
+        _oc->do_oop((oop*)(*curr));
+      }
       ++hr_curr;
     }
     _buffer_curr = _buffer;
@@ -163,17 +172,16 @@
   }
 
 public:
-  virtual void do_oop(narrowOop *p) {
-    guarantee(false, "NYI");
-  }
+  virtual void do_oop(narrowOop* p) { do_oop_work(p); }
+  virtual void do_oop(      oop* p) { do_oop_work(p); }
 
-  virtual void do_oop(oop *p) {
+  template <class T> void do_oop_work(T* p) {
     if (_buffer_curr == _buffer_top) {
       assert(_hr_curr > _hr_buffer, "_hr_curr should be consistent with _buffer_curr");
       process_buffer();
     }
-
-    *_buffer_curr = p;
+    StarTask new_ref(p);
+    *_buffer_curr = new_ref;
     ++_buffer_curr;
     *_hr_curr = _from;
     ++_hr_curr;
diff --git a/hotspot/src/share/vm/gc_implementation/g1/concurrentMark.cpp b/hotspot/src/share/vm/gc_implementation/g1/concurrentMark.cpp
index 20f001d..b19b457 100644
--- a/hotspot/src/share/vm/gc_implementation/g1/concurrentMark.cpp
+++ b/hotspot/src/share/vm/gc_implementation/g1/concurrentMark.cpp
@@ -452,13 +452,10 @@
   _regionStack.allocate(G1MarkRegionStackSize);
 
   // Create & start a ConcurrentMark thread.
-  if (G1ConcMark) {
-    _cmThread = new ConcurrentMarkThread(this);
-    assert(cmThread() != NULL, "CM Thread should have been created");
-    assert(cmThread()->cm() != NULL, "CM Thread should refer to this cm");
-  } else {
-    _cmThread = NULL;
-  }
+  _cmThread = new ConcurrentMarkThread(this);
+  assert(cmThread() != NULL, "CM Thread should have been created");
+  assert(cmThread()->cm() != NULL, "CM Thread should refer to this cm");
+
   _g1h = G1CollectedHeap::heap();
   assert(CGC_lock != NULL, "Where's the CGC_lock?");
   assert(_markBitMap1.covers(rs), "_markBitMap1 inconsistency");
@@ -783,18 +780,18 @@
                      bool do_barrier) : _cm(cm), _g1h(g1h),
                                         _do_barrier(do_barrier) { }
 
-  virtual void do_oop(narrowOop* p) {
-    guarantee(false, "NYI");
-  }
+  virtual void do_oop(narrowOop* p) { do_oop_work(p); }
+  virtual void do_oop(      oop* p) { do_oop_work(p); }
 
-  virtual void do_oop(oop* p) {
-    oop thisOop = *p;
-    if (thisOop != NULL) {
-      assert(thisOop->is_oop() || thisOop->mark() == NULL,
+  template <class T> void do_oop_work(T* p) {
+    T heap_oop = oopDesc::load_heap_oop(p);
+    if (!oopDesc::is_null(heap_oop)) {
+      oop obj = oopDesc::decode_heap_oop_not_null(heap_oop);
+      assert(obj->is_oop() || obj->mark() == NULL,
              "expected an oop, possibly with mark word displaced");
-      HeapWord* addr = (HeapWord*)thisOop;
+      HeapWord* addr = (HeapWord*)obj;
       if (_g1h->is_in_g1_reserved(addr)) {
-        _cm->grayRoot(thisOop);
+        _cm->grayRoot(obj);
       }
     }
     if (_do_barrier) {
@@ -850,16 +847,6 @@
   double start = os::elapsedTime();
   GCOverheadReporter::recordSTWStart(start);
 
-  // If there has not been a GC[n-1] since last GC[n] cycle completed,
-  // precede our marking with a collection of all
-  // younger generations to keep floating garbage to a minimum.
-  // YSR: we won't do this for now -- it's an optimization to be
-  // done post-beta.
-
-  // YSR:    ignoring weak refs for now; will do at bug fixing stage
-  // EVM:    assert(discoveredRefsAreClear());
-
-
   G1CollectorPolicy* g1p = G1CollectedHeap::heap()->g1_policy();
   g1p->record_concurrent_mark_init_start();
   checkpointRootsInitialPre();
@@ -1135,6 +1122,13 @@
     return;
   }
 
+  if (VerifyDuringGC) {
+    HandleMark hm;  // handle scope
+    gclog_or_tty->print(" VerifyDuringGC:(before)");
+    Universe::heap()->prepare_for_verify();
+    Universe::verify(true, false, true);
+  }
+
   G1CollectorPolicy* g1p = g1h->g1_policy();
   g1p->record_concurrent_mark_remark_start();
 
@@ -1159,10 +1153,12 @@
     JavaThread::satb_mark_queue_set().set_active_all_threads(false);
 
     if (VerifyDuringGC) {
-      g1h->prepare_for_verify();
-      g1h->verify(/* allow_dirty */      true,
-                  /* silent */           false,
-                  /* use_prev_marking */ false);
+      HandleMark hm;  // handle scope
+      gclog_or_tty->print(" VerifyDuringGC:(after)");
+      Universe::heap()->prepare_for_verify();
+      Universe::heap()->verify(/* allow_dirty */      true,
+                               /* silent */           false,
+                               /* use_prev_marking */ false);
     }
   }
 
@@ -1658,6 +1654,15 @@
     return;
   }
 
+  if (VerifyDuringGC) {
+    HandleMark hm;  // handle scope
+    gclog_or_tty->print(" VerifyDuringGC:(before)");
+    Universe::heap()->prepare_for_verify();
+    Universe::verify(/* allow dirty  */ true,
+                     /* silent       */ false,
+                     /* prev marking */ true);
+  }
+
   _cleanup_co_tracker.disable();
 
   G1CollectorPolicy* g1p = G1CollectedHeap::heap()->g1_policy();
@@ -1790,10 +1795,12 @@
   g1h->increment_total_collections();
 
   if (VerifyDuringGC) {
-    g1h->prepare_for_verify();
-    g1h->verify(/* allow_dirty */      true,
-                /* silent */           false,
-                /* use_prev_marking */ true);
+    HandleMark hm;  // handle scope
+    gclog_or_tty->print(" VerifyDuringGC:(after)");
+    Universe::heap()->prepare_for_verify();
+    Universe::verify(/* allow dirty  */ true,
+                     /* silent       */ false,
+                     /* prev marking */ true);
   }
 }
 
@@ -1852,12 +1859,11 @@
     _g1(g1), _cm(cm),
     _bitMap(bitMap) {}
 
-  void do_oop(narrowOop* p) {
-    guarantee(false, "NYI");
-  }
+  virtual void do_oop(narrowOop* p) { do_oop_work(p); }
+  virtual void do_oop(      oop* p) { do_oop_work(p); }
 
-  void do_oop(oop* p) {
-    oop thisOop = *p;
+  template <class T> void do_oop_work(T* p) {
+    oop thisOop = oopDesc::load_decode_heap_oop(p);
     HeapWord* addr = (HeapWord*)thisOop;
     if (_g1->is_in_g1_reserved(addr) && _g1->is_obj_ill(thisOop)) {
       _bitMap->mark(addr);
@@ -2016,12 +2022,11 @@
   ReachablePrinterOopClosure(CMBitMapRO* bitmap, outputStream* out) :
     _bitmap(bitmap), _g1h(G1CollectedHeap::heap()), _out(out) { }
 
-  void do_oop(narrowOop* p) {
-    guarantee(false, "NYI");
-  }
+  void do_oop(narrowOop* p) { do_oop_work(p); }
+  void do_oop(      oop* p) { do_oop_work(p); }
 
-  void do_oop(oop* p) {
-    oop         obj = *p;
+  template <class T> void do_oop_work(T* p) {
+    oop         obj = oopDesc::load_decode_heap_oop(p);
     const char* str = NULL;
     const char* str2 = "";
 
@@ -2163,6 +2168,7 @@
 
 
   HeapWord* objAddr = (HeapWord*) obj;
+  assert(obj->is_oop_or_null(true /* ignore mark word */), "Error");
   if (_g1h->is_in_g1_reserved(objAddr)) {
     tmp_guarantee_CM( obj != NULL, "is_in_g1_reserved should ensure this" );
     HeapRegion* hr = _g1h->heap_region_containing(obj);
@@ -2380,7 +2386,7 @@
     }
   }
 
-  bool drain() {
+  template <class T> bool drain() {
     while (_ms_ind > 0) {
       oop obj = pop();
       assert(obj != NULL, "Since index was non-zero.");
@@ -2394,9 +2400,8 @@
         }
         // Now process this portion of this one.
         int lim = MIN2(next_arr_ind, len);
-        assert(!UseCompressedOops, "This needs to be fixed");
         for (int j = arr_ind; j < lim; j++) {
-          do_oop(aobj->obj_at_addr<oop>(j));
+          do_oop(aobj->obj_at_addr<T>(j));
         }
 
       } else {
@@ -2423,13 +2428,13 @@
     FREE_C_HEAP_ARRAY(jint, _array_ind_stack);
   }
 
-  void do_oop(narrowOop* p) {
-    guarantee(false, "NYI");
-  }
+  virtual void do_oop(narrowOop* p) { do_oop_work(p); }
+  virtual void do_oop(      oop* p) { do_oop_work(p); }
 
-  void do_oop(oop* p) {
-    oop obj = *p;
-    if (obj == NULL) return;
+  template <class T> void do_oop_work(T* p) {
+    T heap_oop = oopDesc::load_heap_oop(p);
+    if (oopDesc::is_null(heap_oop)) return;
+    oop obj = oopDesc::decode_heap_oop_not_null(heap_oop);
     if (obj->is_forwarded()) {
       // If the object has already been forwarded, we have to make sure
       // that it's marked.  So follow the forwarding pointer.  Note that
@@ -2478,7 +2483,11 @@
     oop obj = oop(addr);
     if (!obj->is_forwarded()) {
       if (!_oop_cl.push(obj)) return false;
-      if (!_oop_cl.drain()) return false;
+      if (UseCompressedOops) {
+        if (!_oop_cl.drain<narrowOop>()) return false;
+      } else {
+        if (!_oop_cl.drain<oop>()) return false;
+      }
     }
     // Otherwise...
     return true;
@@ -2636,9 +2645,6 @@
 
 // abandon current marking iteration due to a Full GC
 void ConcurrentMark::abort() {
-  // If we're not marking, nothing to do.
-  if (!G1ConcMark) return;
-
   // Clear all marks to force marking thread to do nothing
   _nextMarkBitMap->clearAll();
   // Empty mark stack
@@ -2814,14 +2820,14 @@
   CMTask*            _task;
 
 public:
-  void do_oop(narrowOop* p) {
-    guarantee(false, "NYI");
-  }
+  virtual void do_oop(narrowOop* p) { do_oop_work(p); }
+  virtual void do_oop(      oop* p) { do_oop_work(p); }
 
-  void do_oop(oop* p) {
+  template <class T> void do_oop_work(T* p) {
     tmp_guarantee_CM( _g1h->is_in_g1_reserved((HeapWord*) p), "invariant" );
+    tmp_guarantee_CM( !_g1h->heap_region_containing((HeapWord*) p)->is_on_free_list(), "invariant" );
 
-    oop obj = *p;
+    oop obj = oopDesc::load_decode_heap_oop(p);
     if (_cm->verbose_high())
       gclog_or_tty->print_cr("[%d] we're looking at location "
                              "*"PTR_FORMAT" = "PTR_FORMAT,
@@ -2967,6 +2973,7 @@
   ++_refs_reached;
 
   HeapWord* objAddr = (HeapWord*) obj;
+  assert(obj->is_oop_or_null(true /* ignore mark word */), "Error");
   if (_g1h->is_in_g1_reserved(objAddr)) {
     tmp_guarantee_CM( obj != NULL, "is_in_g1_reserved should ensure this" );
     HeapRegion* hr =  _g1h->heap_region_containing(obj);
@@ -3030,6 +3037,7 @@
 void CMTask::push(oop obj) {
   HeapWord* objAddr = (HeapWord*) obj;
   tmp_guarantee_CM( _g1h->is_in_g1_reserved(objAddr), "invariant" );
+  tmp_guarantee_CM( !_g1h->heap_region_containing(objAddr)->is_on_free_list(), "invariant" );
   tmp_guarantee_CM( !_g1h->is_obj_ill(obj), "invariant" );
   tmp_guarantee_CM( _nextMarkBitMap->isMarked(objAddr), "invariant" );
 
@@ -3275,6 +3283,8 @@
 
       tmp_guarantee_CM( _g1h->is_in_g1_reserved((HeapWord*) obj),
                         "invariant" );
+      tmp_guarantee_CM( !_g1h->heap_region_containing(obj)->is_on_free_list(),
+                        "invariant" );
 
       scan_object(obj);
 
diff --git a/hotspot/src/share/vm/gc_implementation/g1/concurrentMark.hpp b/hotspot/src/share/vm/gc_implementation/g1/concurrentMark.hpp
index 274f8c1..8547717 100644
--- a/hotspot/src/share/vm/gc_implementation/g1/concurrentMark.hpp
+++ b/hotspot/src/share/vm/gc_implementation/g1/concurrentMark.hpp
@@ -763,6 +763,7 @@
   CMBitMap*                   _nextMarkBitMap;
   // the task queue of this task
   CMTaskQueue*                _task_queue;
+private:
   // the task queue set---needed for stealing
   CMTaskQueueSet*             _task_queues;
   // indicates whether the task has been claimed---this is only  for
diff --git a/hotspot/src/share/vm/gc_implementation/g1/g1BlockOffsetTable.cpp b/hotspot/src/share/vm/gc_implementation/g1/g1BlockOffsetTable.cpp
index eddbf86..978367a 100644
--- a/hotspot/src/share/vm/gc_implementation/g1/g1BlockOffsetTable.cpp
+++ b/hotspot/src/share/vm/gc_implementation/g1/g1BlockOffsetTable.cpp
@@ -424,7 +424,7 @@
       while (n <= next_boundary) {
         q = n;
         oop obj = oop(q);
-        if (obj->klass() == NULL) return q;
+        if (obj->klass_or_null() == NULL) return q;
         n += obj->size();
       }
       assert(q <= next_boundary && n > next_boundary, "Consequence of loop");
@@ -436,7 +436,7 @@
       while (n <= next_boundary) {
         q = n;
         oop obj = oop(q);
-        if (obj->klass() == NULL) return q;
+        if (obj->klass_or_null() == NULL) return q;
         n += _sp->block_size(q);
       }
       assert(q <= next_boundary && n > next_boundary, "Consequence of loop");
diff --git a/hotspot/src/share/vm/gc_implementation/g1/g1BlockOffsetTable.inline.hpp b/hotspot/src/share/vm/gc_implementation/g1/g1BlockOffsetTable.inline.hpp
index 45e1485..23f75e6 100644
--- a/hotspot/src/share/vm/gc_implementation/g1/g1BlockOffsetTable.inline.hpp
+++ b/hotspot/src/share/vm/gc_implementation/g1/g1BlockOffsetTable.inline.hpp
@@ -96,14 +96,14 @@
     while (n <= addr) {
       q = n;
       oop obj = oop(q);
-      if (obj->klass() == NULL) return q;
+      if (obj->klass_or_null() == NULL) return q;
       n += obj->size();
     }
   } else {
     while (n <= addr) {
       q = n;
       oop obj = oop(q);
-      if (obj->klass() == NULL) return q;
+      if (obj->klass_or_null() == NULL) return q;
       n += _sp->block_size(q);
     }
   }
@@ -115,7 +115,7 @@
 inline HeapWord*
 G1BlockOffsetArray::forward_to_block_containing_addr(HeapWord* q,
                                                      const void* addr) {
-  if (oop(q)->klass() == NULL) return q;
+  if (oop(q)->klass_or_null() == NULL) return q;
   HeapWord* n = q + _sp->block_size(q);
   // In the normal case, where the query "addr" is a card boundary, and the
   // offset table chunks are the same size as cards, the block starting at
diff --git a/hotspot/src/share/vm/gc_implementation/g1/g1CollectedHeap.cpp b/hotspot/src/share/vm/gc_implementation/g1/g1CollectedHeap.cpp
index a42ace5..f7c53a7 100644
--- a/hotspot/src/share/vm/gc_implementation/g1/g1CollectedHeap.cpp
+++ b/hotspot/src/share/vm/gc_implementation/g1/g1CollectedHeap.cpp
@@ -1655,11 +1655,14 @@
 // Computes the sum of the storage used by the various regions.
 
 size_t G1CollectedHeap::used() const {
-  assert(Heap_lock->owner() != NULL,
-         "Should be owned on this thread's behalf.");
+  // Temporarily, until 6859911 is fixed. XXX
+  // assert(Heap_lock->owner() != NULL,
+  //        "Should be owned on this thread's behalf.");
   size_t result = _summary_bytes_used;
-  if (_cur_alloc_region != NULL)
-    result += _cur_alloc_region->used();
+  // Read only once in case it is set to NULL concurrently
+  HeapRegion* hr = _cur_alloc_region;
+  if (hr != NULL)
+    result += hr->used();
   return result;
 }
 
@@ -2133,13 +2136,13 @@
   VerifyLivenessOopClosure(G1CollectedHeap* _g1h) {
     g1h = _g1h;
   }
-  void do_oop(narrowOop *p) {
-    guarantee(false, "NYI");
-  }
-  void do_oop(oop *p) {
-    oop obj = *p;
-    assert(obj == NULL || !g1h->is_obj_dead(obj),
-           "Dead object referenced by a not dead object");
+  void do_oop(narrowOop *p) { do_oop_work(p); }
+  void do_oop(      oop *p) { do_oop_work(p); }
+
+  template <class T> void do_oop_work(T *p) {
+    oop obj = oopDesc::load_decode_heap_oop(p);
+    guarantee(obj == NULL || !g1h->is_obj_dead(obj),
+              "Dead object referenced by a not dead object");
   }
 };
 
@@ -2206,8 +2209,10 @@
   // use_prev_marking == true  -> use "prev" marking information,
   // use_prev_marking == false -> use "next" marking information
   VerifyRegionClosure(bool allow_dirty, bool par, bool use_prev_marking)
-    : _allow_dirty(allow_dirty), _par(par),
+    : _allow_dirty(allow_dirty),
+      _par(par),
       _use_prev_marking(use_prev_marking) {}
+
   bool doHeapRegion(HeapRegion* r) {
     guarantee(_par || r->claim_value() == HeapRegion::InitialClaimValue,
               "Should be unclaimed at verify points.");
@@ -2231,18 +2236,16 @@
   // use_prev_marking == true  -> use "prev" marking information,
   // use_prev_marking == false -> use "next" marking information
   VerifyRootsClosure(bool use_prev_marking) :
-    _g1h(G1CollectedHeap::heap()), _failures(false),
+    _g1h(G1CollectedHeap::heap()),
+    _failures(false),
     _use_prev_marking(use_prev_marking) { }
 
   bool failures() { return _failures; }
 
-  void do_oop(narrowOop* p) {
-    guarantee(false, "NYI");
-  }
-
-  void do_oop(oop* p) {
-    oop obj = *p;
-    if (obj != NULL) {
+  template <class T> void do_oop_nv(T* p) {
+    T heap_oop = oopDesc::load_heap_oop(p);
+    if (!oopDesc::is_null(heap_oop)) {
+      oop obj = oopDesc::decode_heap_oop_not_null(heap_oop);
       if (_g1h->is_obj_dead_cond(obj, _use_prev_marking)) {
         gclog_or_tty->print_cr("Root location "PTR_FORMAT" "
                                "points to dead obj "PTR_FORMAT, p, (void*) obj);
@@ -2251,6 +2254,9 @@
       }
     }
   }
+
+  void do_oop(oop* p)       { do_oop_nv(p); }
+  void do_oop(narrowOop* p) { do_oop_nv(p); }
 };
 
 // This is the task used for parallel heap verification.
@@ -2267,7 +2273,8 @@
   G1ParVerifyTask(G1CollectedHeap* g1h, bool allow_dirty,
                   bool use_prev_marking) :
     AbstractGangTask("Parallel verify task"),
-    _g1h(g1h), _allow_dirty(allow_dirty),
+    _g1h(g1h),
+    _allow_dirty(allow_dirty),
     _use_prev_marking(use_prev_marking) { }
 
   void work(int worker_i) {
@@ -2479,12 +2486,10 @@
 
 void
 G1CollectedHeap::doConcurrentMark() {
-  if (G1ConcMark) {
-    MutexLockerEx x(CGC_lock, Mutex::_no_safepoint_check_flag);
-    if (!_cmThread->in_progress()) {
-      _cmThread->set_started();
-      CGC_lock->notify();
-    }
+  MutexLockerEx x(CGC_lock, Mutex::_no_safepoint_check_flag);
+  if (!_cmThread->in_progress()) {
+    _cmThread->set_started();
+    CGC_lock->notify();
   }
 }
 
@@ -2561,9 +2566,11 @@
                           "Not enough space for young surv words summary.");
   }
   memset(_surviving_young_words, 0, array_length * sizeof(size_t));
+#ifdef ASSERT
   for (size_t i = 0;  i < array_length; ++i) {
-    guarantee( _surviving_young_words[i] == 0, "invariant" );
+    assert( _surviving_young_words[i] == 0, "memset above" );
   }
+#endif // !ASSERT
 }
 
 void
@@ -2649,7 +2656,7 @@
       COMPILER2_PRESENT(DerivedPointerTable::clear());
 
       // We want to turn off ref discovery, if necessary, and turn it back on
-      // on again later if we do.
+      // on again later if we do. XXX Dubious: why is discovery disabled?
       bool was_enabled = ref_processor()->discovery_enabled();
       if (was_enabled) ref_processor()->disable_discovery();
 
@@ -2662,9 +2669,6 @@
       double start_time_sec = os::elapsedTime();
       GCOverheadReporter::recordSTWStart(start_time_sec);
       size_t start_used_bytes = used();
-      if (!G1ConcMark) {
-        do_sync_mark();
-      }
 
       g1_policy()->record_collection_pause_start(start_time_sec,
                                                  start_used_bytes);
@@ -2775,6 +2779,13 @@
           g1_policy()->should_initiate_conc_mark()) {
         concurrent_mark()->checkpointRootsInitialPost();
         set_marking_started();
+        // CAUTION: after the doConcurrentMark() call below,
+        // the concurrent marking thread(s) could be running
+        // concurrently with us. Make sure that anything after
+        // this point does not assume that we are the only GC thread
+        // running. Note: of course, the actual marking work will
+        // not start until the safepoint itself is released in
+        // ConcurrentGCThread::safepoint_desynchronize().
         doConcurrentMark();
       }
 
@@ -3123,10 +3134,8 @@
   G1CollectedHeap* _g1;
 public:
   G1KeepAliveClosure(G1CollectedHeap* g1) : _g1(g1) {}
-  void do_oop(narrowOop* p) {
-    guarantee(false, "NYI");
-  }
-  void do_oop(oop* p) {
+  void do_oop(narrowOop* p) { guarantee(false, "Not needed"); }
+  void do_oop(      oop* p) {
     oop obj = *p;
 #ifdef G1_DEBUG
     if (PrintGC && Verbose) {
@@ -3138,7 +3147,6 @@
     if (_g1->obj_in_cs(obj)) {
       assert( obj->is_forwarded(), "invariant" );
       *p = obj->forwardee();
-
 #ifdef G1_DEBUG
       gclog_or_tty->print_cr("     in CSet: moved "PTR_FORMAT" -> "PTR_FORMAT,
                              (void*) obj, (void*) *p);
@@ -3155,12 +3163,12 @@
   UpdateRSetImmediate(G1CollectedHeap* g1) :
     _g1(g1), _g1_rem_set(g1->g1_rem_set()) {}
 
-  void do_oop(narrowOop* p) {
-    guarantee(false, "NYI");
-  }
-  void do_oop(oop* p) {
+  virtual void do_oop(narrowOop* p) { do_oop_work(p); }
+  virtual void do_oop(      oop* p) { do_oop_work(p); }
+  template <class T> void do_oop_work(T* p) {
     assert(_from->is_in_reserved(p), "paranoia");
-    if (*p != NULL && !_from->is_survivor()) {
+    T heap_oop = oopDesc::load_heap_oop(p);
+    if (!oopDesc::is_null(heap_oop) && !_from->is_survivor()) {
       _g1_rem_set->par_write_ref(_from, p, 0);
     }
   }
@@ -3176,12 +3184,12 @@
   UpdateRSetDeferred(G1CollectedHeap* g1, DirtyCardQueue* dcq) :
     _g1(g1), _ct_bs((CardTableModRefBS*)_g1->barrier_set()), _dcq(dcq) {}
 
-  void do_oop(narrowOop* p) {
-    guarantee(false, "NYI");
-  }
-  void do_oop(oop* p) {
+  virtual void do_oop(narrowOop* p) { do_oop_work(p); }
+  virtual void do_oop(      oop* p) { do_oop_work(p); }
+  template <class T> void do_oop_work(T* p) {
     assert(_from->is_in_reserved(p), "paranoia");
-    if (!_from->is_in_reserved(*p) && !_from->is_survivor()) {
+    if (!_from->is_in_reserved(oopDesc::load_decode_heap_oop(p)) &&
+        !_from->is_survivor()) {
       size_t card_index = _ct_bs->index_for(p);
       if (_ct_bs->mark_card_deferred(card_index)) {
         _dcq->enqueue((jbyte*)_ct_bs->byte_for_index(card_index));
@@ -3536,614 +3544,66 @@
   fill_with_object(block, free_words);
 }
 
-#define use_local_bitmaps         1
-#define verify_local_bitmaps      0
-
 #ifndef PRODUCT
-
-class GCLabBitMap;
-class GCLabBitMapClosure: public BitMapClosure {
-private:
-  ConcurrentMark* _cm;
-  GCLabBitMap*    _bitmap;
-
-public:
-  GCLabBitMapClosure(ConcurrentMark* cm,
-                     GCLabBitMap* bitmap) {
-    _cm     = cm;
-    _bitmap = bitmap;
-  }
-
-  virtual bool do_bit(size_t offset);
-};
-
-#endif // PRODUCT
-
-#define oop_buffer_length 256
-
-class GCLabBitMap: public BitMap {
-private:
-  ConcurrentMark* _cm;
-
-  int       _shifter;
-  size_t    _bitmap_word_covers_words;
-
-  // beginning of the heap
-  HeapWord* _heap_start;
-
-  // this is the actual start of the GCLab
-  HeapWord* _real_start_word;
-
-  // this is the actual end of the GCLab
-  HeapWord* _real_end_word;
-
-  // this is the first word, possibly located before the actual start
-  // of the GCLab, that corresponds to the first bit of the bitmap
-  HeapWord* _start_word;
-
-  // size of a GCLab in words
-  size_t _gclab_word_size;
-
-  static int shifter() {
-    return MinObjAlignment - 1;
-  }
-
-  // how many heap words does a single bitmap word corresponds to?
-  static size_t bitmap_word_covers_words() {
-    return BitsPerWord << shifter();
-  }
-
-  static size_t gclab_word_size() {
-    return G1ParallelGCAllocBufferSize / HeapWordSize;
-  }
-
-  static size_t bitmap_size_in_bits() {
-    size_t bits_in_bitmap = gclab_word_size() >> shifter();
-    // We are going to ensure that the beginning of a word in this
-    // bitmap also corresponds to the beginning of a word in the
-    // global marking bitmap. To handle the case where a GCLab
-    // starts from the middle of the bitmap, we need to add enough
-    // space (i.e. up to a bitmap word) to ensure that we have
-    // enough bits in the bitmap.
-    return bits_in_bitmap + BitsPerWord - 1;
-  }
-public:
-  GCLabBitMap(HeapWord* heap_start)
-    : BitMap(bitmap_size_in_bits()),
-      _cm(G1CollectedHeap::heap()->concurrent_mark()),
-      _shifter(shifter()),
-      _bitmap_word_covers_words(bitmap_word_covers_words()),
-      _heap_start(heap_start),
-      _gclab_word_size(gclab_word_size()),
-      _real_start_word(NULL),
-      _real_end_word(NULL),
-      _start_word(NULL)
-  {
-    guarantee( size_in_words() >= bitmap_size_in_words(),
-               "just making sure");
-  }
-
-  inline unsigned heapWordToOffset(HeapWord* addr) {
-    unsigned offset = (unsigned) pointer_delta(addr, _start_word) >> _shifter;
-    assert(offset < size(), "offset should be within bounds");
-    return offset;
-  }
-
-  inline HeapWord* offsetToHeapWord(size_t offset) {
-    HeapWord* addr =  _start_word + (offset << _shifter);
-    assert(_real_start_word <= addr && addr < _real_end_word, "invariant");
-    return addr;
-  }
-
-  bool fields_well_formed() {
-    bool ret1 = (_real_start_word == NULL) &&
-                (_real_end_word == NULL) &&
-                (_start_word == NULL);
-    if (ret1)
-      return true;
-
-    bool ret2 = _real_start_word >= _start_word &&
-      _start_word < _real_end_word &&
-      (_real_start_word + _gclab_word_size) == _real_end_word &&
-      (_start_word + _gclab_word_size + _bitmap_word_covers_words)
-                                                              > _real_end_word;
-    return ret2;
-  }
-
-  inline bool mark(HeapWord* addr) {
-    guarantee(use_local_bitmaps, "invariant");
-    assert(fields_well_formed(), "invariant");
-
-    if (addr >= _real_start_word && addr < _real_end_word) {
-      assert(!isMarked(addr), "should not have already been marked");
-
-      // first mark it on the bitmap
-      at_put(heapWordToOffset(addr), true);
-
-      return true;
-    } else {
-      return false;
-    }
-  }
-
-  inline bool isMarked(HeapWord* addr) {
-    guarantee(use_local_bitmaps, "invariant");
-    assert(fields_well_formed(), "invariant");
-
-    return at(heapWordToOffset(addr));
-  }
-
-  void set_buffer(HeapWord* start) {
-    guarantee(use_local_bitmaps, "invariant");
-    clear();
-
-    assert(start != NULL, "invariant");
-    _real_start_word = start;
-    _real_end_word   = start + _gclab_word_size;
-
-    size_t diff =
-      pointer_delta(start, _heap_start) % _bitmap_word_covers_words;
-    _start_word = start - diff;
-
-    assert(fields_well_formed(), "invariant");
-  }
-
-#ifndef PRODUCT
-  void verify() {
-    // verify that the marks have been propagated
-    GCLabBitMapClosure cl(_cm, this);
-    iterate(&cl);
-  }
-#endif // PRODUCT
-
-  void retire() {
-    guarantee(use_local_bitmaps, "invariant");
-    assert(fields_well_formed(), "invariant");
-
-    if (_start_word != NULL) {
-      CMBitMap*       mark_bitmap = _cm->nextMarkBitMap();
-
-      // this means that the bitmap was set up for the GCLab
-      assert(_real_start_word != NULL && _real_end_word != NULL, "invariant");
-
-      mark_bitmap->mostly_disjoint_range_union(this,
-                                0, // always start from the start of the bitmap
-                                _start_word,
-                                size_in_words());
-      _cm->grayRegionIfNecessary(MemRegion(_real_start_word, _real_end_word));
-
-#ifndef PRODUCT
-      if (use_local_bitmaps && verify_local_bitmaps)
-        verify();
-#endif // PRODUCT
-    } else {
-      assert(_real_start_word == NULL && _real_end_word == NULL, "invariant");
-    }
-  }
-
-  static size_t bitmap_size_in_words() {
-    return (bitmap_size_in_bits() + BitsPerWord - 1) / BitsPerWord;
-  }
-};
-
-#ifndef PRODUCT
-
 bool GCLabBitMapClosure::do_bit(size_t offset) {
   HeapWord* addr = _bitmap->offsetToHeapWord(offset);
   guarantee(_cm->isMarked(oop(addr)), "it should be!");
   return true;
 }
-
 #endif // PRODUCT
 
-class G1ParGCAllocBuffer: public ParGCAllocBuffer {
-private:
-  bool        _retired;
-  bool        _during_marking;
-  GCLabBitMap _bitmap;
-
-public:
-  G1ParGCAllocBuffer() :
-    ParGCAllocBuffer(G1ParallelGCAllocBufferSize / HeapWordSize),
-    _during_marking(G1CollectedHeap::heap()->mark_in_progress()),
-    _bitmap(G1CollectedHeap::heap()->reserved_region().start()),
-    _retired(false)
-  { }
-
-  inline bool mark(HeapWord* addr) {
-    guarantee(use_local_bitmaps, "invariant");
-    assert(_during_marking, "invariant");
-    return _bitmap.mark(addr);
-  }
-
-  inline void set_buf(HeapWord* buf) {
-    if (use_local_bitmaps && _during_marking)
-      _bitmap.set_buffer(buf);
-    ParGCAllocBuffer::set_buf(buf);
-    _retired = false;
-  }
-
-  inline void retire(bool end_of_gc, bool retain) {
-    if (_retired)
-      return;
-    if (use_local_bitmaps && _during_marking) {
-      _bitmap.retire();
-    }
-    ParGCAllocBuffer::retire(end_of_gc, retain);
-    _retired = true;
-  }
-};
-
-
-class G1ParScanThreadState : public StackObj {
-protected:
-  G1CollectedHeap* _g1h;
-  RefToScanQueue*  _refs;
-  DirtyCardQueue   _dcq;
-  CardTableModRefBS* _ct_bs;
-  G1RemSet* _g1_rem;
-
-  typedef GrowableArray<oop*> OverflowQueue;
-  OverflowQueue* _overflowed_refs;
-
-  G1ParGCAllocBuffer _alloc_buffers[GCAllocPurposeCount];
-  ageTable           _age_table;
-
-  size_t           _alloc_buffer_waste;
-  size_t           _undo_waste;
-
-  OopsInHeapRegionClosure*      _evac_failure_cl;
-  G1ParScanHeapEvacClosure*     _evac_cl;
-  G1ParScanPartialArrayClosure* _partial_scan_cl;
-
-  int _hash_seed;
-  int _queue_num;
-
-  int _term_attempts;
+G1ParScanThreadState::G1ParScanThreadState(G1CollectedHeap* g1h, int queue_num)
+  : _g1h(g1h),
+    _refs(g1h->task_queue(queue_num)),
+    _dcq(&g1h->dirty_card_queue_set()),
+    _ct_bs((CardTableModRefBS*)_g1h->barrier_set()),
+    _g1_rem(g1h->g1_rem_set()),
+    _hash_seed(17), _queue_num(queue_num),
+    _term_attempts(0),
+    _age_table(false),
 #if G1_DETAILED_STATS
-  int _pushes, _pops, _steals, _steal_attempts;
-  int _overflow_pushes;
+    _pushes(0), _pops(0), _steals(0),
+    _steal_attempts(0),  _overflow_pushes(0),
 #endif
+    _strong_roots_time(0), _term_time(0),
+    _alloc_buffer_waste(0), _undo_waste(0)
+{
+  // we allocate G1YoungSurvRateNumRegions plus one entries, since
+  // we "sacrifice" entry 0 to keep track of surviving bytes for
+  // non-young regions (where the age is -1)
+  // We also add a few elements at the beginning and at the end in
+  // an attempt to eliminate cache contention
+  size_t real_length = 1 + _g1h->g1_policy()->young_cset_length();
+  size_t array_length = PADDING_ELEM_NUM +
+                        real_length +
+                        PADDING_ELEM_NUM;
+  _surviving_young_words_base = NEW_C_HEAP_ARRAY(size_t, array_length);
+  if (_surviving_young_words_base == NULL)
+    vm_exit_out_of_memory(array_length * sizeof(size_t),
+                          "Not enough space for young surv histo.");
+  _surviving_young_words = _surviving_young_words_base + PADDING_ELEM_NUM;
+  memset(_surviving_young_words, 0, real_length * sizeof(size_t));
 
-  double _start;
-  double _start_strong_roots;
-  double _strong_roots_time;
-  double _start_term;
-  double _term_time;
+  _overflowed_refs = new OverflowQueue(10);
 
-  // Map from young-age-index (0 == not young, 1 is youngest) to
-  // surviving words. base is what we get back from the malloc call
-  size_t* _surviving_young_words_base;
-  // this points into the array, as we use the first few entries for padding
-  size_t* _surviving_young_words;
-
-#define PADDING_ELEM_NUM (64 / sizeof(size_t))
-
-  void   add_to_alloc_buffer_waste(size_t waste) { _alloc_buffer_waste += waste; }
-
-  void   add_to_undo_waste(size_t waste)         { _undo_waste += waste; }
-
-  DirtyCardQueue& dirty_card_queue()             { return _dcq;  }
-  CardTableModRefBS* ctbs()                      { return _ct_bs; }
-
-  void immediate_rs_update(HeapRegion* from, oop* p, int tid) {
-    if (!from->is_survivor()) {
-      _g1_rem->par_write_ref(from, p, tid);
-    }
-  }
-
-  void deferred_rs_update(HeapRegion* from, oop* p, int tid) {
-    // If the new value of the field points to the same region or
-    // is the to-space, we don't need to include it in the Rset updates.
-    if (!from->is_in_reserved(*p) && !from->is_survivor()) {
-      size_t card_index = ctbs()->index_for(p);
-      // If the card hasn't been added to the buffer, do it.
-      if (ctbs()->mark_card_deferred(card_index)) {
-        dirty_card_queue().enqueue((jbyte*)ctbs()->byte_for_index(card_index));
-      }
-    }
-  }
-
-public:
-  G1ParScanThreadState(G1CollectedHeap* g1h, int queue_num)
-    : _g1h(g1h),
-      _refs(g1h->task_queue(queue_num)),
-      _dcq(&g1h->dirty_card_queue_set()),
-      _ct_bs((CardTableModRefBS*)_g1h->barrier_set()),
-      _g1_rem(g1h->g1_rem_set()),
-      _hash_seed(17), _queue_num(queue_num),
-      _term_attempts(0),
-      _age_table(false),
-#if G1_DETAILED_STATS
-      _pushes(0), _pops(0), _steals(0),
-      _steal_attempts(0),  _overflow_pushes(0),
-#endif
-      _strong_roots_time(0), _term_time(0),
-      _alloc_buffer_waste(0), _undo_waste(0)
-  {
-    // we allocate G1YoungSurvRateNumRegions plus one entries, since
-    // we "sacrifice" entry 0 to keep track of surviving bytes for
-    // non-young regions (where the age is -1)
-    // We also add a few elements at the beginning and at the end in
-    // an attempt to eliminate cache contention
-    size_t real_length = 1 + _g1h->g1_policy()->young_cset_length();
-    size_t array_length = PADDING_ELEM_NUM +
-                          real_length +
-                          PADDING_ELEM_NUM;
-    _surviving_young_words_base = NEW_C_HEAP_ARRAY(size_t, array_length);
-    if (_surviving_young_words_base == NULL)
-      vm_exit_out_of_memory(array_length * sizeof(size_t),
-                            "Not enough space for young surv histo.");
-    _surviving_young_words = _surviving_young_words_base + PADDING_ELEM_NUM;
-    memset(_surviving_young_words, 0, real_length * sizeof(size_t));
-
-    _overflowed_refs = new OverflowQueue(10);
-
-    _start = os::elapsedTime();
-  }
-
-  ~G1ParScanThreadState() {
-    FREE_C_HEAP_ARRAY(size_t, _surviving_young_words_base);
-  }
-
-  RefToScanQueue*   refs()            { return _refs;             }
-  OverflowQueue*    overflowed_refs() { return _overflowed_refs;  }
-  ageTable*         age_table()       { return &_age_table;       }
-
-  G1ParGCAllocBuffer* alloc_buffer(GCAllocPurpose purpose) {
-    return &_alloc_buffers[purpose];
-  }
-
-  size_t alloc_buffer_waste()                    { return _alloc_buffer_waste; }
-  size_t undo_waste()                            { return _undo_waste; }
-
-  void push_on_queue(oop* ref) {
-    assert(ref != NULL, "invariant");
-    assert(has_partial_array_mask(ref) || _g1h->obj_in_cs(*ref), "invariant");
-
-    if (!refs()->push(ref)) {
-      overflowed_refs()->push(ref);
-      IF_G1_DETAILED_STATS(note_overflow_push());
-    } else {
-      IF_G1_DETAILED_STATS(note_push());
-    }
-  }
-
-  void pop_from_queue(oop*& ref) {
-    if (!refs()->pop_local(ref)) {
-      ref = NULL;
-    } else {
-      assert(ref != NULL, "invariant");
-      assert(has_partial_array_mask(ref) || _g1h->obj_in_cs(*ref),
-             "invariant");
-
-      IF_G1_DETAILED_STATS(note_pop());
-    }
-  }
-
-  void pop_from_overflow_queue(oop*& ref) {
-    ref = overflowed_refs()->pop();
-  }
-
-  int refs_to_scan()                             { return refs()->size();                 }
-  int overflowed_refs_to_scan()                  { return overflowed_refs()->length();    }
-
-  void update_rs(HeapRegion* from, oop* p, int tid) {
-    if (G1DeferredRSUpdate) {
-      deferred_rs_update(from, p, tid);
-    } else {
-      immediate_rs_update(from, p, tid);
-    }
-  }
-
-  HeapWord* allocate_slow(GCAllocPurpose purpose, size_t word_sz) {
-
-    HeapWord* obj = NULL;
-    if (word_sz * 100 <
-        (size_t)(G1ParallelGCAllocBufferSize / HeapWordSize) *
-                                                  ParallelGCBufferWastePct) {
-      G1ParGCAllocBuffer* alloc_buf = alloc_buffer(purpose);
-      add_to_alloc_buffer_waste(alloc_buf->words_remaining());
-      alloc_buf->retire(false, false);
-
-      HeapWord* buf =
-        _g1h->par_allocate_during_gc(purpose, G1ParallelGCAllocBufferSize / HeapWordSize);
-      if (buf == NULL) return NULL; // Let caller handle allocation failure.
-      // Otherwise.
-      alloc_buf->set_buf(buf);
-
-      obj = alloc_buf->allocate(word_sz);
-      assert(obj != NULL, "buffer was definitely big enough...");
-    } else {
-      obj = _g1h->par_allocate_during_gc(purpose, word_sz);
-    }
-    return obj;
-  }
-
-  HeapWord* allocate(GCAllocPurpose purpose, size_t word_sz) {
-    HeapWord* obj = alloc_buffer(purpose)->allocate(word_sz);
-    if (obj != NULL) return obj;
-    return allocate_slow(purpose, word_sz);
-  }
-
-  void undo_allocation(GCAllocPurpose purpose, HeapWord* obj, size_t word_sz) {
-    if (alloc_buffer(purpose)->contains(obj)) {
-      guarantee(alloc_buffer(purpose)->contains(obj + word_sz - 1),
-                "should contain whole object");
-      alloc_buffer(purpose)->undo_allocation(obj, word_sz);
-    } else {
-      CollectedHeap::fill_with_object(obj, word_sz);
-      add_to_undo_waste(word_sz);
-    }
-  }
-
-  void set_evac_failure_closure(OopsInHeapRegionClosure* evac_failure_cl) {
-    _evac_failure_cl = evac_failure_cl;
-  }
-  OopsInHeapRegionClosure* evac_failure_closure() {
-    return _evac_failure_cl;
-  }
-
-  void set_evac_closure(G1ParScanHeapEvacClosure* evac_cl) {
-    _evac_cl = evac_cl;
-  }
-
-  void set_partial_scan_closure(G1ParScanPartialArrayClosure* partial_scan_cl) {
-    _partial_scan_cl = partial_scan_cl;
-  }
-
-  int* hash_seed() { return &_hash_seed; }
-  int  queue_num() { return _queue_num; }
-
-  int term_attempts()   { return _term_attempts; }
-  void note_term_attempt()  { _term_attempts++; }
-
-#if G1_DETAILED_STATS
-  int pushes()          { return _pushes; }
-  int pops()            { return _pops; }
-  int steals()          { return _steals; }
-  int steal_attempts()  { return _steal_attempts; }
-  int overflow_pushes() { return _overflow_pushes; }
-
-  void note_push()          { _pushes++; }
-  void note_pop()           { _pops++; }
-  void note_steal()         { _steals++; }
-  void note_steal_attempt() { _steal_attempts++; }
-  void note_overflow_push() { _overflow_pushes++; }
-#endif
-
-  void start_strong_roots() {
-    _start_strong_roots = os::elapsedTime();
-  }
-  void end_strong_roots() {
-    _strong_roots_time += (os::elapsedTime() - _start_strong_roots);
-  }
-  double strong_roots_time() { return _strong_roots_time; }
-
-  void start_term_time() {
-    note_term_attempt();
-    _start_term = os::elapsedTime();
-  }
-  void end_term_time() {
-    _term_time += (os::elapsedTime() - _start_term);
-  }
-  double term_time() { return _term_time; }
-
-  double elapsed() {
-    return os::elapsedTime() - _start;
-  }
-
-  size_t* surviving_young_words() {
-    // We add on to hide entry 0 which accumulates surviving words for
-    // age -1 regions (i.e. non-young ones)
-    return _surviving_young_words;
-  }
-
-  void retire_alloc_buffers() {
-    for (int ap = 0; ap < GCAllocPurposeCount; ++ap) {
-      size_t waste = _alloc_buffers[ap].words_remaining();
-      add_to_alloc_buffer_waste(waste);
-      _alloc_buffers[ap].retire(true, false);
-    }
-  }
-
-private:
-  void deal_with_reference(oop* ref_to_scan) {
-    if (has_partial_array_mask(ref_to_scan)) {
-      _partial_scan_cl->do_oop_nv(ref_to_scan);
-    } else {
-      // Note: we can use "raw" versions of "region_containing" because
-      // "obj_to_scan" is definitely in the heap, and is not in a
-      // humongous region.
-      HeapRegion* r = _g1h->heap_region_containing_raw(ref_to_scan);
-      _evac_cl->set_region(r);
-      _evac_cl->do_oop_nv(ref_to_scan);
-    }
-  }
-
-public:
-  void trim_queue() {
-    // I've replicated the loop twice, first to drain the overflow
-    // queue, second to drain the task queue. This is better than
-    // having a single loop, which checks both conditions and, inside
-    // it, either pops the overflow queue or the task queue, as each
-    // loop is tighter. Also, the decision to drain the overflow queue
-    // first is not arbitrary, as the overflow queue is not visible
-    // to the other workers, whereas the task queue is. So, we want to
-    // drain the "invisible" entries first, while allowing the other
-    // workers to potentially steal the "visible" entries.
-
-    while (refs_to_scan() > 0 || overflowed_refs_to_scan() > 0) {
-      while (overflowed_refs_to_scan() > 0) {
-        oop *ref_to_scan = NULL;
-        pop_from_overflow_queue(ref_to_scan);
-        assert(ref_to_scan != NULL, "invariant");
-        // We shouldn't have pushed it on the queue if it was not
-        // pointing into the CSet.
-        assert(ref_to_scan != NULL, "sanity");
-        assert(has_partial_array_mask(ref_to_scan) ||
-                                      _g1h->obj_in_cs(*ref_to_scan), "sanity");
-
-        deal_with_reference(ref_to_scan);
-      }
-
-      while (refs_to_scan() > 0) {
-        oop *ref_to_scan = NULL;
-        pop_from_queue(ref_to_scan);
-
-        if (ref_to_scan != NULL) {
-          // We shouldn't have pushed it on the queue if it was not
-          // pointing into the CSet.
-          assert(has_partial_array_mask(ref_to_scan) ||
-                                      _g1h->obj_in_cs(*ref_to_scan), "sanity");
-
-          deal_with_reference(ref_to_scan);
-        }
-      }
-    }
-  }
-};
+  _start = os::elapsedTime();
+}
 
 G1ParClosureSuper::G1ParClosureSuper(G1CollectedHeap* g1, G1ParScanThreadState* par_scan_state) :
   _g1(g1), _g1_rem(_g1->g1_rem_set()), _cm(_g1->concurrent_mark()),
   _par_scan_state(par_scan_state) { }
 
-// This closure is applied to the fields of the objects that have just been copied.
-// Should probably be made inline and moved in g1OopClosures.inline.hpp.
-void G1ParScanClosure::do_oop_nv(oop* p) {
-  oop obj = *p;
-
-  if (obj != NULL) {
-    if (_g1->in_cset_fast_test(obj)) {
-      // We're not going to even bother checking whether the object is
-      // already forwarded or not, as this usually causes an immediate
-      // stall. We'll try to prefetch the object (for write, given that
-      // we might need to install the forwarding reference) and we'll
-      // get back to it when pop it from the queue
-      Prefetch::write(obj->mark_addr(), 0);
-      Prefetch::read(obj->mark_addr(), (HeapWordSize*2));
-
-      // slightly paranoid test; I'm trying to catch potential
-      // problems before we go into push_on_queue to know where the
-      // problem is coming from
-      assert(obj == *p, "the value of *p should not have changed");
-      _par_scan_state->push_on_queue(p);
-    } else {
-      _par_scan_state->update_rs(_from, p, _par_scan_state->queue_num());
-    }
-  }
-}
-
-void G1ParCopyHelper::mark_forwardee(oop* p) {
+template <class T> void G1ParCopyHelper::mark_forwardee(T* p) {
   // This is called _after_ do_oop_work has been called, hence after
   // the object has been relocated to its new location and *p points
   // to its new location.
 
-  oop thisOop = *p;
-  if (thisOop != NULL) {
-    assert((_g1->evacuation_failed()) || (!_g1->obj_in_cs(thisOop)),
+  T heap_oop = oopDesc::load_heap_oop(p);
+  if (!oopDesc::is_null(heap_oop)) {
+    oop obj = oopDesc::decode_heap_oop(heap_oop);
+    assert((_g1->evacuation_failed()) || (!_g1->obj_in_cs(obj)),
            "shouldn't still be in the CSet if evacuation didn't fail.");
-    HeapWord* addr = (HeapWord*)thisOop;
+    HeapWord* addr = (HeapWord*)obj;
     if (_g1->is_in_g1_reserved(addr))
       _cm->grayRoot(oop(addr));
   }
@@ -4226,7 +3686,8 @@
 
     if (obj->is_objArray() && arrayOop(obj)->length() >= ParGCArrayScanChunk) {
       arrayOop(old)->set_length(0);
-      _par_scan_state->push_on_queue(set_partial_array_mask(old));
+      oop* old_p = set_partial_array_mask(old);
+      _par_scan_state->push_on_queue(old_p);
     } else {
       // No point in using the slower heap_region_containing() method,
       // given that we know obj is in the heap.
@@ -4240,11 +3701,11 @@
   return obj;
 }
 
-template<bool do_gen_barrier, G1Barrier barrier,
-         bool do_mark_forwardee, bool skip_cset_test>
-void G1ParCopyClosure<do_gen_barrier, barrier,
-                      do_mark_forwardee, skip_cset_test>::do_oop_work(oop* p) {
-  oop obj = *p;
+template <bool do_gen_barrier, G1Barrier barrier, bool do_mark_forwardee, bool skip_cset_test>
+template <class T>
+void G1ParCopyClosure <do_gen_barrier, barrier, do_mark_forwardee, skip_cset_test>
+::do_oop_work(T* p) {
+  oop obj = oopDesc::load_decode_heap_oop(p);
   assert(barrier != G1BarrierRS || obj != NULL,
          "Precondition: G1BarrierRS implies obj is nonNull");
 
@@ -4261,9 +3722,10 @@
                            "into CS.", p, (void*) obj);
 #endif
     if (obj->is_forwarded()) {
-      *p = obj->forwardee();
+      oopDesc::encode_store_heap_oop(p, obj->forwardee());
     } else {
-      *p = copy_to_survivor_space(obj);
+      oop copy_oop = copy_to_survivor_space(obj);
+      oopDesc::encode_store_heap_oop(p, copy_oop);
     }
     // When scanning the RS, we only care about objs in CS.
     if (barrier == G1BarrierRS) {
@@ -4282,21 +3744,9 @@
 }
 
 template void G1ParCopyClosure<false, G1BarrierEvac, false, true>::do_oop_work(oop* p);
+template void G1ParCopyClosure<false, G1BarrierEvac, false, true>::do_oop_work(narrowOop* p);
 
-template<class T> void G1ParScanPartialArrayClosure::process_array_chunk(
-  oop obj, int start, int end) {
-  // process our set of indices (include header in first chunk)
-  assert(start < end, "invariant");
-  T* const base      = (T*)objArrayOop(obj)->base();
-  T* const start_addr = (start == 0) ? (T*) obj : base + start;
-  T* const end_addr   = base + end;
-  MemRegion mr((HeapWord*)start_addr, (HeapWord*)end_addr);
-  _scanner.set_region(_g1->heap_region_containing(obj));
-  obj->oop_iterate(&_scanner, mr);
-}
-
-void G1ParScanPartialArrayClosure::do_oop_nv(oop* p) {
-  assert(!UseCompressedOops, "Needs to be fixed to work with compressed oops");
+template <class T> void G1ParScanPartialArrayClosure::do_oop_nv(T* p) {
   assert(has_partial_array_mask(p), "invariant");
   oop old = clear_partial_array_mask(p);
   assert(old->is_objArray(), "must be obj array");
@@ -4316,19 +3766,19 @@
     end = start + ParGCArrayScanChunk;
     arrayOop(old)->set_length(end);
     // Push remainder.
-    _par_scan_state->push_on_queue(set_partial_array_mask(old));
+    oop* old_p = set_partial_array_mask(old);
+    assert(arrayOop(old)->length() < obj->length(), "Empty push?");
+    _par_scan_state->push_on_queue(old_p);
   } else {
     // Restore length so that the heap remains parsable in
     // case of evacuation failure.
     arrayOop(old)->set_length(end);
   }
-
+  _scanner.set_region(_g1->heap_region_containing_raw(obj));
   // process our set of indices (include header in first chunk)
-  process_array_chunk<oop>(obj, start, end);
+  obj->oop_iterate_range(&_scanner, start, end);
 }
 
-int G1ScanAndBalanceClosure::_nq = 0;
-
 class G1ParEvacuateFollowersClosure : public VoidClosure {
 protected:
   G1CollectedHeap*              _g1h;
@@ -4351,21 +3801,28 @@
   void do_void() {
     G1ParScanThreadState* pss = par_scan_state();
     while (true) {
-      oop* ref_to_scan;
       pss->trim_queue();
       IF_G1_DETAILED_STATS(pss->note_steal_attempt());
-      if (queues()->steal(pss->queue_num(),
-                          pss->hash_seed(),
-                          ref_to_scan)) {
+
+      StarTask stolen_task;
+      if (queues()->steal(pss->queue_num(), pss->hash_seed(), stolen_task)) {
         IF_G1_DETAILED_STATS(pss->note_steal());
 
         // slightly paranoid tests; I'm trying to catch potential
         // problems before we go into push_on_queue to know where the
         // problem is coming from
-        assert(ref_to_scan != NULL, "invariant");
-        assert(has_partial_array_mask(ref_to_scan) ||
-                                   _g1h->obj_in_cs(*ref_to_scan), "invariant");
-        pss->push_on_queue(ref_to_scan);
+        assert((oop*)stolen_task != NULL, "Error");
+        if (stolen_task.is_narrow()) {
+          assert(UseCompressedOops, "Error");
+          narrowOop* p = (narrowOop*) stolen_task;
+          assert(has_partial_array_mask(p) ||
+                 _g1h->obj_in_cs(oopDesc::load_decode_heap_oop(p)), "Error");
+          pss->push_on_queue(p);
+        } else {
+          oop* p = (oop*) stolen_task;
+          assert(has_partial_array_mask(p) || _g1h->obj_in_cs(*p), "Error");
+          pss->push_on_queue(p);
+        }
         continue;
       }
       pss->start_term_time();
@@ -4382,6 +3839,7 @@
   G1CollectedHeap*       _g1h;
   RefToScanQueueSet      *_queues;
   ParallelTaskTerminator _terminator;
+  int _n_workers;
 
   Mutex _stats_lock;
   Mutex* stats_lock() { return &_stats_lock; }
@@ -4397,7 +3855,8 @@
       _g1h(g1h),
       _queues(task_queues),
       _terminator(workers, _queues),
-      _stats_lock(Mutex::leaf, "parallel G1 stats lock", true)
+      _stats_lock(Mutex::leaf, "parallel G1 stats lock", true),
+      _n_workers(workers)
   {}
 
   RefToScanQueueSet* queues() { return _queues; }
@@ -4407,6 +3866,7 @@
   }
 
   void work(int i) {
+    if (i >= _n_workers) return;  // no work needed this round
     ResourceMark rm;
     HandleMark   hm;
 
@@ -4504,23 +3964,6 @@
 
 // *** Common G1 Evacuation Stuff
 
-class G1CountClosure: public OopsInHeapRegionClosure {
-public:
-  int n;
-  G1CountClosure() : n(0) {}
-  void do_oop(narrowOop* p) {
-    guarantee(false, "NYI");
-  }
-  void do_oop(oop* p) {
-    oop obj = *p;
-    assert(obj != NULL && G1CollectedHeap::heap()->obj_in_cs(obj),
-           "Rem set closure called on non-rem-set pointer.");
-    n++;
-  }
-};
-
-static G1CountClosure count_closure;
-
 void
 G1CollectedHeap::
 g1_process_strong_roots(bool collecting_perm_gen,
@@ -5570,8 +5013,3 @@
 void G1CollectedHeap::g1_unimplemented() {
   // Unimplemented();
 }
-
-
-// Local Variables: ***
-// c-indentation-style: gnu ***
-// End: ***
diff --git a/hotspot/src/share/vm/gc_implementation/g1/g1CollectedHeap.hpp b/hotspot/src/share/vm/gc_implementation/g1/g1CollectedHeap.hpp
index 82cfc84..3dc6017 100644
--- a/hotspot/src/share/vm/gc_implementation/g1/g1CollectedHeap.hpp
+++ b/hotspot/src/share/vm/gc_implementation/g1/g1CollectedHeap.hpp
@@ -56,8 +56,8 @@
 #  define IF_G1_DETAILED_STATS(code)
 #endif
 
-typedef GenericTaskQueue<oop*>    RefToScanQueue;
-typedef GenericTaskQueueSet<oop*> RefToScanQueueSet;
+typedef GenericTaskQueue<StarTask>    RefToScanQueue;
+typedef GenericTaskQueueSet<StarTask> RefToScanQueueSet;
 
 typedef int RegionIdx_t;   // needs to hold [ 0..max_regions() )
 typedef int CardIdx_t;     // needs to hold [ 0..CardsPerRegion )
@@ -1271,6 +1271,552 @@
 
 };
 
-// Local Variables: ***
-// c-indentation-style: gnu ***
-// End: ***
+#define use_local_bitmaps         1
+#define verify_local_bitmaps      0
+#define oop_buffer_length       256
+
+#ifndef PRODUCT
+class GCLabBitMap;
+class GCLabBitMapClosure: public BitMapClosure {
+private:
+  ConcurrentMark* _cm;
+  GCLabBitMap*    _bitmap;
+
+public:
+  GCLabBitMapClosure(ConcurrentMark* cm,
+                     GCLabBitMap* bitmap) {
+    _cm     = cm;
+    _bitmap = bitmap;
+  }
+
+  virtual bool do_bit(size_t offset);
+};
+#endif // !PRODUCT
+
+class GCLabBitMap: public BitMap {
+private:
+  ConcurrentMark* _cm;
+
+  int       _shifter;
+  size_t    _bitmap_word_covers_words;
+
+  // beginning of the heap
+  HeapWord* _heap_start;
+
+  // this is the actual start of the GCLab
+  HeapWord* _real_start_word;
+
+  // this is the actual end of the GCLab
+  HeapWord* _real_end_word;
+
+  // this is the first word, possibly located before the actual start
+  // of the GCLab, that corresponds to the first bit of the bitmap
+  HeapWord* _start_word;
+
+  // size of a GCLab in words
+  size_t _gclab_word_size;
+
+  static int shifter() {
+    return MinObjAlignment - 1;
+  }
+
+  // how many heap words does a single bitmap word corresponds to?
+  static size_t bitmap_word_covers_words() {
+    return BitsPerWord << shifter();
+  }
+
+  static size_t gclab_word_size() {
+    return G1ParallelGCAllocBufferSize / HeapWordSize;
+  }
+
+  static size_t bitmap_size_in_bits() {
+    size_t bits_in_bitmap = gclab_word_size() >> shifter();
+    // We are going to ensure that the beginning of a word in this
+    // bitmap also corresponds to the beginning of a word in the
+    // global marking bitmap. To handle the case where a GCLab
+    // starts from the middle of the bitmap, we need to add enough
+    // space (i.e. up to a bitmap word) to ensure that we have
+    // enough bits in the bitmap.
+    return bits_in_bitmap + BitsPerWord - 1;
+  }
+public:
+  GCLabBitMap(HeapWord* heap_start)
+    : BitMap(bitmap_size_in_bits()),
+      _cm(G1CollectedHeap::heap()->concurrent_mark()),
+      _shifter(shifter()),
+      _bitmap_word_covers_words(bitmap_word_covers_words()),
+      _heap_start(heap_start),
+      _gclab_word_size(gclab_word_size()),
+      _real_start_word(NULL),
+      _real_end_word(NULL),
+      _start_word(NULL)
+  {
+    guarantee( size_in_words() >= bitmap_size_in_words(),
+               "just making sure");
+  }
+
+  inline unsigned heapWordToOffset(HeapWord* addr) {
+    unsigned offset = (unsigned) pointer_delta(addr, _start_word) >> _shifter;
+    assert(offset < size(), "offset should be within bounds");
+    return offset;
+  }
+
+  inline HeapWord* offsetToHeapWord(size_t offset) {
+    HeapWord* addr =  _start_word + (offset << _shifter);
+    assert(_real_start_word <= addr && addr < _real_end_word, "invariant");
+    return addr;
+  }
+
+  bool fields_well_formed() {
+    bool ret1 = (_real_start_word == NULL) &&
+                (_real_end_word == NULL) &&
+                (_start_word == NULL);
+    if (ret1)
+      return true;
+
+    bool ret2 = _real_start_word >= _start_word &&
+      _start_word < _real_end_word &&
+      (_real_start_word + _gclab_word_size) == _real_end_word &&
+      (_start_word + _gclab_word_size + _bitmap_word_covers_words)
+                                                              > _real_end_word;
+    return ret2;
+  }
+
+  inline bool mark(HeapWord* addr) {
+    guarantee(use_local_bitmaps, "invariant");
+    assert(fields_well_formed(), "invariant");
+
+    if (addr >= _real_start_word && addr < _real_end_word) {
+      assert(!isMarked(addr), "should not have already been marked");
+
+      // first mark it on the bitmap
+      at_put(heapWordToOffset(addr), true);
+
+      return true;
+    } else {
+      return false;
+    }
+  }
+
+  inline bool isMarked(HeapWord* addr) {
+    guarantee(use_local_bitmaps, "invariant");
+    assert(fields_well_formed(), "invariant");
+
+    return at(heapWordToOffset(addr));
+  }
+
+  void set_buffer(HeapWord* start) {
+    guarantee(use_local_bitmaps, "invariant");
+    clear();
+
+    assert(start != NULL, "invariant");
+    _real_start_word = start;
+    _real_end_word   = start + _gclab_word_size;
+
+    size_t diff =
+      pointer_delta(start, _heap_start) % _bitmap_word_covers_words;
+    _start_word = start - diff;
+
+    assert(fields_well_formed(), "invariant");
+  }
+
+#ifndef PRODUCT
+  void verify() {
+    // verify that the marks have been propagated
+    GCLabBitMapClosure cl(_cm, this);
+    iterate(&cl);
+  }
+#endif // PRODUCT
+
+  void retire() {
+    guarantee(use_local_bitmaps, "invariant");
+    assert(fields_well_formed(), "invariant");
+
+    if (_start_word != NULL) {
+      CMBitMap*       mark_bitmap = _cm->nextMarkBitMap();
+
+      // this means that the bitmap was set up for the GCLab
+      assert(_real_start_word != NULL && _real_end_word != NULL, "invariant");
+
+      mark_bitmap->mostly_disjoint_range_union(this,
+                                0, // always start from the start of the bitmap
+                                _start_word,
+                                size_in_words());
+      _cm->grayRegionIfNecessary(MemRegion(_real_start_word, _real_end_word));
+
+#ifndef PRODUCT
+      if (use_local_bitmaps && verify_local_bitmaps)
+        verify();
+#endif // PRODUCT
+    } else {
+      assert(_real_start_word == NULL && _real_end_word == NULL, "invariant");
+    }
+  }
+
+  static size_t bitmap_size_in_words() {
+    return (bitmap_size_in_bits() + BitsPerWord - 1) / BitsPerWord;
+  }
+};
+
+class G1ParGCAllocBuffer: public ParGCAllocBuffer {
+private:
+  bool        _retired;
+  bool        _during_marking;
+  GCLabBitMap _bitmap;
+
+public:
+  G1ParGCAllocBuffer() :
+    ParGCAllocBuffer(G1ParallelGCAllocBufferSize / HeapWordSize),
+    _during_marking(G1CollectedHeap::heap()->mark_in_progress()),
+    _bitmap(G1CollectedHeap::heap()->reserved_region().start()),
+    _retired(false)
+  { }
+
+  inline bool mark(HeapWord* addr) {
+    guarantee(use_local_bitmaps, "invariant");
+    assert(_during_marking, "invariant");
+    return _bitmap.mark(addr);
+  }
+
+  inline void set_buf(HeapWord* buf) {
+    if (use_local_bitmaps && _during_marking)
+      _bitmap.set_buffer(buf);
+    ParGCAllocBuffer::set_buf(buf);
+    _retired = false;
+  }
+
+  inline void retire(bool end_of_gc, bool retain) {
+    if (_retired)
+      return;
+    if (use_local_bitmaps && _during_marking) {
+      _bitmap.retire();
+    }
+    ParGCAllocBuffer::retire(end_of_gc, retain);
+    _retired = true;
+  }
+};
+
+class G1ParScanThreadState : public StackObj {
+protected:
+  G1CollectedHeap* _g1h;
+  RefToScanQueue*  _refs;
+  DirtyCardQueue   _dcq;
+  CardTableModRefBS* _ct_bs;
+  G1RemSet* _g1_rem;
+
+  typedef GrowableArray<StarTask> OverflowQueue;
+  OverflowQueue* _overflowed_refs;
+
+  G1ParGCAllocBuffer _alloc_buffers[GCAllocPurposeCount];
+  ageTable           _age_table;
+
+  size_t           _alloc_buffer_waste;
+  size_t           _undo_waste;
+
+  OopsInHeapRegionClosure*      _evac_failure_cl;
+  G1ParScanHeapEvacClosure*     _evac_cl;
+  G1ParScanPartialArrayClosure* _partial_scan_cl;
+
+  int _hash_seed;
+  int _queue_num;
+
+  int _term_attempts;
+#if G1_DETAILED_STATS
+  int _pushes, _pops, _steals, _steal_attempts;
+  int _overflow_pushes;
+#endif
+
+  double _start;
+  double _start_strong_roots;
+  double _strong_roots_time;
+  double _start_term;
+  double _term_time;
+
+  // Map from young-age-index (0 == not young, 1 is youngest) to
+  // surviving words. base is what we get back from the malloc call
+  size_t* _surviving_young_words_base;
+  // this points into the array, as we use the first few entries for padding
+  size_t* _surviving_young_words;
+
+#define PADDING_ELEM_NUM (64 / sizeof(size_t))
+
+  void   add_to_alloc_buffer_waste(size_t waste) { _alloc_buffer_waste += waste; }
+
+  void   add_to_undo_waste(size_t waste)         { _undo_waste += waste; }
+
+  DirtyCardQueue& dirty_card_queue()             { return _dcq;  }
+  CardTableModRefBS* ctbs()                      { return _ct_bs; }
+
+  template <class T> void immediate_rs_update(HeapRegion* from, T* p, int tid) {
+    if (!from->is_survivor()) {
+      _g1_rem->par_write_ref(from, p, tid);
+    }
+  }
+
+  template <class T> void deferred_rs_update(HeapRegion* from, T* p, int tid) {
+    // If the new value of the field points to the same region or
+    // is the to-space, we don't need to include it in the Rset updates.
+    if (!from->is_in_reserved(oopDesc::load_decode_heap_oop(p)) && !from->is_survivor()) {
+      size_t card_index = ctbs()->index_for(p);
+      // If the card hasn't been added to the buffer, do it.
+      if (ctbs()->mark_card_deferred(card_index)) {
+        dirty_card_queue().enqueue((jbyte*)ctbs()->byte_for_index(card_index));
+      }
+    }
+  }
+
+public:
+  G1ParScanThreadState(G1CollectedHeap* g1h, int queue_num);
+
+  ~G1ParScanThreadState() {
+    FREE_C_HEAP_ARRAY(size_t, _surviving_young_words_base);
+  }
+
+  RefToScanQueue*   refs()            { return _refs;             }
+  OverflowQueue*    overflowed_refs() { return _overflowed_refs;  }
+  ageTable*         age_table()       { return &_age_table;       }
+
+  G1ParGCAllocBuffer* alloc_buffer(GCAllocPurpose purpose) {
+    return &_alloc_buffers[purpose];
+  }
+
+  size_t alloc_buffer_waste()                    { return _alloc_buffer_waste; }
+  size_t undo_waste()                            { return _undo_waste; }
+
+  template <class T> void push_on_queue(T* ref) {
+    assert(ref != NULL, "invariant");
+    assert(has_partial_array_mask(ref) ||
+           _g1h->obj_in_cs(oopDesc::load_decode_heap_oop(ref)), "invariant");
+#ifdef ASSERT
+    if (has_partial_array_mask(ref)) {
+      oop p = clear_partial_array_mask(ref);
+      // Verify that we point into the CS
+      assert(_g1h->obj_in_cs(p), "Should be in CS");
+    }
+#endif
+    if (!refs()->push(ref)) {
+      overflowed_refs()->push(ref);
+      IF_G1_DETAILED_STATS(note_overflow_push());
+    } else {
+      IF_G1_DETAILED_STATS(note_push());
+    }
+  }
+
+  void pop_from_queue(StarTask& ref) {
+    if (refs()->pop_local(ref)) {
+      assert((oop*)ref != NULL, "pop_local() returned true");
+      assert(UseCompressedOops || !ref.is_narrow(), "Error");
+      assert(has_partial_array_mask((oop*)ref) ||
+             _g1h->obj_in_cs(ref.is_narrow() ? oopDesc::load_decode_heap_oop((narrowOop*)ref)
+                                             : oopDesc::load_decode_heap_oop((oop*)ref)),
+             "invariant");
+      IF_G1_DETAILED_STATS(note_pop());
+    } else {
+      StarTask null_task;
+      ref = null_task;
+    }
+  }
+
+  void pop_from_overflow_queue(StarTask& ref) {
+    StarTask new_ref = overflowed_refs()->pop();
+    assert((oop*)new_ref != NULL, "pop() from a local non-empty stack");
+    assert(UseCompressedOops || !new_ref.is_narrow(), "Error");
+    assert(has_partial_array_mask((oop*)new_ref) ||
+           _g1h->obj_in_cs(new_ref.is_narrow() ? oopDesc::load_decode_heap_oop((narrowOop*)new_ref)
+                                               : oopDesc::load_decode_heap_oop((oop*)new_ref)),
+             "invariant");
+    ref = new_ref;
+  }
+
+  int refs_to_scan()                             { return refs()->size();                 }
+  int overflowed_refs_to_scan()                  { return overflowed_refs()->length();    }
+
+  template <class T> void update_rs(HeapRegion* from, T* p, int tid) {
+    if (G1DeferredRSUpdate) {
+      deferred_rs_update(from, p, tid);
+    } else {
+      immediate_rs_update(from, p, tid);
+    }
+  }
+
+  HeapWord* allocate_slow(GCAllocPurpose purpose, size_t word_sz) {
+
+    HeapWord* obj = NULL;
+    if (word_sz * 100 <
+        (size_t)(G1ParallelGCAllocBufferSize / HeapWordSize) *
+                                                  ParallelGCBufferWastePct) {
+      G1ParGCAllocBuffer* alloc_buf = alloc_buffer(purpose);
+      add_to_alloc_buffer_waste(alloc_buf->words_remaining());
+      alloc_buf->retire(false, false);
+
+      HeapWord* buf =
+        _g1h->par_allocate_during_gc(purpose, G1ParallelGCAllocBufferSize / HeapWordSize);
+      if (buf == NULL) return NULL; // Let caller handle allocation failure.
+      // Otherwise.
+      alloc_buf->set_buf(buf);
+
+      obj = alloc_buf->allocate(word_sz);
+      assert(obj != NULL, "buffer was definitely big enough...");
+    } else {
+      obj = _g1h->par_allocate_during_gc(purpose, word_sz);
+    }
+    return obj;
+  }
+
+  HeapWord* allocate(GCAllocPurpose purpose, size_t word_sz) {
+    HeapWord* obj = alloc_buffer(purpose)->allocate(word_sz);
+    if (obj != NULL) return obj;
+    return allocate_slow(purpose, word_sz);
+  }
+
+  void undo_allocation(GCAllocPurpose purpose, HeapWord* obj, size_t word_sz) {
+    if (alloc_buffer(purpose)->contains(obj)) {
+      assert(alloc_buffer(purpose)->contains(obj + word_sz - 1),
+             "should contain whole object");
+      alloc_buffer(purpose)->undo_allocation(obj, word_sz);
+    } else {
+      CollectedHeap::fill_with_object(obj, word_sz);
+      add_to_undo_waste(word_sz);
+    }
+  }
+
+  void set_evac_failure_closure(OopsInHeapRegionClosure* evac_failure_cl) {
+    _evac_failure_cl = evac_failure_cl;
+  }
+  OopsInHeapRegionClosure* evac_failure_closure() {
+    return _evac_failure_cl;
+  }
+
+  void set_evac_closure(G1ParScanHeapEvacClosure* evac_cl) {
+    _evac_cl = evac_cl;
+  }
+
+  void set_partial_scan_closure(G1ParScanPartialArrayClosure* partial_scan_cl) {
+    _partial_scan_cl = partial_scan_cl;
+  }
+
+  int* hash_seed() { return &_hash_seed; }
+  int  queue_num() { return _queue_num; }
+
+  int term_attempts()   { return _term_attempts; }
+  void note_term_attempt()  { _term_attempts++; }
+
+#if G1_DETAILED_STATS
+  int pushes()          { return _pushes; }
+  int pops()            { return _pops; }
+  int steals()          { return _steals; }
+  int steal_attempts()  { return _steal_attempts; }
+  int overflow_pushes() { return _overflow_pushes; }
+
+  void note_push()          { _pushes++; }
+  void note_pop()           { _pops++; }
+  void note_steal()         { _steals++; }
+  void note_steal_attempt() { _steal_attempts++; }
+  void note_overflow_push() { _overflow_pushes++; }
+#endif
+
+  void start_strong_roots() {
+    _start_strong_roots = os::elapsedTime();
+  }
+  void end_strong_roots() {
+    _strong_roots_time += (os::elapsedTime() - _start_strong_roots);
+  }
+  double strong_roots_time() { return _strong_roots_time; }
+
+  void start_term_time() {
+    note_term_attempt();
+    _start_term = os::elapsedTime();
+  }
+  void end_term_time() {
+    _term_time += (os::elapsedTime() - _start_term);
+  }
+  double term_time() { return _term_time; }
+
+  double elapsed() {
+    return os::elapsedTime() - _start;
+  }
+
+  size_t* surviving_young_words() {
+    // We add on to hide entry 0 which accumulates surviving words for
+    // age -1 regions (i.e. non-young ones)
+    return _surviving_young_words;
+  }
+
+  void retire_alloc_buffers() {
+    for (int ap = 0; ap < GCAllocPurposeCount; ++ap) {
+      size_t waste = _alloc_buffers[ap].words_remaining();
+      add_to_alloc_buffer_waste(waste);
+      _alloc_buffers[ap].retire(true, false);
+    }
+  }
+
+private:
+  template <class T> void deal_with_reference(T* ref_to_scan) {
+    if (has_partial_array_mask(ref_to_scan)) {
+      _partial_scan_cl->do_oop_nv(ref_to_scan);
+    } else {
+      // Note: we can use "raw" versions of "region_containing" because
+      // "obj_to_scan" is definitely in the heap, and is not in a
+      // humongous region.
+      HeapRegion* r = _g1h->heap_region_containing_raw(ref_to_scan);
+      _evac_cl->set_region(r);
+      _evac_cl->do_oop_nv(ref_to_scan);
+    }
+  }
+
+public:
+  void trim_queue() {
+    // I've replicated the loop twice, first to drain the overflow
+    // queue, second to drain the task queue. This is better than
+    // having a single loop, which checks both conditions and, inside
+    // it, either pops the overflow queue or the task queue, as each
+    // loop is tighter. Also, the decision to drain the overflow queue
+    // first is not arbitrary, as the overflow queue is not visible
+    // to the other workers, whereas the task queue is. So, we want to
+    // drain the "invisible" entries first, while allowing the other
+    // workers to potentially steal the "visible" entries.
+
+    while (refs_to_scan() > 0 || overflowed_refs_to_scan() > 0) {
+      while (overflowed_refs_to_scan() > 0) {
+        StarTask ref_to_scan;
+        assert((oop*)ref_to_scan == NULL, "Constructed above");
+        pop_from_overflow_queue(ref_to_scan);
+        // We shouldn't have pushed it on the queue if it was not
+        // pointing into the CSet.
+        assert((oop*)ref_to_scan != NULL, "Follows from inner loop invariant");
+        if (ref_to_scan.is_narrow()) {
+          assert(UseCompressedOops, "Error");
+          narrowOop* p = (narrowOop*)ref_to_scan;
+          assert(!has_partial_array_mask(p) &&
+                 _g1h->obj_in_cs(oopDesc::load_decode_heap_oop(p)), "sanity");
+          deal_with_reference(p);
+        } else {
+          oop* p = (oop*)ref_to_scan;
+          assert((has_partial_array_mask(p) && _g1h->obj_in_cs(clear_partial_array_mask(p))) ||
+                 _g1h->obj_in_cs(oopDesc::load_decode_heap_oop(p)), "sanity");
+          deal_with_reference(p);
+        }
+      }
+
+      while (refs_to_scan() > 0) {
+        StarTask ref_to_scan;
+        assert((oop*)ref_to_scan == NULL, "Constructed above");
+        pop_from_queue(ref_to_scan);
+        if ((oop*)ref_to_scan != NULL) {
+          if (ref_to_scan.is_narrow()) {
+            assert(UseCompressedOops, "Error");
+            narrowOop* p = (narrowOop*)ref_to_scan;
+            assert(!has_partial_array_mask(p) &&
+                   _g1h->obj_in_cs(oopDesc::load_decode_heap_oop(p)), "sanity");
+            deal_with_reference(p);
+          } else {
+            oop* p = (oop*)ref_to_scan;
+            assert((has_partial_array_mask(p) && _g1h->obj_in_cs(clear_partial_array_mask(p))) ||
+                  _g1h->obj_in_cs(oopDesc::load_decode_heap_oop(p)), "sanity");
+            deal_with_reference(p);
+          }
+        }
+      }
+    }
+  }
+};
diff --git a/hotspot/src/share/vm/gc_implementation/g1/g1CollectorPolicy.cpp b/hotspot/src/share/vm/gc_implementation/g1/g1CollectorPolicy.cpp
index 2833cd7..bb219c5 100644
--- a/hotspot/src/share/vm/gc_implementation/g1/g1CollectorPolicy.cpp
+++ b/hotspot/src/share/vm/gc_implementation/g1/g1CollectorPolicy.cpp
@@ -293,10 +293,6 @@
   if (G1SteadyStateUsed < 50) {
     vm_exit_during_initialization("G1SteadyStateUsed must be at least 50%.");
   }
-  if (UseConcMarkSweepGC) {
-    vm_exit_during_initialization("-XX:+UseG1GC is incompatible with "
-                                  "-XX:+UseConcMarkSweepGC.");
-  }
 
   initialize_gc_policy_counters();
 
diff --git a/hotspot/src/share/vm/gc_implementation/g1/g1OopClosures.hpp b/hotspot/src/share/vm/gc_implementation/g1/g1OopClosures.hpp
index 1619770..f710203 100644
--- a/hotspot/src/share/vm/gc_implementation/g1/g1OopClosures.hpp
+++ b/hotspot/src/share/vm/gc_implementation/g1/g1OopClosures.hpp
@@ -42,18 +42,6 @@
   virtual void set_region(HeapRegion* from) { _from = from; }
 };
 
-
-class G1ScanAndBalanceClosure : public OopClosure {
-  G1CollectedHeap* _g1;
-  static int _nq;
-public:
-  G1ScanAndBalanceClosure(G1CollectedHeap* g1) : _g1(g1) { }
-  inline  void do_oop_nv(oop* p);
-  inline  void do_oop_nv(narrowOop* p) { guarantee(false, "NYI"); }
-  virtual void do_oop(oop* p);
-  virtual void do_oop(narrowOop* p)    { guarantee(false, "NYI"); }
-};
-
 class G1ParClosureSuper : public OopsInHeapRegionClosure {
 protected:
   G1CollectedHeap* _g1;
@@ -69,34 +57,32 @@
 public:
   G1ParScanClosure(G1CollectedHeap* g1, G1ParScanThreadState* par_scan_state) :
     G1ParClosureSuper(g1, par_scan_state) { }
-  void do_oop_nv(oop* p);   // should be made inline
-  inline  void do_oop_nv(narrowOop* p) { guarantee(false, "NYI"); }
+  template <class T> void do_oop_nv(T* p);
   virtual void do_oop(oop* p)          { do_oop_nv(p); }
   virtual void do_oop(narrowOop* p)    { do_oop_nv(p); }
 };
 
-#define G1_PARTIAL_ARRAY_MASK 1
+#define G1_PARTIAL_ARRAY_MASK 0x2
 
-inline bool has_partial_array_mask(oop* ref) {
-  return (intptr_t) ref & G1_PARTIAL_ARRAY_MASK;
+template <class T> inline bool has_partial_array_mask(T* ref) {
+  return ((uintptr_t)ref & G1_PARTIAL_ARRAY_MASK) == G1_PARTIAL_ARRAY_MASK;
 }
 
-inline oop* set_partial_array_mask(oop obj) {
-  return (oop*) ((intptr_t) obj | G1_PARTIAL_ARRAY_MASK);
+template <class T> inline T* set_partial_array_mask(T obj) {
+  assert(((uintptr_t)obj & G1_PARTIAL_ARRAY_MASK) == 0, "Information loss!");
+  return (T*) ((uintptr_t)obj | G1_PARTIAL_ARRAY_MASK);
 }
 
-inline oop clear_partial_array_mask(oop* ref) {
-  return oop((intptr_t) ref & ~G1_PARTIAL_ARRAY_MASK);
+template <class T> inline oop clear_partial_array_mask(T* ref) {
+  return oop((intptr_t)ref & ~G1_PARTIAL_ARRAY_MASK);
 }
 
 class G1ParScanPartialArrayClosure : public G1ParClosureSuper {
   G1ParScanClosure _scanner;
-  template <class T> void process_array_chunk(oop obj, int start, int end);
 public:
   G1ParScanPartialArrayClosure(G1CollectedHeap* g1, G1ParScanThreadState* par_scan_state) :
     G1ParClosureSuper(g1, par_scan_state), _scanner(g1, par_scan_state) { }
-  void do_oop_nv(oop* p);
-  void do_oop_nv(narrowOop* p)      { guarantee(false, "NYI"); }
+  template <class T> void do_oop_nv(T* p);
   virtual void do_oop(oop* p)       { do_oop_nv(p); }
   virtual void do_oop(narrowOop* p) { do_oop_nv(p); }
 };
@@ -105,7 +91,7 @@
 class G1ParCopyHelper : public G1ParClosureSuper {
   G1ParScanClosure *_scanner;
 protected:
-  void mark_forwardee(oop* p);
+  template <class T> void mark_forwardee(T* p);
   oop copy_to_survivor_space(oop obj);
 public:
   G1ParCopyHelper(G1CollectedHeap* g1, G1ParScanThreadState* par_scan_state,
@@ -117,36 +103,35 @@
          bool do_mark_forwardee, bool skip_cset_test>
 class G1ParCopyClosure : public G1ParCopyHelper {
   G1ParScanClosure _scanner;
-  void do_oop_work(oop* p);
-  void do_oop_work(narrowOop* p) { guarantee(false, "NYI"); }
+  template <class T> void do_oop_work(T* p);
 public:
   G1ParCopyClosure(G1CollectedHeap* g1, G1ParScanThreadState* par_scan_state) :
     _scanner(g1, par_scan_state), G1ParCopyHelper(g1, par_scan_state, &_scanner) { }
-  inline void do_oop_nv(oop* p) {
+  template <class T> void do_oop_nv(T* p) {
     do_oop_work(p);
     if (do_mark_forwardee)
       mark_forwardee(p);
   }
-  inline void do_oop_nv(narrowOop* p) { guarantee(false, "NYI"); }
   virtual void do_oop(oop* p)       { do_oop_nv(p); }
   virtual void do_oop(narrowOop* p) { do_oop_nv(p); }
 };
 
 typedef G1ParCopyClosure<false, G1BarrierNone, false, false> G1ParScanExtRootClosure;
 typedef G1ParCopyClosure<true,  G1BarrierNone, false, false> G1ParScanPermClosure;
+typedef G1ParCopyClosure<false, G1BarrierRS,   false, false> G1ParScanHeapRSClosure;
 typedef G1ParCopyClosure<false, G1BarrierNone, true,  false> G1ParScanAndMarkExtRootClosure;
 typedef G1ParCopyClosure<true,  G1BarrierNone, true,  false> G1ParScanAndMarkPermClosure;
-typedef G1ParCopyClosure<false, G1BarrierRS,   false, false> G1ParScanHeapRSClosure;
 typedef G1ParCopyClosure<false, G1BarrierRS,   true,  false> G1ParScanAndMarkHeapRSClosure;
 // This is the only case when we set skip_cset_test. Basically, this
 // closure is (should?) only be called directly while we're draining
 // the overflow and task queues. In that case we know that the
 // reference in question points into the collection set, otherwise we
-// would not have pushed it on the queue.
-typedef G1ParCopyClosure<false, G1BarrierEvac, false, true> G1ParScanHeapEvacClosure;
+// would not have pushed it on the queue. The following is defined in
+// g1_specialized_oop_closures.hpp.
+// typedef G1ParCopyClosure<false, G1BarrierEvac, false, true> G1ParScanHeapEvacClosure;
 // We need a separate closure to handle references during evacuation
-// failure processing, as it cannot asume that the reference already
- // points to the collection set (like G1ParScanHeapEvacClosure does).
+// failure processing, as we cannot asume that the reference already
+// points into the collection set (like G1ParScanHeapEvacClosure does).
 typedef G1ParCopyClosure<false, G1BarrierEvac, false, false> G1ParScanHeapEvacFailureClosure;
 
 class FilterIntoCSClosure: public OopClosure {
@@ -158,10 +143,9 @@
                         G1CollectedHeap* g1, OopClosure* oc) :
     _dcto_cl(dcto_cl), _g1(g1), _oc(oc)
   {}
-  inline void do_oop_nv(oop* p);
-  inline void do_oop_nv(narrowOop* p) { guarantee(false, "NYI"); }
-  virtual void do_oop(oop* p);
-  virtual void do_oop(narrowOop* p)   { guarantee(false, "NYI"); }
+  template <class T> void do_oop_nv(T* p);
+  virtual void do_oop(oop* p)        { do_oop_nv(p); }
+  virtual void do_oop(narrowOop* p)  { do_oop_nv(p); }
   bool apply_to_weak_ref_discovered_field() { return true; }
   bool do_header() { return false; }
 };
@@ -174,10 +158,9 @@
                                      OopsInHeapRegionClosure* oc) :
     _g1(g1), _oc(oc)
   {}
-  inline  void do_oop_nv(oop* p);
-  inline  void do_oop_nv(narrowOop* p) { guarantee(false, "NYI"); }
-  virtual void do_oop(oop* p);
-  virtual void do_oop(narrowOop* p)    { guarantee(false, "NYI"); }
+  template <class T> void do_oop_nv(T* p);
+  virtual void do_oop(oop* p) { do_oop_nv(p); }
+  virtual void do_oop(narrowOop* p) { do_oop_nv(p); }
   bool apply_to_weak_ref_discovered_field() { return true; }
   bool do_header() { return false; }
   void set_region(HeapRegion* from) {
@@ -195,10 +178,9 @@
                                             ConcurrentMark* cm)
   : _g1(g1), _oc(oc), _cm(cm) { }
 
-  inline  void do_oop_nv(oop* p);
-  inline  void do_oop_nv(narrowOop* p) { guarantee(false, "NYI"); }
-  virtual void do_oop(oop* p);
-  virtual void do_oop(narrowOop* p)    { guarantee(false, "NYI"); }
+  template <class T> void do_oop_nv(T* p);
+  virtual void do_oop(oop* p) { do_oop_nv(p); }
+  virtual void do_oop(narrowOop* p) { do_oop_nv(p); }
   bool apply_to_weak_ref_discovered_field() { return true; }
   bool do_header() { return false; }
   void set_region(HeapRegion* from) {
@@ -213,10 +195,9 @@
   int _out_of_region;
 public:
   FilterOutOfRegionClosure(HeapRegion* r, OopClosure* oc);
-  inline  void do_oop_nv(oop* p);
-  inline  void do_oop_nv(narrowOop* p) { guarantee(false, "NYI"); }
-  virtual void do_oop(oop* p);
-  virtual void do_oop(narrowOop* p)   { guarantee(false, "NYI"); }
+  template <class T> void do_oop_nv(T* p);
+  virtual void do_oop(oop* p) { do_oop_nv(p); }
+  virtual void do_oop(narrowOop* p) { do_oop_nv(p); }
   bool apply_to_weak_ref_discovered_field() { return true; }
   bool do_header() { return false; }
   int out_of_region() { return _out_of_region; }
diff --git a/hotspot/src/share/vm/gc_implementation/g1/g1OopClosures.inline.hpp b/hotspot/src/share/vm/gc_implementation/g1/g1OopClosures.inline.hpp
index fdca160..2ba92a7 100644
--- a/hotspot/src/share/vm/gc_implementation/g1/g1OopClosures.inline.hpp
+++ b/hotspot/src/share/vm/gc_implementation/g1/g1OopClosures.inline.hpp
@@ -31,9 +31,10 @@
 // perf-critical inner loop.
 #define FILTERINTOCSCLOSURE_DOHISTOGRAMCOUNT 0
 
-inline void FilterIntoCSClosure::do_oop_nv(oop* p) {
-  oop obj = *p;
-  if (obj != NULL && _g1->obj_in_cs(obj)) {
+template <class T> inline void FilterIntoCSClosure::do_oop_nv(T* p) {
+  T heap_oop = oopDesc::load_heap_oop(p);
+  if (!oopDesc::is_null(heap_oop) &&
+      _g1->obj_in_cs(oopDesc::decode_heap_oop_not_null(heap_oop))) {
     _oc->do_oop(p);
 #if FILTERINTOCSCLOSURE_DOHISTOGRAMCOUNT
     _dcto_cl->incr_count();
@@ -41,44 +42,32 @@
   }
 }
 
-inline void FilterIntoCSClosure::do_oop(oop* p)
-{
-  do_oop_nv(p);
-}
-
 #define FILTEROUTOFREGIONCLOSURE_DOHISTOGRAMCOUNT 0
 
-inline void FilterOutOfRegionClosure::do_oop_nv(oop* p) {
-  oop obj = *p;
-  HeapWord* obj_hw = (HeapWord*)obj;
-  if (obj_hw != NULL && (obj_hw < _r_bottom || obj_hw >= _r_end)) {
-    _oc->do_oop(p);
+template <class T> inline void FilterOutOfRegionClosure::do_oop_nv(T* p) {
+  T heap_oop = oopDesc::load_heap_oop(p);
+  if (!oopDesc::is_null(heap_oop)) {
+    HeapWord* obj_hw = (HeapWord*)oopDesc::decode_heap_oop_not_null(heap_oop);
+    if (obj_hw < _r_bottom || obj_hw >= _r_end) {
+      _oc->do_oop(p);
 #if FILTEROUTOFREGIONCLOSURE_DOHISTOGRAMCOUNT
-    _out_of_region++;
+      _out_of_region++;
 #endif
+    }
   }
 }
 
-inline void FilterOutOfRegionClosure::do_oop(oop* p)
-{
-  do_oop_nv(p);
-}
-
-inline void FilterInHeapRegionAndIntoCSClosure::do_oop_nv(oop* p) {
-  oop obj = *p;
-  if (obj != NULL && _g1->obj_in_cs(obj))
+template <class T> inline void FilterInHeapRegionAndIntoCSClosure::do_oop_nv(T* p) {
+  T heap_oop = oopDesc::load_heap_oop(p);
+  if (!oopDesc::is_null(heap_oop) &&
+      _g1->obj_in_cs(oopDesc::decode_heap_oop_not_null(heap_oop)))
     _oc->do_oop(p);
 }
 
-inline void FilterInHeapRegionAndIntoCSClosure::do_oop(oop* p)
-{
-  do_oop_nv(p);
-}
-
-
-inline void FilterAndMarkInHeapRegionAndIntoCSClosure::do_oop_nv(oop* p) {
-  oop obj = *p;
-  if (obj != NULL) {
+template <class T> inline void FilterAndMarkInHeapRegionAndIntoCSClosure::do_oop_nv(T* p) {
+  T heap_oop = oopDesc::load_heap_oop(p);
+  if (!oopDesc::is_null(heap_oop)) {
+    oop obj = oopDesc::decode_heap_oop_not_null(heap_oop);
     HeapRegion* hr = _g1->heap_region_containing((HeapWord*) obj);
     if (hr != NULL) {
       if (hr->in_collection_set())
@@ -89,24 +78,29 @@
   }
 }
 
-inline void FilterAndMarkInHeapRegionAndIntoCSClosure::do_oop(oop* p)
-{
-  do_oop_nv(p);
-}
+// This closure is applied to the fields of the objects that have just been copied.
+template <class T> inline void G1ParScanClosure::do_oop_nv(T* p) {
+  T heap_oop = oopDesc::load_heap_oop(p);
 
-inline void G1ScanAndBalanceClosure::do_oop_nv(oop* p) {
-  RefToScanQueue* q;
-  if (ParallelGCThreads > 0) {
-    // Deal the work out equally.
-    _nq = (_nq + 1) % ParallelGCThreads;
-    q = _g1->task_queue(_nq);
-  } else {
-    q = _g1->task_queue(0);
+  if (!oopDesc::is_null(heap_oop)) {
+    oop obj = oopDesc::decode_heap_oop_not_null(heap_oop);
+    if (_g1->in_cset_fast_test(obj)) {
+      // We're not going to even bother checking whether the object is
+      // already forwarded or not, as this usually causes an immediate
+      // stall. We'll try to prefetch the object (for write, given that
+      // we might need to install the forwarding reference) and we'll
+      // get back to it when pop it from the queue
+      Prefetch::write(obj->mark_addr(), 0);
+      Prefetch::read(obj->mark_addr(), (HeapWordSize*2));
+
+      // slightly paranoid test; I'm trying to catch potential
+      // problems before we go into push_on_queue to know where the
+      // problem is coming from
+      assert(obj == oopDesc::load_decode_heap_oop(p),
+             "p should still be pointing to obj");
+      _par_scan_state->push_on_queue(p);
+    } else {
+      _par_scan_state->update_rs(_from, p, _par_scan_state->queue_num());
+    }
   }
-  bool nooverflow = q->push(p);
-  guarantee(nooverflow, "Overflow during poplularity region processing");
-}
-
-inline void G1ScanAndBalanceClosure::do_oop(oop* p) {
-  do_oop_nv(p);
 }
diff --git a/hotspot/src/share/vm/gc_implementation/g1/g1RemSet.cpp b/hotspot/src/share/vm/gc_implementation/g1/g1RemSet.cpp
index 86a39a7..9c36026 100644
--- a/hotspot/src/share/vm/gc_implementation/g1/g1RemSet.cpp
+++ b/hotspot/src/share/vm/gc_implementation/g1/g1RemSet.cpp
@@ -65,11 +65,10 @@
   void set_region(HeapRegion* from) {
     _blk->set_region(from);
   }
-  virtual void do_oop(narrowOop* p) {
-    guarantee(false, "NYI");
-  }
-  virtual void do_oop(oop* p) {
-    oop obj = *p;
+  virtual void do_oop(narrowOop* p) { do_oop_work(p); }
+  virtual void do_oop(      oop* p) { do_oop_work(p); }
+  template <class T> void do_oop_work(T* p) {
+    oop obj = oopDesc::load_decode_heap_oop(p);
     if (_g1->obj_in_cs(obj)) _blk->do_oop(p);
   }
   bool apply_to_weak_ref_discovered_field() { return true; }
@@ -110,11 +109,10 @@
 public:
   VerifyRSCleanCardOopClosure(G1CollectedHeap* g1) : _g1(g1) {}
 
-  virtual void do_oop(narrowOop* p) {
-    guarantee(false, "NYI");
-  }
-  virtual void do_oop(oop* p) {
-    oop obj = *p;
+  virtual void do_oop(narrowOop* p) { do_oop_work(p); }
+  virtual void do_oop(      oop* p) { do_oop_work(p); }
+  template <class T> void do_oop_work(T* p) {
+    oop obj = oopDesc::load_decode_heap_oop(p);
     HeapRegion* to = _g1->heap_region_containing(obj);
     guarantee(to == NULL || !to->in_collection_set(),
               "Missed a rem set member.");
@@ -129,9 +127,9 @@
 {
   _seq_task = new SubTasksDone(NumSeqTasks);
   guarantee(n_workers() > 0, "There should be some workers");
-  _new_refs = NEW_C_HEAP_ARRAY(GrowableArray<oop*>*, n_workers());
+  _new_refs = NEW_C_HEAP_ARRAY(GrowableArray<OopOrNarrowOopStar>*, n_workers());
   for (uint i = 0; i < n_workers(); i++) {
-    _new_refs[i] = new (ResourceObj::C_HEAP) GrowableArray<oop*>(8192,true);
+    _new_refs[i] = new (ResourceObj::C_HEAP) GrowableArray<OopOrNarrowOopStar>(8192,true);
   }
 }
 
@@ -140,7 +138,7 @@
   for (uint i = 0; i < n_workers(); i++) {
     delete _new_refs[i];
   }
-  FREE_C_HEAP_ARRAY(GrowableArray<oop*>*, _new_refs);
+  FREE_C_HEAP_ARRAY(GrowableArray<OopOrNarrowOopStar>*, _new_refs);
 }
 
 void CountNonCleanMemRegionClosure::do_MemRegion(MemRegion mr) {
@@ -428,15 +426,15 @@
   }
 };
 
-void
-HRInto_G1RemSet::scanNewRefsRS(OopsInHeapRegionClosure* oc,
-                                             int worker_i) {
+template <class T> void
+HRInto_G1RemSet::scanNewRefsRS_work(OopsInHeapRegionClosure* oc,
+                                    int worker_i) {
   double scan_new_refs_start_sec = os::elapsedTime();
   G1CollectedHeap* g1h = G1CollectedHeap::heap();
   CardTableModRefBS* ct_bs = (CardTableModRefBS*) (g1h->barrier_set());
   for (int i = 0; i < _new_refs[worker_i]->length(); i++) {
-    oop* p = _new_refs[worker_i]->at(i);
-    oop obj = *p;
+    T* p = (T*) _new_refs[worker_i]->at(i);
+    oop obj = oopDesc::load_decode_heap_oop(p);
     // *p was in the collection set when p was pushed on "_new_refs", but
     // another thread may have processed this location from an RS, so it
     // might not point into the CS any longer.  If so, it's obviously been
@@ -549,11 +547,10 @@
   G1CollectedHeap* _g1;
 public:
   UpdateRSetOopsIntoCSImmediate(G1CollectedHeap* g1) : _g1(g1) { }
-  virtual void do_oop(narrowOop* p) {
-    guarantee(false, "NYI");
-  }
-  virtual void do_oop(oop* p) {
-    HeapRegion* to = _g1->heap_region_containing(*p);
+  virtual void do_oop(narrowOop* p) { do_oop_work(p); }
+  virtual void do_oop(      oop* p) { do_oop_work(p); }
+  template <class T> void do_oop_work(T* p) {
+    HeapRegion* to = _g1->heap_region_containing(oopDesc::load_decode_heap_oop(p));
     if (to->in_collection_set()) {
       to->rem_set()->add_reference(p, 0);
     }
@@ -567,11 +564,10 @@
 public:
   UpdateRSetOopsIntoCSDeferred(G1CollectedHeap* g1, DirtyCardQueue* dcq) :
     _g1(g1), _ct_bs((CardTableModRefBS*)_g1->barrier_set()), _dcq(dcq) { }
-  virtual void do_oop(narrowOop* p) {
-    guarantee(false, "NYI");
-  }
-  virtual void do_oop(oop* p) {
-    oop obj = *p;
+  virtual void do_oop(narrowOop* p) { do_oop_work(p); }
+  virtual void do_oop(      oop* p) { do_oop_work(p); }
+  template <class T> void do_oop_work(T* p) {
+    oop obj = oopDesc::load_decode_heap_oop(p);
     if (_g1->obj_in_cs(obj)) {
       size_t card_index = _ct_bs->index_for(p);
       if (_ct_bs->mark_card_deferred(card_index)) {
@@ -581,10 +577,10 @@
   }
 };
 
-void HRInto_G1RemSet::new_refs_iterate(OopClosure* cl) {
+template <class T> void HRInto_G1RemSet::new_refs_iterate_work(OopClosure* cl) {
   for (size_t i = 0; i < n_workers(); i++) {
     for (int j = 0; j < _new_refs[i]->length(); j++) {
-      oop* p = _new_refs[i]->at(j);
+      T* p = (T*) _new_refs[i]->at(j);
       cl->do_oop(p);
     }
   }
diff --git a/hotspot/src/share/vm/gc_implementation/g1/g1RemSet.hpp b/hotspot/src/share/vm/gc_implementation/g1/g1RemSet.hpp
index 4d58e0b..60ba420 100644
--- a/hotspot/src/share/vm/gc_implementation/g1/g1RemSet.hpp
+++ b/hotspot/src/share/vm/gc_implementation/g1/g1RemSet.hpp
@@ -62,10 +62,12 @@
   // If "this" is of the given subtype, return "this", else "NULL".
   virtual HRInto_G1RemSet* as_HRInto_G1RemSet() { return NULL; }
 
-  // Record, if necessary, the fact that *p (where "p" is in region "from")
-  // has changed to its new value.
+  // Record, if necessary, the fact that *p (where "p" is in region "from",
+  // and is, a fortiori, required to be non-NULL) has changed to its new value.
   virtual void write_ref(HeapRegion* from, oop* p) = 0;
+  virtual void write_ref(HeapRegion* from, narrowOop* p) = 0;
   virtual void par_write_ref(HeapRegion* from, oop* p, int tid) = 0;
+  virtual void par_write_ref(HeapRegion* from, narrowOop* p, int tid) = 0;
 
   // Requires "region_bm" and "card_bm" to be bitmaps with 1 bit per region
   // or card, respectively, such that a region or card with a corresponding
@@ -105,7 +107,9 @@
 
   // Nothing is necessary in the version below.
   void write_ref(HeapRegion* from, oop* p) {}
+  void write_ref(HeapRegion* from, narrowOop* p) {}
   void par_write_ref(HeapRegion* from, oop* p, int tid) {}
+  void par_write_ref(HeapRegion* from, narrowOop* p, int tid) {}
 
   void scrub(BitMap* region_bm, BitMap* card_bm) {}
   void scrub_par(BitMap* region_bm, BitMap* card_bm,
@@ -143,8 +147,19 @@
   // their references into the collection summarized in "_new_refs".
   bool _par_traversal_in_progress;
   void set_par_traversal(bool b) { _par_traversal_in_progress = b; }
-  GrowableArray<oop*>** _new_refs;
-  void new_refs_iterate(OopClosure* cl);
+  GrowableArray<OopOrNarrowOopStar>** _new_refs;
+  template <class T> void new_refs_iterate_work(OopClosure* cl);
+  void new_refs_iterate(OopClosure* cl) {
+    if (UseCompressedOops) {
+      new_refs_iterate_work<narrowOop>(cl);
+    } else {
+      new_refs_iterate_work<oop>(cl);
+    }
+  }
+
+protected:
+  template <class T> void write_ref_nv(HeapRegion* from, T* p);
+  template <class T> void par_write_ref_nv(HeapRegion* from, T* p, int tid);
 
 public:
   // This is called to reset dual hash tables after the gc pause
@@ -161,7 +176,14 @@
   void prepare_for_oops_into_collection_set_do();
   void cleanup_after_oops_into_collection_set_do();
   void scanRS(OopsInHeapRegionClosure* oc, int worker_i);
-  void scanNewRefsRS(OopsInHeapRegionClosure* oc, int worker_i);
+  template <class T> void scanNewRefsRS_work(OopsInHeapRegionClosure* oc, int worker_i);
+  void scanNewRefsRS(OopsInHeapRegionClosure* oc, int worker_i) {
+    if (UseCompressedOops) {
+      scanNewRefsRS_work<narrowOop>(oc, worker_i);
+    } else {
+      scanNewRefsRS_work<oop>(oc, worker_i);
+    }
+  }
   void updateRS(int worker_i);
   HeapRegion* calculateStartRegion(int i);
 
@@ -172,12 +194,22 @@
 
   // Record, if necessary, the fact that *p (where "p" is in region "from",
   // which is required to be non-NULL) has changed to a new non-NULL value.
-  inline void write_ref(HeapRegion* from, oop* p);
-  // The "_nv" version is the same; it exists just so that it is not virtual.
-  inline void write_ref_nv(HeapRegion* from, oop* p);
+  // [Below the virtual version calls a non-virtual protected
+  // workhorse that is templatified for narrow vs wide oop.]
+  inline void write_ref(HeapRegion* from, oop* p) {
+    write_ref_nv(from, p);
+  }
+  inline void write_ref(HeapRegion* from, narrowOop* p) {
+    write_ref_nv(from, p);
+  }
+  inline void par_write_ref(HeapRegion* from, oop* p, int tid) {
+    par_write_ref_nv(from, p, tid);
+  }
+  inline void par_write_ref(HeapRegion* from, narrowOop* p, int tid) {
+    par_write_ref_nv(from, p, tid);
+  }
 
-  inline bool self_forwarded(oop obj);
-  inline void par_write_ref(HeapRegion* from, oop* p, int tid);
+  bool self_forwarded(oop obj);
 
   void scrub(BitMap* region_bm, BitMap* card_bm);
   void scrub_par(BitMap* region_bm, BitMap* card_bm,
@@ -208,6 +240,9 @@
   HeapRegion* _from;
   HRInto_G1RemSet* _rs;
   int _worker_i;
+
+  template <class T> void do_oop_work(T* p);
+
 public:
   UpdateRSOopClosure(HRInto_G1RemSet* rs, int worker_i = 0) :
     _from(NULL), _rs(rs), _worker_i(worker_i) {
@@ -219,11 +254,10 @@
     _from = from;
   }
 
-  virtual void do_oop(narrowOop* p);
-  virtual void do_oop(oop* p);
+  virtual void do_oop(narrowOop* p) { do_oop_work(p); }
+  virtual void do_oop(oop* p)       { do_oop_work(p); }
 
   // Override: this closure is idempotent.
   //  bool idempotent() { return true; }
   bool apply_to_weak_ref_discovered_field() { return true; }
 };
-
diff --git a/hotspot/src/share/vm/gc_implementation/g1/g1RemSet.inline.hpp b/hotspot/src/share/vm/gc_implementation/g1/g1RemSet.inline.hpp
index 00aa144..f14c540 100644
--- a/hotspot/src/share/vm/gc_implementation/g1/g1RemSet.inline.hpp
+++ b/hotspot/src/share/vm/gc_implementation/g1/g1RemSet.inline.hpp
@@ -30,12 +30,8 @@
   }
 }
 
-inline void HRInto_G1RemSet::write_ref_nv(HeapRegion* from, oop* p) {
-  par_write_ref(from, p, 0);
-}
-
-inline void HRInto_G1RemSet::write_ref(HeapRegion* from, oop* p) {
-  write_ref_nv(from, p);
+template <class T> inline void HRInto_G1RemSet::write_ref_nv(HeapRegion* from, T* p) {
+  par_write_ref_nv(from, p, 0);
 }
 
 inline bool HRInto_G1RemSet::self_forwarded(oop obj) {
@@ -43,8 +39,8 @@
   return result;
 }
 
-inline void HRInto_G1RemSet::par_write_ref(HeapRegion* from, oop* p, int tid) {
-  oop obj = *p;
+template <class T> inline void HRInto_G1RemSet::par_write_ref_nv(HeapRegion* from, T* p, int tid) {
+  oop obj = oopDesc::load_decode_heap_oop(p);
 #ifdef ASSERT
   // can't do because of races
   // assert(obj == NULL || obj->is_oop(), "expected an oop");
@@ -71,7 +67,7 @@
     // false during the evacuation failure handing.
     if (_par_traversal_in_progress &&
         to->in_collection_set() && !self_forwarded(obj)) {
-      _new_refs[tid]->push(p);
+      _new_refs[tid]->push((void*)p);
       // Deferred updates to the Cset are either discarded (in the normal case),
       // or processed (if an evacuation failure occurs) at the end
       // of the collection.
@@ -89,11 +85,7 @@
   }
 }
 
-inline void UpdateRSOopClosure::do_oop(narrowOop* p) {
-  guarantee(false, "NYI");
-}
-
-inline void UpdateRSOopClosure::do_oop(oop* p) {
+template <class T> inline void UpdateRSOopClosure::do_oop_work(T* p) {
   assert(_from != NULL, "from region must be non-NULL");
   _rs->par_write_ref(_from, p, _worker_i);
 }
diff --git a/hotspot/src/share/vm/gc_implementation/g1/g1SATBCardTableModRefBS.cpp b/hotspot/src/share/vm/gc_implementation/g1/g1SATBCardTableModRefBS.cpp
index 3741498..f82b8ba 100644
--- a/hotspot/src/share/vm/gc_implementation/g1/g1SATBCardTableModRefBS.cpp
+++ b/hotspot/src/share/vm/gc_implementation/g1/g1SATBCardTableModRefBS.cpp
@@ -34,6 +34,7 @@
 
 
 void G1SATBCardTableModRefBS::enqueue(oop pre_val) {
+  assert(pre_val->is_oop_or_null(true), "Error");
   if (!JavaThread::satb_mark_queue_set().active()) return;
   Thread* thr = Thread::current();
   if (thr->is_Java_thread()) {
@@ -46,32 +47,31 @@
 }
 
 // When we know the current java thread:
-void
-G1SATBCardTableModRefBS::write_ref_field_pre_static(void* field,
-                                                    oop newVal,
+template <class T> void
+G1SATBCardTableModRefBS::write_ref_field_pre_static(T* field,
+                                                    oop new_val,
                                                     JavaThread* jt) {
   if (!JavaThread::satb_mark_queue_set().active()) return;
-  assert(!UseCompressedOops, "Else will need to modify this to deal with narrowOop");
-  oop preVal = *(oop*)field;
-  if (preVal != NULL) {
-    jt->satb_mark_queue().enqueue(preVal);
+  T heap_oop = oopDesc::load_heap_oop(field);
+  if (!oopDesc::is_null(heap_oop)) {
+    oop pre_val = oopDesc::decode_heap_oop_not_null(heap_oop);
+    assert(pre_val->is_oop(true /* ignore mark word */), "Error");
+    jt->satb_mark_queue().enqueue(pre_val);
   }
 }
 
-void
-G1SATBCardTableModRefBS::write_ref_array_pre(MemRegion mr) {
+template <class T> void
+G1SATBCardTableModRefBS::write_ref_array_pre_work(T* dst, int count) {
   if (!JavaThread::satb_mark_queue_set().active()) return;
-  assert(!UseCompressedOops, "Else will need to modify this to deal with narrowOop");
-  oop* elem_ptr = (oop*)mr.start();
-  while ((HeapWord*)elem_ptr < mr.end()) {
-    oop elem = *elem_ptr;
-    if (elem != NULL) enqueue(elem);
-    elem_ptr++;
+  T* elem_ptr = dst;
+  for (int i = 0; i < count; i++, elem_ptr++) {
+    T heap_oop = oopDesc::load_heap_oop(elem_ptr);
+    if (!oopDesc::is_null(heap_oop)) {
+      enqueue(oopDesc::decode_heap_oop_not_null(heap_oop));
+    }
   }
 }
 
-
-
 G1SATBCardTableLoggingModRefBS::
 G1SATBCardTableLoggingModRefBS(MemRegion whole_heap,
                                int max_covered_regions) :
diff --git a/hotspot/src/share/vm/gc_implementation/g1/g1SATBCardTableModRefBS.hpp b/hotspot/src/share/vm/gc_implementation/g1/g1SATBCardTableModRefBS.hpp
index 86f8283..5a5283e 100644
--- a/hotspot/src/share/vm/gc_implementation/g1/g1SATBCardTableModRefBS.hpp
+++ b/hotspot/src/share/vm/gc_implementation/g1/g1SATBCardTableModRefBS.hpp
@@ -47,31 +47,41 @@
 
   // This notes that we don't need to access any BarrierSet data
   // structures, so this can be called from a static context.
-  static void write_ref_field_pre_static(void* field, oop newVal) {
-    assert(!UseCompressedOops, "Else needs to be templatized");
-    oop preVal = *((oop*)field);
-    if (preVal != NULL) {
-      enqueue(preVal);
+  template <class T> static void write_ref_field_pre_static(T* field, oop newVal) {
+    T heap_oop = oopDesc::load_heap_oop(field);
+    if (!oopDesc::is_null(heap_oop)) {
+      enqueue(oopDesc::decode_heap_oop(heap_oop));
     }
   }
 
   // When we know the current java thread:
-  static void write_ref_field_pre_static(void* field, oop newVal,
-                                         JavaThread* jt);
+  template <class T> static void write_ref_field_pre_static(T* field, oop newVal,
+                                                            JavaThread* jt);
 
   // We export this to make it available in cases where the static
   // type of the barrier set is known.  Note that it is non-virtual.
-  inline void inline_write_ref_field_pre(void* field, oop newVal) {
+  template <class T> inline void inline_write_ref_field_pre(T* field, oop newVal) {
     write_ref_field_pre_static(field, newVal);
   }
 
-  // This is the more general virtual version.
-  void write_ref_field_pre_work(void* field, oop new_val) {
+  // These are the more general virtual versions.
+  virtual void write_ref_field_pre_work(oop* field, oop new_val) {
     inline_write_ref_field_pre(field, new_val);
   }
+  virtual void write_ref_field_pre_work(narrowOop* field, oop new_val) {
+    inline_write_ref_field_pre(field, new_val);
+  }
+  virtual void write_ref_field_pre_work(void* field, oop new_val) {
+    guarantee(false, "Not needed");
+  }
 
-  virtual void write_ref_array_pre(MemRegion mr);
-
+  template <class T> void write_ref_array_pre_work(T* dst, int count);
+  virtual void write_ref_array_pre(oop* dst, int count) {
+    write_ref_array_pre_work(dst, count);
+  }
+  virtual void write_ref_array_pre(narrowOop* dst, int count) {
+    write_ref_array_pre_work(dst, count);
+  }
 };
 
 // Adds card-table logging to the post-barrier.
diff --git a/hotspot/src/share/vm/gc_implementation/g1/g1_globals.hpp b/hotspot/src/share/vm/gc_implementation/g1/g1_globals.hpp
index 5b296e2..8e134bb 100644
--- a/hotspot/src/share/vm/gc_implementation/g1/g1_globals.hpp
+++ b/hotspot/src/share/vm/gc_implementation/g1/g1_globals.hpp
@@ -80,9 +80,6 @@
   develop(bool, G1TraceConcurrentRefinement, false,                         \
           "Trace G1 concurrent refinement")                                 \
                                                                             \
-  develop(bool, G1ConcMark, true,                                           \
-          "If true, run concurrent marking for G1")                         \
-                                                                            \
   product(intx, G1MarkStackSize, 2 * 1024 * 1024,                           \
           "Size of the mark stack for concurrent marking.")                 \
                                                                             \
diff --git a/hotspot/src/share/vm/gc_implementation/g1/g1_specialized_oop_closures.hpp b/hotspot/src/share/vm/gc_implementation/g1/g1_specialized_oop_closures.hpp
index 5171bdb..101d330 100644
--- a/hotspot/src/share/vm/gc_implementation/g1/g1_specialized_oop_closures.hpp
+++ b/hotspot/src/share/vm/gc_implementation/g1/g1_specialized_oop_closures.hpp
@@ -37,14 +37,12 @@
 class G1ParCopyClosure;
 class G1ParScanClosure;
 
-typedef G1ParCopyClosure<false, G1BarrierEvac, false, true>
-                                                      G1ParScanHeapEvacClosure;
+typedef G1ParCopyClosure<false, G1BarrierEvac, false, true> G1ParScanHeapEvacClosure;
 
 class FilterIntoCSClosure;
 class FilterOutOfRegionClosure;
 class FilterInHeapRegionAndIntoCSClosure;
 class FilterAndMarkInHeapRegionAndIntoCSClosure;
-class G1ScanAndBalanceClosure;
 
 #ifdef FURTHER_SPECIALIZED_OOP_OOP_ITERATE_CLOSURES
 #error "FURTHER_SPECIALIZED_OOP_OOP_ITERATE_CLOSURES already defined."
@@ -56,8 +54,7 @@
       f(FilterIntoCSClosure,_nv)                        \
       f(FilterOutOfRegionClosure,_nv)                   \
       f(FilterInHeapRegionAndIntoCSClosure,_nv)         \
-      f(FilterAndMarkInHeapRegionAndIntoCSClosure,_nv)  \
-      f(G1ScanAndBalanceClosure,_nv)
+      f(FilterAndMarkInHeapRegionAndIntoCSClosure,_nv)
 
 #ifdef FURTHER_SPECIALIZED_SINCE_SAVE_MARKS_CLOSURES
 #error "FURTHER_SPECIALIZED_SINCE_SAVE_MARKS_CLOSURES already defined."
diff --git a/hotspot/src/share/vm/gc_implementation/g1/heapRegion.cpp b/hotspot/src/share/vm/gc_implementation/g1/heapRegion.cpp
index 43af2b5..20876c6 100644
--- a/hotspot/src/share/vm/gc_implementation/g1/heapRegion.cpp
+++ b/hotspot/src/share/vm/gc_implementation/g1/heapRegion.cpp
@@ -66,16 +66,16 @@
   bool failures() { return _failures; }
   int n_failures() { return _n_failures; }
 
-  virtual void do_oop(narrowOop* p) {
-    guarantee(false, "NYI");
-  }
+  virtual void do_oop(narrowOop* p) { do_oop_work(p); }
+  virtual void do_oop(      oop* p) { do_oop_work(p); }
 
-  void do_oop(oop* p) {
+  template <class T> void do_oop_work(T* p) {
     assert(_containing_obj != NULL, "Precondition");
     assert(!_g1h->is_obj_dead_cond(_containing_obj, _use_prev_marking),
            "Precondition");
-    oop obj = *p;
-    if (obj != NULL) {
+    T heap_oop = oopDesc::load_heap_oop(p);
+    if (!oopDesc::is_null(heap_oop)) {
+      oop obj = oopDesc::decode_heap_oop_not_null(heap_oop);
       bool failed = false;
       if (!_g1h->is_in_closed_subset(obj) ||
           _g1h->is_obj_dead_cond(obj, _use_prev_marking)) {
@@ -106,8 +106,8 @@
       }
 
       if (!_g1h->full_collection()) {
-        HeapRegion* from = _g1h->heap_region_containing(p);
-        HeapRegion* to   = _g1h->heap_region_containing(*p);
+        HeapRegion* from = _g1h->heap_region_containing((HeapWord*)p);
+        HeapRegion* to   = _g1h->heap_region_containing(obj);
         if (from != NULL && to != NULL &&
             from != to &&
             !to->isHumongous()) {
@@ -534,13 +534,13 @@
   // Otherwise, find the obj that extends onto mr.start().
 
   assert(cur <= mr.start()
-         && (oop(cur)->klass() == NULL ||
+         && (oop(cur)->klass_or_null() == NULL ||
              cur + oop(cur)->size() > mr.start()),
          "postcondition of block_start");
   oop obj;
   while (cur < mr.end()) {
     obj = oop(cur);
-    if (obj->klass() == NULL) {
+    if (obj->klass_or_null() == NULL) {
       // Ran into an unparseable point.
       return cur;
     } else if (!g1h->is_obj_dead(obj)) {
@@ -577,7 +577,7 @@
   assert(cur <= mr.start(), "Postcondition");
 
   while (cur <= mr.start()) {
-    if (oop(cur)->klass() == NULL) {
+    if (oop(cur)->klass_or_null() == NULL) {
       // Ran into an unparseable point.
       return cur;
     }
@@ -591,7 +591,7 @@
   obj = oop(cur);
   // If we finish this loop...
   assert(cur <= mr.start()
-         && obj->klass() != NULL
+         && obj->klass_or_null() != NULL
          && cur + obj->size() > mr.start(),
          "Loop postcondition");
   if (!g1h->is_obj_dead(obj)) {
@@ -601,7 +601,7 @@
   HeapWord* next;
   while (cur < mr.end()) {
     obj = oop(cur);
-    if (obj->klass() == NULL) {
+    if (obj->klass_or_null() == NULL) {
       // Ran into an unparseable point.
       return cur;
     };
@@ -781,8 +781,13 @@
     // will pick up the right saved_mark_word() as the high water mark
     // of the region. Either way, the behaviour will be correct.
     ContiguousSpace::set_saved_mark();
+    OrderAccess::storestore();
     _gc_time_stamp = curr_gc_time_stamp;
-    OrderAccess::fence();
+    // The following fence is to force a flush of the writes above, but
+    // is strictly not needed because when an allocating worker thread
+    // calls set_saved_mark() it does so under the ParGCRareEvent_lock;
+    // when the lock is released, the write will be flushed.
+    // OrderAccess::fence();
   }
 }
 
diff --git a/hotspot/src/share/vm/gc_implementation/g1/heapRegionRemSet.cpp b/hotspot/src/share/vm/gc_implementation/g1/heapRegionRemSet.cpp
index 5fb5762..f52a1ff 100644
--- a/hotspot/src/share/vm/gc_implementation/g1/heapRegionRemSet.cpp
+++ b/hotspot/src/share/vm/gc_implementation/g1/heapRegionRemSet.cpp
@@ -126,7 +126,7 @@
     }
   }
 
-  void add_reference_work(oop* from, bool par) {
+  void add_reference_work(OopOrNarrowOopStar from, bool par) {
     // Must make this robust in case "from" is not in "_hr", because of
     // concurrency.
 
@@ -173,11 +173,11 @@
     _bm.clear();
   }
 
-  void add_reference(oop* from) {
+  void add_reference(OopOrNarrowOopStar from) {
     add_reference_work(from, /*parallel*/ true);
   }
 
-  void seq_add_reference(oop* from) {
+  void seq_add_reference(OopOrNarrowOopStar from) {
     add_reference_work(from, /*parallel*/ false);
   }
 
@@ -220,7 +220,7 @@
   }
 
   // Requires "from" to be in "hr()".
-  bool contains_reference(oop* from) const {
+  bool contains_reference(OopOrNarrowOopStar from) const {
     assert(hr()->is_in_reserved(from), "Precondition.");
     size_t card_ind = pointer_delta(from, hr()->bottom(),
                                     CardTableModRefBS::card_size);
@@ -394,7 +394,7 @@
   void set_next(PosParPRT* nxt) { _next = nxt; }
   PosParPRT** next_addr() { return &_next; }
 
-  void add_reference(oop* from, int tid) {
+  void add_reference(OopOrNarrowOopStar from, int tid) {
     // Expand if necessary.
     PerRegionTable** pt = par_tables();
     if (par_tables() == NULL && tid > 0 && hr()->is_gc_alloc_region()) {
@@ -447,7 +447,7 @@
     return res;
   }
 
-  bool contains_reference(oop* from) const {
+  bool contains_reference(OopOrNarrowOopStar from) const {
     if (PerRegionTable::contains_reference(from)) return true;
     if (_par_tables != NULL) {
       for (int i = 0; i < HeapRegionRemSet::num_par_rem_sets()-1; i++) {
@@ -564,12 +564,15 @@
 }
 #endif
 
-void OtherRegionsTable::add_reference(oop* from, int tid) {
+void OtherRegionsTable::add_reference(OopOrNarrowOopStar from, int tid) {
   size_t cur_hrs_ind = hr()->hrs_index();
 
 #if HRRS_VERBOSE
   gclog_or_tty->print_cr("ORT::add_reference_work(" PTR_FORMAT "->" PTR_FORMAT ").",
-                                                  from, *from);
+                                                  from,
+                                                  UseCompressedOops
+                                                  ? oopDesc::load_decode_heap_oop((narrowOop*)from)
+                                                  : oopDesc::load_decode_heap_oop((oop*)from));
 #endif
 
   int from_card = (int)(uintptr_t(from) >> CardTableModRefBS::card_shift);
@@ -1021,13 +1024,13 @@
   }
 }
 
-bool OtherRegionsTable::contains_reference(oop* from) const {
+bool OtherRegionsTable::contains_reference(OopOrNarrowOopStar from) const {
   // Cast away const in this case.
   MutexLockerEx x((Mutex*)&_m, Mutex::_no_safepoint_check_flag);
   return contains_reference_locked(from);
 }
 
-bool OtherRegionsTable::contains_reference_locked(oop* from) const {
+bool OtherRegionsTable::contains_reference_locked(OopOrNarrowOopStar from) const {
   HeapRegion* hr = _g1h->heap_region_containing_raw(from);
   if (hr == NULL) return false;
   RegionIdx_t hr_ind = (RegionIdx_t) hr->hrs_index();
@@ -1288,24 +1291,24 @@
 
 
 
-oop**        HeapRegionRemSet::_recorded_oops = NULL;
-HeapWord**   HeapRegionRemSet::_recorded_cards = NULL;
-HeapRegion** HeapRegionRemSet::_recorded_regions = NULL;
-int          HeapRegionRemSet::_n_recorded = 0;
+OopOrNarrowOopStar* HeapRegionRemSet::_recorded_oops = NULL;
+HeapWord**          HeapRegionRemSet::_recorded_cards = NULL;
+HeapRegion**        HeapRegionRemSet::_recorded_regions = NULL;
+int                 HeapRegionRemSet::_n_recorded = 0;
 
 HeapRegionRemSet::Event* HeapRegionRemSet::_recorded_events = NULL;
 int*         HeapRegionRemSet::_recorded_event_index = NULL;
 int          HeapRegionRemSet::_n_recorded_events = 0;
 
-void HeapRegionRemSet::record(HeapRegion* hr, oop* f) {
+void HeapRegionRemSet::record(HeapRegion* hr, OopOrNarrowOopStar f) {
   if (_recorded_oops == NULL) {
     assert(_n_recorded == 0
            && _recorded_cards == NULL
            && _recorded_regions == NULL,
            "Inv");
-    _recorded_oops = NEW_C_HEAP_ARRAY(oop*, MaxRecorded);
-    _recorded_cards = NEW_C_HEAP_ARRAY(HeapWord*, MaxRecorded);
-    _recorded_regions = NEW_C_HEAP_ARRAY(HeapRegion*, MaxRecorded);
+    _recorded_oops    = NEW_C_HEAP_ARRAY(OopOrNarrowOopStar, MaxRecorded);
+    _recorded_cards   = NEW_C_HEAP_ARRAY(HeapWord*,          MaxRecorded);
+    _recorded_regions = NEW_C_HEAP_ARRAY(HeapRegion*,        MaxRecorded);
   }
   if (_n_recorded == MaxRecorded) {
     gclog_or_tty->print_cr("Filled up 'recorded' (%d).", MaxRecorded);
@@ -1408,21 +1411,21 @@
   HeapRegionRemSet* hrrs = hr0->rem_set();
 
   // Make three references from region 0x101...
-  hrrs->add_reference((oop*)hr1_start);
-  hrrs->add_reference((oop*)hr1_mid);
-  hrrs->add_reference((oop*)hr1_last);
+  hrrs->add_reference((OopOrNarrowOopStar)hr1_start);
+  hrrs->add_reference((OopOrNarrowOopStar)hr1_mid);
+  hrrs->add_reference((OopOrNarrowOopStar)hr1_last);
 
-  hrrs->add_reference((oop*)hr2_start);
-  hrrs->add_reference((oop*)hr2_mid);
-  hrrs->add_reference((oop*)hr2_last);
+  hrrs->add_reference((OopOrNarrowOopStar)hr2_start);
+  hrrs->add_reference((OopOrNarrowOopStar)hr2_mid);
+  hrrs->add_reference((OopOrNarrowOopStar)hr2_last);
 
-  hrrs->add_reference((oop*)hr3_start);
-  hrrs->add_reference((oop*)hr3_mid);
-  hrrs->add_reference((oop*)hr3_last);
+  hrrs->add_reference((OopOrNarrowOopStar)hr3_start);
+  hrrs->add_reference((OopOrNarrowOopStar)hr3_mid);
+  hrrs->add_reference((OopOrNarrowOopStar)hr3_last);
 
   // Now cause a coarsening.
-  hrrs->add_reference((oop*)hr4->bottom());
-  hrrs->add_reference((oop*)hr5->bottom());
+  hrrs->add_reference((OopOrNarrowOopStar)hr4->bottom());
+  hrrs->add_reference((OopOrNarrowOopStar)hr5->bottom());
 
   // Now, does iteration yield these three?
   HeapRegionRemSetIterator iter;
diff --git a/hotspot/src/share/vm/gc_implementation/g1/heapRegionRemSet.hpp b/hotspot/src/share/vm/gc_implementation/g1/heapRegionRemSet.hpp
index d630981..d525b5a 100644
--- a/hotspot/src/share/vm/gc_implementation/g1/heapRegionRemSet.hpp
+++ b/hotspot/src/share/vm/gc_implementation/g1/heapRegionRemSet.hpp
@@ -116,9 +116,9 @@
 
   // For now.  Could "expand" some tables in the future, so that this made
   // sense.
-  void add_reference(oop* from, int tid);
+  void add_reference(OopOrNarrowOopStar from, int tid);
 
-  void add_reference(oop* from) {
+  void add_reference(OopOrNarrowOopStar from) {
     return add_reference(from, 0);
   }
 
@@ -140,8 +140,8 @@
   static size_t static_mem_size();
   static size_t fl_mem_size();
 
-  bool contains_reference(oop* from) const;
-  bool contains_reference_locked(oop* from) const;
+  bool contains_reference(OopOrNarrowOopStar from) const;
+  bool contains_reference_locked(OopOrNarrowOopStar from) const;
 
   void clear();
 
@@ -192,10 +192,10 @@
   // Unused unless G1RecordHRRSOops is true.
 
   static const int MaxRecorded = 1000000;
-  static oop**        _recorded_oops;
-  static HeapWord**   _recorded_cards;
-  static HeapRegion** _recorded_regions;
-  static int          _n_recorded;
+  static OopOrNarrowOopStar* _recorded_oops;
+  static HeapWord**          _recorded_cards;
+  static HeapRegion**        _recorded_regions;
+  static int                 _n_recorded;
 
   static const int MaxRecordedEvents = 1000;
   static Event*       _recorded_events;
@@ -231,13 +231,13 @@
 
   /* Used in the sequential case.  Returns "true" iff this addition causes
      the size limit to be reached. */
-  void add_reference(oop* from) {
+  void add_reference(OopOrNarrowOopStar from) {
     _other_regions.add_reference(from);
   }
 
   /* Used in the parallel case.  Returns "true" iff this addition causes
      the size limit to be reached. */
-  void add_reference(oop* from, int tid) {
+  void add_reference(OopOrNarrowOopStar from, int tid) {
     _other_regions.add_reference(from, tid);
   }
 
@@ -301,7 +301,7 @@
     return OtherRegionsTable::fl_mem_size();
   }
 
-  bool contains_reference(oop* from) const {
+  bool contains_reference(OopOrNarrowOopStar from) const {
     return _other_regions.contains_reference(from);
   }
   void print() const;
@@ -329,7 +329,7 @@
   }
 #endif
 
-  static void record(HeapRegion* hr, oop* f);
+  static void record(HeapRegion* hr, OopOrNarrowOopStar f);
   static void print_recorded();
   static void record_event(Event evnt);
 
diff --git a/hotspot/src/share/vm/gc_implementation/g1/satbQueue.cpp b/hotspot/src/share/vm/gc_implementation/g1/satbQueue.cpp
index a3e74a4..3cf402c 100644
--- a/hotspot/src/share/vm/gc_implementation/g1/satbQueue.cpp
+++ b/hotspot/src/share/vm/gc_implementation/g1/satbQueue.cpp
@@ -43,6 +43,18 @@
     }
   }
 }
+
+#ifdef ASSERT
+void ObjPtrQueue::verify_oops_in_buffer() {
+  if (_buf == NULL) return;
+  for (size_t i = _index; i < _sz; i += oopSize) {
+    oop obj = (oop)_buf[byte_index_to_index((int)i)];
+    assert(obj != NULL && obj->is_oop(true /* ignore mark word */),
+           "Not an oop");
+  }
+}
+#endif
+
 #ifdef _MSC_VER // the use of 'this' below gets a warning, make it go away
 #pragma warning( disable:4355 ) // 'this' : used in base member initializer list
 #endif // _MSC_VER
@@ -66,6 +78,7 @@
 
 
 void SATBMarkQueueSet::handle_zero_index_for_thread(JavaThread* t) {
+  DEBUG_ONLY(t->satb_mark_queue().verify_oops_in_buffer();)
   t->satb_mark_queue().handle_zero_index();
 }
 
@@ -143,7 +156,7 @@
     }
     _completed_buffers_tail = NULL;
     _n_completed_buffers = 0;
-    debug_only(assert_completed_buffer_list_len_correct_locked());
+    DEBUG_ONLY(assert_completed_buffer_list_len_correct_locked());
   }
   while (buffers_to_delete != NULL) {
     CompletedBufferNode* nd = buffers_to_delete;
diff --git a/hotspot/src/share/vm/gc_implementation/g1/satbQueue.hpp b/hotspot/src/share/vm/gc_implementation/g1/satbQueue.hpp
index ab8bf5f..ed1181d 100644
--- a/hotspot/src/share/vm/gc_implementation/g1/satbQueue.hpp
+++ b/hotspot/src/share/vm/gc_implementation/g1/satbQueue.hpp
@@ -39,6 +39,7 @@
   static void apply_closure_to_buffer(ObjectClosure* cl,
                                       void** buf, size_t index, size_t sz);
 
+  void verify_oops_in_buffer() NOT_DEBUG_RETURN;
 };
 
 
diff --git a/hotspot/src/share/vm/gc_implementation/includeDB_gc_g1 b/hotspot/src/share/vm/gc_implementation/includeDB_gc_g1
index f7236ed..7e2df3f 100644
--- a/hotspot/src/share/vm/gc_implementation/includeDB_gc_g1
+++ b/hotspot/src/share/vm/gc_implementation/includeDB_gc_g1
@@ -27,6 +27,7 @@
 bufferingOopClosure.hpp			genOopClosures.hpp
 bufferingOopClosure.hpp			generation.hpp
 bufferingOopClosure.hpp			os.hpp
+bufferingOopClosure.hpp			taskqueue.hpp
 
 cardTableRS.cpp				concurrentMark.hpp
 cardTableRS.cpp				g1SATBCardTableModRefBS.hpp
@@ -139,7 +140,7 @@
 g1CollectedHeap.cpp                     g1CollectedHeap.inline.hpp
 g1CollectedHeap.cpp                     g1CollectorPolicy.hpp
 g1CollectedHeap.cpp                     g1MarkSweep.hpp
-g1CollectedHeap.cpp                     g1RemSet.hpp
+g1CollectedHeap.cpp                     g1RemSet.inline.hpp
 g1CollectedHeap.cpp                     g1OopClosures.inline.hpp
 g1CollectedHeap.cpp                     genOopClosures.inline.hpp
 g1CollectedHeap.cpp                     gcLocker.inline.hpp
@@ -151,13 +152,14 @@
 g1CollectedHeap.cpp                     isGCActiveMark.hpp
 g1CollectedHeap.cpp			oop.inline.hpp
 g1CollectedHeap.cpp			oop.pcgc.inline.hpp
-g1CollectedHeap.cpp			parGCAllocBuffer.hpp
 g1CollectedHeap.cpp                     vm_operations_g1.hpp
 g1CollectedHeap.cpp                     vmThread.hpp
 
 g1CollectedHeap.hpp                     barrierSet.hpp
+g1CollectedHeap.hpp                     g1RemSet.hpp
 g1CollectedHeap.hpp                     heapRegion.hpp
 g1CollectedHeap.hpp                     memRegion.hpp
+g1CollectedHeap.hpp			parGCAllocBuffer.hpp
 g1CollectedHeap.hpp                     sharedHeap.hpp
 
 g1CollectedHeap.inline.hpp              concurrentMark.hpp
@@ -245,6 +247,7 @@
 g1RemSet.cpp				iterator.hpp
 g1RemSet.cpp				oop.inline.hpp
 
+g1RemSet.inline.hpp			oop.inline.hpp
 g1RemSet.inline.hpp			g1RemSet.hpp
 g1RemSet.inline.hpp			heapRegionRemSet.hpp
 
@@ -255,6 +258,7 @@
 g1SATBCardTableModRefBS.cpp		thread_<os_family>.inline.hpp
 g1SATBCardTableModRefBS.cpp		satbQueue.hpp
 
+g1SATBCardTableModRefBS.hpp		oop.inline.hpp
 g1SATBCardTableModRefBS.hpp		cardTableModRefBS.hpp
 g1SATBCardTableModRefBS.hpp		memRegion.hpp
 
diff --git a/hotspot/src/share/vm/gc_implementation/parNew/parCardTableModRefBS.cpp b/hotspot/src/share/vm/gc_implementation/parNew/parCardTableModRefBS.cpp
index fcbd6cc..ce7ba20 100644
--- a/hotspot/src/share/vm/gc_implementation/parNew/parCardTableModRefBS.cpp
+++ b/hotspot/src/share/vm/gc_implementation/parNew/parCardTableModRefBS.cpp
@@ -31,9 +31,10 @@
                                                         bool clear,
                                                         int n_threads) {
   if (n_threads > 0) {
-    assert(n_threads == (int)ParallelGCThreads, "# worker threads != # requested!");
-
-      // Make sure the LNC array is valid for the space.
+    assert((n_threads == 1 && ParallelGCThreads == 0) ||
+           n_threads <= (int)ParallelGCThreads,
+           "# worker threads != # requested!");
+    // Make sure the LNC array is valid for the space.
     jbyte**   lowest_non_clean;
     uintptr_t lowest_non_clean_base_chunk_index;
     size_t    lowest_non_clean_chunk_size;
diff --git a/hotspot/src/share/vm/gc_implementation/parallelScavenge/parallelScavengeHeap.cpp b/hotspot/src/share/vm/gc_implementation/parallelScavenge/parallelScavengeHeap.cpp
index d0fa3a4..b48344f 100644
--- a/hotspot/src/share/vm/gc_implementation/parallelScavenge/parallelScavengeHeap.cpp
+++ b/hotspot/src/share/vm/gc_implementation/parallelScavenge/parallelScavengeHeap.cpp
@@ -885,7 +885,7 @@
 }
 
 
-void ParallelScavengeHeap::verify(bool allow_dirty, bool silent) {
+void ParallelScavengeHeap::verify(bool allow_dirty, bool silent, bool option /* ignored */) {
   // Why do we need the total_collections()-filter below?
   if (total_collections() > 0) {
     if (!silent) {
diff --git a/hotspot/src/share/vm/gc_implementation/parallelScavenge/parallelScavengeHeap.hpp b/hotspot/src/share/vm/gc_implementation/parallelScavenge/parallelScavengeHeap.hpp
index 2b8904c..350c043 100644
--- a/hotspot/src/share/vm/gc_implementation/parallelScavenge/parallelScavengeHeap.hpp
+++ b/hotspot/src/share/vm/gc_implementation/parallelScavenge/parallelScavengeHeap.hpp
@@ -217,7 +217,7 @@
   virtual void gc_threads_do(ThreadClosure* tc) const;
   virtual void print_tracing_info() const;
 
-  void verify(bool allow_dirty, bool silent);
+  void verify(bool allow_dirty, bool silent, bool /* option */);
 
   void print_heap_change(size_t prev_used);
 
diff --git a/hotspot/src/share/vm/gc_implementation/parallelScavenge/psPromotionManager.inline.hpp b/hotspot/src/share/vm/gc_implementation/parallelScavenge/psPromotionManager.inline.hpp
index f1a9e64..e1f0994 100644
--- a/hotspot/src/share/vm/gc_implementation/parallelScavenge/psPromotionManager.inline.hpp
+++ b/hotspot/src/share/vm/gc_implementation/parallelScavenge/psPromotionManager.inline.hpp
@@ -117,6 +117,7 @@
     process_array_chunk(old);
   } else {
     if (p.is_narrow()) {
+      assert(UseCompressedOops, "Error");
       PSScavenge::copy_and_push_safe_barrier(this, (narrowOop*)p);
     } else {
       PSScavenge::copy_and_push_safe_barrier(this, (oop*)p);
diff --git a/hotspot/src/share/vm/gc_interface/collectedHeap.hpp b/hotspot/src/share/vm/gc_interface/collectedHeap.hpp
index 3db5224..11da8de 100644
--- a/hotspot/src/share/vm/gc_interface/collectedHeap.hpp
+++ b/hotspot/src/share/vm/gc_interface/collectedHeap.hpp
@@ -533,7 +533,7 @@
   virtual void print_tracing_info() const = 0;
 
   // Heap verification
-  virtual void verify(bool allow_dirty, bool silent) = 0;
+  virtual void verify(bool allow_dirty, bool silent, bool option) = 0;
 
   // Non product verification and debugging.
 #ifndef PRODUCT
diff --git a/hotspot/src/share/vm/includeDB_core b/hotspot/src/share/vm/includeDB_core
index f595248..bd61099 100644
--- a/hotspot/src/share/vm/includeDB_core
+++ b/hotspot/src/share/vm/includeDB_core
@@ -554,7 +554,6 @@
 ciEnv.cpp                               linkResolver.hpp
 ciEnv.cpp                               methodDataOop.hpp
 ciEnv.cpp                               objArrayKlass.hpp
-ciEnv.cpp                               oop.hpp
 ciEnv.cpp                               oop.inline.hpp
 ciEnv.cpp                               oop.inline2.hpp
 ciEnv.cpp                               oopFactory.hpp
@@ -785,7 +784,6 @@
 ciSignature.cpp                         allocation.inline.hpp
 ciSignature.cpp                         ciSignature.hpp
 ciSignature.cpp                         ciUtilities.hpp
-ciSignature.cpp                         oop.hpp
 ciSignature.cpp                         oop.inline.hpp
 ciSignature.cpp                         signature.hpp
 
@@ -950,7 +948,6 @@
 classify.cpp                            classify.hpp
 classify.cpp                            systemDictionary.hpp
 
-classify.hpp                            oop.hpp
 classify.hpp                            oop.inline.hpp
 
 codeBlob.cpp                            allocation.inline.hpp
@@ -1185,7 +1182,6 @@
 compilerOracle.cpp                      jniHandles.hpp
 compilerOracle.cpp                      klass.hpp
 compilerOracle.cpp                      methodOop.hpp
-compilerOracle.cpp                      oop.hpp
 compilerOracle.cpp                      oop.inline.hpp
 compilerOracle.cpp                      oopFactory.hpp
 compilerOracle.cpp                      resourceArea.hpp
@@ -1629,7 +1625,6 @@
 frame.cpp                               methodOop.hpp
 frame.cpp                               monitorChunk.hpp
 frame.cpp                               nativeInst_<arch>.hpp
-frame.cpp                               oop.hpp
 frame.cpp                               oop.inline.hpp
 frame.cpp                               oop.inline2.hpp
 frame.cpp                               oopMapCache.hpp
@@ -1797,7 +1792,6 @@
 generation.cpp                          generation.hpp
 generation.cpp                          generation.inline.hpp
 generation.cpp                          java.hpp
-generation.cpp                          oop.hpp
 generation.cpp                          oop.inline.hpp
 generation.cpp                          spaceDecorator.hpp
 generation.cpp                          space.inline.hpp
@@ -2270,7 +2264,6 @@
 java.cpp                                memprofiler.hpp
 java.cpp                                methodOop.hpp
 java.cpp                                objArrayOop.hpp
-java.cpp                                oop.hpp
 java.cpp                                oop.inline.hpp
 java.cpp                                oopFactory.hpp
 java.cpp                                sharedRuntime.hpp
@@ -2947,7 +2940,7 @@
 nativeInst_<arch>.cpp                   assembler_<arch>.inline.hpp
 nativeInst_<arch>.cpp                   handles.hpp
 nativeInst_<arch>.cpp                   nativeInst_<arch>.hpp
-nativeInst_<arch>.cpp                   oop.hpp
+nativeInst_<arch>.cpp                   oop.inline.hpp
 nativeInst_<arch>.cpp                   ostream.hpp
 nativeInst_<arch>.cpp                   resourceArea.hpp
 nativeInst_<arch>.cpp                   sharedRuntime.hpp
@@ -3842,7 +3835,7 @@
 stackValue.cpp                          debugInfo.hpp
 stackValue.cpp                          frame.inline.hpp
 stackValue.cpp                          handles.inline.hpp
-stackValue.cpp                          oop.hpp
+stackValue.cpp                          oop.inline.hpp
 stackValue.cpp                          stackValue.hpp
 
 stackValue.hpp                          handles.hpp
@@ -4329,7 +4322,6 @@
 unhandledOops.cpp                       collectedHeap.hpp
 unhandledOops.cpp                       gcLocker.inline.hpp
 unhandledOops.cpp                       globalDefinitions.hpp
-unhandledOops.cpp                       oop.hpp
 unhandledOops.cpp                       oop.inline.hpp
 unhandledOops.cpp                       thread.hpp
 unhandledOops.cpp                       unhandledOops.hpp
@@ -4465,7 +4457,6 @@
 vframe.cpp                              nmethod.hpp
 vframe.cpp                              objectMonitor.hpp
 vframe.cpp                              objectMonitor.inline.hpp
-vframe.cpp                              oop.hpp
 vframe.cpp                              oop.inline.hpp
 vframe.cpp                              oopMapCache.hpp
 vframe.cpp                              pcDesc.hpp
@@ -4577,7 +4568,6 @@
 vmThread.cpp                            interfaceSupport.hpp
 vmThread.cpp                            methodOop.hpp
 vmThread.cpp                            mutexLocker.hpp
-vmThread.cpp                            oop.hpp
 vmThread.cpp                            oop.inline.hpp
 vmThread.cpp                            os.hpp
 vmThread.cpp                            resourceArea.hpp
diff --git a/hotspot/src/share/vm/includeDB_features b/hotspot/src/share/vm/includeDB_features
index 863f651..c0b02eb 100644
--- a/hotspot/src/share/vm/includeDB_features
+++ b/hotspot/src/share/vm/includeDB_features
@@ -47,7 +47,7 @@
 dump.cpp                                javaClasses.hpp
 dump.cpp                                loaderConstraints.hpp
 dump.cpp                                methodDataOop.hpp
-dump.cpp                                oop.hpp
+dump.cpp                                oop.inline.hpp
 dump.cpp                                oopFactory.hpp
 dump.cpp                                resourceArea.hpp
 dump.cpp                                signature.hpp
@@ -237,7 +237,7 @@
 serialize.cpp                           compiledICHolderOop.hpp
 serialize.cpp                           methodDataOop.hpp
 serialize.cpp                           objArrayOop.hpp
-serialize.cpp                           oop.hpp
+serialize.cpp                           oop.inline.hpp
 serialize.cpp                           symbolTable.hpp
 serialize.cpp                           systemDictionary.hpp
 
@@ -295,7 +295,7 @@
 vmStructs.cpp                           objArrayKlass.hpp
 vmStructs.cpp                           objArrayKlassKlass.hpp
 vmStructs.cpp                           objArrayOop.hpp
-vmStructs.cpp                           oop.hpp
+vmStructs.cpp                           oop.inline.hpp
 vmStructs.cpp                           oopMap.hpp
 vmStructs.cpp                           pcDesc.hpp
 vmStructs.cpp                           perfMemory.hpp
diff --git a/hotspot/src/share/vm/memory/barrierSet.cpp b/hotspot/src/share/vm/memory/barrierSet.cpp
index 2330880..07ff7a0 100644
--- a/hotspot/src/share/vm/memory/barrierSet.cpp
+++ b/hotspot/src/share/vm/memory/barrierSet.cpp
@@ -25,12 +25,27 @@
 # include "incls/_precompiled.incl"
 # include "incls/_barrierSet.cpp.incl"
 
-// count is in HeapWord's
+// count is number of array elements being written
 void BarrierSet::static_write_ref_array_pre(HeapWord* start, size_t count) {
-   Universe::heap()->barrier_set()->write_ref_array_pre(MemRegion(start, start + count));
+  assert(count <= (size_t)max_intx, "count too large");
+#if 0
+  warning("Pre: \t" INTPTR_FORMAT "[" SIZE_FORMAT "]\t",
+                   start,            count);
+#endif
+  if (UseCompressedOops) {
+    Universe::heap()->barrier_set()->write_ref_array_pre((narrowOop*)start, (int)count);
+  } else {
+    Universe::heap()->barrier_set()->write_ref_array_pre(      (oop*)start, (int)count);
+  }
 }
 
-// count is in HeapWord's
+// count is number of array elements being written
 void BarrierSet::static_write_ref_array_post(HeapWord* start, size_t count) {
-   Universe::heap()->barrier_set()->write_ref_array_work(MemRegion(start, start + count));
+  assert(count <= (size_t)max_intx, "count too large");
+  HeapWord* end = start + objArrayOopDesc::array_size((int)count);
+#if 0
+  warning("Post:\t" INTPTR_FORMAT "[" SIZE_FORMAT "] : [" INTPTR_FORMAT","INTPTR_FORMAT")\t",
+                   start,            count,              start,          end);
+#endif
+  Universe::heap()->barrier_set()->write_ref_array_work(MemRegion(start, end));
 }
diff --git a/hotspot/src/share/vm/memory/barrierSet.hpp b/hotspot/src/share/vm/memory/barrierSet.hpp
index bff929b..0fc6a00 100644
--- a/hotspot/src/share/vm/memory/barrierSet.hpp
+++ b/hotspot/src/share/vm/memory/barrierSet.hpp
@@ -81,9 +81,13 @@
   // barrier types.  Semantically, it should be thought of as a call to the
   // virtual "_work" function below, which must implement the barrier.)
   // First the pre-write versions...
-  inline void write_ref_field_pre(void* field, oop new_val);
+  template <class T> inline void write_ref_field_pre(T* field, oop new_val);
+private:
+  // Keep this private so as to catch violations at build time.
+  virtual void write_ref_field_pre_work(     void* field, oop new_val) { guarantee(false, "Not needed"); };
 protected:
-  virtual void write_ref_field_pre_work(void* field, oop new_val) {};
+  virtual void write_ref_field_pre_work(      oop* field, oop new_val) {};
+  virtual void write_ref_field_pre_work(narrowOop* field, oop new_val) {};
 public:
 
   // ...then the post-write version.
@@ -117,12 +121,17 @@
   virtual void read_ref_array(MemRegion mr) = 0;
   virtual void read_prim_array(MemRegion mr) = 0;
 
-  virtual void write_ref_array_pre(MemRegion mr) {}
+  virtual void write_ref_array_pre(      oop* dst, int length) {}
+  virtual void write_ref_array_pre(narrowOop* dst, int length) {}
   inline void write_ref_array(MemRegion mr);
 
   // Static versions, suitable for calling from generated code.
   static void static_write_ref_array_pre(HeapWord* start, size_t count);
   static void static_write_ref_array_post(HeapWord* start, size_t count);
+  // Narrow oop versions of the above; count is # of array elements being written,
+  // starting with "start", which is HeapWord-aligned.
+  static void static_write_ref_array_pre_narrow(HeapWord* start, size_t count);
+  static void static_write_ref_array_post_narrow(HeapWord* start, size_t count);
 
 protected:
   virtual void write_ref_array_work(MemRegion mr) = 0;
diff --git a/hotspot/src/share/vm/memory/barrierSet.inline.hpp b/hotspot/src/share/vm/memory/barrierSet.inline.hpp
index 50382c9..edcb551 100644
--- a/hotspot/src/share/vm/memory/barrierSet.inline.hpp
+++ b/hotspot/src/share/vm/memory/barrierSet.inline.hpp
@@ -23,10 +23,10 @@
  */
 
 // Inline functions of BarrierSet, which de-virtualize certain
-// performance-critical calls when when the barrier is the most common
+// performance-critical calls when the barrier is the most common
 // card-table kind.
 
-void BarrierSet::write_ref_field_pre(void* field, oop new_val) {
+template <class T> void BarrierSet::write_ref_field_pre(T* field, oop new_val) {
   if (kind() == CardTableModRef) {
     ((CardTableModRefBS*)this)->inline_write_ref_field_pre(field, new_val);
   } else {
diff --git a/hotspot/src/share/vm/memory/cardTableModRefBS.hpp b/hotspot/src/share/vm/memory/cardTableModRefBS.hpp
index 46a5dd9..d31dc6e 100644
--- a/hotspot/src/share/vm/memory/cardTableModRefBS.hpp
+++ b/hotspot/src/share/vm/memory/cardTableModRefBS.hpp
@@ -287,7 +287,7 @@
   // these functions here for performance.
 protected:
   void write_ref_field_work(oop obj, size_t offset, oop newVal);
-  void write_ref_field_work(void* field, oop newVal);
+  virtual void write_ref_field_work(void* field, oop newVal);
 public:
 
   bool has_write_ref_array_opt() { return true; }
@@ -317,10 +317,10 @@
 
   // *** Card-table-barrier-specific things.
 
-  inline void inline_write_ref_field_pre(void* field, oop newVal) {}
+  template <class T> inline void inline_write_ref_field_pre(T* field, oop newVal) {}
 
-  inline void inline_write_ref_field(void* field, oop newVal) {
-    jbyte* byte = byte_for(field);
+  template <class T> inline void inline_write_ref_field(T* field, oop newVal) {
+    jbyte* byte = byte_for((void*)field);
     *byte = dirty_card;
   }
 
diff --git a/hotspot/src/share/vm/memory/genCollectedHeap.cpp b/hotspot/src/share/vm/memory/genCollectedHeap.cpp
index 048bd0e..e25cc51 100644
--- a/hotspot/src/share/vm/memory/genCollectedHeap.cpp
+++ b/hotspot/src/share/vm/memory/genCollectedHeap.cpp
@@ -1194,7 +1194,7 @@
   return _gens[level]->gc_stats();
 }
 
-void GenCollectedHeap::verify(bool allow_dirty, bool silent) {
+void GenCollectedHeap::verify(bool allow_dirty, bool silent, bool option /* ignored */) {
   if (!silent) {
     gclog_or_tty->print("permgen ");
   }
diff --git a/hotspot/src/share/vm/memory/genCollectedHeap.hpp b/hotspot/src/share/vm/memory/genCollectedHeap.hpp
index 0630eec..0c14c7f 100644
--- a/hotspot/src/share/vm/memory/genCollectedHeap.hpp
+++ b/hotspot/src/share/vm/memory/genCollectedHeap.hpp
@@ -325,7 +325,7 @@
   void prepare_for_verify();
 
   // Override.
-  void verify(bool allow_dirty, bool silent);
+  void verify(bool allow_dirty, bool silent, bool /* option */);
 
   // Override.
   void print() const;
diff --git a/hotspot/src/share/vm/memory/genOopClosures.hpp b/hotspot/src/share/vm/memory/genOopClosures.hpp
index 4cb2ca6..8f8fe5c 100644
--- a/hotspot/src/share/vm/memory/genOopClosures.hpp
+++ b/hotspot/src/share/vm/memory/genOopClosures.hpp
@@ -57,7 +57,7 @@
   template <class T> void do_barrier(T* p);
 
   // Version for use by closures that may be called in parallel code.
-  void par_do_barrier(oop* p);
+  template <class T> void par_do_barrier(T* p);
 
  public:
   OopsInGenClosure() : OopClosure(NULL),
diff --git a/hotspot/src/share/vm/memory/genOopClosures.inline.hpp b/hotspot/src/share/vm/memory/genOopClosures.inline.hpp
index 5f13668..8971b43 100644
--- a/hotspot/src/share/vm/memory/genOopClosures.inline.hpp
+++ b/hotspot/src/share/vm/memory/genOopClosures.inline.hpp
@@ -40,18 +40,20 @@
 
 template <class T> inline void OopsInGenClosure::do_barrier(T* p) {
   assert(generation()->is_in_reserved(p), "expected ref in generation");
-  assert(!oopDesc::is_null(*p), "expected non-null object");
-  oop obj = oopDesc::load_decode_heap_oop_not_null(p);
+  T heap_oop = oopDesc::load_heap_oop(p);
+  assert(!oopDesc::is_null(heap_oop), "expected non-null oop");
+  oop obj = oopDesc::decode_heap_oop_not_null(heap_oop);
   // If p points to a younger generation, mark the card.
   if ((HeapWord*)obj < _gen_boundary) {
     _rs->inline_write_ref_field_gc(p, obj);
   }
 }
 
-inline void OopsInGenClosure::par_do_barrier(oop* p) {
+template <class T> inline void OopsInGenClosure::par_do_barrier(T* p) {
   assert(generation()->is_in_reserved(p), "expected ref in generation");
-  oop obj = *p;
-  assert(obj != NULL, "expected non-null object");
+  T heap_oop = oopDesc::load_heap_oop(p);
+  assert(!oopDesc::is_null(heap_oop), "expected non-null oop");
+  oop obj = oopDesc::decode_heap_oop_not_null(heap_oop);
   // If p points to a younger generation, mark the card.
   if ((HeapWord*)obj < gen_boundary()) {
     rs()->write_ref_field_gc_par(p, obj);
diff --git a/hotspot/src/share/vm/memory/referenceProcessor.cpp b/hotspot/src/share/vm/memory/referenceProcessor.cpp
index 36d9514..616ca5d 100644
--- a/hotspot/src/share/vm/memory/referenceProcessor.cpp
+++ b/hotspot/src/share/vm/memory/referenceProcessor.cpp
@@ -1013,12 +1013,19 @@
   // discovered_addr.
   oop current_head = refs_list.head();
 
-  // Note: In the case of G1, this pre-barrier is strictly
+  // Note: In the case of G1, this specific pre-barrier is strictly
   // not necessary because the only case we are interested in
-  // here is when *discovered_addr is NULL, so this will expand to
-  // nothing. As a result, I am just manually eliding this out for G1.
+  // here is when *discovered_addr is NULL (see the CAS further below),
+  // so this will expand to nothing. As a result, we have manually
+  // elided this out for G1, but left in the test for some future
+  // collector that might have need for a pre-barrier here.
   if (_discovered_list_needs_barrier && !UseG1GC) {
-    _bs->write_ref_field_pre((void*)discovered_addr, current_head); guarantee(false, "Needs to be fixed: YSR");
+    if (UseCompressedOops) {
+      _bs->write_ref_field_pre((narrowOop*)discovered_addr, current_head);
+    } else {
+      _bs->write_ref_field_pre((oop*)discovered_addr, current_head);
+    }
+    guarantee(false, "Need to check non-G1 collector");
   }
   oop retest = oopDesc::atomic_compare_exchange_oop(current_head, discovered_addr,
                                                     NULL);
@@ -1029,9 +1036,8 @@
     refs_list.set_head(obj);
     refs_list.inc_length(1);
     if (_discovered_list_needs_barrier) {
-      _bs->write_ref_field((void*)discovered_addr, current_head); guarantee(false, "Needs to be fixed: YSR");
+      _bs->write_ref_field((void*)discovered_addr, current_head);
     }
-
   } else {
     // If retest was non NULL, another thread beat us to it:
     // The reference has already been discovered...
@@ -1177,11 +1183,16 @@
     // pre-value, we can safely elide the pre-barrier here for the case of G1.
     assert(discovered == NULL, "control point invariant");
     if (_discovered_list_needs_barrier && !UseG1GC) { // safe to elide for G1
-      _bs->write_ref_field_pre((oop*)discovered_addr, current_head);
+      if (UseCompressedOops) {
+        _bs->write_ref_field_pre((narrowOop*)discovered_addr, current_head);
+      } else {
+        _bs->write_ref_field_pre((oop*)discovered_addr, current_head);
+      }
+      guarantee(false, "Need to check non-G1 collector");
     }
     oop_store_raw(discovered_addr, current_head);
     if (_discovered_list_needs_barrier) {
-      _bs->write_ref_field((oop*)discovered_addr, current_head);
+      _bs->write_ref_field((void*)discovered_addr, current_head);
     }
     list->set_head(obj);
     list->inc_length(1);
diff --git a/hotspot/src/share/vm/memory/space.hpp b/hotspot/src/share/vm/memory/space.hpp
index e6164ba..98fc8a2 100644
--- a/hotspot/src/share/vm/memory/space.hpp
+++ b/hotspot/src/share/vm/memory/space.hpp
@@ -106,6 +106,7 @@
   virtual void set_end(HeapWord* value)    { _end = value; }
 
   virtual HeapWord* saved_mark_word() const  { return _saved_mark_word; }
+
   void set_saved_mark_word(HeapWord* p) { _saved_mark_word = p; }
 
   MemRegionClosure* preconsumptionDirtyCardClosure() const {
diff --git a/hotspot/src/share/vm/memory/universe.cpp b/hotspot/src/share/vm/memory/universe.cpp
index b854d64..adcc3ec 100644
--- a/hotspot/src/share/vm/memory/universe.cpp
+++ b/hotspot/src/share/vm/memory/universe.cpp
@@ -1170,7 +1170,7 @@
   st->print_cr("}");
 }
 
-void Universe::verify(bool allow_dirty, bool silent) {
+void Universe::verify(bool allow_dirty, bool silent, bool option) {
   if (SharedSkipVerify) {
     return;
   }
@@ -1194,7 +1194,7 @@
   if (!silent) gclog_or_tty->print("[Verifying ");
   if (!silent) gclog_or_tty->print("threads ");
   Threads::verify();
-  heap()->verify(allow_dirty, silent);
+  heap()->verify(allow_dirty, silent, option);
 
   if (!silent) gclog_or_tty->print("syms ");
   SymbolTable::verify();
diff --git a/hotspot/src/share/vm/memory/universe.hpp b/hotspot/src/share/vm/memory/universe.hpp
index 9e17a18..53cf780 100644
--- a/hotspot/src/share/vm/memory/universe.hpp
+++ b/hotspot/src/share/vm/memory/universe.hpp
@@ -398,7 +398,7 @@
 
   // Debugging
   static bool verify_in_progress() { return _verify_in_progress; }
-  static void verify(bool allow_dirty = true, bool silent = false);
+  static void verify(bool allow_dirty = true, bool silent = false, bool option = true);
   static int  verify_count()                  { return _verify_count; }
   static void print();
   static void print_on(outputStream* st);
diff --git a/hotspot/src/share/vm/oops/instanceRefKlass.cpp b/hotspot/src/share/vm/oops/instanceRefKlass.cpp
index c6d9f75..5e6d40f 100644
--- a/hotspot/src/share/vm/oops/instanceRefKlass.cpp
+++ b/hotspot/src/share/vm/oops/instanceRefKlass.cpp
@@ -28,13 +28,14 @@
 template <class T>
 static void specialized_oop_follow_contents(instanceRefKlass* ref, oop obj) {
   T* referent_addr = (T*)java_lang_ref_Reference::referent_addr(obj);
-  oop referent = oopDesc::load_decode_heap_oop(referent_addr);
+  T heap_oop = oopDesc::load_heap_oop(referent_addr);
   debug_only(
     if(TraceReferenceGC && PrintGCDetails) {
       gclog_or_tty->print_cr("instanceRefKlass::oop_follow_contents " INTPTR_FORMAT, obj);
     }
   )
-  if (referent != NULL) {
+  if (!oopDesc::is_null(heap_oop)) {
+    oop referent = oopDesc::decode_heap_oop_not_null(heap_oop);
     if (!referent->is_gc_marked() &&
         MarkSweep::ref_processor()->
           discover_reference(obj, ref->reference_type())) {
@@ -81,13 +82,14 @@
                                             ParCompactionManager* cm,
                                             oop obj) {
   T* referent_addr = (T*)java_lang_ref_Reference::referent_addr(obj);
-  oop referent = oopDesc::load_decode_heap_oop(referent_addr);
+  T heap_oop = oopDesc::load_heap_oop(referent_addr);
   debug_only(
     if(TraceReferenceGC && PrintGCDetails) {
       gclog_or_tty->print_cr("instanceRefKlass::oop_follow_contents " INTPTR_FORMAT, obj);
     }
   )
-  if (referent != NULL) {
+  if (!oopDesc::is_null(heap_oop)) {
+    oop referent = oopDesc::decode_heap_oop_not_null(heap_oop);
     if (PSParallelCompact::mark_bitmap()->is_unmarked(referent) &&
         PSParallelCompact::ref_processor()->
           discover_reference(obj, ref->reference_type())) {
@@ -182,9 +184,10 @@
   }                                                                             \
                                                                                 \
   T* referent_addr = (T*)java_lang_ref_Reference::referent_addr(obj);           \
-  oop referent = oopDesc::load_decode_heap_oop(referent_addr);                  \
-  if (referent != NULL && contains(referent_addr)) {                            \
+  T heap_oop = oopDesc::load_heap_oop(referent_addr);                           \
+  if (!oopDesc::is_null(heap_oop) && contains(referent_addr)) {                 \
     ReferenceProcessor* rp = closure->_ref_processor;                           \
+    oop referent = oopDesc::decode_heap_oop_not_null(heap_oop);                 \
     if (!referent->is_gc_marked() && (rp != NULL) &&                            \
         rp->discover_reference(obj, reference_type())) {                        \
       return size;                                                              \
diff --git a/hotspot/src/share/vm/oops/objArrayKlass.cpp b/hotspot/src/share/vm/oops/objArrayKlass.cpp
index 3f56d9c..a97cb0a 100644
--- a/hotspot/src/share/vm/oops/objArrayKlass.cpp
+++ b/hotspot/src/share/vm/oops/objArrayKlass.cpp
@@ -84,8 +84,6 @@
 template <class T> void objArrayKlass::do_copy(arrayOop s, T* src,
                                arrayOop d, T* dst, int length, TRAPS) {
 
-  const size_t word_len = objArrayOopDesc::array_size(length);
-
   BarrierSet* bs = Universe::heap()->barrier_set();
   // For performance reasons, we assume we are that the write barrier we
   // are using has optimized modes for arrays of references.  At least one
@@ -93,11 +91,10 @@
   assert(bs->has_write_ref_array_opt(), "Barrier set must have ref array opt");
   assert(bs->has_write_ref_array_pre_opt(), "For pre-barrier as well.");
 
-  MemRegion dst_mr = MemRegion((HeapWord*)dst, word_len);
   if (s == d) {
     // since source and destination are equal we do not need conversion checks.
     assert(length > 0, "sanity check");
-    bs->write_ref_array_pre(dst_mr);
+    bs->write_ref_array_pre(dst, length);
     Copy::conjoint_oops_atomic(src, dst, length);
   } else {
     // We have to make sure all elements conform to the destination array
@@ -105,7 +102,7 @@
     klassOop stype = objArrayKlass::cast(s->klass())->element_klass();
     if (stype == bound || Klass::cast(stype)->is_subtype_of(bound)) {
       // elements are guaranteed to be subtypes, so no check necessary
-      bs->write_ref_array_pre(dst_mr);
+      bs->write_ref_array_pre(dst, length);
       Copy::conjoint_oops_atomic(src, dst, length);
     } else {
       // slow case: need individual subtype checks
@@ -137,6 +134,7 @@
       }
     }
   }
+  const size_t word_len = objArrayOopDesc::array_size(length);
   bs->write_ref_array(MemRegion((HeapWord*)dst, word_len));
 }
 
diff --git a/hotspot/src/share/vm/oops/oop.inline.hpp b/hotspot/src/share/vm/oops/oop.inline.hpp
index 5c4ad5a..962a075 100644
--- a/hotspot/src/share/vm/oops/oop.inline.hpp
+++ b/hotspot/src/share/vm/oops/oop.inline.hpp
@@ -148,12 +148,14 @@
 
 inline narrowOop oopDesc::encode_heap_oop_not_null(oop v) {
   assert(!is_null(v), "oop value can never be zero");
+  assert(Universe::heap()->is_in_reserved(v), "Address not in heap");
   address base = Universe::narrow_oop_base();
   int    shift = Universe::narrow_oop_shift();
   uint64_t  pd = (uint64_t)(pointer_delta((void*)v, (void*)base, 1));
   assert(OopEncodingHeapMax > pd, "change encoding max if new encoding");
   uint64_t result = pd >> shift;
   assert((result & CONST64(0xffffffff00000000)) == 0, "narrow oop overflow");
+  assert(decode_heap_oop(result) == v, "reversibility");
   return (narrowOop)result;
 }
 
@@ -449,7 +451,7 @@
   oopDesc::bs()->write_ref_field(p, v);
 }
 
-inline void update_barrier_set_pre(void* p, oop v) {
+template <class T> inline void update_barrier_set_pre(T* p, oop v) {
   oopDesc::bs()->write_ref_field_pre(p, v);
 }
 
@@ -459,15 +461,15 @@
   } else {
     update_barrier_set_pre(p, v);
     oopDesc::encode_store_heap_oop(p, v);
-    update_barrier_set(p, v);
+    update_barrier_set((void*)p, v);  // cast away type
   }
 }
 
 template <class T> inline void oop_store(volatile T* p, oop v) {
-  update_barrier_set_pre((void*)p, v);
+  update_barrier_set_pre((T*)p, v);   // cast away volatile
   // Used by release_obj_field_put, so use release_store_ptr.
   oopDesc::release_encode_store_heap_oop(p, v);
-  update_barrier_set((void*)p, v);
+  update_barrier_set((void*)p, v);    // cast away type
 }
 
 template <class T> inline void oop_store_without_check(T* p, oop v) {
diff --git a/hotspot/src/share/vm/oops/oopsHierarchy.hpp b/hotspot/src/share/vm/oops/oopsHierarchy.hpp
index 1e86452..4609f0e 100644
--- a/hotspot/src/share/vm/oops/oopsHierarchy.hpp
+++ b/hotspot/src/share/vm/oops/oopsHierarchy.hpp
@@ -29,6 +29,7 @@
 typedef juint narrowOop; // Offset instead of address for an oop within a java object
 typedef class klassOopDesc* wideKlassOop; // to keep SA happy and unhandled oop
                                           // detector happy.
+typedef void* OopOrNarrowOopStar;
 
 #ifndef CHECK_UNHANDLED_OOPS
 
diff --git a/hotspot/src/share/vm/opto/cfgnode.cpp b/hotspot/src/share/vm/opto/cfgnode.cpp
index e48c15e..a81859c 100644
--- a/hotspot/src/share/vm/opto/cfgnode.cpp
+++ b/hotspot/src/share/vm/opto/cfgnode.cpp
@@ -1789,7 +1789,7 @@
 #ifdef _LP64
   // Push DecodeN down through phi.
   // The rest of phi graph will transform by split EncodeP node though phis up.
-  if (UseCompressedOops && can_reshape && progress == NULL) {
+  if (UseNewCode && UseCompressedOops && can_reshape && progress == NULL) {
     bool may_push = true;
     bool has_decodeN = false;
     Node* in_decodeN = NULL;
diff --git a/hotspot/src/share/vm/prims/unsafe.cpp b/hotspot/src/share/vm/prims/unsafe.cpp
index 081f19f..8bf06ef 100644
--- a/hotspot/src/share/vm/prims/unsafe.cpp
+++ b/hotspot/src/share/vm/prims/unsafe.cpp
@@ -1048,7 +1048,11 @@
   oop e = JNIHandles::resolve(e_h);
   oop p = JNIHandles::resolve(obj);
   HeapWord* addr = (HeapWord *)index_oop_from_field_offset_long(p, offset);
-  update_barrier_set_pre((void*)addr, e);
+  if (UseCompressedOops) {
+    update_barrier_set_pre((narrowOop*)addr, e);
+  } else {
+    update_barrier_set_pre((oop*)addr, e);
+  }
   oop res = oopDesc::atomic_compare_exchange_oop(x, addr, e);
   jboolean success  = (res == e);
   if (success)
diff --git a/hotspot/src/share/vm/runtime/arguments.cpp b/hotspot/src/share/vm/runtime/arguments.cpp
index d86e7a9..6eaa8ce 100644
--- a/hotspot/src/share/vm/runtime/arguments.cpp
+++ b/hotspot/src/share/vm/runtime/arguments.cpp
@@ -1202,18 +1202,13 @@
   }
 
 #ifdef _LP64
-  // Compressed Headers do not work with CMS, which uses a bit in the klass
-  // field offset to determine free list chunk markers.
   // Check that UseCompressedOops can be set with the max heap size allocated
   // by ergonomics.
   if (MaxHeapSize <= max_heap_for_compressed_oops()) {
-    if (FLAG_IS_DEFAULT(UseCompressedOops) && !UseG1GC) {
+    if (FLAG_IS_DEFAULT(UseCompressedOops)) {
       // Turn off until bug is fixed.
       // the following line to return it to default status.
       // FLAG_SET_ERGO(bool, UseCompressedOops, true);
-    } else if (UseCompressedOops && UseG1GC) {
-      warning(" UseCompressedOops does not currently work with UseG1GC; switching off UseCompressedOops. ");
-      FLAG_SET_DEFAULT(UseCompressedOops, false);
     }
 #ifdef _WIN64
     if (UseLargePages && UseCompressedOops) {
@@ -1454,6 +1449,7 @@
   if (UseSerialGC)                       i++;
   if (UseConcMarkSweepGC || UseParNewGC) i++;
   if (UseParallelGC || UseParallelOldGC) i++;
+  if (UseG1GC)                           i++;
   if (i > 1) {
     jio_fprintf(defaultStream::error_stream(),
                 "Conflicting collector combinations in option list; "
@@ -2603,22 +2599,6 @@
     return result;
   }
 
-  // These are hacks until G1 is fully supported and tested
-  // but lets you force -XX:+UseG1GC in PRT and get it where it (mostly) works
-  if (UseG1GC) {
-    if (UseConcMarkSweepGC || UseParNewGC || UseParallelGC || UseParallelOldGC || UseSerialGC) {
-#ifndef PRODUCT
-      tty->print_cr("-XX:+UseG1GC is incompatible with other collectors, using UseG1GC");
-#endif // PRODUCT
-      UseConcMarkSweepGC = false;
-      UseParNewGC        = false;
-      UseParallelGC      = false;
-      UseParallelOldGC   = false;
-      UseSerialGC        = false;
-    }
-    no_shared_spaces();
-  }
-
 #ifndef PRODUCT
   if (TraceBytecodesAt != 0) {
     TraceBytecodes = true;
@@ -2676,10 +2656,7 @@
   } else if (UseParNewGC) {
     // Set some flags for ParNew
     set_parnew_gc_flags();
-  }
-  // Temporary; make the "if" an "else-if" before
-  // we integrate G1. XXX
-  if (UseG1GC) {
+  } else if (UseG1GC) {
     // Set some flags for garbage-first, if needed.
     set_g1_gc_flags();
   }
diff --git a/hotspot/src/share/vm/runtime/safepoint.cpp b/hotspot/src/share/vm/runtime/safepoint.cpp
index eed2aff..01bd705 100644
--- a/hotspot/src/share/vm/runtime/safepoint.cpp
+++ b/hotspot/src/share/vm/runtime/safepoint.cpp
@@ -49,7 +49,7 @@
     // In the future we should investigate whether CMS can use the
     // more-general mechanism below.  DLD (01/05).
     ConcurrentMarkSweepThread::synchronize(false);
-  } else {
+  } else if (UseG1GC) {
     ConcurrentGCThread::safepoint_synchronize();
   }
 #endif // SERIALGC
@@ -400,7 +400,7 @@
   // If there are any concurrent GC threads resume them.
   if (UseConcMarkSweepGC) {
     ConcurrentMarkSweepThread::desynchronize(false);
-  } else {
+  } else if (UseG1GC) {
     ConcurrentGCThread::safepoint_desynchronize();
   }
 #endif // SERIALGC
diff --git a/hotspot/src/share/vm/runtime/sharedRuntime.cpp b/hotspot/src/share/vm/runtime/sharedRuntime.cpp
index 4709b6b..b729eef 100644
--- a/hotspot/src/share/vm/runtime/sharedRuntime.cpp
+++ b/hotspot/src/share/vm/runtime/sharedRuntime.cpp
@@ -119,6 +119,7 @@
     assert(false, "should be optimized out");
     return;
   }
+  assert(orig->is_oop(true /* ignore mark word */), "Error");
   // store the original value that was in the field reference
   thread->satb_mark_queue().enqueue(orig);
 JRT_END
diff --git a/hotspot/src/share/vm/utilities/taskqueue.cpp b/hotspot/src/share/vm/utilities/taskqueue.cpp
index 768f9b5..3a41b57 100644
--- a/hotspot/src/share/vm/utilities/taskqueue.cpp
+++ b/hotspot/src/share/vm/utilities/taskqueue.cpp
@@ -64,15 +64,18 @@
 }
 
 void ParallelTaskTerminator::yield() {
+  assert(_offered_termination <= _n_threads, "Invariant");
   os::yield();
 }
 
 void ParallelTaskTerminator::sleep(uint millis) {
+  assert(_offered_termination <= _n_threads, "Invariant");
   os::sleep(Thread::current(), millis, false);
 }
 
 bool
 ParallelTaskTerminator::offer_termination(TerminatorTerminator* terminator) {
+  assert(_offered_termination < _n_threads, "Invariant");
   Atomic::inc(&_offered_termination);
 
   uint yield_count = 0;
@@ -96,6 +99,7 @@
   // Loop waiting for all threads to offer termination or
   // more work.
   while (true) {
+    assert(_offered_termination <= _n_threads, "Invariant");
     // Are all threads offering termination?
     if (_offered_termination == _n_threads) {
       return true;
@@ -151,6 +155,7 @@
       if (peek_in_queue_set() ||
           (terminator != NULL && terminator->should_exit_termination())) {
         Atomic::dec(&_offered_termination);
+        assert(_offered_termination < _n_threads, "Invariant");
         return false;
       }
     }
diff --git a/hotspot/src/share/vm/utilities/taskqueue.hpp b/hotspot/src/share/vm/utilities/taskqueue.hpp
index 6b83bc0..489122a 100644
--- a/hotspot/src/share/vm/utilities/taskqueue.hpp
+++ b/hotspot/src/share/vm/utilities/taskqueue.hpp
@@ -560,8 +560,14 @@
 class StarTask {
   void*  _holder;        // either union oop* or narrowOop*
  public:
-  StarTask(narrowOop *p) { _holder = (void *)((uintptr_t)p | COMPRESSED_OOP_MASK); }
-  StarTask(oop *p)       { _holder = (void*)p; }
+  StarTask(narrowOop* p) {
+    assert(((uintptr_t)p & COMPRESSED_OOP_MASK) == 0, "Information loss!");
+    _holder = (void *)((uintptr_t)p | COMPRESSED_OOP_MASK);
+  }
+  StarTask(oop* p)       {
+    assert(((uintptr_t)p & COMPRESSED_OOP_MASK) == 0, "Information loss!");
+    _holder = (void*)p;
+  }
   StarTask()             { _holder = NULL; }
   operator oop*()        { return (oop*)_holder; }
   operator narrowOop*()  {