Merge

32 files changed, 849 insertions, 228 deletions

diff --git a/src/share/vm/c1/c1_CodeStubs.hpp b/src/share/vm/c1/c1_CodeStubs.hpp
index 1367f080d..128a6d48c 100644
--- a/src/share/vm/c1/c1_CodeStubs.hpp
+++ b/src/share/vm/c1/c1_CodeStubs.hpp
@@ -519,42 +519,126 @@ class ArrayCopyStub: public CodeStub {
 // Code stubs for Garbage-First barriers.
 class G1PreBarrierStub: public CodeStub {
  private:
+  bool _do_load;
   LIR_Opr _addr;
   LIR_Opr _pre_val;
   LIR_PatchCode _patch_code;
   CodeEmitInfo* _info;
 
  public:
-  // pre_val (a temporary register) must be a register;
+  // Version that _does_ generate a load of the previous value from addr.
   // addr (the address of the field to be read) must be a LIR_Address
+  // pre_val (a temporary register) must be a register;
   G1PreBarrierStub(LIR_Opr addr, LIR_Opr pre_val, LIR_PatchCode patch_code, CodeEmitInfo* info) :
-    _addr(addr), _pre_val(pre_val), _patch_code(patch_code), _info(info)
+    _addr(addr), _pre_val(pre_val), _do_load(true),
+    _patch_code(patch_code), _info(info)
   {
     assert(_pre_val->is_register(), "should be temporary register");
     assert(_addr->is_address(), "should be the address of the field");
   }
 
+  // Version that _does not_ generate load of the previous value; the
+  // previous value is assumed to have already been loaded into pre_val.
+  G1PreBarrierStub(LIR_Opr pre_val) :
+    _addr(LIR_OprFact::illegalOpr), _pre_val(pre_val), _do_load(false),
+    _patch_code(lir_patch_none), _info(NULL)
+  {
+    assert(_pre_val->is_register(), "should be a register");
+  }
+
   LIR_Opr addr() const { return _addr; }
   LIR_Opr pre_val() const { return _pre_val; }
   LIR_PatchCode patch_code() const { return _patch_code; }
   CodeEmitInfo* info() const { return _info; }
+  bool do_load() const { return _do_load; }
 
   virtual void emit_code(LIR_Assembler* e);
   virtual void visit(LIR_OpVisitState* visitor) {
-    // don't pass in the code emit info since it's processed in the fast
-    // path
-    if (_info != NULL)
-      visitor->do_slow_case(_info);
-    else
+    if (_do_load) {
+      // don't pass in the code emit info since it's processed in the fast
+      // path
+      if (_info != NULL)
+        visitor->do_slow_case(_info);
+      else
+        visitor->do_slow_case();
+
+      visitor->do_input(_addr);
+      visitor->do_temp(_pre_val);
+    } else {
       visitor->do_slow_case();
-    visitor->do_input(_addr);
-    visitor->do_temp(_pre_val);
+      visitor->do_input(_pre_val);
+    }
   }
 #ifndef PRODUCT
   virtual void print_name(outputStream* out) const { out->print("G1PreBarrierStub"); }
 #endif // PRODUCT
 };
 
+// This G1 barrier code stub is used in Unsafe.getObject.
+// It generates a sequence of guards around the SATB
+// barrier code that are used to detect when we have
+// the referent field of a Reference object.
+// The first check is assumed to have been generated
+// in the code generated for Unsafe.getObject().
+
+class G1UnsafeGetObjSATBBarrierStub: public CodeStub {
+ private:
+  LIR_Opr _val;
+  LIR_Opr _src;
+
+  LIR_Opr _tmp;
+  LIR_Opr _thread;
+
+  bool _gen_src_check;
+
+ public:
+  // A G1 barrier that is guarded by generated guards that determine whether
+  // val (which is the result of Unsafe.getObject() should be recorded in an
+  // SATB log buffer. We could be reading the referent field of a Reference object
+  // using Unsafe.getObject() and we need to record the referent.
+  //
+  // * val is the operand returned by the unsafe.getObject routine.
+  // * src is the base object
+  // * tmp is a temp used to load the klass of src, and then reference type
+  // * thread is the thread object.
+
+  G1UnsafeGetObjSATBBarrierStub(LIR_Opr val, LIR_Opr src,
+                                LIR_Opr tmp, LIR_Opr thread,
+                                bool gen_src_check) :
+    _val(val), _src(src),
+    _tmp(tmp), _thread(thread),
+    _gen_src_check(gen_src_check)
+  {
+    assert(_val->is_register(), "should have already been loaded");
+    assert(_src->is_register(), "should have already been loaded");
+
+    assert(_tmp->is_register(), "should be a temporary register");
+  }
+
+  LIR_Opr val() const { return _val; }
+  LIR_Opr src() const { return _src; }
+
+  LIR_Opr tmp() const { return _tmp; }
+  LIR_Opr thread() const { return _thread; }
+
+  bool gen_src_check() const { return _gen_src_check; }
+
+  virtual void emit_code(LIR_Assembler* e);
+
+  virtual void visit(LIR_OpVisitState* visitor) {
+    visitor->do_slow_case();
+    visitor->do_input(_val);
+    visitor->do_input(_src);
+    visitor->do_input(_thread);
+
+    visitor->do_temp(_tmp);
+  }
+
+#ifndef PRODUCT
+  virtual void print_name(outputStream* out) const { out->print("G1UnsafeGetObjSATBBarrierStub"); }
+#endif // PRODUCT
+};
+
 class G1PostBarrierStub: public CodeStub {
  private:
   LIR_Opr _addr;
diff --git a/src/share/vm/c1/c1_GraphBuilder.cpp b/src/share/vm/c1/c1_GraphBuilder.cpp
index 9bac474f5..ffe9e1248 100644
--- a/src/share/vm/c1/c1_GraphBuilder.cpp
+++ b/src/share/vm/c1/c1_GraphBuilder.cpp
@@ -2913,6 +2913,46 @@ GraphBuilder::GraphBuilder(Compilation* compilation, IRScope* scope)
       block()->set_end(end);
       break;
     }
+
+  case vmIntrinsics::_Reference_get:
+    {
+      if (UseG1GC) {
+        // With java.lang.ref.reference.get() we must go through the
+        // intrinsic - when G1 is enabled - even when get() is the root
+        // method of the compile so that, if necessary, the value in
+        // the referent field of the reference object gets recorded by
+        // the pre-barrier code.
+        // Specifically, if G1 is enabled, the value in the referent
+        // field is recorded by the G1 SATB pre barrier. This will
+        // result in the referent being marked live and the reference
+        // object removed from the list of discovered references during
+        // reference processing.
+
+        // Set up a stream so that appending instructions works properly.
+        ciBytecodeStream s(scope->method());
+        s.reset_to_bci(0);
+        scope_data()->set_stream(&s);
+        s.next();
+
+        // setup the initial block state
+        _block = start_block;
+        _state = start_block->state()->copy_for_parsing();
+        _last  = start_block;
+        load_local(objectType, 0);
+
+        // Emit the intrinsic node.
+        bool result = try_inline_intrinsics(scope->method());
+        if (!result) BAILOUT("failed to inline intrinsic");
+        method_return(apop());
+
+        // connect the begin and end blocks and we're all done.
+        BlockEnd* end = last()->as_BlockEnd();
+        block()->set_end(end);
+        break;
+      }
+      // Otherwise, fall thru
+    }
+
   default:
     scope_data()->add_to_work_list(start_block);
     iterate_all_blocks();
@@ -3150,6 +3190,15 @@ bool GraphBuilder::try_inline_intrinsics(ciMethod* callee) {
       append_unsafe_CAS(callee);
       return true;
 
+    case vmIntrinsics::_Reference_get:
+      // It is only when G1 is enabled that we absolutely
+      // need to use the intrinsic version of Reference.get()
+      // so that the value in the referent field, if necessary,
+      // can be registered by the pre-barrier code.
+      if (!UseG1GC) return false;
+      preserves_state = true;
+      break;
+
     default                       : return false; // do not inline
   }
   // create intrinsic node
diff --git a/src/share/vm/c1/c1_LIRGenerator.cpp b/src/share/vm/c1/c1_LIRGenerator.cpp
index 33278ef50..8226935f0 100644
--- a/src/share/vm/c1/c1_LIRGenerator.cpp
+++ b/src/share/vm/c1/c1_LIRGenerator.cpp
@@ -1209,6 +1209,38 @@ void LIRGenerator::do_Return(Return* x) {
   set_no_result(x);
 }
 
+// Examble: ref.get()
+// Combination of LoadField and g1 pre-write barrier
+void LIRGenerator::do_Reference_get(Intrinsic* x) {
+
+  const int referent_offset = java_lang_ref_Reference::referent_offset;
+  guarantee(referent_offset > 0, "referent offset not initialized");
+
+  assert(x->number_of_arguments() == 1, "wrong type");
+
+  LIRItem reference(x->argument_at(0), this);
+  reference.load_item();
+
+  // need to perform the null check on the reference objecy
+  CodeEmitInfo* info = NULL;
+  if (x->needs_null_check()) {
+    info = state_for(x);
+  }
+
+  LIR_Address* referent_field_adr =
+    new LIR_Address(reference.result(), referent_offset, T_OBJECT);
+
+  LIR_Opr result = rlock_result(x);
+
+  __ load(referent_field_adr, result, info);
+
+  // Register the value in the referent field with the pre-barrier
+  pre_barrier(LIR_OprFact::illegalOpr /* addr_opr */,
+              result /* pre_val */,
+              false  /* do_load */,
+              false  /* patch */,
+              NULL   /* info */);
+}
 
 // Example: object.getClass ()
 void LIRGenerator::do_getClass(Intrinsic* x) {
@@ -1351,13 +1383,14 @@ LIR_Opr LIRGenerator::load_constant(LIR_Const* c) {
 
 // Various barriers
 
-void LIRGenerator::pre_barrier(LIR_Opr addr_opr, bool patch,  CodeEmitInfo* info) {
+void LIRGenerator::pre_barrier(LIR_Opr addr_opr, LIR_Opr pre_val,
+                               bool do_load, bool patch, CodeEmitInfo* info) {
   // Do the pre-write barrier, if any.
   switch (_bs->kind()) {
 #ifndef SERIALGC
     case BarrierSet::G1SATBCT:
     case BarrierSet::G1SATBCTLogging:
-      G1SATBCardTableModRef_pre_barrier(addr_opr, patch, info);
+      G1SATBCardTableModRef_pre_barrier(addr_opr, pre_val, do_load, patch, info);
       break;
 #endif // SERIALGC
     case BarrierSet::CardTableModRef:
@@ -1398,9 +1431,8 @@ void LIRGenerator::post_barrier(LIR_OprDesc* addr, LIR_OprDesc* new_val) {
 ////////////////////////////////////////////////////////////////////////
 #ifndef SERIALGC
 
-void LIRGenerator::G1SATBCardTableModRef_pre_barrier(LIR_Opr addr_opr, bool patch,  CodeEmitInfo* info) {
-  if (G1DisablePreBarrier) return;
-
+void LIRGenerator::G1SATBCardTableModRef_pre_barrier(LIR_Opr addr_opr, LIR_Opr pre_val,
+                                                     bool do_load, bool patch, CodeEmitInfo* info) {
   // First we test whether marking is in progress.
   BasicType flag_type;
   if (in_bytes(PtrQueue::byte_width_of_active()) == 4) {
@@ -1419,26 +1451,40 @@ void LIRGenerator::G1SATBCardTableModRef_pre_barrier(LIR_Opr addr_opr, bool patc
   // Read the marking-in-progress flag.
   LIR_Opr flag_val = new_register(T_INT);
   __ load(mark_active_flag_addr, flag_val);
+  __ cmp(lir_cond_notEqual, flag_val, LIR_OprFact::intConst(0));
 
-  LIR_PatchCode pre_val_patch_code =
-    patch ? lir_patch_normal : lir_patch_none;
+  LIR_PatchCode pre_val_patch_code = lir_patch_none;
 
-  LIR_Opr pre_val = new_register(T_OBJECT);
+  CodeStub* slow;
 
-  __ cmp(lir_cond_notEqual, flag_val, LIR_OprFact::intConst(0));
-  if (!addr_opr->is_address()) {
-    assert(addr_opr->is_register(), "must be");
-    addr_opr = LIR_OprFact::address(new LIR_Address(addr_opr, T_OBJECT));
+  if (do_load) {
+    assert(pre_val == LIR_OprFact::illegalOpr, "sanity");
+    assert(addr_opr != LIR_OprFact::illegalOpr, "sanity");
+
+    if (patch)
+      pre_val_patch_code = lir_patch_normal;
+
+    pre_val = new_register(T_OBJECT);
+
+    if (!addr_opr->is_address()) {
+      assert(addr_opr->is_register(), "must be");
+      addr_opr = LIR_OprFact::address(new LIR_Address(addr_opr, T_OBJECT));
+    }
+    slow = new G1PreBarrierStub(addr_opr, pre_val, pre_val_patch_code, info);
+  } else {
+    assert(addr_opr == LIR_OprFact::illegalOpr, "sanity");
+    assert(pre_val->is_register(), "must be");
+    assert(pre_val->type() == T_OBJECT, "must be an object");
+    assert(info == NULL, "sanity");
+
+    slow = new G1PreBarrierStub(pre_val);
   }
-  CodeStub* slow = new G1PreBarrierStub(addr_opr, pre_val, pre_val_patch_code,
-                                        info);
+
   __ branch(lir_cond_notEqual, T_INT, slow);
   __ branch_destination(slow->continuation());
 }
 
 void LIRGenerator::G1SATBCardTableModRef_post_barrier(LIR_OprDesc* addr, LIR_OprDesc* new_val) {
-  if (G1DisablePostBarrier) return;
-
   // If the "new_val" is a constant NULL, no barrier is necessary.
   if (new_val->is_constant() &&
       new_val->as_constant_ptr()->as_jobject() == NULL) return;
@@ -1662,6 +1708,8 @@ void LIRGenerator::do_StoreField(StoreField* x) {
   if (is_oop) {
     // Do the pre-write barrier, if any.
     pre_barrier(LIR_OprFact::address(address),
+                LIR_OprFact::illegalOpr /* pre_val */,
+                true /* do_load*/,
                 needs_patching,
                 (info ? new CodeEmitInfo(info) : NULL));
   }
@@ -2091,9 +2139,144 @@ void LIRGenerator::do_UnsafeGetObject(UnsafeGetObject* x) {
   off.load_item();
   src.load_item();
 
-  LIR_Opr reg = reg = rlock_result(x, x->basic_type());
+  LIR_Opr reg = rlock_result(x, x->basic_type());
 
   get_Object_unsafe(reg, src.result(), off.result(), type, x->is_volatile());
+
+#ifndef SERIALGC
+  // We might be reading the value of the referent field of a
+  // Reference object in order to attach it back to the live
+  // object graph. If G1 is enabled then we need to record
+  // the value that is being returned in an SATB log buffer.
+  //
+  // We need to generate code similar to the following...
+  //
+  // if (offset == java_lang_ref_Reference::referent_offset) {
+  //   if (src != NULL) {
+  //     if (klass(src)->reference_type() != REF_NONE) {
+  //       pre_barrier(..., reg, ...);
+  //     }
+  //   }
+  // }
+  //
+  // The first non-constant check of either the offset or
+  // the src operand will be done here; the remainder
+  // will take place in the generated code stub.
+
+  if (UseG1GC && type == T_OBJECT) {
+    bool gen_code_stub = true;       // Assume we need to generate the slow code stub.
+    bool gen_offset_check = true;       // Assume the code stub has to generate the offset guard.
+    bool gen_source_check = true;       // Assume the code stub has to check the src object for null.
+
+    if (off.is_constant()) {
+      jlong off_con = (off.type()->is_int() ?
+                        (jlong) off.get_jint_constant() :
+                        off.get_jlong_constant());
+
+
+      if (off_con != (jlong) java_lang_ref_Reference::referent_offset) {
+        // The constant offset is something other than referent_offset.
+        // We can skip generating/checking the remaining guards and
+        // skip generation of the code stub.
+        gen_code_stub = false;
+      } else {
+        // The constant offset is the same as referent_offset -
+        // we do not need to generate a runtime offset check.
+        gen_offset_check = false;
+      }
+    }
+
+    // We don't need to generate stub if the source object is an array
+    if (gen_code_stub && src.type()->is_array()) {
+      gen_code_stub = false;
+    }
+
+    if (gen_code_stub) {
+      // We still need to continue with the checks.
+      if (src.is_constant()) {
+        ciObject* src_con = src.get_jobject_constant();
+
+        if (src_con->is_null_object()) {
+          // The constant src object is null - We can skip
+          // generating the code stub.
+          gen_code_stub = false;
+        } else {
+          // Non-null constant source object. We still have to generate
+          // the slow stub - but we don't need to generate the runtime
+          // null object check.
+          gen_source_check = false;
+        }
+      }
+    }
+
+    if (gen_code_stub) {
+      // Temoraries.
+      LIR_Opr src_klass = new_register(T_OBJECT);
+
+      // Get the thread pointer for the pre-barrier
+      LIR_Opr thread = getThreadPointer();
+
+      CodeStub* stub;
+
+      // We can have generate one runtime check here. Let's start with
+      // the offset check.
+      if (gen_offset_check) {
+        // if (offset == referent_offset) -> slow code stub
+        // If offset is an int then we can do the comparison with the
+        // referent_offset constant; otherwise we need to move
+        // referent_offset into a temporary register and generate
+        // a reg-reg compare.
+
+        LIR_Opr referent_off;
+
+        if (off.type()->is_int()) {
+          referent_off = LIR_OprFact::intConst(java_lang_ref_Reference::referent_offset);
+        } else {
+          assert(off.type()->is_long(), "what else?");
+          referent_off = new_register(T_LONG);
+          __ move(LIR_OprFact::longConst(java_lang_ref_Reference::referent_offset), referent_off);
+        }
+
+        __ cmp(lir_cond_equal, off.result(), referent_off);
+
+        // Optionally generate "src == null" check.
+        stub = new G1UnsafeGetObjSATBBarrierStub(reg, src.result(),
+                                                    src_klass, thread,
+                                                    gen_source_check);
+
+        __ branch(lir_cond_equal, as_BasicType(off.type()), stub);
+      } else {
+        if (gen_source_check) {
+          // offset is a const and equals referent offset
+          // if (source != null) -> slow code stub
+          __ cmp(lir_cond_notEqual, src.result(), LIR_OprFact::oopConst(NULL));
+
+          // Since we are generating the "if src == null" guard here,
+          // there is no need to generate the "src == null" check again.
+          stub = new G1UnsafeGetObjSATBBarrierStub(reg, src.result(),
+                                                    src_klass, thread,
+                                                    false);
+
+          __ branch(lir_cond_notEqual, T_OBJECT, stub);
+        } else {
+          // We have statically determined that offset == referent_offset
+          // && src != null so we unconditionally branch to code stub
+          // to perform the guards and record reg in the SATB log buffer.
+
+          stub = new G1UnsafeGetObjSATBBarrierStub(reg, src.result(),
+                                                    src_klass, thread,
+                                                    false);
+
+          __ branch(lir_cond_always, T_ILLEGAL, stub);
+        }
+      }
+
+      // Continuation point
+      __ branch_destination(stub->continuation());
+    }
+  }
+#endif // SERIALGC
+
   if (x->is_volatile() && os::is_MP()) __ membar_acquire();
 }
 
@@ -2759,6 +2942,10 @@ void LIRGenerator::do_Intrinsic(Intrinsic* x) {
     do_AttemptUpdate(x);
     break;
 
+  case vmIntrinsics::_Reference_get:
+    do_Reference_get(x);
+    break;
+
   default: ShouldNotReachHere(); break;
   }
 }
diff --git a/src/share/vm/c1/c1_LIRGenerator.hpp b/src/share/vm/c1/c1_LIRGenerator.hpp
index 4b9c4aeca..2b8fc23e6 100644
--- a/src/share/vm/c1/c1_LIRGenerator.hpp
+++ b/src/share/vm/c1/c1_LIRGenerator.hpp
@@ -246,6 +246,7 @@ class LIRGenerator: public InstructionVisitor, public BlockClosure {
   void do_AttemptUpdate(Intrinsic* x);
   void do_NIOCheckIndex(Intrinsic* x);
   void do_FPIntrinsics(Intrinsic* x);
+  void do_Reference_get(Intrinsic* x);
 
   void do_UnsafePrefetch(UnsafePrefetch* x, bool is_store);
 
@@ -260,13 +261,14 @@ class LIRGenerator: public InstructionVisitor, public BlockClosure {
 
   // generic interface
 
-  void pre_barrier(LIR_Opr addr_opr, bool patch,  CodeEmitInfo* info);
+  void pre_barrier(LIR_Opr addr_opr, LIR_Opr pre_val, bool do_load, bool patch, CodeEmitInfo* info);
   void post_barrier(LIR_OprDesc* addr, LIR_OprDesc* new_val);
 
   // specific implementations
   // pre barriers
 
-  void G1SATBCardTableModRef_pre_barrier(LIR_Opr addr_opr, bool patch,  CodeEmitInfo* info);
+  void G1SATBCardTableModRef_pre_barrier(LIR_Opr addr_opr, LIR_Opr pre_val,
+                                         bool do_load, bool patch, CodeEmitInfo* info);
 
   // post barriers
 
diff --git a/src/share/vm/classfile/vmSymbols.hpp b/src/share/vm/classfile/vmSymbols.hpp
index 00bd191c8..6b72a3ff3 100644
--- a/src/share/vm/classfile/vmSymbols.hpp
+++ b/src/share/vm/classfile/vmSymbols.hpp
@@ -677,6 +677,10 @@
   do_intrinsic(_checkIndex,               java_nio_Buffer,        checkIndex_name, int_int_signature,            F_R)   \
    do_name(     checkIndex_name,                                 "checkIndex")                                          \
                                                                                                                         \
+  /* java/lang/ref/Reference */                                                                                         \
+  do_intrinsic(_Reference_get,            java_lang_ref_Reference, get_name,    void_object_signature, F_R)             \
+                                                                                                                        \
+                                                                                                                        \
   do_class(sun_misc_AtomicLongCSImpl,     "sun/misc/AtomicLongCSImpl")                                                  \
   do_intrinsic(_get_AtomicLong,           sun_misc_AtomicLongCSImpl, get_name, void_long_signature,              F_R)   \
   /*   (symbols get_name and void_long_signature defined above) */                                                      \
diff --git a/src/share/vm/gc_implementation/g1/g1SATBCardTableModRefBS.cpp b/src/share/vm/gc_implementation/g1/g1SATBCardTableModRefBS.cpp
index 8145aa9fc..59d259065 100644
--- a/src/share/vm/gc_implementation/g1/g1SATBCardTableModRefBS.cpp
+++ b/src/share/vm/gc_implementation/g1/g1SATBCardTableModRefBS.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2001, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2001, 2011, Oracle and/or its affiliates. All rights reserved.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
  * This code is free software; you can redistribute it and/or modify it
@@ -47,7 +47,9 @@ G1SATBCardTableModRefBS::G1SATBCardTableModRefBS(MemRegion whole_heap,
 
 
 void G1SATBCardTableModRefBS::enqueue(oop pre_val) {
-  assert(pre_val->is_oop_or_null(true), "Error");
+  // Nulls should have been already filtered.
+  assert(pre_val->is_oop(true), "Error");
+
   if (!JavaThread::satb_mark_queue_set().is_active()) return;
   Thread* thr = Thread::current();
   if (thr->is_Java_thread()) {
@@ -59,20 +61,6 @@ void G1SATBCardTableModRefBS::enqueue(oop pre_val) {
   }
 }
 
-// When we know the current java thread:
-template <class T> void
-G1SATBCardTableModRefBS::write_ref_field_pre_static(T* field,
-                                                    oop new_val,
-                                                    JavaThread* jt) {
-  if (!JavaThread::satb_mark_queue_set().is_active()) return;
-  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);
-  }
-}
-
 template <class T> void
 G1SATBCardTableModRefBS::write_ref_array_pre_work(T* dst, int count) {
   if (!JavaThread::satb_mark_queue_set().is_active()) return;
diff --git a/src/share/vm/gc_implementation/g1/g1SATBCardTableModRefBS.hpp b/src/share/vm/gc_implementation/g1/g1SATBCardTableModRefBS.hpp
index 50cc99d09..dde3ba4be 100644
--- a/src/share/vm/gc_implementation/g1/g1SATBCardTableModRefBS.hpp
+++ b/src/share/vm/gc_implementation/g1/g1SATBCardTableModRefBS.hpp
@@ -37,12 +37,11 @@ class DirtyCardQueueSet;
 // snapshot-at-the-beginning marking.
 
 class G1SATBCardTableModRefBS: public CardTableModRefBSForCTRS {
-private:
+public:
   // Add "pre_val" to a set of objects that may have been disconnected from the
   // pre-marking object graph.
   static void enqueue(oop pre_val);
 
-public:
   G1SATBCardTableModRefBS(MemRegion whole_heap,
                           int max_covered_regions);
 
@@ -61,10 +60,6 @@ public:
     }
   }
 
-  // When we know the current java thread:
-  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.
   template <class T> inline void inline_write_ref_field_pre(T* field, oop newVal) {
diff --git a/src/share/vm/gc_implementation/g1/g1_globals.hpp b/src/share/vm/gc_implementation/g1/g1_globals.hpp
index b42949f22..5fd535533 100644
--- a/src/share/vm/gc_implementation/g1/g1_globals.hpp
+++ b/src/share/vm/gc_implementation/g1/g1_globals.hpp
@@ -89,13 +89,9 @@
           "The number of discovered reference objects to process before "   \
           "draining concurrent marking work queues.")                       \
                                                                             \
-  experimental(bool, G1UseConcMarkReferenceProcessing, false,               \
+  experimental(bool, G1UseConcMarkReferenceProcessing, true,                \
           "If true, enable reference discovery during concurrent "          \
-          "marking and reference processing at the end of remark "          \
-          "(unsafe).")                                                      \
-                                                                            \
-  develop(bool, G1SATBBarrierPrintNullPreVals, false,                       \
-          "If true, count frac of ptr writes with null pre-vals.")          \
+          "marking and reference processing at the end of remark.")         \
                                                                             \
   product(intx, G1SATBBufferSize, 1*K,                                      \
           "Number of entries in an SATB log buffer.")                       \
@@ -150,12 +146,6 @@
   develop(bool, G1PrintParCleanupStats, false,                              \
           "When true, print extra stats about parallel cleanup.")           \
                                                                             \
-  develop(bool, G1DisablePreBarrier, false,                                 \
-          "Disable generation of pre-barrier (i.e., marking barrier)   ")   \
-                                                                            \
-  develop(bool, G1DisablePostBarrier, false,                                \
-          "Disable generation of post-barrier (i.e., RS barrier)   ")       \
-                                                                            \
   product(intx, G1UpdateBufferSize, 256,                                    \
           "Size of an update buffer")                                       \
                                                                             \
diff --git a/src/share/vm/gc_implementation/parallelScavenge/psOldGen.cpp b/src/share/vm/gc_implementation/parallelScavenge/psOldGen.cpp
index df0865ee3..acbe76f94 100644
--- a/src/share/vm/gc_implementation/parallelScavenge/psOldGen.cpp
+++ b/src/share/vm/gc_implementation/parallelScavenge/psOldGen.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2001, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2001, 2011, Oracle and/or its affiliates. All rights reserved.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
  * This code is free software; you can redistribute it and/or modify it
@@ -176,10 +176,6 @@ void PSOldGen::compact() {
   object_mark_sweep()->compact(ZapUnusedHeapArea);
 }
 
-void PSOldGen::move_and_update(ParCompactionManager* cm) {
-  PSParallelCompact::move_and_update(cm, PSParallelCompact::old_space_id);
-}
-
 size_t PSOldGen::contiguous_available() const {
   return object_space()->free_in_bytes() + virtual_space()->uncommitted_size();
 }
diff --git a/src/share/vm/gc_implementation/parallelScavenge/psOldGen.hpp b/src/share/vm/gc_implementation/parallelScavenge/psOldGen.hpp
index 0d9cdd75d..76f43f51a 100644
--- a/src/share/vm/gc_implementation/parallelScavenge/psOldGen.hpp
+++ b/src/share/vm/gc_implementation/parallelScavenge/psOldGen.hpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2001, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2001, 2011, Oracle and/or its affiliates. All rights reserved.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
  * This code is free software; you can redistribute it and/or modify it
@@ -143,9 +143,6 @@ class PSOldGen : public CHeapObj {
   void adjust_pointers();
   void compact();
 
-  // Parallel old
-  virtual void move_and_update(ParCompactionManager* cm);
-
   // Size info
   size_t capacity_in_bytes() const        { return object_space()->capacity_in_bytes(); }
   size_t used_in_bytes() const            { return object_space()->used_in_bytes(); }
diff --git a/src/share/vm/gc_implementation/parallelScavenge/psParallelCompact.cpp b/src/share/vm/gc_implementation/parallelScavenge/psParallelCompact.cpp
index 5f7504ded..7ff75bd8d 100644
--- a/src/share/vm/gc_implementation/parallelScavenge/psParallelCompact.cpp
+++ b/src/share/vm/gc_implementation/parallelScavenge/psParallelCompact.cpp
@@ -2104,11 +2104,7 @@ void PSParallelCompact::invoke_no_policy(bool maximum_heap_compaction) {
     // klasses are used in the update of an object?
     compact_perm(vmthread_cm);
 
-    if (UseParallelOldGCCompacting) {
-      compact();
-    } else {
-      compact_serial(vmthread_cm);
-    }
+    compact();
 
     // Reset the mark bitmap, summary data, and do other bookkeeping.  Must be
     // done before resizing.
@@ -2582,18 +2578,16 @@ void PSParallelCompact::enqueue_dense_prefix_tasks(GCTaskQueue* q,
     // each thread?
     if (total_dense_prefix_regions > 0) {
       uint tasks_for_dense_prefix = 1;
-      if (UseParallelDensePrefixUpdate) {
-        if (total_dense_prefix_regions <=
-            (parallel_gc_threads * PAR_OLD_DENSE_PREFIX_OVER_PARTITIONING)) {
-          // Don't over partition.  This assumes that
-          // PAR_OLD_DENSE_PREFIX_OVER_PARTITIONING is a small integer value
-          // so there are not many regions to process.
-          tasks_for_dense_prefix = parallel_gc_threads;
-        } else {
-          // Over partition
-          tasks_for_dense_prefix = parallel_gc_threads *
-            PAR_OLD_DENSE_PREFIX_OVER_PARTITIONING;
-        }
+      if (total_dense_prefix_regions <=
+          (parallel_gc_threads * PAR_OLD_DENSE_PREFIX_OVER_PARTITIONING)) {
+        // Don't over partition.  This assumes that
+        // PAR_OLD_DENSE_PREFIX_OVER_PARTITIONING is a small integer value
+        // so there are not many regions to process.
+        tasks_for_dense_prefix = parallel_gc_threads;
+      } else {
+        // Over partition
+        tasks_for_dense_prefix = parallel_gc_threads *
+          PAR_OLD_DENSE_PREFIX_OVER_PARTITIONING;
       }
       size_t regions_per_thread = total_dense_prefix_regions /
         tasks_for_dense_prefix;
@@ -2733,21 +2727,6 @@ void PSParallelCompact::verify_complete(SpaceId space_id) {
 }
 #endif  // #ifdef ASSERT
 
-void PSParallelCompact::compact_serial(ParCompactionManager* cm) {
-  EventMark m("5 compact serial");
-  TraceTime tm("compact serial", print_phases(), true, gclog_or_tty);
-
-  ParallelScavengeHeap* heap = (ParallelScavengeHeap*)Universe::heap();
-  assert(heap->kind() == CollectedHeap::ParallelScavengeHeap, "Sanity");
-
-  PSYoungGen* young_gen = heap->young_gen();
-  PSOldGen* old_gen = heap->old_gen();
-
-  old_gen->start_array()->reset();
-  old_gen->move_and_update(cm);
-  young_gen->move_and_update(cm);
-}
-
 void
 PSParallelCompact::follow_weak_klass_links() {
   // All klasses on the revisit stack are marked at this point.
@@ -3530,11 +3509,8 @@ PSParallelCompact::VerifyUpdateClosure::do_addr(HeapWord* addr, size_t words) {
            "Object liveness is wrong.");
     return ParMarkBitMap::incomplete;
   }
-  assert(UseParallelOldGCDensePrefix ||
-         (HeapMaximumCompactionInterval > 1) ||
-         (MarkSweepAlwaysCompactCount > 1) ||
-         (forwarding_ptr == new_pointer),
-    "Calculation of new location is incorrect");
+  assert(HeapMaximumCompactionInterval > 1 || MarkSweepAlwaysCompactCount > 1 ||
+         forwarding_ptr == new_pointer, "new location is incorrect");
   return ParMarkBitMap::incomplete;
 }
 
diff --git a/src/share/vm/gc_implementation/parallelScavenge/psParallelCompact.hpp b/src/share/vm/gc_implementation/parallelScavenge/psParallelCompact.hpp
index 0e2c6477a..f47bff5c2 100644
--- a/src/share/vm/gc_implementation/parallelScavenge/psParallelCompact.hpp
+++ b/src/share/vm/gc_implementation/parallelScavenge/psParallelCompact.hpp
@@ -1027,9 +1027,6 @@ class PSParallelCompact : AllStatic {
                                        ParallelTaskTerminator* terminator_ptr,
                                        uint parallel_gc_threads);
 
-  // For debugging only - compacts the old gen serially
-  static void compact_serial(ParCompactionManager* cm);
-
   // If objects are left in eden after a collection, try to move the boundary
   // and absorb them into the old gen.  Returns true if eden was emptied.
   static bool absorb_live_data_from_eden(PSAdaptiveSizePolicy* size_policy,
diff --git a/src/share/vm/gc_implementation/parallelScavenge/psPermGen.cpp b/src/share/vm/gc_implementation/parallelScavenge/psPermGen.cpp
index 60fea3165..eacb47fb2 100644
--- a/src/share/vm/gc_implementation/parallelScavenge/psPermGen.cpp
+++ b/src/share/vm/gc_implementation/parallelScavenge/psPermGen.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2001, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2001, 2011, Oracle and/or its affiliates. All rights reserved.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
  * This code is free software; you can redistribute it and/or modify it
@@ -121,12 +121,6 @@ void PSPermGen::compute_new_size(size_t used_before_collection) {
   }
 }
 
-
-
-void PSPermGen::move_and_update(ParCompactionManager* cm) {
-  PSParallelCompact::move_and_update(cm, PSParallelCompact::perm_space_id);
-}
-
 void PSPermGen::precompact() {
   // Reset start array first.
   _start_array.reset();
diff --git a/src/share/vm/gc_implementation/parallelScavenge/psPermGen.hpp b/src/share/vm/gc_implementation/parallelScavenge/psPermGen.hpp
index 7a890fc8d..ba384d429 100644
--- a/src/share/vm/gc_implementation/parallelScavenge/psPermGen.hpp
+++ b/src/share/vm/gc_implementation/parallelScavenge/psPermGen.hpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2001, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2001, 2011, Oracle and/or its affiliates. All rights reserved.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
  * This code is free software; you can redistribute it and/or modify it
@@ -51,9 +51,6 @@ class PSPermGen : public PSOldGen {
   // MarkSweep code
   virtual void precompact();
 
-  // Parallel old
-  virtual void move_and_update(ParCompactionManager* cm);
-
   virtual const char* name() const { return "PSPermGen"; }
 };
 
diff --git a/src/share/vm/gc_implementation/parallelScavenge/psYoungGen.cpp b/src/share/vm/gc_implementation/parallelScavenge/psYoungGen.cpp
index b5e6462e7..5355abe0a 100644
--- a/src/share/vm/gc_implementation/parallelScavenge/psYoungGen.cpp
+++ b/src/share/vm/gc_implementation/parallelScavenge/psYoungGen.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2001, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2001, 2011, Oracle and/or its affiliates. All rights reserved.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
  * This code is free software; you can redistribute it and/or modify it
@@ -792,12 +792,6 @@ void PSYoungGen::compact() {
   to_mark_sweep()->compact(false);
 }
 
-void PSYoungGen::move_and_update(ParCompactionManager* cm) {
-  PSParallelCompact::move_and_update(cm, PSParallelCompact::eden_space_id);
-  PSParallelCompact::move_and_update(cm, PSParallelCompact::from_space_id);
-  PSParallelCompact::move_and_update(cm, PSParallelCompact::to_space_id);
-}
-
 void PSYoungGen::print() const { print_on(tty); }
 void PSYoungGen::print_on(outputStream* st) const {
   st->print(" %-15s", "PSYoungGen");
diff --git a/src/share/vm/gc_implementation/parallelScavenge/psYoungGen.hpp b/src/share/vm/gc_implementation/parallelScavenge/psYoungGen.hpp
index 3d4c56637..898b56f57 100644
--- a/src/share/vm/gc_implementation/parallelScavenge/psYoungGen.hpp
+++ b/src/share/vm/gc_implementation/parallelScavenge/psYoungGen.hpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2001, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2001, 2011, Oracle and/or its affiliates. All rights reserved.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
  * This code is free software; you can redistribute it and/or modify it
@@ -127,9 +127,6 @@ class PSYoungGen : public CHeapObj {
   void adjust_pointers();
   void compact();
 
-  // Parallel Old
-  void move_and_update(ParCompactionManager* cm);
-
   // Called during/after gc
   void swap_spaces();
 
diff --git a/src/share/vm/interpreter/abstractInterpreter.hpp b/src/share/vm/interpreter/abstractInterpreter.hpp
index a1908a2c9..c407fbaba 100644
--- a/src/share/vm/interpreter/abstractInterpreter.hpp
+++ b/src/share/vm/interpreter/abstractInterpreter.hpp
@@ -104,6 +104,7 @@ class AbstractInterpreter: AllStatic {
     java_lang_math_sqrt,                                        // implementation of java.lang.Math.sqrt  (x)
     java_lang_math_log,                                         // implementation of java.lang.Math.log   (x)
     java_lang_math_log10,                                       // implementation of java.lang.Math.log10 (x)
+    java_lang_ref_reference_get,                                // implementation of java.lang.ref.Reference.get()
     number_of_method_entries,
     invalid = -1
   };
@@ -140,7 +141,7 @@ class AbstractInterpreter: AllStatic {
   // Method activation
   static MethodKind method_kind(methodHandle m);
   static address    entry_for_kind(MethodKind k)                { assert(0 <= k && k < number_of_method_entries, "illegal kind"); return _entry_table[k]; }
-  static address    entry_for_method(methodHandle m)            { return _entry_table[method_kind(m)]; }
+  static address    entry_for_method(methodHandle m)            { return entry_for_kind(method_kind(m)); }
 
   static void       print_method_kind(MethodKind kind)          PRODUCT_RETURN;
 
diff --git a/src/share/vm/interpreter/cppInterpreter.cpp b/src/share/vm/interpreter/cppInterpreter.cpp
index 6ce16e271..9a6669519 100644
--- a/src/share/vm/interpreter/cppInterpreter.cpp
+++ b/src/share/vm/interpreter/cppInterpreter.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 1997, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1997, 2011, Oracle and/or its affiliates. All rights reserved.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
  * This code is free software; you can redistribute it and/or modify it
@@ -125,6 +125,7 @@ void CppInterpreterGenerator::generate_all() {
     method_entry(java_lang_math_sqrt  );
     method_entry(java_lang_math_log   );
     method_entry(java_lang_math_log10 );
+    method_entry(java_lang_ref_reference_get);
     Interpreter::_native_entry_begin = Interpreter::code()->code_end();
     method_entry(native);
     method_entry(native_synchronized);
diff --git a/src/share/vm/interpreter/interpreter.cpp b/src/share/vm/interpreter/interpreter.cpp
index 04a9d68f4..137f50ee3 100644
--- a/src/share/vm/interpreter/interpreter.cpp
+++ b/src/share/vm/interpreter/interpreter.cpp
@@ -208,12 +208,6 @@ AbstractInterpreter::MethodKind AbstractInterpreter::method_kind(methodHandle m)
     return empty;
   }
 
-  // Accessor method?
-  if (m->is_accessor()) {
-    assert(m->size_of_parameters() == 1, "fast code for accessors assumes parameter size = 1");
-    return accessor;
-  }
-
   // Special intrinsic method?
   // Note: This test must come _after_ the test for native methods,
   //       otherwise we will run into problems with JDK 1.2, see also
@@ -227,6 +221,15 @@ AbstractInterpreter::MethodKind AbstractInterpreter::method_kind(methodHandle m)
     case vmIntrinsics::_dsqrt : return java_lang_math_sqrt ;
     case vmIntrinsics::_dlog  : return java_lang_math_log  ;
     case vmIntrinsics::_dlog10: return java_lang_math_log10;
+
+    case vmIntrinsics::_Reference_get:
+                                return java_lang_ref_reference_get;
+  }
+
+  // Accessor method?
+  if (m->is_accessor()) {
+    assert(m->size_of_parameters() == 1, "fast code for accessors assumes parameter size = 1");
+    return accessor;
   }
 
   // Note: for now: zero locals for all non-empty methods
diff --git a/src/share/vm/interpreter/templateInterpreter.cpp b/src/share/vm/interpreter/templateInterpreter.cpp
index 0d1125ecb..6b22dc1eb 100644
--- a/src/share/vm/interpreter/templateInterpreter.cpp
+++ b/src/share/vm/interpreter/templateInterpreter.cpp
@@ -372,6 +372,7 @@ void TemplateInterpreterGenerator::generate_all() {
   method_entry(java_lang_math_sqrt )
   method_entry(java_lang_math_log  )
   method_entry(java_lang_math_log10)
+  method_entry(java_lang_ref_reference_get)
 
   // all native method kinds (must be one contiguous block)
   Interpreter::_native_entry_begin = Interpreter::code()->code_end();
diff --git a/src/share/vm/oops/instanceKlass.hpp b/src/share/vm/oops/instanceKlass.hpp
index e0d70ae68..ab9ddf326 100644
--- a/src/share/vm/oops/instanceKlass.hpp
+++ b/src/share/vm/oops/instanceKlass.hpp
@@ -401,6 +401,8 @@ class instanceKlass: public Klass {
   ReferenceType reference_type() const     { return _reference_type; }
   void set_reference_type(ReferenceType t) { _reference_type = t; }
 
+  static int reference_type_offset_in_bytes() { return offset_of(instanceKlass, _reference_type); }
+
   // find local field, returns true if found
   bool find_local_field(Symbol* name, Symbol* sig, fieldDescriptor* fd) const;
   // find field in direct superinterfaces, returns the interface in which the field is defined
diff --git a/src/share/vm/opto/compile.cpp b/src/share/vm/opto/compile.cpp
index e6084be0e..7117f7ccc 100644
--- a/src/share/vm/opto/compile.cpp
+++ b/src/share/vm/opto/compile.cpp
@@ -629,7 +629,7 @@ Compile::Compile( ciEnv* ci_env, C2Compiler* compiler, ciMethod* target, int osr
     initial_gvn()->transform_no_reclaim(top());
 
     // Set up tf(), start(), and find a CallGenerator.
-    CallGenerator* cg;
+    CallGenerator* cg = NULL;
     if (is_osr_compilation()) {
       const TypeTuple *domain = StartOSRNode::osr_domain();
       const TypeTuple *range = TypeTuple::make_range(method()->signature());
@@ -644,9 +644,24 @@ Compile::Compile( ciEnv* ci_env, C2Compiler* compiler, ciMethod* target, int osr
       StartNode* s = new (this, 2) StartNode(root(), tf()->domain());
       initial_gvn()->set_type_bottom(s);
       init_start(s);
-      float past_uses = method()->interpreter_invocation_count();
-      float expected_uses = past_uses;
-      cg = CallGenerator::for_inline(method(), expected_uses);
+      if (method()->intrinsic_id() == vmIntrinsics::_Reference_get && UseG1GC) {
+        // With java.lang.ref.reference.get() we must go through the
+        // intrinsic when G1 is enabled - even when get() is the root
+        // method of the compile - so that, if necessary, the value in
+        // the referent field of the reference object gets recorded by
+        // the pre-barrier code.
+        // Specifically, if G1 is enabled, the value in the referent
+        // field is recorded by the G1 SATB pre barrier. This will
+        // result in the referent being marked live and the reference
+        // object removed from the list of discovered references during
+        // reference processing.
+        cg = find_intrinsic(method(), false);
+      }
+      if (cg == NULL) {
+        float past_uses = method()->interpreter_invocation_count();
+        float expected_uses = past_uses;
+        cg = CallGenerator::for_inline(method(), expected_uses);
+      }
     }
     if (failing())  return;
     if (cg == NULL) {
@@ -2041,6 +2056,52 @@ static bool oop_offset_is_sane(const TypeInstPtr* tp) {
   // Note that OffsetBot and OffsetTop are very negative.
 }
 
+// Eliminate trivially redundant StoreCMs and accumulate their
+// precedence edges.
+static void eliminate_redundant_card_marks(Node* n) {
+  assert(n->Opcode() == Op_StoreCM, "expected StoreCM");
+  if (n->in(MemNode::Address)->outcnt() > 1) {
+    // There are multiple users of the same address so it might be
+    // possible to eliminate some of the StoreCMs
+    Node* mem = n->in(MemNode::Memory);
+    Node* adr = n->in(MemNode::Address);
+    Node* val = n->in(MemNode::ValueIn);
+    Node* prev = n;
+    bool done = false;
+    // Walk the chain of StoreCMs eliminating ones that match.  As
+    // long as it's a chain of single users then the optimization is
+    // safe.  Eliminating partially redundant StoreCMs would require
+    // cloning copies down the other paths.
+    while (mem->Opcode() == Op_StoreCM && mem->outcnt() == 1 && !done) {
+      if (adr == mem->in(MemNode::Address) &&
+          val == mem->in(MemNode::ValueIn)) {
+        // redundant StoreCM
+        if (mem->req() > MemNode::OopStore) {
+          // Hasn't been processed by this code yet.
+          n->add_prec(mem->in(MemNode::OopStore));
+        } else {
+          // Already converted to precedence edge
+          for (uint i = mem->req(); i < mem->len(); i++) {
+            // Accumulate any precedence edges
+            if (mem->in(i) != NULL) {
+              n->add_prec(mem->in(i));
+            }
+          }
+          // Everything above this point has been processed.
+          done = true;
+        }
+        // Eliminate the previous StoreCM
+        prev->set_req(MemNode::Memory, mem->in(MemNode::Memory));
+        assert(mem->outcnt() == 0, "should be dead");
+        mem->disconnect_inputs(NULL);
+      } else {
+        prev = mem;
+      }
+      mem = prev->in(MemNode::Memory);
+    }
+  }
+}
+
 //------------------------------final_graph_reshaping_impl----------------------
 // Implement items 1-5 from final_graph_reshaping below.
 static void final_graph_reshaping_impl( Node *n, Final_Reshape_Counts &frc ) {
@@ -2167,9 +2228,19 @@ static void final_graph_reshaping_impl( Node *n, Final_Reshape_Counts &frc ) {
     frc.inc_float_count();
     goto handle_mem;
 
+  case Op_StoreCM:
+    {
+      // Convert OopStore dependence into precedence edge
+      Node* prec = n->in(MemNode::OopStore);
+      n->del_req(MemNode::OopStore);
+      n->add_prec(prec);
+      eliminate_redundant_card_marks(n);
+    }
+
+    // fall through
+
   case Op_StoreB:
   case Op_StoreC:
-  case Op_StoreCM:
   case Op_StorePConditional:
   case Op_StoreI:
   case Op_StoreL:
diff --git a/src/share/vm/opto/graphKit.cpp b/src/share/vm/opto/graphKit.cpp
index afa2efff9..f2405cabe 100644
--- a/src/share/vm/opto/graphKit.cpp
+++ b/src/share/vm/opto/graphKit.cpp
@@ -1457,19 +1457,22 @@ Node* GraphKit::store_to_memory(Node* ctl, Node* adr, Node *val, BasicType bt,
 }
 
 
-void GraphKit::pre_barrier(Node* ctl,
+void GraphKit::pre_barrier(bool do_load,
+                           Node* ctl,
                            Node* obj,
                            Node* adr,
                            uint  adr_idx,
                            Node* val,
                            const TypeOopPtr* val_type,
+                           Node* pre_val,
                            BasicType bt) {
+
   BarrierSet* bs = Universe::heap()->barrier_set();
   set_control(ctl);
   switch (bs->kind()) {
     case BarrierSet::G1SATBCT:
     case BarrierSet::G1SATBCTLogging:
-      g1_write_barrier_pre(obj, adr, adr_idx, val, val_type, bt);
+      g1_write_barrier_pre(do_load, obj, adr, adr_idx, val, val_type, pre_val, bt);
       break;
 
     case BarrierSet::CardTableModRef:
@@ -1532,7 +1535,11 @@ Node* GraphKit::store_oop(Node* ctl,
   uint adr_idx = C->get_alias_index(adr_type);
   assert(adr_idx != Compile::AliasIdxTop, "use other store_to_memory factory" );
 
-  pre_barrier(control(), obj, adr, adr_idx, val, val_type, bt);
+  pre_barrier(true /* do_load */,
+              control(), obj, adr, adr_idx, val, val_type,
+              NULL /* pre_val */,
+              bt);
+
   Node* store = store_to_memory(control(), adr, val, bt, adr_idx);
   post_barrier(control(), store, obj, adr, adr_idx, val, bt, use_precise);
   return store;
@@ -3470,12 +3477,31 @@ void GraphKit::write_barrier_post(Node* oop_store,
 }
 
 // G1 pre/post barriers
-void GraphKit::g1_write_barrier_pre(Node* obj,
+void GraphKit::g1_write_barrier_pre(bool do_load,
+                                    Node* obj,
                                     Node* adr,
                                     uint alias_idx,
                                     Node* val,
                                     const TypeOopPtr* val_type,
+                                    Node* pre_val,
                                     BasicType bt) {
+
+  // Some sanity checks
+  // Note: val is unused in this routine.
+
+  if (do_load) {
+    // We need to generate the load of the previous value
+    assert(obj != NULL, "must have a base");
+    assert(adr != NULL, "where are loading from?");
+    assert(pre_val == NULL, "loaded already?");
+    assert(val_type != NULL, "need a type");
+  } else {
+    // In this case both val_type and alias_idx are unused.
+    assert(pre_val != NULL, "must be loaded already");
+    assert(pre_val->bottom_type()->basic_type() == T_OBJECT, "or we shouldn't be here");
+  }
+  assert(bt == T_OBJECT, "or we shouldn't be here");
+
   IdealKit ideal(this, true);
 
   Node* tls = __ thread(); // ThreadLocalStorage
@@ -3497,32 +3523,28 @@ void GraphKit::g1_write_barrier_pre(Node* obj,
                                           PtrQueue::byte_offset_of_index());
   const int buffer_offset  = in_bytes(JavaThread::satb_mark_queue_offset() +  // 652
                                           PtrQueue::byte_offset_of_buf());
-  // Now the actual pointers into the thread
-
-  // set_control( ctl);
 
+  // Now the actual pointers into the thread
   Node* marking_adr = __ AddP(no_base, tls, __ ConX(marking_offset));
   Node* buffer_adr  = __ AddP(no_base, tls, __ ConX(buffer_offset));
   Node* index_adr   = __ AddP(no_base, tls, __ ConX(index_offset));
 
   // Now some of the values
-
   Node* marking = __ load(__ ctrl(), marking_adr, TypeInt::INT, active_type, Compile::AliasIdxRaw);
 
   // if (!marking)
   __ if_then(marking, BoolTest::ne, zero); {
     Node* index   = __ load(__ ctrl(), index_adr, TypeInt::INT, T_INT, Compile::AliasIdxRaw);
 
-    const Type* t1 = adr->bottom_type();
-    const Type* t2 = val->bottom_type();
-
-    Node* orig = __ load(no_ctrl, adr, val_type, bt, alias_idx);
-    // if (orig != NULL)
-    __ if_then(orig, BoolTest::ne, null()); {
-      Node* buffer  = __ load(__ ctrl(), buffer_adr, TypeRawPtr::NOTNULL, T_ADDRESS, Compile::AliasIdxRaw);
-
+    if (do_load) {
       // load original value
       // alias_idx correct??
+      pre_val = __ load(no_ctrl, adr, val_type, bt, alias_idx);
+    }
+
+    // if (pre_val != NULL)
+    __ if_then(pre_val, BoolTest::ne, null()); {
+      Node* buffer  = __ load(__ ctrl(), buffer_adr, TypeRawPtr::NOTNULL, T_ADDRESS, Compile::AliasIdxRaw);
 
       // is the queue for this thread full?
       __ if_then(index, BoolTest::ne, zero, likely); {
@@ -3536,10 +3558,9 @@ void GraphKit::g1_write_barrier_pre(Node* obj,
         next_indexX = _gvn.transform( new (C, 2) ConvI2LNode(next_index, TypeLong::make(0, max_jlong, Type::WidenMax)) );
 #endif
 
-        // Now get the buffer location we will log the original value into and store it
+        // Now get the buffer location we will log the previous value into and store it
         Node *log_addr = __ AddP(no_base, buffer, next_indexX);
-        __ store(__ ctrl(), log_addr, orig, T_OBJECT, Compile::AliasIdxRaw);
-
+        __ store(__ ctrl(), log_addr, pre_val, T_OBJECT, Compile::AliasIdxRaw);
         // update the index
         __ store(__ ctrl(), index_adr, next_index, T_INT, Compile::AliasIdxRaw);
 
@@ -3547,9 +3568,9 @@ void GraphKit::g1_write_barrier_pre(Node* obj,
 
         // logging buffer is full, call the runtime
         const TypeFunc *tf = OptoRuntime::g1_wb_pre_Type();
-        __ make_leaf_call(tf, CAST_FROM_FN_PTR(address, SharedRuntime::g1_wb_pre), "g1_wb_pre", orig, tls);
+        __ make_leaf_call(tf, CAST_FROM_FN_PTR(address, SharedRuntime::g1_wb_pre), "g1_wb_pre", pre_val, tls);
       } __ end_if();  // (!index)
-    } __ end_if();  // (orig != NULL)
+    } __ end_if();  // (pre_val != NULL)
   } __ end_if();  // (!marking)
 
   // Final sync IdealKit and GraphKit.
diff --git a/src/share/vm/opto/graphKit.hpp b/src/share/vm/opto/graphKit.hpp
index 4523326db..fdddd66e7 100644
--- a/src/share/vm/opto/graphKit.hpp
+++ b/src/share/vm/opto/graphKit.hpp
@@ -544,8 +544,10 @@ class GraphKit : public Phase {
                              BasicType bt);
 
   // For the few case where the barriers need special help
-  void pre_barrier(Node* ctl, Node* obj, Node* adr, uint adr_idx,
-                   Node* val, const TypeOopPtr* val_type, BasicType bt);
+  void pre_barrier(bool do_load, Node* ctl,
+                   Node* obj, Node* adr, uint adr_idx, Node* val, const TypeOopPtr* val_type,
+                   Node* pre_val,
+                   BasicType bt);
 
   void post_barrier(Node* ctl, Node* store, Node* obj, Node* adr, uint adr_idx,
                     Node* val, BasicType bt, bool use_precise);
@@ -671,11 +673,13 @@ class GraphKit : public Phase {
                           Node* adr,  uint adr_idx, Node* val, bool use_precise);
 
   // G1 pre/post barriers
-  void g1_write_barrier_pre(Node* obj,
+  void g1_write_barrier_pre(bool do_load,
+                            Node* obj,
                             Node* adr,
                             uint alias_idx,
                             Node* val,
                             const TypeOopPtr* val_type,
+                            Node* pre_val,
                             BasicType bt);
 
   void g1_write_barrier_post(Node* store,
diff --git a/src/share/vm/opto/lcm.cpp b/src/share/vm/opto/lcm.cpp
index 1767bce2d..5729c7eea 100644
--- a/src/share/vm/opto/lcm.cpp
+++ b/src/share/vm/opto/lcm.cpp
@@ -688,20 +688,22 @@ bool Block::schedule_local(PhaseCFG *cfg, Matcher &matcher, int *ready_cnt, Vect
       }
       ready_cnt[n->_idx] = local; // Count em up
 
-      // A few node types require changing a required edge to a precedence edge
-      // before allocation.
+#ifdef ASSERT
       if( UseConcMarkSweepGC || UseG1GC ) {
         if( n->is_Mach() && n->as_Mach()->ideal_Opcode() == Op_StoreCM ) {
-          // Note: Required edges with an index greater than oper_input_base
-          // are not supported by the allocator.
-          // Note2: Can only depend on unmatched edge being last,
-          // can not depend on its absolute position.
-          Node *oop_store = n->in(n->req() - 1);
-          n->del_req(n->req() - 1);
-          n->add_prec(oop_store);
-          assert(cfg->_bbs[oop_store->_idx]->_dom_depth <= this->_dom_depth, "oop_store must dominate card-mark");
+          // Check the precedence edges
+          for (uint prec = n->req(); prec < n->len(); prec++) {
+            Node* oop_store = n->in(prec);
+            if (oop_store != NULL) {
+              assert(cfg->_bbs[oop_store->_idx]->_dom_depth <= this->_dom_depth, "oop_store must dominate card-mark");
+            }
+          }
         }
       }
+#endif
+
+      // A few node types require changing a required edge to a precedence edge
+      // before allocation.
       if( n->is_Mach() && n->req() > TypeFunc::Parms &&
           (n->as_Mach()->ideal_Opcode() == Op_MemBarAcquire ||
            n->as_Mach()->ideal_Opcode() == Op_MemBarVolatile) ) {
diff --git a/src/share/vm/opto/library_call.cpp b/src/share/vm/opto/library_call.cpp
index 0f73e144d..8f86f5fdc 100644
--- a/src/share/vm/opto/library_call.cpp
+++ b/src/share/vm/opto/library_call.cpp
@@ -166,6 +166,10 @@ class LibraryCallKit : public GraphKit {
   // This returns Type::AnyPtr, RawPtr, or OopPtr.
   int classify_unsafe_addr(Node* &base, Node* &offset);
   Node* make_unsafe_address(Node* base, Node* offset);
+  // Helper for inline_unsafe_access.
+  // Generates the guards that check whether the result of
+  // Unsafe.getObject should be recorded in an SATB log buffer.
+  void insert_g1_pre_barrier(Node* base_oop, Node* offset, Node* pre_val);
   bool inline_unsafe_access(bool is_native_ptr, bool is_store, BasicType type, bool is_volatile);
   bool inline_unsafe_prefetch(bool is_native_ptr, bool is_store, bool is_static);
   bool inline_unsafe_allocate();
@@ -240,6 +244,8 @@ class LibraryCallKit : public GraphKit {
   bool inline_numberOfTrailingZeros(vmIntrinsics::ID id);
   bool inline_bitCount(vmIntrinsics::ID id);
   bool inline_reverseBytes(vmIntrinsics::ID id);
+
+  bool inline_reference_get();
 };
 
 
@@ -336,6 +342,14 @@ CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) {
     if (!UsePopCountInstruction)  return NULL;
     break;
 
+  case vmIntrinsics::_Reference_get:
+    // It is only when G1 is enabled that we absolutely
+    // need to use the intrinsic version of Reference.get()
+    // so that the value in the referent field, if necessary,
+    // can be registered by the pre-barrier code.
+    if (!UseG1GC) return NULL;
+    break;
+
  default:
     assert(id <= vmIntrinsics::LAST_COMPILER_INLINE, "caller responsibility");
     assert(id != vmIntrinsics::_Object_init && id != vmIntrinsics::_invoke, "enum out of order?");
@@ -387,6 +401,7 @@ JVMState* LibraryIntrinsic::generate(JVMState* jvms) {
     tty->print_cr("Intrinsic %s", str);
   }
 #endif
+
   if (kit.try_to_inline()) {
     if (PrintIntrinsics || PrintInlining NOT_PRODUCT( || PrintOptoInlining) ) {
       CompileTask::print_inlining(kit.callee(), jvms->depth() - 1, kit.bci(), is_virtual() ? "(intrinsic, virtual)" : "(intrinsic)");
@@ -402,11 +417,19 @@ JVMState* LibraryIntrinsic::generate(JVMState* jvms) {
   }
 
   if (PrintIntrinsics) {
-    tty->print("Did not inline intrinsic %s%s at bci:%d in",
+    if (jvms->has_method()) {
+      // Not a root compile.
+      tty->print("Did not inline intrinsic %s%s at bci:%d in",
+                 vmIntrinsics::name_at(intrinsic_id()),
+                 (is_virtual() ? " (virtual)" : ""), kit.bci());
+      kit.caller()->print_short_name(tty);
+      tty->print_cr(" (%d bytes)", kit.caller()->code_size());
+    } else {
+      // Root compile
+      tty->print("Did not generate intrinsic %s%s at bci:%d in",
                vmIntrinsics::name_at(intrinsic_id()),
                (is_virtual() ? " (virtual)" : ""), kit.bci());
-    kit.caller()->print_short_name(tty);
-    tty->print_cr(" (%d bytes)", kit.caller()->code_size());
+    }
   }
   C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
   return NULL;
@@ -418,6 +441,14 @@ bool LibraryCallKit::try_to_inline() {
   const bool is_native_ptr  = true;
   const bool is_static      = true;
 
+  if (!jvms()->has_method()) {
+    // Root JVMState has a null method.
+    assert(map()->memory()->Opcode() == Op_Parm, "");
+    // Insert the memory aliasing node
+    set_all_memory(reset_memory());
+  }
+  assert(merged_memory(), "");
+
   switch (intrinsic_id()) {
   case vmIntrinsics::_hashCode:
     return inline_native_hashcode(intrinsic()->is_virtual(), !is_static);
@@ -658,6 +689,9 @@ bool LibraryCallKit::try_to_inline() {
   case vmIntrinsics::_getCallerClass:
     return inline_native_Reflection_getCallerClass();
 
+  case vmIntrinsics::_Reference_get:
+    return inline_reference_get();
+
   default:
     // If you get here, it may be that someone has added a new intrinsic
     // to the list in vmSymbols.hpp without implementing it here.
@@ -2076,6 +2110,106 @@ bool LibraryCallKit::inline_reverseBytes(vmIntrinsics::ID id) {
 
 const static BasicType T_ADDRESS_HOLDER = T_LONG;
 
+// Helper that guards and inserts a G1 pre-barrier.
+void LibraryCallKit::insert_g1_pre_barrier(Node* base_oop, Node* offset, Node* pre_val) {
+  assert(UseG1GC, "should not call this otherwise");
+
+  // We could be accessing the referent field of a reference object. If so, when G1
+  // is enabled, we need to log the value in the referent field in an SATB buffer.
+  // This routine performs some compile time filters and generates suitable
+  // runtime filters that guard the pre-barrier code.
+
+  // Some compile time checks.
+
+  // If offset is a constant, is it java_lang_ref_Reference::_reference_offset?
+  const TypeX* otype = offset->find_intptr_t_type();
+  if (otype != NULL && otype->is_con() &&
+      otype->get_con() != java_lang_ref_Reference::referent_offset) {
+    // Constant offset but not the reference_offset so just return
+    return;
+  }
+
+  // We only need to generate the runtime guards for instances.
+  const TypeOopPtr* btype = base_oop->bottom_type()->isa_oopptr();
+  if (btype != NULL) {
+    if (btype->isa_aryptr()) {
+      // Array type so nothing to do
+      return;
+    }
+
+    const TypeInstPtr* itype = btype->isa_instptr();
+    if (itype != NULL) {
+      // Can the klass of base_oop be statically determined
+      // to be _not_ a sub-class of Reference?
+      ciKlass* klass = itype->klass();
+      if (klass->is_subtype_of(env()->Reference_klass()) &&
+          !env()->Reference_klass()->is_subtype_of(klass)) {
+        return;
+      }
+    }
+  }
+
+  // The compile time filters did not reject base_oop/offset so
+  // we need to generate the following runtime filters
+  //
+  // if (offset == java_lang_ref_Reference::_reference_offset) {
+  //   if (base != null) {
+  //     if (klass(base)->reference_type() != REF_NONE)) {
+  //       pre_barrier(_, pre_val, ...);
+  //     }
+  //   }
+  // }
+
+  float likely  = PROB_LIKELY(0.999);
+  float unlikely  = PROB_UNLIKELY(0.999);
+
+  IdealKit ideal(this);
+#define __ ideal.
+
+  const int reference_type_offset = instanceKlass::reference_type_offset_in_bytes() +
+                                        sizeof(oopDesc);
+
+  Node* referent_off = __ ConX(java_lang_ref_Reference::referent_offset);
+
+  __ if_then(offset, BoolTest::eq, referent_off, unlikely); {
+    __ if_then(base_oop, BoolTest::ne, null(), likely); {
+
+      // Update graphKit memory and control from IdealKit.
+      sync_kit(ideal);
+
+      Node* ref_klass_con = makecon(TypeKlassPtr::make(env()->Reference_klass()));
+      Node* is_instof = gen_instanceof(base_oop, ref_klass_con);
+
+      // Update IdealKit memory and control from graphKit.
+      __ sync_kit(this);
+
+      Node* one = __ ConI(1);
+
+      __ if_then(is_instof, BoolTest::eq, one, unlikely); {
+
+        // Update graphKit from IdeakKit.
+        sync_kit(ideal);
+
+        // Use the pre-barrier to record the value in the referent field
+        pre_barrier(false /* do_load */,
+                    __ ctrl(),
+                    NULL /* obj */, NULL /* adr */, -1 /* alias_idx */, NULL /* val */, NULL /* val_type */,
+                    pre_val /* pre_val */,
+                    T_OBJECT);
+
+        // Update IdealKit from graphKit.
+        __ sync_kit(this);
+
+      } __ end_if(); // _ref_type != ref_none
+    } __ end_if(); // base  != NULL
+  } __ end_if(); // offset == referent_offset
+
+  // Final sync IdealKit and GraphKit.
+  final_sync(ideal);
+#undef __
+}
+
+
 // Interpret Unsafe.fieldOffset cookies correctly:
 extern jlong Unsafe_field_offset_to_byte_offset(jlong field_offset);
 
@@ -2152,9 +2286,11 @@ bool LibraryCallKit::inline_unsafe_access(bool is_native_ptr, bool is_store, Bas
   // Build address expression.  See the code in inline_unsafe_prefetch.
   Node *adr;
   Node *heap_base_oop = top();
+  Node* offset = top();
+
   if (!is_native_ptr) {
     // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
-    Node* offset = pop_pair();
+    offset = pop_pair();
     // The base is either a Java object or a value produced by Unsafe.staticFieldBase
     Node* base   = pop();
     // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
@@ -2195,6 +2331,13 @@ bool LibraryCallKit::inline_unsafe_access(bool is_native_ptr, bool is_store, Bas
   // or Compile::must_alias will throw a diagnostic assert.)
   bool need_mem_bar = (alias_type->adr_type() == TypeOopPtr::BOTTOM);
 
+  // If we are reading the value of the referent field of a Reference
+  // object (either by using Unsafe directly or through reflection)
+  // then, if G1 is enabled, we need to record the referent in an
+  // SATB log buffer using the pre-barrier mechanism.
+  bool need_read_barrier = UseG1GC && !is_native_ptr && !is_store &&
+                           offset != top() && heap_base_oop != top();
+
   if (!is_store && type == T_OBJECT) {
     // Attempt to infer a sharper value type from the offset and base type.
     ciKlass* sharpened_klass = NULL;
@@ -2278,8 +2421,13 @@ bool LibraryCallKit::inline_unsafe_access(bool is_native_ptr, bool is_store, Bas
     case T_SHORT:
     case T_INT:
     case T_FLOAT:
+      push(p);
+      break;
     case T_OBJECT:
-      push( p );
+      if (need_read_barrier) {
+        insert_g1_pre_barrier(heap_base_oop, offset, p);
+      }
+      push(p);
       break;
     case T_ADDRESS:
       // Cast to an int type.
@@ -2534,7 +2682,10 @@ bool LibraryCallKit::inline_unsafe_CAS(BasicType type) {
   case T_OBJECT:
      // reference stores need a store barrier.
     // (They don't if CAS fails, but it isn't worth checking.)
-    pre_barrier(control(), base, adr, alias_idx, newval, value_type->make_oopptr(), T_OBJECT);
+    pre_barrier(true /* do_load*/,
+                control(), base, adr, alias_idx, newval, value_type->make_oopptr(),
+                NULL /* pre_val*/,
+                T_OBJECT);
 #ifdef _LP64
     if (adr->bottom_type()->is_ptr_to_narrowoop()) {
       Node *newval_enc = _gvn.transform(new (C, 2) EncodePNode(newval, newval->bottom_type()->make_narrowoop()));
@@ -5235,3 +5386,44 @@ LibraryCallKit::generate_unchecked_arraycopy(const TypePtr* adr_type,
                     copyfunc_addr, copyfunc_name, adr_type,
                     src_start, dest_start, copy_length XTOP);
 }
+
+//----------------------------inline_reference_get----------------------------
+
+bool LibraryCallKit::inline_reference_get() {
+  const int nargs = 1; // self
+
+  guarantee(java_lang_ref_Reference::referent_offset > 0,
+            "should have already been set");
+
+  int referent_offset = java_lang_ref_Reference::referent_offset;
+
+  // Restore the stack and pop off the argument
+  _sp += nargs;
+  Node *reference_obj = pop();
+
+  // Null check on self without removing any arguments.
+  _sp += nargs;
+  reference_obj = do_null_check(reference_obj, T_OBJECT);
+  _sp -= nargs;;
+
+  if (stopped()) return true;
+
+  Node *adr = basic_plus_adr(reference_obj, reference_obj, referent_offset);
+
+  ciInstanceKlass* klass = env()->Object_klass();
+  const TypeOopPtr* object_type = TypeOopPtr::make_from_klass(klass);
+
+  Node* no_ctrl = NULL;
+  Node *result = make_load(no_ctrl, adr, object_type, T_OBJECT);
+
+  // Use the pre-barrier to record the value in the referent field
+  pre_barrier(false /* do_load */,
+              control(),
+              NULL /* obj */, NULL /* adr */, -1 /* alias_idx */, NULL /* val */, NULL /* val_type */,
+              result /* pre_val */,
+              T_OBJECT);
+
+  push(result);
+  return true;
+}
+
diff --git a/src/share/vm/opto/memnode.cpp b/src/share/vm/opto/memnode.cpp
index d5d53132a..7ba8df53e 100644
--- a/src/share/vm/opto/memnode.cpp
+++ b/src/share/vm/opto/memnode.cpp
@@ -2159,9 +2159,12 @@ Node *StoreNode::Ideal(PhaseGVN *phase, bool can_reshape) {
   Node* mem     = in(MemNode::Memory);
   Node* address = in(MemNode::Address);
 
-  // Back-to-back stores to same address?  Fold em up.
-  // Generally unsafe if I have intervening uses...
-  if (mem->is_Store() && phase->eqv_uncast(mem->in(MemNode::Address), address)) {
+  // Back-to-back stores to same address?  Fold em up.  Generally
+  // unsafe if I have intervening uses...  Also disallowed for StoreCM
+  // since they must follow each StoreP operation.  Redundant StoreCMs
+  // are eliminated just before matching in final_graph_reshape.
+  if (mem->is_Store() && phase->eqv_uncast(mem->in(MemNode::Address), address) &&
+      mem->Opcode() != Op_StoreCM) {
     // Looking at a dead closed cycle of memory?
     assert(mem != mem->in(MemNode::Memory), "dead loop in StoreNode::Ideal");
 
diff --git a/src/share/vm/opto/output.cpp b/src/share/vm/opto/output.cpp
index 1205cc263..2c185d5b3 100644
--- a/src/share/vm/opto/output.cpp
+++ b/src/share/vm/opto/output.cpp
@@ -1354,15 +1354,20 @@ void Compile::Fill_buffer() {
         // Check that oop-store precedes the card-mark
         else if( mach->ideal_Opcode() == Op_StoreCM ) {
           uint storeCM_idx = j;
-          Node *oop_store = mach->in(mach->_cnt);  // First precedence edge
-          assert( oop_store != NULL, "storeCM expects a precedence edge");
-          uint i4;
-          for( i4 = 0; i4 < last_inst; ++i4 ) {
-            if( b->_nodes[i4] == oop_store ) break;
+          int count = 0;
+          for (uint prec = mach->req(); prec < mach->len(); prec++) {
+            Node *oop_store = mach->in(prec);  // Precedence edge
+            if (oop_store == NULL) continue;
+            count++;
+            uint i4;
+            for( i4 = 0; i4 < last_inst; ++i4 ) {
+              if( b->_nodes[i4] == oop_store ) break;
+            }
+            // Note: This test can provide a false failure if other precedence
+            // edges have been added to the storeCMNode.
+            assert( i4 == last_inst || i4 < storeCM_idx, "CM card-mark executes before oop-store");
           }
-          // Note: This test can provide a false failure if other precedence
-          // edges have been added to the storeCMNode.
-          assert( i4 == last_inst || i4 < storeCM_idx, "CM card-mark executes before oop-store");
+          assert(count > 0, "storeCM expects at least one precedence edge");
         }
 #endif
 
diff --git a/src/share/vm/prims/jni.cpp b/src/share/vm/prims/jni.cpp
index d77c7ce9e..56fa2e167 100644
--- a/src/share/vm/prims/jni.cpp
+++ b/src/share/vm/prims/jni.cpp
@@ -29,6 +29,9 @@
 #include "classfile/systemDictionary.hpp"
 #include "classfile/vmSymbols.hpp"
 #include "interpreter/linkResolver.hpp"
+#ifndef SERIALGC
+#include "gc_implementation/g1/g1SATBCardTableModRefBS.hpp"
+#endif // SERIALGC
 #include "memory/allocation.inline.hpp"
 #include "memory/gcLocker.inline.hpp"
 #include "memory/oopFactory.hpp"
@@ -1724,6 +1727,26 @@ JNI_ENTRY(jobject, jni_GetObjectField(JNIEnv *env, jobject obj, jfieldID fieldID
     o = JvmtiExport::jni_GetField_probe(thread, obj, o, k, fieldID, false);
   }
   jobject ret = JNIHandles::make_local(env, o->obj_field(offset));
+#ifndef SERIALGC
+  // If G1 is enabled and we are accessing the value of the referent
+  // field in a reference object then we need to register a non-null
+  // referent with the SATB barrier.
+  if (UseG1GC) {
+    bool needs_barrier = false;
+
+    if (ret != NULL &&
+        offset == java_lang_ref_Reference::referent_offset &&
+        instanceKlass::cast(k)->reference_type() != REF_NONE) {
+      assert(instanceKlass::cast(k)->is_subclass_of(SystemDictionary::Reference_klass()), "sanity");
+      needs_barrier = true;
+    }
+
+    if (needs_barrier) {
+      oop referent = JNIHandles::resolve(ret);
+      G1SATBCardTableModRefBS::enqueue(referent);
+    }
+  }
+#endif // SERIALGC
   DTRACE_PROBE1(hotspot_jni, GetObjectField__return, ret);
   return ret;
 JNI_END
diff --git a/src/share/vm/prims/unsafe.cpp b/src/share/vm/prims/unsafe.cpp
index 5458bf70e..e1a21f030 100644
--- a/src/share/vm/prims/unsafe.cpp
+++ b/src/share/vm/prims/unsafe.cpp
@@ -24,6 +24,9 @@
 
 #include "precompiled.hpp"
 #include "classfile/vmSymbols.hpp"
+#ifndef SERIALGC
+#include "gc_implementation/g1/g1SATBCardTableModRefBS.hpp"
+#endif // SERIALGC
 #include "memory/allocation.inline.hpp"
 #include "prims/jni.h"
 #include "prims/jvm.h"
@@ -193,7 +196,32 @@ UNSAFE_ENTRY(jobject, Unsafe_GetObject140(JNIEnv *env, jobject unsafe, jobject o
   UnsafeWrapper("Unsafe_GetObject");
   if (obj == NULL)  THROW_0(vmSymbols::java_lang_NullPointerException());
   GET_OOP_FIELD(obj, offset, v)
-  return JNIHandles::make_local(env, v);
+  jobject ret = JNIHandles::make_local(env, v);
+#ifndef SERIALGC
+  // We could be accessing the referent field in a reference
+  // object. If G1 is enabled then we need to register a non-null
+  // referent with the SATB barrier.
+  if (UseG1GC) {
+    bool needs_barrier = false;
+
+    if (ret != NULL) {
+      if (offset == java_lang_ref_Reference::referent_offset) {
+        oop o = JNIHandles::resolve_non_null(obj);
+        klassOop k = o->klass();
+        if (instanceKlass::cast(k)->reference_type() != REF_NONE) {
+          assert(instanceKlass::cast(k)->is_subclass_of(SystemDictionary::Reference_klass()), "sanity");
+          needs_barrier = true;
+        }
+      }
+    }
+
+    if (needs_barrier) {
+      oop referent = JNIHandles::resolve(ret);
+      G1SATBCardTableModRefBS::enqueue(referent);
+    }
+  }
+#endif // SERIALGC
+  return ret;
 UNSAFE_END
 
 UNSAFE_ENTRY(void, Unsafe_SetObject140(JNIEnv *env, jobject unsafe, jobject obj, jint offset, jobject x_h))
@@ -226,7 +254,32 @@ UNSAFE_END
 UNSAFE_ENTRY(jobject, Unsafe_GetObject(JNIEnv *env, jobject unsafe, jobject obj, jlong offset))
   UnsafeWrapper("Unsafe_GetObject");
   GET_OOP_FIELD(obj, offset, v)
-  return JNIHandles::make_local(env, v);
+  jobject ret = JNIHandles::make_local(env, v);
+#ifndef SERIALGC
+  // We could be accessing the referent field in a reference
+  // object. If G1 is enabled then we need to register non-null
+  // referent with the SATB barrier.
+  if (UseG1GC) {
+    bool needs_barrier = false;
+
+    if (ret != NULL) {
+      if (offset == java_lang_ref_Reference::referent_offset && obj != NULL) {
+        oop o = JNIHandles::resolve(obj);
+        klassOop k = o->klass();
+        if (instanceKlass::cast(k)->reference_type() != REF_NONE) {
+          assert(instanceKlass::cast(k)->is_subclass_of(SystemDictionary::Reference_klass()), "sanity");
+          needs_barrier = true;
+        }
+      }
+    }
+
+    if (needs_barrier) {
+      oop referent = JNIHandles::resolve(ret);
+      G1SATBCardTableModRefBS::enqueue(referent);
+    }
+  }
+#endif // SERIALGC
+  return ret;
 UNSAFE_END
 
 UNSAFE_ENTRY(void, Unsafe_SetObject(JNIEnv *env, jobject unsafe, jobject obj, jlong offset, jobject x_h))
diff --git a/src/share/vm/runtime/arguments.cpp b/src/share/vm/runtime/arguments.cpp
index fdf271c80..6cb615c4b 100644
--- a/src/share/vm/runtime/arguments.cpp
+++ b/src/share/vm/runtime/arguments.cpp
@@ -243,6 +243,12 @@ static ObsoleteFlag obsolete_jvm_flags[] = {
   { "MaxLiveObjectEvacuationRatio",
                            JDK_Version::jdk_update(6,24), JDK_Version::jdk(8) },
   { "ForceSharedSpaces",   JDK_Version::jdk_update(6,25), JDK_Version::jdk(8) },
+  { "UseParallelOldGCCompacting",
+                           JDK_Version::jdk_update(6,27), JDK_Version::jdk(8) },
+  { "UseParallelDensePrefixUpdate",
+                           JDK_Version::jdk_update(6,27), JDK_Version::jdk(8) },
+  { "UseParallelOldGCDensePrefix",
+                           JDK_Version::jdk_update(6,27), JDK_Version::jdk(8) },
   { "AllowTransitionalJSR292",       JDK_Version::jdk(7), JDK_Version::jdk(8) },
   { NULL, JDK_Version(0), JDK_Version(0) }
 };
@@ -800,26 +806,22 @@ bool Arguments::process_argument(const char* arg,
 
   JDK_Version since = JDK_Version();
 
-  if (parse_argument(arg, origin)) {
-    // do nothing
-  } else if (is_newly_obsolete(arg, &since)) {
-    enum { bufsize = 256 };
-    char buffer[bufsize];
-    since.to_string(buffer, bufsize);
-    jio_fprintf(defaultStream::error_stream(),
-      "Warning: The flag %s has been EOL'd as of %s and will"
-      " be ignored\n", arg, buffer);
-  } else {
-    if (!ignore_unrecognized) {
-      jio_fprintf(defaultStream::error_stream(),
-                  "Unrecognized VM option '%s'\n", arg);
-      // allow for commandline "commenting out" options like -XX:#+Verbose
-      if (strlen(arg) == 0 || arg[0] != '#') {
-        return false;
-      }
-    }
+  if (parse_argument(arg, origin) || ignore_unrecognized) {
+    return true;
   }
-  return true;
+
+  const char * const argname = *arg == '+' || *arg == '-' ? arg + 1 : arg;
+  if (is_newly_obsolete(arg, &since)) {
+    char version[256];
+    since.to_string(version, sizeof(version));
+    warning("ignoring option %s; support was removed in %s", argname, version);
+    return true;
+  }
+
+  jio_fprintf(defaultStream::error_stream(),
+              "Unrecognized VM option '%s'\n", argname);
+  // allow for commandline "commenting out" options like -XX:#+Verbose
+  return arg[0] == '#';
 }
 
 bool Arguments::process_settings_file(const char* file_name, bool should_exist, jboolean ignore_unrecognized) {
diff --git a/src/share/vm/runtime/globals.hpp b/src/share/vm/runtime/globals.hpp
index 96f0abbb0..c16f88660 100644
--- a/src/share/vm/runtime/globals.hpp
+++ b/src/share/vm/runtime/globals.hpp
@@ -1355,13 +1355,6 @@ class CommandLineFlags {
   product(bool, UseParallelOldGC, false,                                    \
           "Use the Parallel Old garbage collector")                         \
                                                                             \
-  product(bool, UseParallelOldGCCompacting, true,                           \
-          "In the Parallel Old garbage collector use parallel compaction")  \
-                                                                            \
-  product(bool, UseParallelDensePrefixUpdate, true,                         \
-          "In the Parallel Old garbage collector use parallel dense"        \
-          " prefix update")                                                 \
-                                                                            \
   product(uintx, HeapMaximumCompactionInterval, 20,                         \
           "How often should we maximally compact the heap (not allowing "   \
           "any dead space)")                                                \
@@ -1381,9 +1374,6 @@ class CommandLineFlags {
           "The standard deviation used by the par compact dead wood"        \
           "limiter (a number between 0-100).")                              \
                                                                             \
-  product(bool, UseParallelOldGCDensePrefix, true,                          \
-          "Use a dense prefix with the Parallel Old garbage collector")     \
-                                                                            \
   product(uintx, ParallelGCThreads, 0,                                      \
           "Number of parallel threads parallel gc will use")                \
                                                                             \