changeset 2050:9a5762f44859 jdk7-b128

Merge
author trims
date Tue, 01 Feb 2011 18:57:53 -0800
parents 907c1aed0f8c d535bf4c1235
children 6ecdca5709df
files .hgtags src/share/vm/gc_implementation/g1/concurrentZFThread.cpp src/share/vm/gc_implementation/g1/concurrentZFThread.hpp
diffstat 144 files changed, 4051 insertions(+), 3679 deletions(-) [+]
line wrap: on
line diff
--- a/.hgtags	Thu Jan 27 17:28:34 2011 -0800
+++ b/.hgtags	Tue Feb 01 18:57:53 2011 -0800
@@ -142,4 +142,5 @@
 0a8e0d4345b37b71ec49dda08ee03b68c4f1b592 hs20-b05
 e24ab3fa6aafad3efabbe7dba9918c5f461a20b1 jdk7-b125
 4c851c931d001a882cab809aaf3a55371b919244 jdk7-b126
+e24ab3fa6aafad3efabbe7dba9918c5f461a20b1 hs20-b06
 102466e70debc4b907afbd7624e34ddb1aafee9f jdk7-b127
--- a/make/hotspot_version	Thu Jan 27 17:28:34 2011 -0800
+++ b/make/hotspot_version	Tue Feb 01 18:57:53 2011 -0800
@@ -35,7 +35,7 @@
 
 HS_MAJOR_VER=20
 HS_MINOR_VER=0
-HS_BUILD_NUMBER=06
+HS_BUILD_NUMBER=07
 
 JDK_MAJOR_VER=1
 JDK_MINOR_VER=7
--- a/src/cpu/sparc/vm/assembler_sparc.cpp	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/cpu/sparc/vm/assembler_sparc.cpp	Tue Feb 01 18:57:53 2011 -0800
@@ -4104,7 +4104,7 @@
 
   ld_ptr(G2_thread, in_bytes(JavaThread::tlab_start_offset()), t1);
   sub(top, t1, t1); // size of tlab's allocated portion
-  incr_allocated_bytes(t1, 0, t2);
+  incr_allocated_bytes(t1, t2, t3);
 
   // refill the tlab with an eden allocation
   bind(do_refill);
@@ -4138,19 +4138,14 @@
   delayed()->nop();
 }
 
-void MacroAssembler::incr_allocated_bytes(Register var_size_in_bytes,
-                                          int con_size_in_bytes,
-                                          Register t1) {
+void MacroAssembler::incr_allocated_bytes(RegisterOrConstant size_in_bytes,
+                                          Register t1, Register t2) {
   // Bump total bytes allocated by this thread
   assert(t1->is_global(), "must be global reg"); // so all 64 bits are saved on a context switch
-  assert_different_registers(var_size_in_bytes, t1);
+  assert_different_registers(size_in_bytes.register_or_noreg(), t1, t2);
   // v8 support has gone the way of the dodo
   ldx(G2_thread, in_bytes(JavaThread::allocated_bytes_offset()), t1);
-  if (var_size_in_bytes->is_valid()) {
-    add(t1, var_size_in_bytes, t1);
-  } else {
-    add(t1, con_size_in_bytes, t1);
-  }
+  add(t1, ensure_simm13_or_reg(size_in_bytes, t2), t1);
   stx(t1, G2_thread, in_bytes(JavaThread::allocated_bytes_offset()));
 }
 
--- a/src/cpu/sparc/vm/assembler_sparc.hpp	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/cpu/sparc/vm/assembler_sparc.hpp	Tue Feb 01 18:57:53 2011 -0800
@@ -823,15 +823,23 @@
   };
 
   // test if x is within signed immediate range for nbits
-  static bool is_simm(int x, int nbits) { return -( 1 << nbits-1 )  <= x   &&   x  <  ( 1 << nbits-1 ); }
+  static bool is_simm(intptr_t x, int nbits) { return -( intptr_t(1) << nbits-1 )  <= x   &&   x  <  ( intptr_t(1) << nbits-1 ); }
 
   // test if -4096 <= x <= 4095
-  static bool is_simm13(int x) { return is_simm(x, 13); }
+  static bool is_simm13(intptr_t x) { return is_simm(x, 13); }
+
+  static bool is_in_wdisp_range(address a, address b, int nbits) {
+    intptr_t d = intptr_t(b) - intptr_t(a);
+    return is_simm(d, nbits + 2);
+  }
 
   // test if label is in simm16 range in words (wdisp16).
   bool is_in_wdisp16_range(Label& L) {
-    intptr_t d = intptr_t(pc()) - intptr_t(target(L));
-    return is_simm(d, 18);
+    return is_in_wdisp_range(target(L), pc(), 16);
+  }
+  // test if the distance between two addresses fits in simm30 range in words
+  static bool is_in_wdisp30_range(address a, address b) {
+    return is_in_wdisp_range(a, b, 30);
   }
 
   enum ASIs { // page 72, v9
@@ -1843,6 +1851,8 @@
   inline void jmp( Register s1, Register s2 );
   inline void jmp( Register s1, int simm13a, RelocationHolder const& rspec = RelocationHolder() );
 
+  // Check if the call target is out of wdisp30 range (relative to the code cache)
+  static inline bool is_far_target(address d);
   inline void call( address d,  relocInfo::relocType rt = relocInfo::runtime_call_type );
   inline void call( Label& L,   relocInfo::relocType rt = relocInfo::runtime_call_type );
   inline void callr( Register s1, Register s2 );
@@ -2389,7 +2399,8 @@
     Label&   slow_case                 // continuation point if fast allocation fails
   );
   void tlab_refill(Label& retry_tlab, Label& try_eden, Label& slow_case);
-  void incr_allocated_bytes(Register var_size_in_bytes, int con_size_in_bytes, Register t1);
+  void incr_allocated_bytes(RegisterOrConstant size_in_bytes,
+                            Register t1, Register t2);
 
   // interface method calling
   void lookup_interface_method(Register recv_klass,
--- a/src/cpu/sparc/vm/assembler_sparc.inline.hpp	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/cpu/sparc/vm/assembler_sparc.inline.hpp	Tue Feb 01 18:57:53 2011 -0800
@@ -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
@@ -588,10 +588,13 @@
 inline void MacroAssembler::jmp( Register s1, Register s2 ) { jmpl( s1, s2, G0 ); }
 inline void MacroAssembler::jmp( Register s1, int simm13a, RelocationHolder const& rspec ) { jmpl( s1, simm13a, G0, rspec); }
 
+inline bool MacroAssembler::is_far_target(address d) {
+  return !is_in_wdisp30_range(d, CodeCache::low_bound()) || !is_in_wdisp30_range(d, CodeCache::high_bound());
+}
+
 // Call with a check to see if we need to deal with the added
 // expense of relocation and if we overflow the displacement
-// of the quick call instruction./
-// Check to see if we have to deal with relocations
+// of the quick call instruction.
 inline void MacroAssembler::call( address d, relocInfo::relocType rt ) {
 #ifdef _LP64
   intptr_t disp;
@@ -603,14 +606,12 @@
 
   // Is this address within range of the call instruction?
   // If not, use the expensive instruction sequence
-  disp = (intptr_t)d - (intptr_t)pc();
-  if ( disp != (intptr_t)(int32_t)disp ) {
+  if (is_far_target(d)) {
     relocate(rt);
     AddressLiteral dest(d);
     jumpl_to(dest, O7, O7);
-  }
-  else {
-    Assembler::call( d, rt );
+  } else {
+    Assembler::call(d, rt);
   }
 #else
   Assembler::call( d, rt );
--- a/src/cpu/sparc/vm/c1_LIRAssembler_sparc.cpp	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/cpu/sparc/vm/c1_LIRAssembler_sparc.cpp	Tue Feb 01 18:57:53 2011 -0800
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2000, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2000, 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
@@ -2358,6 +2358,8 @@
          op->tmp3()->as_register()  == G4 &&
          op->tmp4()->as_register()  == O1 &&
          op->klass()->as_register() == G5, "must be");
+
+  LP64_ONLY( __ signx(op->len()->as_register()); )
   if (UseSlowPath ||
       (!UseFastNewObjectArray && (op->type() == T_OBJECT || op->type() == T_ARRAY)) ||
       (!UseFastNewTypeArray   && (op->type() != T_OBJECT && op->type() != T_ARRAY))) {
--- a/src/cpu/sparc/vm/c1_MacroAssembler_sparc.cpp	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/cpu/sparc/vm/c1_MacroAssembler_sparc.cpp	Tue Feb 01 18:57:53 2011 -0800
@@ -170,11 +170,13 @@
   Register t2,                         // temp register
   Label&   slow_case                   // continuation point if fast allocation fails
 ) {
+  RegisterOrConstant size_in_bytes = var_size_in_bytes->is_valid()
+    ? RegisterOrConstant(var_size_in_bytes) : RegisterOrConstant(con_size_in_bytes);
   if (UseTLAB) {
     tlab_allocate(obj, var_size_in_bytes, con_size_in_bytes, t1, slow_case);
   } else {
     eden_allocate(obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
-    incr_allocated_bytes(var_size_in_bytes, con_size_in_bytes, t1);
+    incr_allocated_bytes(size_in_bytes, t1, t2);
   }
 }
 
--- a/src/cpu/sparc/vm/c1_Runtime1_sparc.cpp	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/cpu/sparc/vm/c1_Runtime1_sparc.cpp	Tue Feb 01 18:57:53 2011 -0800
@@ -343,9 +343,10 @@
   // returned.
 
   restore_live_registers(sasm);
-  __ restore();
-  __ br(Assembler::always, false, Assembler::pt, deopt_blob->unpack_with_reexecution(), relocInfo::runtime_call_type);
-  __ delayed()->nop();
+
+  AddressLiteral dest(deopt_blob->unpack_with_reexecution());
+  __ jump_to(dest, O0);
+  __ delayed()->restore();
 
   __ bind(no_deopt);
   restore_live_registers(sasm);
@@ -461,7 +462,7 @@
           // get the instance size
           __ ld(G5_klass, klassOopDesc::header_size() * HeapWordSize + Klass::layout_helper_offset_in_bytes(), G1_obj_size);
           __ eden_allocate(O0_obj, G1_obj_size, 0, G3_t1, G4_t2, slow_path);
-          __ incr_allocated_bytes(G1_obj_size, 0, G3_t1);
+          __ incr_allocated_bytes(G1_obj_size, G3_t1, G4_t2);
 
           __ initialize_object(O0_obj, G5_klass, G1_obj_size, 0, G3_t1, G4_t2);
           __ verify_oop(O0_obj);
@@ -577,7 +578,7 @@
           __ and3(G1_arr_size, ~MinObjAlignmentInBytesMask, G1_arr_size);
 
           __ eden_allocate(O0_obj, G1_arr_size, 0, G3_t1, O1_t2, slow_path);  // preserves G1_arr_size
-          __ incr_allocated_bytes(G1_arr_size, 0, G3_t1);
+          __ incr_allocated_bytes(G1_arr_size, G3_t1, O1_t2);
 
           __ initialize_header(O0_obj, G5_klass, G4_length, G3_t1, O1_t2);
           __ ldub(klass_lh, G3_t1, klass_lh_header_size_offset);
--- a/src/cpu/sparc/vm/interp_masm_sparc.cpp	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/cpu/sparc/vm/interp_masm_sparc.cpp	Tue Feb 01 18:57:53 2011 -0800
@@ -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
@@ -1295,16 +1295,13 @@
 // Get the method data pointer from the methodOop and set the
 // specified register to its value.
 
-void InterpreterMacroAssembler::set_method_data_pointer_offset(Register Roff) {
+void InterpreterMacroAssembler::set_method_data_pointer() {
   assert(ProfileInterpreter, "must be profiling interpreter");
   Label get_continue;
 
   ld_ptr(Lmethod, in_bytes(methodOopDesc::method_data_offset()), ImethodDataPtr);
   test_method_data_pointer(get_continue);
   add(ImethodDataPtr, in_bytes(methodDataOopDesc::data_offset()), ImethodDataPtr);
-  if (Roff != noreg)
-    // Roff contains a method data index ("mdi").  It defaults to zero.
-    add(ImethodDataPtr, Roff, ImethodDataPtr);
   bind(get_continue);
 }
 
@@ -1315,10 +1312,11 @@
   Label zero_continue;
 
   // Test MDO to avoid the call if it is NULL.
-  ld_ptr(Lmethod, methodOopDesc::method_data_offset(), ImethodDataPtr);
+  ld_ptr(Lmethod, in_bytes(methodOopDesc::method_data_offset()), ImethodDataPtr);
   test_method_data_pointer(zero_continue);
   call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), Lmethod, Lbcp);
-  set_method_data_pointer_offset(O0);
+  add(ImethodDataPtr, in_bytes(methodDataOopDesc::data_offset()), ImethodDataPtr);
+  add(ImethodDataPtr, O0, ImethodDataPtr);
   bind(zero_continue);
 }
 
@@ -1369,7 +1367,6 @@
 }
 
 void InterpreterMacroAssembler::test_invocation_counter_for_mdp(Register invocation_count,
-                                                                Register cur_bcp,
                                                                 Register Rtmp,
                                                                 Label &profile_continue) {
   assert(ProfileInterpreter, "must be profiling interpreter");
@@ -1400,8 +1397,8 @@
   delayed()->nop();
 
   // Build it now.
-  call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method), cur_bcp);
-  set_method_data_pointer_offset(O0);
+  call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
+  set_method_data_pointer_for_bcp();
   ba(false, profile_continue);
   delayed()->nop();
   bind(done);
--- a/src/cpu/sparc/vm/interp_masm_sparc.hpp	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/cpu/sparc/vm/interp_masm_sparc.hpp	Tue Feb 01 18:57:53 2011 -0800
@@ -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
@@ -269,12 +269,11 @@
 
 #ifndef CC_INTERP
   // Interpreter profiling operations
-  void set_method_data_pointer() { set_method_data_pointer_offset(noreg); }
+  void set_method_data_pointer();
   void set_method_data_pointer_for_bcp();
-  void set_method_data_pointer_offset(Register mdi_reg);
   void test_method_data_pointer(Label& zero_continue);
   void verify_method_data_pointer();
-  void test_invocation_counter_for_mdp(Register invocation_count, Register cur_bcp, Register Rtmp, Label &profile_continue);
+  void test_invocation_counter_for_mdp(Register invocation_count, Register Rtmp, Label &profile_continue);
 
   void set_mdp_data_at(int constant, Register value);
   void increment_mdp_data_at(Address counter, Register bumped_count,
--- a/src/cpu/sparc/vm/methodHandles_sparc.cpp	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/cpu/sparc/vm/methodHandles_sparc.cpp	Tue Feb 01 18:57:53 2011 -0800
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2008, 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
@@ -395,7 +395,7 @@
 //
 // Generate an "entry" field for a method handle.
 // This determines how the method handle will respond to calls.
-void MethodHandles::generate_method_handle_stub(MacroAssembler* _masm, MethodHandles::EntryKind ek, TRAPS) {
+void MethodHandles::generate_method_handle_stub(MacroAssembler* _masm, MethodHandles::EntryKind ek) {
   // Here is the register state during an interpreted call,
   // as set up by generate_method_handle_interpreter_entry():
   // - G5: garbage temp (was MethodHandle.invoke methodOop, unused)
@@ -447,8 +447,9 @@
       // exception.  Since we use a C2I adapter to set up the
       // interpreter state, arguments are expected in compiler
       // argument registers.
-      methodHandle mh(raise_exception_method());
-      address c2i_entry = methodOopDesc::make_adapters(mh, CATCH);
+      assert(raise_exception_method(), "must be set");
+      address c2i_entry = raise_exception_method()->get_c2i_entry();
+      assert(c2i_entry, "method must be linked");
 
       __ mov(O5_savedSP, SP);  // Cut the stack back to where the caller started.
 
--- a/src/cpu/sparc/vm/sparc.ad	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/cpu/sparc/vm/sparc.ad	Tue Feb 01 18:57:53 2011 -0800
@@ -1,5 +1,5 @@
 //
-// Copyright (c) 1998, 2010, Oracle and/or its affiliates. All rights reserved.
+// Copyright (c) 1998, 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
@@ -575,7 +575,11 @@
 
 int MachCallRuntimeNode::ret_addr_offset() {
 #ifdef _LP64
-  return NativeFarCall::instruction_size;  // farcall; delay slot
+  if (MacroAssembler::is_far_target(entry_point())) {
+    return NativeFarCall::instruction_size;
+  } else {
+    return NativeCall::instruction_size;
+  }
 #else
   return NativeCall::instruction_size;  // call; delay slot
 #endif
@@ -941,7 +945,7 @@
 #endif
 }
 
-void emit_call_reloc(CodeBuffer &cbuf, intptr_t entry_point, relocInfo::relocType rtype, bool preserve_g2 = false, bool force_far_call = false) {
+void emit_call_reloc(CodeBuffer &cbuf, intptr_t entry_point, relocInfo::relocType rtype, bool preserve_g2 = false) {
   // The method which records debug information at every safepoint
   // expects the call to be the first instruction in the snippet as
   // it creates a PcDesc structure which tracks the offset of a call
@@ -963,20 +967,7 @@
   int startpos = __ offset();
 #endif /* ASSERT */
 
-#ifdef _LP64
-  // Calls to the runtime or native may not be reachable from compiled code,
-  // so we generate the far call sequence on 64 bit sparc.
-  // This code sequence is relocatable to any address, even on LP64.
-  if ( force_far_call ) {
-    __ relocate(rtype);
-    AddressLiteral dest(entry_point);
-    __ jumpl_to(dest, O7, O7);
-  }
-  else
-#endif
-  {
-     __ call((address)entry_point, rtype);
-  }
+  __ call((address)entry_point, rtype);
 
   if (preserve_g2)   __ delayed()->mov(G2, L7);
   else __ delayed()->nop();
@@ -2507,7 +2498,7 @@
     // CALL directly to the runtime
     // The user of this is responsible for ensuring that R_L7 is empty (killed).
     emit_call_reloc(cbuf, $meth$$method, relocInfo::runtime_call_type,
-                    /*preserve_g2=*/true, /*force far call*/true);
+                    /*preserve_g2=*/true);
   %}
 
   enc_class preserve_SP %{
--- a/src/cpu/sparc/vm/templateInterpreter_sparc.cpp	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/cpu/sparc/vm/templateInterpreter_sparc.cpp	Tue Feb 01 18:57:53 2011 -0800
@@ -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
@@ -1364,15 +1364,8 @@
       // We have decided to profile this method in the interpreter
       __ bind(profile_method);
 
-      __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method), Lbcp, true);
-
-#ifdef ASSERT
-      __ tst(O0);
-      __ breakpoint_trap(Assembler::notEqual);
-#endif
-
-      __ set_method_data_pointer();
-
+      __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
+      __ set_method_data_pointer_for_bcp();
       __ ba(false, profile_method_continue);
       __ delayed()->nop();
     }
--- a/src/cpu/sparc/vm/templateTable_sparc.cpp	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/cpu/sparc/vm/templateTable_sparc.cpp	Tue Feb 01 18:57:53 2011 -0800
@@ -1689,7 +1689,7 @@
       const Register G4_invoke_ctr = G4;
       __ increment_backedge_counter(G4_invoke_ctr, G1_scratch);
       if (ProfileInterpreter) {
-        __ test_invocation_counter_for_mdp(G4_invoke_ctr, Lbcp, G3_scratch, Lforward);
+        __ test_invocation_counter_for_mdp(G4_invoke_ctr, G3_scratch, Lforward);
         if (UseOnStackReplacement) {
           __ test_backedge_count_for_osr(O2_bumped_count, O0_cur_bcp, G3_scratch);
         }
@@ -3447,7 +3447,8 @@
     __ delayed()->nop();
 
     // bump total bytes allocated by this thread
-    __ incr_allocated_bytes(Roffset, 0, G1_scratch);
+    // RoldTopValue and RtopAddr are dead, so can use G1 and G3
+    __ incr_allocated_bytes(Roffset, G1_scratch, G3_scratch);
   }
 
   if (UseTLAB || Universe::heap()->supports_inline_contig_alloc()) {
--- a/src/cpu/x86/vm/c1_LIRAssembler_x86.cpp	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/cpu/x86/vm/c1_LIRAssembler_x86.cpp	Tue Feb 01 18:57:53 2011 -0800
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2000, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2000, 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
@@ -1641,12 +1641,14 @@
 }
 
 void LIR_Assembler::emit_alloc_array(LIR_OpAllocArray* op) {
+  Register len =  op->len()->as_register();
+  LP64_ONLY( __ movslq(len, len); )
+
   if (UseSlowPath ||
       (!UseFastNewObjectArray && (op->type() == T_OBJECT || op->type() == T_ARRAY)) ||
       (!UseFastNewTypeArray   && (op->type() != T_OBJECT && op->type() != T_ARRAY))) {
     __ jmp(*op->stub()->entry());
   } else {
-    Register len =  op->len()->as_register();
     Register tmp1 = op->tmp1()->as_register();
     Register tmp2 = op->tmp2()->as_register();
     Register tmp3 = op->tmp3()->as_register();
--- a/src/cpu/x86/vm/globals_x86.hpp	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/cpu/x86/vm/globals_x86.hpp	Tue Feb 01 18:57:53 2011 -0800
@@ -62,7 +62,7 @@
 // due to lack of optimization caused by C++ compiler bugs
 define_pd_global(intx, StackShadowPages, SOLARIS_ONLY(20) NOT_SOLARIS(6) DEBUG_ONLY(+2));
 #else
-define_pd_global(intx, StackShadowPages, 3 DEBUG_ONLY(+1));
+define_pd_global(intx, StackShadowPages, 3 DEBUG_ONLY(+5));
 #endif // AMD64
 
 define_pd_global(intx, PreInflateSpin,           10);
--- a/src/cpu/x86/vm/interp_masm_x86_32.cpp	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/cpu/x86/vm/interp_masm_x86_32.cpp	Tue Feb 01 18:57:53 2011 -0800
@@ -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
@@ -819,7 +819,7 @@
 // Set the method data pointer for the current bcp.
 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
   assert(ProfileInterpreter, "must be profiling interpreter");
-  Label zero_continue;
+  Label set_mdp;
   push(rax);
   push(rbx);
 
@@ -827,21 +827,17 @@
   // Test MDO to avoid the call if it is NULL.
   movptr(rax, Address(rbx, in_bytes(methodOopDesc::method_data_offset())));
   testptr(rax, rax);
-  jcc(Assembler::zero, zero_continue);
-
+  jcc(Assembler::zero, set_mdp);
   // rbx,: method
   // rsi: bcp
   call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), rbx, rsi);
   // rax,: mdi
-
+  // mdo is guaranteed to be non-zero here, we checked for it before the call.
   movptr(rbx, Address(rbx, in_bytes(methodOopDesc::method_data_offset())));
-  testptr(rbx, rbx);
-  jcc(Assembler::zero, zero_continue);
   addptr(rbx, in_bytes(methodDataOopDesc::data_offset()));
-  addptr(rbx, rax);
-  movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), rbx);
-
-  bind(zero_continue);
+  addptr(rax, rbx);
+  bind(set_mdp);
+  movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), rax);
   pop(rbx);
   pop(rax);
 }
--- a/src/cpu/x86/vm/interp_masm_x86_64.cpp	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/cpu/x86/vm/interp_masm_x86_64.cpp	Tue Feb 01 18:57:53 2011 -0800
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2003, 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
@@ -855,7 +855,7 @@
 // Set the method data pointer for the current bcp.
 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
   assert(ProfileInterpreter, "must be profiling interpreter");
-  Label zero_continue;
+  Label set_mdp;
   push(rax);
   push(rbx);
 
@@ -863,21 +863,17 @@
   // Test MDO to avoid the call if it is NULL.
   movptr(rax, Address(rbx, in_bytes(methodOopDesc::method_data_offset())));
   testptr(rax, rax);
-  jcc(Assembler::zero, zero_continue);
-
+  jcc(Assembler::zero, set_mdp);
   // rbx: method
   // r13: bcp
   call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), rbx, r13);
   // rax: mdi
-
+  // mdo is guaranteed to be non-zero here, we checked for it before the call.
   movptr(rbx, Address(rbx, in_bytes(methodOopDesc::method_data_offset())));
-  testptr(rbx, rbx);
-  jcc(Assembler::zero, zero_continue);
   addptr(rbx, in_bytes(methodDataOopDesc::data_offset()));
-  addptr(rbx, rax);
-  movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), rbx);
-
-  bind(zero_continue);
+  addptr(rax, rbx);
+  bind(set_mdp);
+  movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), rax);
   pop(rbx);
   pop(rax);
 }
--- a/src/cpu/x86/vm/methodHandles_x86.cpp	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/cpu/x86/vm/methodHandles_x86.cpp	Tue Feb 01 18:57:53 2011 -0800
@@ -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
@@ -390,7 +390,7 @@
 //
 // Generate an "entry" field for a method handle.
 // This determines how the method handle will respond to calls.
-void MethodHandles::generate_method_handle_stub(MacroAssembler* _masm, MethodHandles::EntryKind ek, TRAPS) {
+void MethodHandles::generate_method_handle_stub(MacroAssembler* _masm, MethodHandles::EntryKind ek) {
   // Here is the register state during an interpreted call,
   // as set up by generate_method_handle_interpreter_entry():
   // - rbx: garbage temp (was MethodHandle.invoke methodOop, unused)
@@ -451,8 +451,9 @@
       // exception.  Since we use a C2I adapter to set up the
       // interpreter state, arguments are expected in compiler
       // argument registers.
-      methodHandle mh(raise_exception_method());
-      address c2i_entry = methodOopDesc::make_adapters(mh, CHECK);
+      assert(raise_exception_method(), "must be set");
+      address c2i_entry = raise_exception_method()->get_c2i_entry();
+      assert(c2i_entry, "method must be linked");
 
       const Register rdi_pc = rax;
       __ pop(rdi_pc);  // caller PC
--- a/src/cpu/x86/vm/templateInterpreter_x86_32.cpp	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/cpu/x86/vm/templateInterpreter_x86_32.cpp	Tue Feb 01 18:57:53 2011 -0800
@@ -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
@@ -1367,15 +1367,9 @@
     if (ProfileInterpreter) {
       // We have decided to profile this method in the interpreter
       __ bind(profile_method);
-
-      __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method), rsi, true);
-
-      __ movptr(rbx, Address(rbp, method_offset));   // restore methodOop
-      __ movptr(rax, Address(rbx, in_bytes(methodOopDesc::method_data_offset())));
-      __ movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), rax);
-      __ test_method_data_pointer(rax, profile_method_continue);
-      __ addptr(rax, in_bytes(methodDataOopDesc::data_offset()));
-      __ movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), rax);
+      __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
+      __ set_method_data_pointer_for_bcp();
+      __ get_method(rbx);
       __ jmp(profile_method_continue);
     }
     // Handle overflow of counter and compile method
--- a/src/cpu/x86/vm/templateInterpreter_x86_64.cpp	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/cpu/x86/vm/templateInterpreter_x86_64.cpp	Tue Feb 01 18:57:53 2011 -0800
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2003, 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
@@ -1383,20 +1383,9 @@
     if (ProfileInterpreter) {
       // We have decided to profile this method in the interpreter
       __ bind(profile_method);
-
-      __ call_VM(noreg,
-                 CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method),
-                 r13, true);
-
-      __ movptr(rbx, Address(rbp, method_offset)); // restore methodOop
-      __ movptr(rax, Address(rbx,
-                             in_bytes(methodOopDesc::method_data_offset())));
-      __ movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize),
-                rax);
-      __ test_method_data_pointer(rax, profile_method_continue);
-      __ addptr(rax, in_bytes(methodDataOopDesc::data_offset()));
-      __ movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize),
-              rax);
+      __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
+      __ set_method_data_pointer_for_bcp();
+      __ get_method(rbx);
       __ jmp(profile_method_continue);
     }
     // Handle overflow of counter and compile method
--- a/src/cpu/x86/vm/templateTable_x86_32.cpp	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/cpu/x86/vm/templateTable_x86_32.cpp	Tue Feb 01 18:57:53 2011 -0800
@@ -1665,16 +1665,9 @@
     if (ProfileInterpreter) {
       // Out-of-line code to allocate method data oop.
       __ bind(profile_method);
-      __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method), rsi);
+      __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
       __ load_unsigned_byte(rbx, Address(rsi, 0));  // restore target bytecode
-      __ movptr(rcx, Address(rbp, method_offset));
-      __ movptr(rcx, Address(rcx, in_bytes(methodOopDesc::method_data_offset())));
-      __ movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), rcx);
-      __ test_method_data_pointer(rcx, dispatch);
-      // offset non-null mdp by MDO::data_offset() + IR::profile_method()
-      __ addptr(rcx, in_bytes(methodDataOopDesc::data_offset()));
-      __ addptr(rcx, rax);
-      __ movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), rcx);
+      __ set_method_data_pointer_for_bcp();
       __ jmp(dispatch);
     }
 
--- a/src/cpu/x86/vm/templateTable_x86_64.cpp	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/cpu/x86/vm/templateTable_x86_64.cpp	Tue Feb 01 18:57:53 2011 -0800
@@ -1695,21 +1695,9 @@
     if (ProfileInterpreter) {
       // Out-of-line code to allocate method data oop.
       __ bind(profile_method);
-      __ call_VM(noreg,
-                 CAST_FROM_FN_PTR(address,
-                                  InterpreterRuntime::profile_method), r13);
+      __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
       __ load_unsigned_byte(rbx, Address(r13, 0));  // restore target bytecode
-      __ movptr(rcx, Address(rbp, method_offset));
-      __ movptr(rcx, Address(rcx,
-                             in_bytes(methodOopDesc::method_data_offset())));
-      __ movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize),
-                rcx);
-      __ test_method_data_pointer(rcx, dispatch);
-      // offset non-null mdp by MDO::data_offset() + IR::profile_method()
-      __ addptr(rcx, in_bytes(methodDataOopDesc::data_offset()));
-      __ addptr(rcx, rax);
-      __ movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize),
-                rcx);
+      __ set_method_data_pointer_for_bcp();
       __ jmp(dispatch);
     }
 
--- a/src/os/linux/vm/os_linux.cpp	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/os/linux/vm/os_linux.cpp	Tue Feb 01 18:57:53 2011 -0800
@@ -1610,10 +1610,9 @@
 
 const char* os::dll_file_extension() { return ".so"; }
 
-const char* os::get_temp_directory() {
-  const char *prop = Arguments::get_property("java.io.tmpdir");
-  return prop == NULL ? "/tmp" : prop;
-}
+// This must be hard coded because it's the system's temporary
+// directory not the java application's temp directory, ala java.io.tmpdir.
+const char* os::get_temp_directory() { return "/tmp"; }
 
 static bool file_exists(const char* filename) {
   struct stat statbuf;
--- a/src/os/solaris/vm/os_solaris.cpp	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/os/solaris/vm/os_solaris.cpp	Tue Feb 01 18:57:53 2011 -0800
@@ -1884,10 +1884,9 @@
 
 const char* os::dll_file_extension() { return ".so"; }
 
-const char* os::get_temp_directory() {
-  const char *prop = Arguments::get_property("java.io.tmpdir");
-  return prop == NULL ? "/tmp" : prop;
-}
+// This must be hard coded because it's the system's temporary
+// directory not the java application's temp directory, ala java.io.tmpdir.
+const char* os::get_temp_directory() { return "/tmp"; }
 
 static bool file_exists(const char* filename) {
   struct stat statbuf;
--- a/src/os/windows/vm/os_windows.cpp	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/os/windows/vm/os_windows.cpp	Tue Feb 01 18:57:53 2011 -0800
@@ -1044,9 +1044,9 @@
     return 0;
 }
 
+// This must be hard coded because it's the system's temporary
+// directory not the java application's temp directory, ala java.io.tmpdir.
 const char* os::get_temp_directory() {
-  const char *prop = Arguments::get_property("java.io.tmpdir");
-  if (prop != 0) return prop;
   static char path_buf[MAX_PATH];
   if (GetTempPath(MAX_PATH, path_buf)>0)
     return path_buf;
--- a/src/os_cpu/linux_sparc/vm/atomic_linux_sparc.inline.hpp	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/os_cpu/linux_sparc/vm/atomic_linux_sparc.inline.hpp	Tue Feb 01 18:57:53 2011 -0800
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 1999, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1999, 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
@@ -54,6 +54,8 @@
 inline void Atomic::dec_ptr(volatile intptr_t* dest) { (void)add_ptr(-1, dest); }
 inline void Atomic::dec_ptr(volatile void*     dest) { (void)add_ptr(-1, dest); }
 
+inline jlong Atomic::load(volatile jlong* src) { return *src; }
+
 inline jint     Atomic::add    (jint     add_value, volatile jint*     dest) {
   intptr_t rv;
   __asm__ volatile(
--- a/src/os_cpu/linux_x86/vm/atomic_linux_x86.inline.hpp	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/os_cpu/linux_x86/vm/atomic_linux_x86.inline.hpp	Tue Feb 01 18:57:53 2011 -0800
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 1999, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1999, 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
@@ -100,11 +100,6 @@
   return exchange_value;
 }
 
-extern "C" {
-  // defined in linux_x86.s
-  jlong _Atomic_cmpxchg_long(jlong, volatile jlong*, jlong, bool);
-}
-
 #ifdef AMD64
 inline void Atomic::store    (jlong    store_value, jlong*    dest) { *dest = store_value; }
 inline void Atomic::store    (jlong    store_value, volatile jlong*    dest) { *dest = store_value; }
@@ -164,9 +159,9 @@
   return (void*)cmpxchg((jlong)exchange_value, (volatile jlong*)dest, (jlong)compare_value);
 }
 
-#else
-//inline void Atomic::store    (jlong    store_value, jlong*    dest) { *dest = store_value; }
-//inline void Atomic::store  (jlong    store_value, volatile jlong*    dest) { *dest = store_value; }
+inline jlong Atomic::load(volatile jlong* src) { return *src; }
+
+#else // !AMD64
 
 inline intptr_t Atomic::add_ptr(intptr_t add_value, volatile intptr_t* dest) {
   return (intptr_t)Atomic::add((jint)add_value, (volatile jint*)dest);
@@ -189,6 +184,12 @@
   return (intptr_t)xchg((jint)exchange_value, (volatile jint*)dest);
 }
 
+extern "C" {
+  // defined in linux_x86.s
+  jlong _Atomic_cmpxchg_long(jlong, volatile jlong*, jlong, bool);
+  void _Atomic_move_long(volatile jlong* src, volatile jlong* dst);
+}
+
 inline jlong    Atomic::cmpxchg    (jlong    exchange_value, volatile jlong*    dest, jlong    compare_value) {
   return _Atomic_cmpxchg_long(exchange_value, dest, compare_value, os::is_MP());
 }
@@ -200,6 +201,21 @@
 inline void*    Atomic::cmpxchg_ptr(void*    exchange_value, volatile void*     dest, void*    compare_value) {
   return (void*)cmpxchg((jint)exchange_value, (volatile jint*)dest, (jint)compare_value);
 }
+
+inline jlong Atomic::load(volatile jlong* src) {
+  volatile jlong dest;
+  _Atomic_move_long(src, &dest);
+  return dest;
+}
+
+inline void Atomic::store(jlong store_value, jlong* dest) {
+  _Atomic_move_long((volatile jlong*)&store_value, (volatile jlong*)dest);
+}
+
+inline void Atomic::store(jlong store_value, volatile jlong* dest) {
+  _Atomic_move_long((volatile jlong*)&store_value, dest);
+}
+
 #endif // AMD64
 
 #endif // OS_CPU_LINUX_X86_VM_ATOMIC_LINUX_X86_INLINE_HPP
--- a/src/os_cpu/linux_x86/vm/linux_x86_32.s	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/os_cpu/linux_x86/vm/linux_x86_32.s	Tue Feb 01 18:57:53 2011 -0800
@@ -1,5 +1,5 @@
 # 
-# Copyright (c) 2004, 2010, Oracle and/or its affiliates. All rights reserved.
+# Copyright (c) 2004, 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
@@ -38,6 +38,7 @@
 	.globl _mmx_Copy_arrayof_conjoint_jshorts
 
         .globl _Atomic_cmpxchg_long
+        .globl _Atomic_move_long
 
 	.text
 
@@ -653,3 +654,15 @@
         popl     %ebx
         ret
 
+
+        # Support for jlong Atomic::load and Atomic::store.
+        # void _Atomic_move_long(volatile jlong* src, volatile jlong* dst)
+        .p2align 4,,15
+	.type    _Atomic_move_long,@function
+_Atomic_move_long:
+        movl     4(%esp), %eax   # src
+        fildll    (%eax)
+        movl     8(%esp), %eax   # dest
+        fistpll   (%eax)
+        ret
+
--- a/src/os_cpu/linux_x86/vm/orderAccess_linux_x86.inline.hpp	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/os_cpu/linux_x86/vm/orderAccess_linux_x86.inline.hpp	Tue Feb 01 18:57:53 2011 -0800
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2003, 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
@@ -25,6 +25,7 @@
 #ifndef OS_CPU_LINUX_X86_VM_ORDERACCESS_LINUX_X86_INLINE_HPP
 #define OS_CPU_LINUX_X86_VM_ORDERACCESS_LINUX_X86_INLINE_HPP
 
+#include "runtime/atomic.hpp"
 #include "runtime/orderAccess.hpp"
 #include "vm_version_x86.hpp"
 
@@ -64,11 +65,11 @@
 inline jbyte    OrderAccess::load_acquire(volatile jbyte*   p) { return *p; }
 inline jshort   OrderAccess::load_acquire(volatile jshort*  p) { return *p; }
 inline jint     OrderAccess::load_acquire(volatile jint*    p) { return *p; }
-inline jlong    OrderAccess::load_acquire(volatile jlong*   p) { return *p; }
+inline jlong    OrderAccess::load_acquire(volatile jlong*   p) { return Atomic::load(p); }
 inline jubyte   OrderAccess::load_acquire(volatile jubyte*  p) { return *p; }
 inline jushort  OrderAccess::load_acquire(volatile jushort* p) { return *p; }
 inline juint    OrderAccess::load_acquire(volatile juint*   p) { return *p; }
-inline julong   OrderAccess::load_acquire(volatile julong*  p) { return *p; }
+inline julong   OrderAccess::load_acquire(volatile julong*  p) { return Atomic::load((volatile jlong*)p); }
 inline jfloat   OrderAccess::load_acquire(volatile jfloat*  p) { return *p; }
 inline jdouble  OrderAccess::load_acquire(volatile jdouble* p) { return *p; }
 
@@ -79,11 +80,11 @@
 inline void     OrderAccess::release_store(volatile jbyte*   p, jbyte   v) { *p = v; }
 inline void     OrderAccess::release_store(volatile jshort*  p, jshort  v) { *p = v; }
 inline void     OrderAccess::release_store(volatile jint*    p, jint    v) { *p = v; }
-inline void     OrderAccess::release_store(volatile jlong*   p, jlong   v) { *p = v; }
+inline void     OrderAccess::release_store(volatile jlong*   p, jlong   v) { Atomic::store(v, p); }
 inline void     OrderAccess::release_store(volatile jubyte*  p, jubyte  v) { *p = v; }
 inline void     OrderAccess::release_store(volatile jushort* p, jushort v) { *p = v; }
 inline void     OrderAccess::release_store(volatile juint*   p, juint   v) { *p = v; }
-inline void     OrderAccess::release_store(volatile julong*  p, julong  v) { *p = v; }
+inline void     OrderAccess::release_store(volatile julong*  p, julong  v) { Atomic::store((jlong)v, (volatile jlong*)p); }
 inline void     OrderAccess::release_store(volatile jfloat*  p, jfloat  v) { *p = v; }
 inline void     OrderAccess::release_store(volatile jdouble* p, jdouble v) { *p = v; }
 
@@ -178,7 +179,7 @@
                           : "0" (v), "r" (p)
                           : "memory");
 #else
-  *p = v; fence();
+  release_store(p, v); fence();
 #endif // AMD64
 }
 
--- a/src/os_cpu/solaris_sparc/vm/atomic_solaris_sparc.inline.hpp	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/os_cpu/solaris_sparc/vm/atomic_solaris_sparc.inline.hpp	Tue Feb 01 18:57:53 2011 -0800
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 1999, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1999, 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
@@ -35,14 +35,12 @@
 inline void Atomic::store    (jbyte    store_value, jbyte*    dest) { *dest = store_value; }
 inline void Atomic::store    (jshort   store_value, jshort*   dest) { *dest = store_value; }
 inline void Atomic::store    (jint     store_value, jint*     dest) { *dest = store_value; }
-inline void Atomic::store    (jlong    store_value, jlong*    dest) { *dest = store_value; }
 inline void Atomic::store_ptr(intptr_t store_value, intptr_t* dest) { *dest = store_value; }
 inline void Atomic::store_ptr(void*    store_value, void*     dest) { *(void**)dest = store_value; }
 
 inline void Atomic::store    (jbyte    store_value, volatile jbyte*    dest) { *dest = store_value; }
 inline void Atomic::store    (jshort   store_value, volatile jshort*   dest) { *dest = store_value; }
 inline void Atomic::store    (jint     store_value, volatile jint*     dest) { *dest = store_value; }
-inline void Atomic::store    (jlong    store_value, volatile jlong*    dest) { *dest = store_value; }
 inline void Atomic::store_ptr(intptr_t store_value, volatile intptr_t* dest) { *dest = store_value; }
 inline void Atomic::store_ptr(void*    store_value, volatile void*     dest) { *(void* volatile *)dest = store_value; }
 
@@ -54,8 +52,49 @@
 inline void Atomic::dec_ptr(volatile intptr_t* dest) { (void)add_ptr(-1, dest); }
 inline void Atomic::dec_ptr(volatile void*     dest) { (void)add_ptr(-1, dest); }
 
+
+#ifdef _LP64
+
+inline void Atomic::store(jlong store_value, jlong* dest) { *dest = store_value; }
+inline void Atomic::store(jlong store_value, volatile jlong* dest) { *dest = store_value; }
 inline jlong Atomic::load(volatile jlong* src) { return *src; }
 
+#else
+
+extern "C" void _Atomic_move_long_v8(volatile jlong* src, volatile jlong* dst);
+extern "C" void _Atomic_move_long_v9(volatile jlong* src, volatile jlong* dst);
+
+inline void Atomic_move_long(volatile jlong* src, volatile jlong* dst) {
+#ifdef COMPILER2
+  // Compiler2 does not support v8, it is used only for v9.
+  assert (VM_Version::v9_instructions_work(), "only supported on v9");
+  _Atomic_move_long_v9(src, dst);
+#else
+  // The branch is cheaper then emulated LDD.
+  if (VM_Version::v9_instructions_work()) {
+    _Atomic_move_long_v9(src, dst);
+  } else {
+    _Atomic_move_long_v8(src, dst);
+  }
+#endif
+}
+
+inline jlong Atomic::load(volatile jlong* src) {
+  volatile jlong dest;
+  Atomic_move_long(src, &dest);
+  return dest;
+}
+
+inline void Atomic::store(jlong store_value, jlong* dest) {
+  Atomic_move_long((volatile jlong*)&store_value, (volatile jlong*)dest);
+}
+
+inline void Atomic::store(jlong store_value, volatile jlong* dest) {
+  Atomic_move_long((volatile jlong*)&store_value, dest);
+}
+
+#endif
+
 #ifdef _GNU_SOURCE
 
 inline jint     Atomic::add    (jint     add_value, volatile jint*     dest) {
--- a/src/os_cpu/solaris_sparc/vm/orderAccess_solaris_sparc.inline.hpp	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/os_cpu/solaris_sparc/vm/orderAccess_solaris_sparc.inline.hpp	Tue Feb 01 18:57:53 2011 -0800
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2003, 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
@@ -77,11 +77,11 @@
 inline jbyte    OrderAccess::load_acquire(volatile jbyte*   p) { return *p; }
 inline jshort   OrderAccess::load_acquire(volatile jshort*  p) { return *p; }
 inline jint     OrderAccess::load_acquire(volatile jint*    p) { return *p; }
-inline jlong    OrderAccess::load_acquire(volatile jlong*   p) { return *p; }
+inline jlong    OrderAccess::load_acquire(volatile jlong*   p) { return Atomic::load(p); }
 inline jubyte   OrderAccess::load_acquire(volatile jubyte*  p) { return *p; }
 inline jushort  OrderAccess::load_acquire(volatile jushort* p) { return *p; }
 inline juint    OrderAccess::load_acquire(volatile juint*   p) { return *p; }
-inline julong   OrderAccess::load_acquire(volatile julong*  p) { return *p; }
+inline julong   OrderAccess::load_acquire(volatile julong*  p) { return Atomic::load((volatile jlong*)p); }
 inline jfloat   OrderAccess::load_acquire(volatile jfloat*  p) { return *p; }
 inline jdouble  OrderAccess::load_acquire(volatile jdouble* p) { return *p; }
 
@@ -92,11 +92,11 @@
 inline void     OrderAccess::release_store(volatile jbyte*   p, jbyte   v) { *p = v; }
 inline void     OrderAccess::release_store(volatile jshort*  p, jshort  v) { *p = v; }
 inline void     OrderAccess::release_store(volatile jint*    p, jint    v) { *p = v; }
-inline void     OrderAccess::release_store(volatile jlong*   p, jlong   v) { *p = v; }
+inline void     OrderAccess::release_store(volatile jlong*   p, jlong   v) { Atomic::store(v, p); }
 inline void     OrderAccess::release_store(volatile jubyte*  p, jubyte  v) { *p = v; }
 inline void     OrderAccess::release_store(volatile jushort* p, jushort v) { *p = v; }
 inline void     OrderAccess::release_store(volatile juint*   p, juint   v) { *p = v; }
-inline void     OrderAccess::release_store(volatile julong*  p, julong  v) { *p = v; }
+inline void     OrderAccess::release_store(volatile julong*  p, julong  v) { Atomic::store((jlong)v, (volatile jlong*)p); }
 inline void     OrderAccess::release_store(volatile jfloat*  p, jfloat  v) { *p = v; }
 inline void     OrderAccess::release_store(volatile jdouble* p, jdouble v) { *p = v; }
 
@@ -120,11 +120,11 @@
 inline void     OrderAccess::release_store_fence(volatile jbyte*   p, jbyte   v) { *p = v; fence(); }
 inline void     OrderAccess::release_store_fence(volatile jshort*  p, jshort  v) { *p = v; fence(); }
 inline void     OrderAccess::release_store_fence(volatile jint*    p, jint    v) { *p = v; fence(); }
-inline void     OrderAccess::release_store_fence(volatile jlong*   p, jlong   v) { *p = v; fence(); }
+inline void     OrderAccess::release_store_fence(volatile jlong*   p, jlong   v) { release_store(p, v); fence(); }
 inline void     OrderAccess::release_store_fence(volatile jubyte*  p, jubyte  v) { *p = v; fence(); }
 inline void     OrderAccess::release_store_fence(volatile jushort* p, jushort v) { *p = v; fence(); }
 inline void     OrderAccess::release_store_fence(volatile juint*   p, juint   v) { *p = v; fence(); }
-inline void     OrderAccess::release_store_fence(volatile julong*  p, julong  v) { *p = v; fence(); }
+inline void     OrderAccess::release_store_fence(volatile julong*  p, julong  v) { release_store(p, v); fence(); }
 inline void     OrderAccess::release_store_fence(volatile jfloat*  p, jfloat  v) { *p = v; fence(); }
 inline void     OrderAccess::release_store_fence(volatile jdouble* p, jdouble v) { *p = v; fence(); }
 
--- a/src/os_cpu/solaris_sparc/vm/solaris_sparc.il	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/os_cpu/solaris_sparc/vm/solaris_sparc.il	Tue Feb 01 18:57:53 2011 -0800
@@ -1,5 +1,5 @@
 //
-// Copyright (c) 2002, 2005, Oracle and/or its affiliates. All rights reserved.
+// Copyright (c) 2002, 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
@@ -152,6 +152,39 @@
         .nonvolatile
         .end
 
+  // Support for jlong Atomic::load and Atomic::store on v8.
+  //
+  // void _Atomic_move_long_v8(volatile jlong* src, volatile jlong* dst)
+  //
+  // Arguments:
+  //      src:  O0
+  //      dest: O1
+  //
+  // Overwrites O2 and O3
+
+        .inline _Atomic_move_long_v8,2
+        .volatile
+        ldd     [%o0], %o2
+        std     %o2, [%o1]
+        .nonvolatile
+        .end
+
+  // Support for jlong Atomic::load and Atomic::store on v9.
+  //
+  // void _Atomic_move_long_v9(volatile jlong* src, volatile jlong* dst)
+  //
+  // Arguments:
+  //      src:  O0
+  //      dest: O1
+  //
+  // Overwrites O2
+
+        .inline _Atomic_move_long_v9,2
+        .volatile
+        ldx     [%o0], %o2
+        stx     %o2, [%o1]
+        .nonvolatile
+        .end
 
   // Support for jint Atomic::add(jint add_value, volatile jint* dest).
   //
--- a/src/os_cpu/solaris_x86/vm/atomic_solaris_x86.inline.hpp	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/os_cpu/solaris_x86/vm/atomic_solaris_x86.inline.hpp	Tue Feb 01 18:57:53 2011 -0800
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 1999, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1999, 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
@@ -151,14 +151,22 @@
   return (void*)cmpxchg((jint)exchange_value, (volatile jint*)dest, (jint)compare_value);
 }
 
-extern "C" void _Atomic_load_long(volatile jlong* src, volatile jlong* dst);
+extern "C" void _Atomic_move_long(volatile jlong* src, volatile jlong* dst);
 
 inline jlong Atomic::load(volatile jlong* src) {
   volatile jlong dest;
-  _Atomic_load_long(src, &dest);
+  _Atomic_move_long(src, &dest);
   return dest;
 }
 
+inline void Atomic::store(jlong store_value, jlong* dest) {
+  _Atomic_move_long((volatile jlong*)&store_value, (volatile jlong*)dest);
+}
+
+inline void Atomic::store(jlong store_value, volatile jlong* dest) {
+  _Atomic_move_long((volatile jlong*)&store_value, dest);
+}
+
 #endif // AMD64
 
 #ifdef _GNU_SOURCE
--- a/src/os_cpu/solaris_x86/vm/orderAccess_solaris_x86.inline.hpp	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/os_cpu/solaris_x86/vm/orderAccess_solaris_x86.inline.hpp	Tue Feb 01 18:57:53 2011 -0800
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2003, 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
@@ -25,6 +25,7 @@
 #ifndef OS_CPU_SOLARIS_X86_VM_ORDERACCESS_SOLARIS_X86_INLINE_HPP
 #define OS_CPU_SOLARIS_X86_VM_ORDERACCESS_SOLARIS_X86_INLINE_HPP
 
+#include "runtime/atomic.hpp"
 #include "runtime/orderAccess.hpp"
 #include "vm_version_x86.hpp"
 
@@ -80,11 +81,11 @@
 inline jbyte    OrderAccess::load_acquire(volatile jbyte*   p) { return *p; }
 inline jshort   OrderAccess::load_acquire(volatile jshort*  p) { return *p; }
 inline jint     OrderAccess::load_acquire(volatile jint*    p) { return *p; }
-inline jlong    OrderAccess::load_acquire(volatile jlong*   p) { return *p; }
+inline jlong    OrderAccess::load_acquire(volatile jlong*   p) { return Atomic::load(p); }
 inline jubyte   OrderAccess::load_acquire(volatile jubyte*  p) { return *p; }
 inline jushort  OrderAccess::load_acquire(volatile jushort* p) { return *p; }
 inline juint    OrderAccess::load_acquire(volatile juint*   p) { return *p; }
-inline julong   OrderAccess::load_acquire(volatile julong*  p) { return *p; }
+inline julong   OrderAccess::load_acquire(volatile julong*  p) { return Atomic::load((volatile jlong*)p); }
 inline jfloat   OrderAccess::load_acquire(volatile jfloat*  p) { return *p; }
 inline jdouble  OrderAccess::load_acquire(volatile jdouble* p) { return *p; }
 
@@ -95,11 +96,11 @@
 inline void     OrderAccess::release_store(volatile jbyte*   p, jbyte   v) { *p = v; }
 inline void     OrderAccess::release_store(volatile jshort*  p, jshort  v) { *p = v; }
 inline void     OrderAccess::release_store(volatile jint*    p, jint    v) { *p = v; }
-inline void     OrderAccess::release_store(volatile jlong*   p, jlong   v) { *p = v; }
+inline void     OrderAccess::release_store(volatile jlong*   p, jlong   v) { Atomic::store(v, p); }
 inline void     OrderAccess::release_store(volatile jubyte*  p, jubyte  v) { *p = v; }
 inline void     OrderAccess::release_store(volatile jushort* p, jushort v) { *p = v; }
 inline void     OrderAccess::release_store(volatile juint*   p, juint   v) { *p = v; }
-inline void     OrderAccess::release_store(volatile julong*  p, julong  v) { *p = v; }
+inline void     OrderAccess::release_store(volatile julong*  p, julong  v) { Atomic::store((jlong)v, (volatile jlong*)p); }
 inline void     OrderAccess::release_store(volatile jfloat*  p, jfloat  v) { *p = v; }
 inline void     OrderAccess::release_store(volatile jdouble* p, jdouble v) { *p = v; }
 
@@ -123,11 +124,11 @@
 inline void     OrderAccess::release_store_fence(volatile jbyte*   p, jbyte   v) { *p = v; fence(); }
 inline void     OrderAccess::release_store_fence(volatile jshort*  p, jshort  v) { *p = v; fence(); }
 inline void     OrderAccess::release_store_fence(volatile jint*    p, jint    v) { *p = v; fence(); }
-inline void     OrderAccess::release_store_fence(volatile jlong*   p, jlong   v) { *p = v; fence(); }
+inline void     OrderAccess::release_store_fence(volatile jlong*   p, jlong   v) { release_store(p, v); fence(); }
 inline void     OrderAccess::release_store_fence(volatile jubyte*  p, jubyte  v) { *p = v; fence(); }
 inline void     OrderAccess::release_store_fence(volatile jushort* p, jushort v) { *p = v; fence(); }
 inline void     OrderAccess::release_store_fence(volatile juint*   p, juint   v) { *p = v; fence(); }
-inline void     OrderAccess::release_store_fence(volatile julong*  p, julong  v) { *p = v; fence(); }
+inline void     OrderAccess::release_store_fence(volatile julong*  p, julong  v) { release_store(p, v); fence(); }
 inline void     OrderAccess::release_store_fence(volatile jfloat*  p, jfloat  v) { *p = v; fence(); }
 inline void     OrderAccess::release_store_fence(volatile jdouble* p, jdouble v) { *p = v; fence(); }
 
--- a/src/os_cpu/solaris_x86/vm/solaris_x86_32.il	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/os_cpu/solaris_x86/vm/solaris_x86_32.il	Tue Feb 01 18:57:53 2011 -0800
@@ -1,5 +1,5 @@
 //
-// Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
+// Copyright (c) 2003, 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
@@ -104,8 +104,9 @@
       popl     %ebx
       .end
 
-  // Support for void Atomic::load(volatile jlong* src, volatile jlong* dest).
-      .inline _Atomic_load_long,2
+  // Support for jlong Atomic::load and Atomic::store.
+  // void _Atomic_move_long(volatile jlong* src, volatile jlong* dst)
+      .inline _Atomic_move_long,2
       movl     0(%esp), %eax   // src
       fildll    (%eax)
       movl     4(%esp), %eax   // dest
--- a/src/os_cpu/windows_x86/vm/atomic_windows_x86.inline.hpp	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/os_cpu/windows_x86/vm/atomic_windows_x86.inline.hpp	Tue Feb 01 18:57:53 2011 -0800
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 1999, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1999, 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
@@ -137,10 +137,10 @@
   return (void*)cmpxchg((jlong)exchange_value, (volatile jlong*)dest, (jlong)compare_value);
 }
 
+inline jlong Atomic::load(volatile jlong* src) { return *src; }
+
 #else // !AMD64
 
-//inline void Atomic::store    (jlong    store_value, jlong*    dest) { *dest = store_value; }
-//inline void Atomic::store  (jlong    store_value, volatile jlong*    dest) { *dest = store_value; }
 inline jint     Atomic::add    (jint     add_value, volatile jint*     dest) {
   int mp = os::is_MP();
   __asm {
@@ -254,6 +254,33 @@
 inline void*    Atomic::cmpxchg_ptr(void*    exchange_value, volatile void*     dest, void*    compare_value) {
   return (void*)cmpxchg((jint)exchange_value, (volatile jint*)dest, (jint)compare_value);
 }
+
+inline jlong Atomic::load(volatile jlong* src) {
+  volatile jlong dest;
+  volatile jlong* pdest = &dest;
+  __asm {
+    mov eax, src
+    fild     qword ptr [eax]
+    mov eax, pdest
+    fistp    qword ptr [eax]
+  }
+  return dest;
+}
+
+inline void Atomic::store(jlong store_value, volatile jlong* dest) {
+  volatile jlong* src = &store_value;
+  __asm {
+    mov eax, src
+    fild     qword ptr [eax]
+    mov eax, dest
+    fistp    qword ptr [eax]
+  }
+}
+
+inline void Atomic::store(jlong store_value, jlong* dest) {
+  Atomic::store(store_value, (volatile jlong*)dest);
+}
+
 #endif // AMD64
 
 #pragma warning(default: 4035) // Enables warnings reporting missing return statement
--- a/src/os_cpu/windows_x86/vm/orderAccess_windows_x86.inline.hpp	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/os_cpu/windows_x86/vm/orderAccess_windows_x86.inline.hpp	Tue Feb 01 18:57:53 2011 -0800
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2003, 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
@@ -25,6 +25,7 @@
 #ifndef OS_CPU_WINDOWS_X86_VM_ORDERACCESS_WINDOWS_X86_INLINE_HPP
 #define OS_CPU_WINDOWS_X86_VM_ORDERACCESS_WINDOWS_X86_INLINE_HPP
 
+#include "runtime/atomic.hpp"
 #include "runtime/orderAccess.hpp"
 #include "vm_version_x86.hpp"
 
@@ -65,11 +66,11 @@
 inline jbyte    OrderAccess::load_acquire(volatile jbyte*   p) { return *p; }
 inline jshort   OrderAccess::load_acquire(volatile jshort*  p) { return *p; }
 inline jint     OrderAccess::load_acquire(volatile jint*    p) { return *p; }
-inline jlong    OrderAccess::load_acquire(volatile jlong*   p) { return *p; }
+inline jlong    OrderAccess::load_acquire(volatile jlong*   p) { return Atomic::load(p); }
 inline jubyte   OrderAccess::load_acquire(volatile jubyte*  p) { return *p; }
 inline jushort  OrderAccess::load_acquire(volatile jushort* p) { return *p; }
 inline juint    OrderAccess::load_acquire(volatile juint*   p) { return *p; }
-inline julong   OrderAccess::load_acquire(volatile julong*  p) { return *p; }
+inline julong   OrderAccess::load_acquire(volatile julong*  p) { return Atomic::load((volatile jlong*)p); }
 inline jfloat   OrderAccess::load_acquire(volatile jfloat*  p) { return *p; }
 inline jdouble  OrderAccess::load_acquire(volatile jdouble* p) { return *p; }
 
@@ -80,11 +81,11 @@
 inline void     OrderAccess::release_store(volatile jbyte*   p, jbyte   v) { *p = v; }
 inline void     OrderAccess::release_store(volatile jshort*  p, jshort  v) { *p = v; }
 inline void     OrderAccess::release_store(volatile jint*    p, jint    v) { *p = v; }
-inline void     OrderAccess::release_store(volatile jlong*   p, jlong   v) { *p = v; }
+inline void     OrderAccess::release_store(volatile jlong*   p, jlong   v) { Atomic::store(v, p); }
 inline void     OrderAccess::release_store(volatile jubyte*  p, jubyte  v) { *p = v; }
 inline void     OrderAccess::release_store(volatile jushort* p, jushort v) { *p = v; }
 inline void     OrderAccess::release_store(volatile juint*   p, juint   v) { *p = v; }
-inline void     OrderAccess::release_store(volatile julong*  p, julong  v) { *p = v; }
+inline void     OrderAccess::release_store(volatile julong*  p, julong  v) { Atomic::store((jlong)v, (volatile jlong*)p); }
 inline void     OrderAccess::release_store(volatile jfloat*  p, jfloat  v) { *p = v; }
 inline void     OrderAccess::release_store(volatile jdouble* p, jdouble v) { *p = v; }
 
@@ -188,7 +189,7 @@
 #endif // AMD64
 }
 
-inline void     OrderAccess::release_store_fence(volatile jlong*   p, jlong   v) { *p = v; fence(); }
+inline void     OrderAccess::release_store_fence(volatile jlong*   p, jlong   v) { release_store(p, v); fence(); }
 
 inline void     OrderAccess::release_store_fence(volatile jubyte*  p, jubyte  v) { release_store_fence((volatile jbyte*)p,  (jbyte)v);  }
 inline void     OrderAccess::release_store_fence(volatile jushort* p, jushort v) { release_store_fence((volatile jshort*)p, (jshort)v); }
--- a/src/share/vm/c1/c1_GraphBuilder.cpp	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/share/vm/c1/c1_GraphBuilder.cpp	Tue Feb 01 18:57:53 2011 -0800
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 1999, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1999, 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
@@ -319,24 +319,24 @@
 
       case Bytecodes::_tableswitch: {
         // set block for each case
-        Bytecode_tableswitch *switch_ = Bytecode_tableswitch_at(s.cur_bcp());
-        int l = switch_->length();
+        Bytecode_tableswitch sw(&s);
+        int l = sw.length();
         for (int i = 0; i < l; i++) {
-          make_block_at(cur_bci + switch_->dest_offset_at(i), current);
+          make_block_at(cur_bci + sw.dest_offset_at(i), current);
         }
-        make_block_at(cur_bci + switch_->default_offset(), current);
+        make_block_at(cur_bci + sw.default_offset(), current);
         current = NULL;
         break;
       }
 
       case Bytecodes::_lookupswitch: {
         // set block for each case
-        Bytecode_lookupswitch *switch_ = Bytecode_lookupswitch_at(s.cur_bcp());
-        int l = switch_->number_of_pairs();
+        Bytecode_lookupswitch sw(&s);
+        int l = sw.number_of_pairs();
         for (int i = 0; i < l; i++) {
-          make_block_at(cur_bci + switch_->pair_at(i)->offset(), current);
+          make_block_at(cur_bci + sw.pair_at(i).offset(), current);
         }
-        make_block_at(cur_bci + switch_->default_offset(), current);
+        make_block_at(cur_bci + sw.default_offset(), current);
         current = NULL;
         break;
       }
@@ -1275,15 +1275,15 @@
 
 
 void GraphBuilder::table_switch() {
-  Bytecode_tableswitch* switch_ = Bytecode_tableswitch_at(method()->code() + bci());
-  const int l = switch_->length();
+  Bytecode_tableswitch sw(stream());
+  const int l = sw.length();
   if (CanonicalizeNodes && l == 1) {
     // total of 2 successors => use If instead of switch
     // Note: This code should go into the canonicalizer as soon as it can
     //       can handle canonicalized forms that contain more than one node.
-    Value key = append(new Constant(new IntConstant(switch_->low_key())));
-    BlockBegin* tsux = block_at(bci() + switch_->dest_offset_at(0));
-    BlockBegin* fsux = block_at(bci() + switch_->default_offset());
+    Value key = append(new Constant(new IntConstant(sw.low_key())));
+    BlockBegin* tsux = block_at(bci() + sw.dest_offset_at(0));
+    BlockBegin* fsux = block_at(bci() + sw.default_offset());
     bool is_bb = tsux->bci() < bci() || fsux->bci() < bci();
     ValueStack* state_before = is_bb ? copy_state_before() : NULL;
     append(new If(ipop(), If::eql, true, key, tsux, fsux, state_before, is_bb));
@@ -1293,29 +1293,29 @@
     int i;
     bool has_bb = false;
     for (i = 0; i < l; i++) {
-      sux->at_put(i, block_at(bci() + switch_->dest_offset_at(i)));
-      if (switch_->dest_offset_at(i) < 0) has_bb = true;
+      sux->at_put(i, block_at(bci() + sw.dest_offset_at(i)));
+      if (sw.dest_offset_at(i) < 0) has_bb = true;
     }
     // add default successor
-    sux->at_put(i, block_at(bci() + switch_->default_offset()));
+    sux->at_put(i, block_at(bci() + sw.default_offset()));
     ValueStack* state_before = has_bb ? copy_state_before() : NULL;
-    append(new TableSwitch(ipop(), sux, switch_->low_key(), state_before, has_bb));
+    append(new TableSwitch(ipop(), sux, sw.low_key(), state_before, has_bb));
   }
 }
 
 
 void GraphBuilder::lookup_switch() {
-  Bytecode_lookupswitch* switch_ = Bytecode_lookupswitch_at(method()->code() + bci());
-  const int l = switch_->number_of_pairs();
+  Bytecode_lookupswitch sw(stream());
+  const int l = sw.number_of_pairs();
   if (CanonicalizeNodes && l == 1) {
     // total of 2 successors => use If instead of switch
     // Note: This code should go into the canonicalizer as soon as it can
     //       can handle canonicalized forms that contain more than one node.
     // simplify to If
-    LookupswitchPair* pair = switch_->pair_at(0);
-    Value key = append(new Constant(new IntConstant(pair->match())));
-    BlockBegin* tsux = block_at(bci() + pair->offset());
-    BlockBegin* fsux = block_at(bci() + switch_->default_offset());
+    LookupswitchPair pair = sw.pair_at(0);
+    Value key = append(new Constant(new IntConstant(pair.match())));
+    BlockBegin* tsux = block_at(bci() + pair.offset());
+    BlockBegin* fsux = block_at(bci() + sw.default_offset());
     bool is_bb = tsux->bci() < bci() || fsux->bci() < bci();
     ValueStack* state_before = is_bb ? copy_state_before() : NULL;
     append(new If(ipop(), If::eql, true, key, tsux, fsux, state_before, is_bb));
@@ -1326,13 +1326,13 @@
     int i;
     bool has_bb = false;
     for (i = 0; i < l; i++) {
-      LookupswitchPair* pair = switch_->pair_at(i);
-      if (pair->offset() < 0) has_bb = true;
-      sux->at_put(i, block_at(bci() + pair->offset()));
-      keys->at_put(i, pair->match());
+      LookupswitchPair pair = sw.pair_at(i);
+      if (pair.offset() < 0) has_bb = true;
+      sux->at_put(i, block_at(bci() + pair.offset()));
+      keys->at_put(i, pair.match());
     }
     // add default successor
-    sux->at_put(i, block_at(bci() + switch_->default_offset()));
+    sux->at_put(i, block_at(bci() + sw.default_offset()));
     ValueStack* state_before = has_bb ? copy_state_before() : NULL;
     append(new LookupSwitch(ipop(), sux, keys, state_before, has_bb));
   }
--- a/src/share/vm/c1/c1_LIRGenerator.cpp	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/share/vm/c1/c1_LIRGenerator.cpp	Tue Feb 01 18:57:53 2011 -0800
@@ -1990,9 +1990,8 @@
 
   LIR_Opr reg = reg = rlock_result(x, x->basic_type());
 
+  get_Object_unsafe(reg, src.result(), off.result(), type, x->is_volatile());
   if (x->is_volatile() && os::is_MP()) __ membar_acquire();
-  get_Object_unsafe(reg, src.result(), off.result(), type, x->is_volatile());
-  if (x->is_volatile() && os::is_MP()) __ membar();
 }
 
 
@@ -2014,6 +2013,7 @@
 
   if (x->is_volatile() && os::is_MP()) __ membar_release();
   put_Object_unsafe(src.result(), off.result(), data.result(), type, x->is_volatile());
+  if (x->is_volatile() && os::is_MP()) __ membar();
 }
 
 
--- a/src/share/vm/c1/c1_Runtime1.cpp	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/share/vm/c1/c1_Runtime1.cpp	Tue Feb 01 18:57:53 2011 -0800
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 1999, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1999, 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
@@ -369,7 +369,7 @@
   if (branch_bci != InvocationEntryBci) {
     // Compute desination bci
     address pc = method()->code_base() + branch_bci;
-    Bytecodes::Code branch = Bytecodes::code_at(pc, method());
+    Bytecodes::Code branch = Bytecodes::code_at(method(), pc);
     int offset = 0;
     switch (branch) {
       case Bytecodes::_if_icmplt: case Bytecodes::_iflt:
@@ -659,14 +659,14 @@
 
 
 static klassOop resolve_field_return_klass(methodHandle caller, int bci, TRAPS) {
-  Bytecode_field* field_access = Bytecode_field_at(caller, bci);
+  Bytecode_field field_access(caller, bci);
   // This can be static or non-static field access
-  Bytecodes::Code code       = field_access->code();
+  Bytecodes::Code code       = field_access.code();
 
   // We must load class, initialize class and resolvethe field
   FieldAccessInfo result; // initialize class if needed
   constantPoolHandle constants(THREAD, caller->constants());
-  LinkResolver::resolve_field(result, constants, field_access->index(), Bytecodes::java_code(code), false, CHECK_NULL);
+  LinkResolver::resolve_field(result, constants, field_access.index(), Bytecodes::java_code(code), false, CHECK_NULL);
   return result.klass()();
 }
 
@@ -767,7 +767,7 @@
 
   Events::log("patch_code @ " INTPTR_FORMAT , caller_frame.pc());
 
-  Bytecodes::Code code = Bytecode_at(caller_method->bcp_from(bci))->java_code();
+  Bytecodes::Code code = caller_method()->java_code_at(bci);
 
 #ifndef PRODUCT
   // this is used by assertions in the access_field_patching_id
@@ -779,11 +779,11 @@
   Handle load_klass(THREAD, NULL);                // oop needed by load_klass_patching code
   if (stub_id == Runtime1::access_field_patching_id) {
 
-    Bytecode_field* field_access = Bytecode_field_at(caller_method, bci);
+    Bytecode_field field_access(caller_method, bci);
     FieldAccessInfo result; // initialize class if needed
-    Bytecodes::Code code = field_access->code();
+    Bytecodes::Code code = field_access.code();
     constantPoolHandle constants(THREAD, caller_method->constants());
-    LinkResolver::resolve_field(result, constants, field_access->index(), Bytecodes::java_code(code), false, CHECK);
+    LinkResolver::resolve_field(result, constants, field_access.index(), Bytecodes::java_code(code), false, CHECK);
     patch_field_offset = result.field_offset();
 
     // If we're patching a field which is volatile then at compile it
@@ -811,36 +811,36 @@
         }
         break;
       case Bytecodes::_new:
-        { Bytecode_new* bnew = Bytecode_new_at(caller_method->bcp_from(bci));
-          k = caller_method->constants()->klass_at(bnew->index(), CHECK);
+        { Bytecode_new bnew(caller_method(), caller_method->bcp_from(bci));
+          k = caller_method->constants()->klass_at(bnew.index(), CHECK);
         }
         break;
       case Bytecodes::_multianewarray:
-        { Bytecode_multianewarray* mna = Bytecode_multianewarray_at(caller_method->bcp_from(bci));
-          k = caller_method->constants()->klass_at(mna->index(), CHECK);
+        { Bytecode_multianewarray mna(caller_method(), caller_method->bcp_from(bci));
+          k = caller_method->constants()->klass_at(mna.index(), CHECK);
         }
         break;
       case Bytecodes::_instanceof:
-        { Bytecode_instanceof* io = Bytecode_instanceof_at(caller_method->bcp_from(bci));
-          k = caller_method->constants()->klass_at(io->index(), CHECK);
+        { Bytecode_instanceof io(caller_method(), caller_method->bcp_from(bci));
+          k = caller_method->constants()->klass_at(io.index(), CHECK);
         }
         break;
       case Bytecodes::_checkcast:
-        { Bytecode_checkcast* cc = Bytecode_checkcast_at(caller_method->bcp_from(bci));
-          k = caller_method->constants()->klass_at(cc->index(), CHECK);
+        { Bytecode_checkcast cc(caller_method(), caller_method->bcp_from(bci));
+          k = caller_method->constants()->klass_at(cc.index(), CHECK);
         }
         break;
       case Bytecodes::_anewarray:
-        { Bytecode_anewarray* anew = Bytecode_anewarray_at(caller_method->bcp_from(bci));
-          klassOop ek = caller_method->constants()->klass_at(anew->index(), CHECK);
+        { Bytecode_anewarray anew(caller_method(), caller_method->bcp_from(bci));
+          klassOop ek = caller_method->constants()->klass_at(anew.index(), CHECK);
           k = Klass::cast(ek)->array_klass(CHECK);
         }
         break;
       case Bytecodes::_ldc:
       case Bytecodes::_ldc_w:
         {
-          Bytecode_loadconstant* cc = Bytecode_loadconstant_at(caller_method, bci);
-          k = cc->resolve_constant(CHECK);
+          Bytecode_loadconstant cc(caller_method, bci);
+          k = cc.resolve_constant(CHECK);
           assert(k != NULL && !k->is_klass(), "must be class mirror or other Java constant");
         }
         break;
--- a/src/share/vm/ci/bcEscapeAnalyzer.cpp	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/share/vm/ci/bcEscapeAnalyzer.cpp	Tue Feb 01 18:57:53 2011 -0800
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2005, 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
@@ -761,15 +761,15 @@
       case Bytecodes::_tableswitch:
         {
           state.spop();
-          Bytecode_tableswitch* switch_ = Bytecode_tableswitch_at(s.cur_bcp());
-          int len = switch_->length();
+          Bytecode_tableswitch sw(&s);
+          int len = sw.length();
           int dest_bci;
           for (int i = 0; i < len; i++) {
-            dest_bci = s.cur_bci() + switch_->dest_offset_at(i);
+            dest_bci = s.cur_bci() + sw.dest_offset_at(i);
             assert(_methodBlocks->is_block_start(dest_bci), "branch destination must start a block");
             successors.push(_methodBlocks->block_containing(dest_bci));
           }
-          dest_bci = s.cur_bci() + switch_->default_offset();
+          dest_bci = s.cur_bci() + sw.default_offset();
           assert(_methodBlocks->is_block_start(dest_bci), "branch destination must start a block");
           successors.push(_methodBlocks->block_containing(dest_bci));
           assert(s.next_bci() == limit_bci, "branch must end block");
@@ -779,15 +779,15 @@
       case Bytecodes::_lookupswitch:
         {
           state.spop();
-          Bytecode_lookupswitch* switch_ = Bytecode_lookupswitch_at(s.cur_bcp());
-          int len = switch_->number_of_pairs();
+          Bytecode_lookupswitch sw(&s);
+          int len = sw.number_of_pairs();
           int dest_bci;
           for (int i = 0; i < len; i++) {
-            dest_bci = s.cur_bci() + switch_->pair_at(i)->offset();
+            dest_bci = s.cur_bci() + sw.pair_at(i).offset();
             assert(_methodBlocks->is_block_start(dest_bci), "branch destination must start a block");
             successors.push(_methodBlocks->block_containing(dest_bci));
           }
-          dest_bci = s.cur_bci() + switch_->default_offset();
+          dest_bci = s.cur_bci() + sw.default_offset();
           assert(_methodBlocks->is_block_start(dest_bci), "branch destination must start a block");
           successors.push(_methodBlocks->block_containing(dest_bci));
           fall_through = false;
--- a/src/share/vm/ci/ciEnv.cpp	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/share/vm/ci/ciEnv.cpp	Tue Feb 01 18:57:53 2011 -0800
@@ -409,15 +409,15 @@
   } else {
     fail_type = _unloaded_ciinstance_klass;
   }
-  klassOop found_klass;
+  KlassHandle found_klass;
   if (!require_local) {
-    found_klass =
-      SystemDictionary::find_constrained_instance_or_array_klass(sym, loader,
-                                                                 KILL_COMPILE_ON_FATAL_(fail_type));
+    klassOop kls = SystemDictionary::find_constrained_instance_or_array_klass(
+        sym, loader, KILL_COMPILE_ON_FATAL_(fail_type));
+    found_klass = KlassHandle(THREAD, kls);
   } else {
-    found_klass =
-      SystemDictionary::find_instance_or_array_klass(sym, loader, domain,
-                                                     KILL_COMPILE_ON_FATAL_(fail_type));
+    klassOop kls = SystemDictionary::find_instance_or_array_klass(
+        sym, loader, domain, KILL_COMPILE_ON_FATAL_(fail_type));
+    found_klass = KlassHandle(THREAD, kls);
   }
 
   // If we fail to find an array klass, look again for its element type.
@@ -444,9 +444,9 @@
     }
   }
 
-  if (found_klass != NULL) {
+  if (found_klass() != NULL) {
     // Found it.  Build a CI handle.
-    return get_object(found_klass)->as_klass();
+    return get_object(found_klass())->as_klass();
   }
 
   if (require_local)  return NULL;
--- a/src/share/vm/ci/ciMethod.hpp	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/share/vm/ci/ciMethod.hpp	Tue Feb 01 18:57:53 2011 -0800
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 1999, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1999, 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
@@ -144,7 +144,7 @@
 
   Bytecodes::Code java_code_at_bci(int bci) {
     address bcp = code() + bci;
-    return Bytecodes::java_code_at(bcp);
+    return Bytecodes::java_code_at(NULL, bcp);
   }
   BCEscapeAnalyzer  *get_bcea();
   ciMethodBlocks    *get_method_blocks();
--- a/src/share/vm/ci/ciMethodBlocks.cpp	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/share/vm/ci/ciMethodBlocks.cpp	Tue Feb 01 18:57:53 2011 -0800
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2006, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2006, 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
@@ -175,15 +175,15 @@
       case Bytecodes::_tableswitch :
         {
           cur_block->set_control_bci(bci);
-          Bytecode_tableswitch* switch_ = Bytecode_tableswitch_at(s.cur_bcp());
-          int len = switch_->length();
+          Bytecode_tableswitch sw(&s);
+          int len = sw.length();
           ciBlock *dest;
           int dest_bci;
           for (int i = 0; i < len; i++) {
-            dest_bci = s.cur_bci() + switch_->dest_offset_at(i);
+            dest_bci = s.cur_bci() + sw.dest_offset_at(i);
             dest = make_block_at(dest_bci);
           }
-          dest_bci = s.cur_bci() + switch_->default_offset();
+          dest_bci = s.cur_bci() + sw.default_offset();
           make_block_at(dest_bci);
           if (s.next_bci() < limit_bci) {
             dest = make_block_at(s.next_bci());
@@ -194,15 +194,15 @@
       case Bytecodes::_lookupswitch:
         {
           cur_block->set_control_bci(bci);
-          Bytecode_lookupswitch* switch_ = Bytecode_lookupswitch_at(s.cur_bcp());
-          int len = switch_->number_of_pairs();
+          Bytecode_lookupswitch sw(&s);
+          int len = sw.number_of_pairs();
           ciBlock *dest;
           int dest_bci;
           for (int i = 0; i < len; i++) {
-            dest_bci = s.cur_bci() + switch_->pair_at(i)->offset();
+            dest_bci = s.cur_bci() + sw.pair_at(i).offset();
             dest = make_block_at(dest_bci);
           }
-          dest_bci = s.cur_bci() + switch_->default_offset();
+          dest_bci = s.cur_bci() + sw.default_offset();
           dest = make_block_at(dest_bci);
           if (s.next_bci() < limit_bci) {
             dest = make_block_at(s.next_bci());
--- a/src/share/vm/ci/ciMethodHandle.cpp	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/share/vm/ci/ciMethodHandle.cpp	Tue Feb 01 18:57:53 2011 -0800
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2009, 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
@@ -38,11 +38,12 @@
 // Return an adapter for this MethodHandle.
 ciMethod* ciMethodHandle::get_adapter(bool is_invokedynamic) const {
   VM_ENTRY_MARK;
-
   Handle h(get_oop());
   methodHandle callee(_callee->get_methodOop());
-  MethodHandleCompiler mhc(h, callee, is_invokedynamic, THREAD);
-  methodHandle m = mhc.compile(CHECK_NULL);
+  // We catch all exceptions here that could happen in the method
+  // handle compiler and stop the VM.
+  MethodHandleCompiler mhc(h, callee, is_invokedynamic, CATCH);
+  methodHandle m = mhc.compile(CATCH);
   return CURRENT_ENV->get_object(m())->as_method();
 }
 
--- a/src/share/vm/ci/ciStreams.hpp	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/share/vm/ci/ciStreams.hpp	Tue Feb 01 18:57:53 2011 -0800
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 1999, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1999, 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
@@ -78,8 +78,8 @@
     else { assert(!is_wide(), "must not be a wide instruction"); }
   }
 
-  Bytecode* bytecode() const { return Bytecode_at(_bc_start); }
-  Bytecode* next_bytecode() const { return Bytecode_at(_pc); }
+  Bytecode bytecode() const { return Bytecode(this, _bc_start); }
+  Bytecode next_bytecode() const { return Bytecode(this, _pc); }
 
 public:
   // End-Of-Bytecodes
@@ -151,11 +151,11 @@
   bool has_cache_index() const { return Bytecodes::uses_cp_cache(cur_bc_raw()); }
 
   int get_index_u1() const {
-    return bytecode()->get_index_u1(cur_bc_raw());
+    return bytecode().get_index_u1(cur_bc_raw());
   }
 
   int get_index_u1_cpcache() const {
-    return bytecode()->get_index_u1_cpcache(cur_bc_raw());
+    return bytecode().get_index_u1_cpcache(cur_bc_raw());
   }
 
   // Get a byte index following this bytecode.
@@ -169,29 +169,29 @@
 
   // Get 2-byte index (byte swapping depending on which bytecode)
   int get_index_u2(bool is_wide = false) const {
-    return bytecode()->get_index_u2(cur_bc_raw(), is_wide);
+    return bytecode().get_index_u2(cur_bc_raw(), is_wide);
   }
 
   // Get 2-byte index in native byte order.  (Rewriter::rewrite makes these.)
   int get_index_u2_cpcache() const {
-    return bytecode()->get_index_u2_cpcache(cur_bc_raw());
+    return bytecode().get_index_u2_cpcache(cur_bc_raw());
   }
 
   // Get 4-byte index, for invokedynamic.
   int get_index_u4() const {
-    return bytecode()->get_index_u4(cur_bc_raw());
+    return bytecode().get_index_u4(cur_bc_raw());
   }
 
   bool has_index_u4() const {
-    return bytecode()->has_index_u4(cur_bc_raw());
+    return bytecode().has_index_u4(cur_bc_raw());
   }
 
   // Get dimensions byte (multinewarray)
   int get_dimensions() const { return *(unsigned char*)(_pc-1); }
 
   // Sign-extended index byte/short, no widening
-  int get_constant_u1()                     const { return bytecode()->get_constant_u1(instruction_size()-1, cur_bc_raw()); }
-  int get_constant_u2(bool is_wide = false) const { return bytecode()->get_constant_u2(instruction_size()-2, cur_bc_raw(), is_wide); }
+  int get_constant_u1()                     const { return bytecode().get_constant_u1(instruction_size()-1, cur_bc_raw()); }
+  int get_constant_u2(bool is_wide = false) const { return bytecode().get_constant_u2(instruction_size()-2, cur_bc_raw(), is_wide); }
 
   // Get a byte signed constant for "iinc".  Invalid for other bytecodes.
   // If prefixed with a wide bytecode, get a wide constant
@@ -199,18 +199,18 @@
 
   // 2-byte branch offset from current pc
   int get_dest() const {
-    return cur_bci() + bytecode()->get_offset_s2(cur_bc_raw());
+    return cur_bci() + bytecode().get_offset_s2(cur_bc_raw());
   }
 
   // 2-byte branch offset from next pc
   int next_get_dest() const {
     assert(_pc < _end, "");
-    return next_bci() + next_bytecode()->get_offset_s2(Bytecodes::_ifeq);
+    return next_bci() + next_bytecode().get_offset_s2(Bytecodes::_ifeq);
   }
 
   // 4-byte branch offset from current pc
   int get_far_dest() const {
-    return cur_bci() + bytecode()->get_offset_s4(cur_bc_raw());
+    return cur_bci() + bytecode().get_offset_s4(cur_bc_raw());
   }
 
   // For a lookup or switch table, return target destination
@@ -407,4 +407,11 @@
   }
 };
 
+
+
+// Implementation for declarations in bytecode.hpp
+Bytecode::Bytecode(const ciBytecodeStream* stream, address bcp): _bcp(bcp != NULL ? bcp : stream->cur_bcp()), _code(Bytecodes::code_at(NULL, addr_at(0))) {}
+Bytecode_lookupswitch::Bytecode_lookupswitch(const ciBytecodeStream* stream): Bytecode(stream) { verify(); }
+Bytecode_tableswitch::Bytecode_tableswitch(const ciBytecodeStream* stream): Bytecode(stream) { verify(); }
+
 #endif // SHARE_VM_CI_CISTREAMS_HPP
--- a/src/share/vm/ci/ciTypeFlow.cpp	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/share/vm/ci/ciTypeFlow.cpp	Tue Feb 01 18:57:53 2011 -0800
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2000, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2000, 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
@@ -1698,18 +1698,17 @@
         break;
 
       case Bytecodes::_tableswitch:  {
-        Bytecode_tableswitch *tableswitch =
-          Bytecode_tableswitch_at(str->cur_bcp());
+        Bytecode_tableswitch tableswitch(str);
 
-        int len = tableswitch->length();
+        int len = tableswitch.length();
         _successors =
           new (arena) GrowableArray<Block*>(arena, len+1, 0, NULL);
-        int bci = current_bci + tableswitch->default_offset();
+        int bci = current_bci + tableswitch.default_offset();
         Block* block = analyzer->block_at(bci, jsrs);
         assert(_successors->length() == SWITCH_DEFAULT, "");
         _successors->append(block);
         while (--len >= 0) {
-          int bci = current_bci + tableswitch->dest_offset_at(len);
+          int bci = current_bci + tableswitch.dest_offset_at(len);
           block = analyzer->block_at(bci, jsrs);
           assert(_successors->length() >= SWITCH_CASES, "");
           _successors->append_if_missing(block);
@@ -1718,19 +1717,18 @@
       }
 
       case Bytecodes::_lookupswitch: {
-        Bytecode_lookupswitch *lookupswitch =
-          Bytecode_lookupswitch_at(str->cur_bcp());
+        Bytecode_lookupswitch lookupswitch(str);
 
-        int npairs = lookupswitch->number_of_pairs();
+        int npairs = lookupswitch.number_of_pairs();
         _successors =
           new (arena) GrowableArray<Block*>(arena, npairs+1, 0, NULL);
-        int bci = current_bci + lookupswitch->default_offset();
+        int bci = current_bci + lookupswitch.default_offset();
         Block* block = analyzer->block_at(bci, jsrs);
         assert(_successors->length() == SWITCH_DEFAULT, "");
         _successors->append(block);
         while(--npairs >= 0) {
-          LookupswitchPair *pair = lookupswitch->pair_at(npairs);
-          int bci = current_bci + pair->offset();
+          LookupswitchPair pair = lookupswitch.pair_at(npairs);
+          int bci = current_bci + pair.offset();
           Block* block = analyzer->block_at(bci, jsrs);
           assert(_successors->length() >= SWITCH_CASES, "");
           _successors->append_if_missing(block);
--- a/src/share/vm/classfile/classLoader.cpp	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/share/vm/classfile/classLoader.cpp	Tue Feb 01 18:57:53 2011 -0800
@@ -1382,3 +1382,61 @@
 }
 
 #endif //PRODUCT
+
+// Please keep following two functions at end of this file. With them placed at top or in middle of the file,
+// they could get inlined by agressive compiler, an unknown trick, see bug 6966589.
+void PerfClassTraceTime::initialize() {
+  if (!UsePerfData) return;
+
+  if (_eventp != NULL) {
+    // increment the event counter
+    _eventp->inc();
+  }
+
+  // stop the current active thread-local timer to measure inclusive time
+  _prev_active_event = -1;
+  for (int i=0; i < EVENT_TYPE_COUNT; i++) {
+     if (_timers[i].is_active()) {
+       assert(_prev_active_event == -1, "should have only one active timer");
+       _prev_active_event = i;
+       _timers[i].stop();
+     }
+  }
+
+  if (_recursion_counters == NULL || (_recursion_counters[_event_type])++ == 0) {
+    // start the inclusive timer if not recursively called
+    _t.start();
+  }
+
+  // start thread-local timer of the given event type
+   if (!_timers[_event_type].is_active()) {
+    _timers[_event_type].start();
+  }
+}
+
+PerfClassTraceTime::~PerfClassTraceTime() {
+  if (!UsePerfData) return;
+
+  // stop the thread-local timer as the event completes
+  // and resume the thread-local timer of the event next on the stack
+  _timers[_event_type].stop();
+  jlong selftime = _timers[_event_type].ticks();
+
+  if (_prev_active_event >= 0) {
+    _timers[_prev_active_event].start();
+  }
+
+  if (_recursion_counters != NULL && --(_recursion_counters[_event_type]) > 0) return;
+
+  // increment the counters only on the leaf call
+  _t.stop();
+  _timep->inc(_t.ticks());
+  if (_selftimep != NULL) {
+    _selftimep->inc(selftime);
+  }
+  // add all class loading related event selftime to the accumulated time counter
+  ClassLoader::perf_accumulated_time()->inc(selftime);
+
+  // reset the timer
+  _timers[_event_type].reset();
+}
--- a/src/share/vm/classfile/classLoader.hpp	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/share/vm/classfile/classLoader.hpp	Tue Feb 01 18:57:53 2011 -0800
@@ -356,111 +356,57 @@
 // (i.e. only one event type) are active at a time even multiple PerfClassTraceTime
 // instances have been created as multiple events are happening.
 class PerfClassTraceTime {
-  public:
-    enum {
-       CLASS_LOAD   = 0,
-       PARSE_CLASS  = 1,
-       CLASS_LINK   = 2,
-       CLASS_VERIFY = 3,
-       CLASS_CLINIT = 4,
-       DEFINE_CLASS = 5,
-       EVENT_TYPE_COUNT = 6
-    };
-  protected:
-    // _t tracks time from initialization to destruction of this timer instance
-    // including time for all other event types, and recursive calls of this type.
-    // When a timer is called recursively, the elapsedTimer _t would not be used.
-    elapsedTimer     _t;
-    PerfLongCounter* _timep;
-    PerfLongCounter* _selftimep;
-    PerfLongCounter* _eventp;
-    // pointer to thread-local recursion counter and timer array
-    // The thread_local timers track cumulative time for specific event types
-    // exclusive of time for other event types, but including recursive calls
-    // of the same type.
-    int*             _recursion_counters;
-    elapsedTimer*    _timers;
-    int              _event_type;
-    int              _prev_active_event;
+ public:
+  enum {
+    CLASS_LOAD   = 0,
+    PARSE_CLASS  = 1,
+    CLASS_LINK   = 2,
+    CLASS_VERIFY = 3,
+    CLASS_CLINIT = 4,
+    DEFINE_CLASS = 5,
+    EVENT_TYPE_COUNT = 6
+  };
+ protected:
+  // _t tracks time from initialization to destruction of this timer instance
+  // including time for all other event types, and recursive calls of this type.
+  // When a timer is called recursively, the elapsedTimer _t would not be used.
+  elapsedTimer     _t;
+  PerfLongCounter* _timep;
+  PerfLongCounter* _selftimep;
+  PerfLongCounter* _eventp;
+  // pointer to thread-local recursion counter and timer array
+  // The thread_local timers track cumulative time for specific event types
+  // exclusive of time for other event types, but including recursive calls
+  // of the same type.
+  int*             _recursion_counters;
+  elapsedTimer*    _timers;
+  int              _event_type;
+  int              _prev_active_event;
 
-  public:
+ public:
 
-    inline PerfClassTraceTime(PerfLongCounter* timep,     /* counter incremented with inclusive time */
-                              PerfLongCounter* selftimep, /* counter incremented with exclusive time */
-                              PerfLongCounter* eventp,    /* event counter */
-                              int* recursion_counters,    /* thread-local recursion counter array */
-                              elapsedTimer* timers,       /* thread-local timer array */
-                              int type                    /* event type */ ) :
-        _timep(timep), _selftimep(selftimep), _eventp(eventp), _recursion_counters(recursion_counters), _timers(timers), _event_type(type) {
-      initialize();
-    }
+  inline PerfClassTraceTime(PerfLongCounter* timep,     /* counter incremented with inclusive time */
+                            PerfLongCounter* selftimep, /* counter incremented with exclusive time */
+                            PerfLongCounter* eventp,    /* event counter */
+                            int* recursion_counters,    /* thread-local recursion counter array */
+                            elapsedTimer* timers,       /* thread-local timer array */
+                            int type                    /* event type */ ) :
+      _timep(timep), _selftimep(selftimep), _eventp(eventp), _recursion_counters(recursion_counters), _timers(timers), _event_type(type) {
+    initialize();
+  }
 
-    inline PerfClassTraceTime(PerfLongCounter* timep,     /* counter incremented with inclusive time */
-                              elapsedTimer* timers,       /* thread-local timer array */
-                              int type                    /* event type */ ) :
-        _timep(timep), _selftimep(NULL), _eventp(NULL), _recursion_counters(NULL), _timers(timers), _event_type(type) {
-      initialize();
-    }
+  inline PerfClassTraceTime(PerfLongCounter* timep,     /* counter incremented with inclusive time */
+                            elapsedTimer* timers,       /* thread-local timer array */
+                            int type                    /* event type */ ) :
+      _timep(timep), _selftimep(NULL), _eventp(NULL), _recursion_counters(NULL), _timers(timers), _event_type(type) {
+    initialize();
+  }
 
-    void initialize() {
-      if (!UsePerfData) return;
+  inline void suspend() { _t.stop(); _timers[_event_type].stop(); }
+  inline void resume()  { _t.start(); _timers[_event_type].start(); }
 
-      if (_eventp != NULL) {
-        // increment the event counter
-        _eventp->inc();
-      }
-
-      // stop the current active thread-local timer to measure inclusive time
-      _prev_active_event = -1;
-      for (int i=0; i < EVENT_TYPE_COUNT; i++) {
-         if (_timers[i].is_active()) {
-           assert(_prev_active_event == -1, "should have only one active timer");
-           _prev_active_event = i;
-           _timers[i].stop();
-         }
-      }
-
-      if (_recursion_counters == NULL || (_recursion_counters[_event_type])++ == 0) {
-        // start the inclusive timer if not recursively called
-        _t.start();
-      }
-
-      // start thread-local timer of the given event type
-      if (!_timers[_event_type].is_active()) {
-        _timers[_event_type].start();
-      }
-    }
-
-    inline void suspend() { _t.stop(); _timers[_event_type].stop(); }
-    inline void resume()  { _t.start(); _timers[_event_type].start(); }
-
-    ~PerfClassTraceTime() {
-      if (!UsePerfData) return;
-
-      // stop the thread-local timer as the event completes
-      // and resume the thread-local timer of the event next on the stack
-      _timers[_event_type].stop();
-      jlong selftime = _timers[_event_type].ticks();
-
-      if (_prev_active_event >= 0) {
-        _timers[_prev_active_event].start();
-      }
-
-      if (_recursion_counters != NULL && --(_recursion_counters[_event_type]) > 0) return;
-
-      // increment the counters only on the leaf call
-      _t.stop();
-      _timep->inc(_t.ticks());
-      if (_selftimep != NULL) {
-        _selftimep->inc(selftime);
-      }
-      // add all class loading related event selftime to the accumulated time counter
-      ClassLoader::perf_accumulated_time()->inc(selftime);
-
-      // reset the timer
-      _timers[_event_type].reset();
-    }
+  ~PerfClassTraceTime();
+  void initialize();
 };
 
-
 #endif // SHARE_VM_CLASSFILE_CLASSLOADER_HPP
--- a/src/share/vm/classfile/systemDictionary.hpp	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/share/vm/classfile/systemDictionary.hpp	Tue Feb 01 18:57:53 2011 -0800
@@ -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
@@ -172,6 +172,8 @@
                                                                               \
   template(sun_jkernel_DownloadManager_klass, sun_jkernel_DownloadManager, Opt_Kernel) \
                                                                               \
+  template(sun_misc_PostVMInitHook_klass, sun_misc_PostVMInitHook, Opt)       \
+                                                                              \
   /* Preload boxing klasses */                                                \
   template(Boolean_klass,                java_lang_Boolean,              Pre) \
   template(Character_klass,              java_lang_Character,            Pre) \
--- a/src/share/vm/classfile/vmSymbols.hpp	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/share/vm/classfile/vmSymbols.hpp	Tue Feb 01 18:57:53 2011 -0800
@@ -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
@@ -111,6 +111,7 @@
   template(sun_jkernel_DownloadManager,               "sun/jkernel/DownloadManager")              \
   template(getBootClassPathEntryForClass_name,        "getBootClassPathEntryForClass")            \
   template(setBootClassLoaderHook_name,               "setBootClassLoaderHook")                   \
+  template(sun_misc_PostVMInitHook,                   "sun/misc/PostVMInitHook")                  \
                                                                                                   \
   /* class file format tags */                                                                    \
   template(tag_source_file,                           "SourceFile")                               \
--- a/src/share/vm/code/nmethod.cpp	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/share/vm/code/nmethod.cpp	Tue Feb 01 18:57:53 2011 -0800
@@ -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
@@ -1863,9 +1863,9 @@
 #ifndef SHARK
   if (!method()->is_native()) {
     SimpleScopeDesc ssd(this, fr.pc());
-    Bytecode_invoke* call = Bytecode_invoke_at(ssd.method(), ssd.bci());
-    bool has_receiver = call->has_receiver();
-    symbolOop signature = call->signature();
+    Bytecode_invoke call(ssd.method(), ssd.bci());
+    bool has_receiver = call.has_receiver();
+    symbolOop signature = call.signature();
     fr.oops_compiled_arguments_do(signature, has_receiver, reg_map, f);
   }
 #endif // !SHARK
@@ -2698,8 +2698,7 @@
       } else if (sd->method()->is_native()) {
         st->print("method is native");
       } else {
-        address bcp  = sd->method()->bcp_from(sd->bci());
-        Bytecodes::Code bc = Bytecodes::java_code_at(bcp);
+        Bytecodes::Code bc = sd->method()->java_code_at(sd->bci());
         st->print(";*%s", Bytecodes::name(bc));
         switch (bc) {
         case Bytecodes::_invokevirtual:
@@ -2707,10 +2706,10 @@
         case Bytecodes::_invokestatic:
         case Bytecodes::_invokeinterface:
           {
-            Bytecode_invoke* invoke = Bytecode_invoke_at(sd->method(), sd->bci());
+            Bytecode_invoke invoke(sd->method(), sd->bci());
             st->print(" ");
-            if (invoke->name() != NULL)
-              invoke->name()->print_symbol_on(st);
+            if (invoke.name() != NULL)
+              invoke.name()->print_symbol_on(st);
             else
               st->print("<UNKNOWN>");
             break;
@@ -2720,10 +2719,10 @@
         case Bytecodes::_getstatic:
         case Bytecodes::_putstatic:
           {
-            Bytecode_field* field = Bytecode_field_at(sd->method(), sd->bci());
+            Bytecode_field field(sd->method(), sd->bci());
             st->print(" ");
-            if (field->name() != NULL)
-              field->name()->print_symbol_on(st);
+            if (field.name() != NULL)
+              field.name()->print_symbol_on(st);
             else
               st->print("<UNKNOWN>");
           }
--- a/src/share/vm/compiler/methodLiveness.cpp	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/share/vm/compiler/methodLiveness.cpp	Tue Feb 01 18:57:53 2011 -0800
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 1998, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1998, 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
@@ -286,16 +286,15 @@
         break;
       case Bytecodes::_tableswitch:
         {
-          Bytecode_tableswitch *tableswitch =
-            Bytecode_tableswitch_at(bytes.cur_bcp());
+          Bytecode_tableswitch tableswitch(&bytes);
 
-          int len = tableswitch->length();
+          int len = tableswitch.length();
 
-          dest = _block_map->at(bci + tableswitch->default_offset());
+          dest = _block_map->at(bci + tableswitch.default_offset());
           assert(dest != NULL, "branch desination must start a block.");
           dest->add_normal_predecessor(current_block);
           while (--len >= 0) {
-            dest = _block_map->at(bci + tableswitch->dest_offset_at(len));
+            dest = _block_map->at(bci + tableswitch.dest_offset_at(len));
             assert(dest != NULL, "branch desination must start a block.");
             dest->add_normal_predecessor(current_block);
           }
@@ -304,17 +303,16 @@
 
       case Bytecodes::_lookupswitch:
         {
-          Bytecode_lookupswitch *lookupswitch =
-            Bytecode_lookupswitch_at(bytes.cur_bcp());
+          Bytecode_lookupswitch lookupswitch(&bytes);
 
-          int npairs = lookupswitch->number_of_pairs();
+          int npairs = lookupswitch.number_of_pairs();
 
-          dest = _block_map->at(bci + lookupswitch->default_offset());
+          dest = _block_map->at(bci + lookupswitch.default_offset());
           assert(dest != NULL, "branch desination must start a block.");
           dest->add_normal_predecessor(current_block);
           while(--npairs >= 0) {
-            LookupswitchPair *pair = lookupswitch->pair_at(npairs);
-            dest = _block_map->at( bci + pair->offset());
+            LookupswitchPair pair = lookupswitch.pair_at(npairs);
+            dest = _block_map->at( bci + pair.offset());
             assert(dest != NULL, "branch desination must start a block.");
             dest->add_normal_predecessor(current_block);
           }
--- a/src/share/vm/gc_implementation/concurrentMarkSweep/concurrentMarkSweepGeneration.cpp	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/share/vm/gc_implementation/concurrentMarkSweep/concurrentMarkSweepGeneration.cpp	Tue Feb 01 18:57:53 2011 -0800
@@ -3478,6 +3478,7 @@
   assert(_collectorState == InitialMarking, "Wrong collector state");
   check_correct_thread_executing();
   TraceCMSMemoryManagerStats tms(_collectorState);
+
   ReferenceProcessor* rp = ref_processor();
   SpecializationStats::clear();
   assert(_restart_addr == NULL, "Control point invariant");
@@ -4978,6 +4979,7 @@
   if (should_unload_classes()) {
     CodeCache::gc_epilogue();
   }
+  JvmtiExport::gc_epilogue();
 
   // If we encountered any (marking stack / work queue) overflow
   // events during the current CMS cycle, take appropriate
@@ -5940,11 +5942,6 @@
   }
   rp->verify_no_references_recorded();
   assert(!rp->discovery_enabled(), "should have been disabled");
-
-  // JVMTI object tagging is based on JNI weak refs. If any of these
-  // refs were cleared then JVMTI needs to update its maps and
-  // maybe post ObjectFrees to agents.
-  JvmtiExport::cms_ref_processing_epilogue();
 }
 
 #ifndef PRODUCT
@@ -6305,6 +6302,7 @@
 
   switch (op) {
     case CMS_op_checkpointRootsInitial: {
+      SvcGCMarker sgcm(SvcGCMarker::OTHER);
       checkpointRootsInitial(true);       // asynch
       if (PrintGC) {
         _cmsGen->printOccupancy("initial-mark");
@@ -6312,6 +6310,7 @@
       break;
     }
     case CMS_op_checkpointRootsFinal: {
+      SvcGCMarker sgcm(SvcGCMarker::OTHER);
       checkpointRootsFinal(true,    // asynch
                            false,   // !clear_all_soft_refs
                            false);  // !init_mark_was_synchronous
@@ -7881,25 +7880,23 @@
 }
 
 // We need this destructor to reclaim any space at the end
-// of the space, which do_blk below may not have added back to
-// the free lists. [basically dealing with the "fringe effect"]
+// of the space, which do_blk below may not yet have added back to
+// the free lists.
 SweepClosure::~SweepClosure() {
   assert_lock_strong(_freelistLock);
-  // this should be treated as the end of a free run if any
-  // The current free range should be returned to the free lists
-  // as one coalesced chunk.
+  assert(_limit >= _sp->bottom() && _limit <= _sp->end(),
+         "sweep _limit out of bounds");
+  // Flush any remaining coterminal free run as a single
+  // coalesced chunk to the appropriate free list.
   if (inFreeRange()) {
-    flushCurFreeChunk(freeFinger(),
-      pointer_delta(_limit, freeFinger()));
-    assert(freeFinger() < _limit, "the finger pointeth off base");
+    assert(freeFinger() < _limit, "freeFinger points too high");
+    flush_cur_free_chunk(freeFinger(), pointer_delta(_limit, freeFinger()));
     if (CMSTraceSweeper) {
-      gclog_or_tty->print("destructor:");
-      gclog_or_tty->print("Sweep:put_free_blk 0x%x ("SIZE_FORMAT") "
-                 "[coalesced:"SIZE_FORMAT"]\n",
-                 freeFinger(), pointer_delta(_limit, freeFinger()),
-                 lastFreeRangeCoalesced());
-    }
-  }
+      gclog_or_tty->print("Sweep: last chunk: ");
+      gclog_or_tty->print("put_free_blk 0x%x ("SIZE_FORMAT") [coalesced:"SIZE_FORMAT"]\n",
+                          freeFinger(), pointer_delta(_limit, freeFinger()), lastFreeRangeCoalesced());
+    }
+  } // else nothing to flush
   NOT_PRODUCT(
     if (Verbose && PrintGC) {
       gclog_or_tty->print("Collected "SIZE_FORMAT" objects, "
@@ -7936,9 +7933,8 @@
 void SweepClosure::initialize_free_range(HeapWord* freeFinger,
     bool freeRangeInFreeLists) {
   if (CMSTraceSweeper) {
-    gclog_or_tty->print("---- Start free range 0x%x with free block [%d] (%d)\n",
-               freeFinger, _sp->block_size(freeFinger),
-               freeRangeInFreeLists);
+    gclog_or_tty->print("---- Start free range at 0x%x with free block (%d)\n",
+               freeFinger, freeRangeInFreeLists);
   }
   assert(!inFreeRange(), "Trampling existing free range");
   set_inFreeRange(true);
@@ -7993,21 +7989,36 @@
   // may have caused us to coalesce the block ending at the address _limit
   // with a newly expanded chunk (this happens when _limit was set to the
   // previous _end of the space), so we may have stepped past _limit; see CR 6977970.
-  if (addr >= _limit) { // we have swept up to or past the limit, do nothing more
+  if (addr >= _limit) { // we have swept up to or past the limit: finish up
     assert(_limit >= _sp->bottom() && _limit <= _sp->end(),
            "sweep _limit out of bounds");
     assert(addr < _sp->end(), "addr out of bounds");
-    // help the closure application finish
+    // Flush any remaining coterminal free run as a single
+    // coalesced chunk to the appropriate free list.
+    if (inFreeRange()) {
+      assert(freeFinger() < _limit, "finger points too high");
+      flush_cur_free_chunk(freeFinger(),
+                           pointer_delta(addr, freeFinger()));
+      if (CMSTraceSweeper) {
+        gclog_or_tty->print("Sweep: last chunk: ");
+        gclog_or_tty->print("put_free_blk 0x%x ("SIZE_FORMAT") "
+                   "[coalesced:"SIZE_FORMAT"]\n",
+                   freeFinger(), pointer_delta(addr, freeFinger()),
+                   lastFreeRangeCoalesced());
+      }
+    }
+
+    // help the iterator loop finish
     return pointer_delta(_sp->end(), addr);
   }
+
   assert(addr < _limit, "sweep invariant");
-
   // check if we should yield
   do_yield_check(addr);
   if (fc->isFree()) {
     // Chunk that is already free
     res = fc->size();
-    doAlreadyFreeChunk(fc);
+    do_already_free_chunk(fc);
     debug_only(_sp->verifyFreeLists());
     assert(res == fc->size(), "Don't expect the size to change");
     NOT_PRODUCT(
@@ -8017,7 +8028,7 @@
     NOT_PRODUCT(_last_fc = fc;)
   } else if (!_bitMap->isMarked(addr)) {
     // Chunk is fresh garbage
-    res = doGarbageChunk(fc);
+    res = do_garbage_chunk(fc);
     debug_only(_sp->verifyFreeLists());
     NOT_PRODUCT(
       _numObjectsFreed++;
@@ -8025,7 +8036,7 @@
     )
   } else {
     // Chunk that is alive.
-    res = doLiveChunk(fc);
+    res = do_live_chunk(fc);
     debug_only(_sp->verifyFreeLists());
     NOT_PRODUCT(
         _numObjectsLive++;
@@ -8078,7 +8089,7 @@
 // to a free list which may be overpopulated.
 //
 
-void SweepClosure::doAlreadyFreeChunk(FreeChunk* fc) {
+void SweepClosure::do_already_free_chunk(FreeChunk* fc) {
   size_t size = fc->size();
   // Chunks that cannot be coalesced are not in the
   // free lists.
@@ -8094,23 +8105,23 @@
   // addr and purported end of this block.
   _bitMap->verifyNoOneBitsInRange(addr + 1, addr + size);
 
-  // Some chunks cannot be coalesced in under any circumstances.
+  // Some chunks cannot be coalesced under any circumstances.
   // See the definition of cantCoalesce().
   if (!fc->cantCoalesce()) {
     // This chunk can potentially be coalesced.
     if (_sp->adaptive_freelists()) {
       // All the work is done in
-      doPostIsFreeOrGarbageChunk(fc, size);
+      do_post_free_or_garbage_chunk(fc, size);
     } else {  // Not adaptive free lists
       // this is a free chunk that can potentially be coalesced by the sweeper;
       if (!inFreeRange()) {
         // if the next chunk is a free block that can't be coalesced
         // it doesn't make sense to remove this chunk from the free lists
         FreeChunk* nextChunk = (FreeChunk*)(addr + size);
-        assert((HeapWord*)nextChunk <= _limit, "sweep invariant");
-        if ((HeapWord*)nextChunk < _limit  &&    // there's a next chunk...
-            nextChunk->isFree()    &&            // which is free...
-            nextChunk->cantCoalesce()) {         // ... but cant be coalesced
+        assert((HeapWord*)nextChunk <= _sp->end(), "Chunk size out of bounds?");
+        if ((HeapWord*)nextChunk < _sp->end() &&     // There is another free chunk to the right ...
+            nextChunk->isFree()               &&     // ... which is free...
+            nextChunk->cantCoalesce()) {             // ... but can't be coalesced
           // nothing to do
         } else {
           // Potentially the start of a new free range:
@@ -8156,14 +8167,14 @@
     // as the end of a free run if any
     if (inFreeRange()) {
       // we kicked some butt; time to pick up the garbage
-      assert(freeFinger() < addr, "the finger pointeth off base");
-      flushCurFreeChunk(freeFinger(), pointer_delta(addr, freeFinger()));
+      assert(freeFinger() < addr, "freeFinger points too high");
+      flush_cur_free_chunk(freeFinger(), pointer_delta(addr, freeFinger()));
     }
     // else, nothing to do, just continue
   }
 }
 
-size_t SweepClosure::doGarbageChunk(FreeChunk* fc) {
+size_t SweepClosure::do_garbage_chunk(FreeChunk* fc) {
   // This is a chunk of garbage.  It is not in any free list.
   // Add it to a free list or let it possibly be coalesced into
   // a larger chunk.
@@ -8175,7 +8186,7 @@
     // addr and purported end of just dead object.
     _bitMap->verifyNoOneBitsInRange(addr + 1, addr + size);
 
-    doPostIsFreeOrGarbageChunk(fc, size);
+    do_post_free_or_garbage_chunk(fc, size);
   } else {
     if (!inFreeRange()) {
       // start of a new free range
@@ -8214,35 +8225,16 @@
   return size;
 }
 
-size_t SweepClosure::doLiveChunk(FreeChunk* fc) {
+size_t SweepClosure::do_live_chunk(FreeChunk* fc) {
   HeapWord* addr = (HeapWord*) fc;
   // The sweeper has just found a live object. Return any accumulated
   // left hand chunk to the free lists.
   if (inFreeRange()) {
-    if (_sp->adaptive_freelists()) {
-      flushCurFreeChunk(freeFinger(),
-                        pointer_delta(addr, freeFinger()));
-    } else { // not adaptive freelists
-      set_inFreeRange(false);
-      // Add the free range back to the free list if it is not already
-      // there.
-      if (!freeRangeInFreeLists()) {
-        assert(freeFinger() < addr, "the finger pointeth off base");
-        if (CMSTraceSweeper) {
-          gclog_or_tty->print("Sweep:put_free_blk 0x%x (%d) "
-            "[coalesced:%d]\n",
-            freeFinger(), pointer_delta(addr, freeFinger()),
-            lastFreeRangeCoalesced());
-        }
-        _sp->addChunkAndRepairOffsetTable(freeFinger(),
-          pointer_delta(addr, freeFinger()), lastFreeRangeCoalesced());
-      }
-    }
-  }
-
-  // Common code path for original and adaptive free lists.
-
-  // this object is live: we'd normally expect this to be
+    assert(freeFinger() < addr, "freeFinger points too high");
+    flush_cur_free_chunk(freeFinger(), pointer_delta(addr, freeFinger()));
+  }
+
+  // This object is live: we'd normally expect this to be
   // an oop, and like to assert the following:
   // assert(oop(addr)->is_oop(), "live block should be an oop");
   // However, as we commented above, this may be an object whose
@@ -8257,7 +8249,7 @@
     assert(size == CompactibleFreeListSpace::adjustObjectSize(size),
            "alignment problem");
 
-    #ifdef DEBUG
+#ifdef DEBUG
       if (oop(addr)->klass_or_null() != NULL &&
           (   !_collector->should_unload_classes()
            || (oop(addr)->is_parsable()) &&
@@ -8271,7 +8263,7 @@
                CompactibleFreeListSpace::adjustObjectSize(oop(addr)->size()),
                "P-mark and computed size do not agree");
       }
-    #endif
+#endif
 
   } else {
     // This should be an initialized object that's alive.
@@ -8298,18 +8290,16 @@
   return size;
 }
 
-void SweepClosure::doPostIsFreeOrGarbageChunk(FreeChunk* fc,
-                                            size_t chunkSize) {
-  // doPostIsFreeOrGarbageChunk() should only be called in the smart allocation
-  // scheme.
+void SweepClosure::do_post_free_or_garbage_chunk(FreeChunk* fc,
+                                                 size_t chunkSize) {
+  // do_post_free_or_garbage_chunk() should only be called in the case
+  // of the adaptive free list allocator.
   bool fcInFreeLists = fc->isFree();
   assert(_sp->adaptive_freelists(), "Should only be used in this case.");
   assert((HeapWord*)fc <= _limit, "sweep invariant");
   if (CMSTestInFreeList && fcInFreeLists) {
-    assert(_sp->verifyChunkInFreeLists(fc),
-      "free chunk is not in free lists");
-  }
-
+    assert(_sp->verifyChunkInFreeLists(fc), "free chunk is not in free lists");
+  }
 
   if (CMSTraceSweeper) {
     gclog_or_tty->print_cr("  -- pick up another chunk at 0x%x (%d)", fc, chunkSize);
@@ -8382,20 +8372,21 @@
     if (inFreeRange()) {
       // In a free range but cannot coalesce with the right hand chunk.
       // Put the current free range into the free lists.
-      flushCurFreeChunk(freeFinger(),
-        pointer_delta(addr, freeFinger()));
+      flush_cur_free_chunk(freeFinger(),
+                           pointer_delta(addr, freeFinger()));
     }
     // Set up for new free range.  Pass along whether the right hand
     // chunk is in the free lists.
     initialize_free_range((HeapWord*)fc, fcInFreeLists);
   }
 }
-void SweepClosure::flushCurFreeChunk(HeapWord* chunk, size_t size) {
+
+void SweepClosure::flush_cur_free_chunk(HeapWord* chunk, size_t size) {
   assert(inFreeRange(), "Should only be called if currently in a free range.");
   assert(size > 0,
     "A zero sized chunk cannot be added to the free lists.");
   if (!freeRangeInFreeLists()) {
-    if(CMSTestInFreeList) {
+    if (CMSTestInFreeList) {
       FreeChunk* fc = (FreeChunk*) chunk;
       fc->setSize(size);
       assert(!_sp->verifyChunkInFreeLists(fc),
@@ -8430,7 +8421,7 @@
   // chunk just flushed, they will need to wait for the next
   // sweep to be coalesced.
   if (inFreeRange()) {
-    flushCurFreeChunk(freeFinger(), pointer_delta(addr, freeFinger()));
+    flush_cur_free_chunk(freeFinger(), pointer_delta(addr, freeFinger()));
   }
 
   // First give up the locks, then yield, then re-lock.
--- a/src/share/vm/gc_implementation/concurrentMarkSweep/concurrentMarkSweepGeneration.hpp	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/share/vm/gc_implementation/concurrentMarkSweep/concurrentMarkSweepGeneration.hpp	Tue Feb 01 18:57:53 2011 -0800
@@ -1701,7 +1701,9 @@
   CMSCollector*                  _collector;  // collector doing the work
   ConcurrentMarkSweepGeneration* _g;    // Generation being swept
   CompactibleFreeListSpace*      _sp;   // Space being swept
-  HeapWord*                      _limit;
+  HeapWord*                      _limit;// the address at which the sweep should stop because
+                                        // we do not expect blocks eligible for sweeping past
+                                        // that address.
   Mutex*                         _freelistLock; // Free list lock (in space)
   CMSBitMap*                     _bitMap;       // Marking bit map (in
                                                 // generation)
@@ -1745,14 +1747,13 @@
  private:
   // Code that is common to a free chunk or garbage when
   // encountered during sweeping.
-  void doPostIsFreeOrGarbageChunk(FreeChunk *fc,
-                                  size_t chunkSize);
+  void do_post_free_or_garbage_chunk(FreeChunk *fc, size_t chunkSize);
   // Process a free chunk during sweeping.
-  void doAlreadyFreeChunk(FreeChunk *fc);
+  void do_already_free_chunk(FreeChunk *fc);
   // Process a garbage chunk during sweeping.
-  size_t doGarbageChunk(FreeChunk *fc);
+  size_t do_garbage_chunk(FreeChunk *fc);
   // Process a live chunk during sweeping.
-  size_t doLiveChunk(FreeChunk* fc);
+  size_t do_live_chunk(FreeChunk* fc);
 
   // Accessors.
   HeapWord* freeFinger() const          { return _freeFinger; }
@@ -1769,7 +1770,7 @@
   // Initialize a free range.
   void initialize_free_range(HeapWord* freeFinger, bool freeRangeInFreeLists);
   // Return this chunk to the free lists.
-  void flushCurFreeChunk(HeapWord* chunk, size_t size);
+  void flush_cur_free_chunk(HeapWord* chunk, size_t size);
 
   // Check if we should yield and do so when necessary.
   inline void do_yield_check(HeapWord* addr);
--- a/src/share/vm/gc_implementation/g1/concurrentMark.cpp	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/share/vm/gc_implementation/g1/concurrentMark.cpp	Tue Feb 01 18:57:53 2011 -0800
@@ -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
@@ -31,6 +31,7 @@
 #include "gc_implementation/g1/g1RemSet.hpp"
 #include "gc_implementation/g1/heapRegionRemSet.hpp"
 #include "gc_implementation/g1/heapRegionSeq.inline.hpp"
+#include "gc_implementation/shared/vmGCOperations.hpp"
 #include "memory/genOopClosures.inline.hpp"
 #include "memory/referencePolicy.hpp"
 #include "memory/resourceArea.hpp"
@@ -457,6 +458,7 @@
   _marking_task_overhead(1.0),
   _cleanup_sleep_factor(0.0),
   _cleanup_task_overhead(1.0),
+  _cleanup_list("Cleanup List"),
   _region_bm(max_regions, false /* in_resource_area*/),
   _card_bm((rs.size() + CardTableModRefBS::card_size - 1) >>
            CardTableModRefBS::card_shift,
@@ -520,12 +522,6 @@
   SATBMarkQueueSet& satb_qs = JavaThread::satb_mark_queue_set();
   satb_qs.set_buffer_size(G1SATBBufferSize);
 
-  int size = (int) MAX2(ParallelGCThreads, (size_t)1);
-  _par_cleanup_thread_state = NEW_C_HEAP_ARRAY(ParCleanupThreadState*, size);
-  for (int i = 0 ; i < size; i++) {
-    _par_cleanup_thread_state[i] = new ParCleanupThreadState;
-  }
-
   _tasks = NEW_C_HEAP_ARRAY(CMTask*, _max_task_num);
   _accum_task_vtime = NEW_C_HEAP_ARRAY(double, _max_task_num);
 
@@ -710,11 +706,6 @@
 }
 
 ConcurrentMark::~ConcurrentMark() {
-  int size = (int) MAX2(ParallelGCThreads, (size_t)1);
-  for (int i = 0; i < size; i++) delete _par_cleanup_thread_state[i];
-  FREE_C_HEAP_ARRAY(ParCleanupThreadState*,
-                    _par_cleanup_thread_state);
-
   for (int i = 0; i < (int) _max_task_num; ++i) {
     delete _task_queues->queue(i);
     delete _tasks[i];
@@ -1142,6 +1133,8 @@
     return;
   }
 
+  SvcGCMarker sgcm(SvcGCMarker::OTHER);
+
   if (VerifyDuringGC) {
     HandleMark hm;  // handle scope
     gclog_or_tty->print(" VerifyDuringGC:(before)");
@@ -1168,12 +1161,12 @@
     if (G1TraceMarkStackOverflow)
       gclog_or_tty->print_cr("\nRemark led to restart for overflow.");
   } else {
+    SATBMarkQueueSet& satb_mq_set = JavaThread::satb_mark_queue_set();
     // We're done with marking.
     // This is the end of  the marking cycle, we're expected all
     // threads to have SATB queues with active set to true.
-    JavaThread::satb_mark_queue_set().set_active_all_threads(
-                                                  false, /* new active value */
-                                                  true /* expected_active */);
+    satb_mq_set.set_active_all_threads(false, /* new active value */
+                                       true /* expected_active */);
 
     if (VerifyDuringGC) {
       HandleMark hm;  // handle scope
@@ -1507,21 +1500,20 @@
   size_t _max_live_bytes;
   size_t _regions_claimed;
   size_t _freed_bytes;
-  size_t _cleared_h_regions;
-  size_t _freed_regions;
-  UncleanRegionList* _unclean_region_list;
+  FreeRegionList _local_cleanup_list;
+  HumongousRegionSet _humongous_proxy_set;
   double _claimed_region_time;
   double _max_region_time;
 
 public:
   G1NoteEndOfConcMarkClosure(G1CollectedHeap* g1,
-                             UncleanRegionList* list,
                              int worker_num);
   size_t freed_bytes() { return _freed_bytes; }
-  size_t cleared_h_regions() { return _cleared_h_regions; }
-  size_t freed_regions() { return  _freed_regions; }
-  UncleanRegionList* unclean_region_list() {
-    return _unclean_region_list;
+  FreeRegionList* local_cleanup_list() {
+    return &_local_cleanup_list;
+  }
+  HumongousRegionSet* humongous_proxy_set() {
+    return &_humongous_proxy_set;
   }
 
   bool doHeapRegion(HeapRegion *r);
@@ -1534,25 +1526,22 @@
 
 class G1ParNoteEndTask: public AbstractGangTask {
   friend class G1NoteEndOfConcMarkClosure;
+
 protected:
   G1CollectedHeap* _g1h;
   size_t _max_live_bytes;
   size_t _freed_bytes;
-  ConcurrentMark::ParCleanupThreadState** _par_cleanup_thread_state;
+  FreeRegionList* _cleanup_list;
+
 public:
   G1ParNoteEndTask(G1CollectedHeap* g1h,
-                   ConcurrentMark::ParCleanupThreadState**
-                   par_cleanup_thread_state) :
+                   FreeRegionList* cleanup_list) :
     AbstractGangTask("G1 note end"), _g1h(g1h),
-    _max_live_bytes(0), _freed_bytes(0),
-    _par_cleanup_thread_state(par_cleanup_thread_state)
-  {}
+    _max_live_bytes(0), _freed_bytes(0), _cleanup_list(cleanup_list) { }
 
   void work(int i) {
     double start = os::elapsedTime();
-    G1NoteEndOfConcMarkClosure g1_note_end(_g1h,
-                                           &_par_cleanup_thread_state[i]->list,
-                                           i);
+    G1NoteEndOfConcMarkClosure g1_note_end(_g1h, i);
     if (G1CollectedHeap::use_parallel_gc_threads()) {
       _g1h->heap_region_par_iterate_chunked(&g1_note_end, i,
                                             HeapRegion::NoteEndClaimValue);
@@ -1561,14 +1550,18 @@
     }
     assert(g1_note_end.complete(), "Shouldn't have yielded!");
 
-    // Now finish up freeing the current thread's regions.
-    _g1h->finish_free_region_work(g1_note_end.freed_bytes(),
-                                  g1_note_end.cleared_h_regions(),
-                                  0, NULL);
+    // Now update the lists
+    _g1h->update_sets_after_freeing_regions(g1_note_end.freed_bytes(),
+                                            NULL /* free_list */,
+                                            g1_note_end.humongous_proxy_set(),
+                                            true /* par */);
     {
       MutexLockerEx x(ParGCRareEvent_lock, Mutex::_no_safepoint_check_flag);
       _max_live_bytes += g1_note_end.max_live_bytes();
       _freed_bytes += g1_note_end.freed_bytes();
+
+      _cleanup_list->add_as_tail(g1_note_end.local_cleanup_list());
+      assert(g1_note_end.local_cleanup_list()->is_empty(), "post-condition");
     }
     double end = os::elapsedTime();
     if (G1PrintParCleanupStats) {
@@ -1609,30 +1602,28 @@
 
 G1NoteEndOfConcMarkClosure::
 G1NoteEndOfConcMarkClosure(G1CollectedHeap* g1,
-                           UncleanRegionList* list,
                            int worker_num)
   : _g1(g1), _worker_num(worker_num),
     _max_live_bytes(0), _regions_claimed(0),
-    _freed_bytes(0), _cleared_h_regions(0), _freed_regions(0),
+    _freed_bytes(0),
     _claimed_region_time(0.0), _max_region_time(0.0),
-    _unclean_region_list(list)
-{}
-
-bool G1NoteEndOfConcMarkClosure::doHeapRegion(HeapRegion *r) {
+    _local_cleanup_list("Local Cleanup List"),
+    _humongous_proxy_set("Local Cleanup Humongous Proxy Set") { }
+
+bool G1NoteEndOfConcMarkClosure::doHeapRegion(HeapRegion *hr) {
   // We use a claim value of zero here because all regions
   // were claimed with value 1 in the FinalCount task.
-  r->reset_gc_time_stamp();
-  if (!r->continuesHumongous()) {
+  hr->reset_gc_time_stamp();
+  if (!hr->continuesHumongous()) {
     double start = os::elapsedTime();
     _regions_claimed++;
-    r->note_end_of_marking();
-    _max_live_bytes += r->max_live_bytes();
-    _g1->free_region_if_totally_empty_work(r,
-                                           _freed_bytes,
-                                           _cleared_h_regions,
-                                           _freed_regions,
-                                           _unclean_region_list,
-                                           true /*par*/);
+    hr->note_end_of_marking();
+    _max_live_bytes += hr->max_live_bytes();
+    _g1->free_region_if_totally_empty(hr,
+                                      &_freed_bytes,
+                                      &_local_cleanup_list,
+                                      &_humongous_proxy_set,
+                                      true /* par */);
     double region_time = (os::elapsedTime() - start);
     _claimed_region_time += region_time;
     if (region_time > _max_region_time) _max_region_time = region_time;
@@ -1652,6 +1643,8 @@
     return;
   }
 
+  g1h->verify_region_sets_optional();
+
   if (VerifyDuringGC) {
     HandleMark hm;  // handle scope
     gclog_or_tty->print(" VerifyDuringGC:(before)");
@@ -1716,7 +1709,7 @@
 
   // Note end of marking in all heap regions.
   double note_end_start = os::elapsedTime();
-  G1ParNoteEndTask g1_par_note_end_task(g1h, _par_cleanup_thread_state);
+  G1ParNoteEndTask g1_par_note_end_task(g1h, &_cleanup_list);
   if (G1CollectedHeap::use_parallel_gc_threads()) {
     int n_workers = g1h->workers()->total_workers();
     g1h->set_par_threads(n_workers);
@@ -1728,9 +1721,14 @@
   } else {
     g1_par_note_end_task.work(0);
   }
-  g1h->set_unclean_regions_coming(true);
+
+  if (!cleanup_list_is_empty()) {
+    // The cleanup list is not empty, so we'll have to process it
+    // concurrently. Notify anyone else that might be wanting free
+    // regions that there will be more free regions coming soon.
+    g1h->set_free_regions_coming();
+  }
   double note_end_end = os::elapsedTime();
-  // Tell the mutators that there might be unclean regions coming...
   if (G1PrintParCleanupStats) {
     gclog_or_tty->print_cr("  note end of marking: %8.3f ms.",
                            (note_end_end - note_end_start)*1000.0);
@@ -1796,33 +1794,63 @@
                      /* silent       */ false,
                      /* prev marking */ true);
   }
+
+  g1h->verify_region_sets_optional();
 }
 
 void ConcurrentMark::completeCleanup() {
-  // A full collection intervened.
   if (has_aborted()) return;
 
-  int first = 0;
-  int last = (int)MAX2(ParallelGCThreads, (size_t)1);
-  for (int t = 0; t < last; t++) {
-    UncleanRegionList* list = &_par_cleanup_thread_state[t]->list;
-    assert(list->well_formed(), "Inv");
-    HeapRegion* hd = list->hd();
-    while (hd != NULL) {
-      // Now finish up the other stuff.
-      hd->rem_set()->clear();
-      HeapRegion* next_hd = hd->next_from_unclean_list();
-      (void)list->pop();
-      assert(list->hd() == next_hd, "how not?");
-      _g1h->put_region_on_unclean_list(hd);
-      if (!hd->isHumongous()) {
-        // Add this to the _free_regions count by 1.
-        _g1h->finish_free_region_work(0, 0, 1, NULL);
+  G1CollectedHeap* g1h = G1CollectedHeap::heap();
+
+  _cleanup_list.verify_optional();
+  FreeRegionList local_free_list("Local Cleanup List");
+
+  if (G1ConcRegionFreeingVerbose) {
+    gclog_or_tty->print_cr("G1ConcRegionFreeing [complete cleanup] : "
+                           "cleanup list has "SIZE_FORMAT" entries",
+                           _cleanup_list.length());
+  }
+
+  // Noone else should be accessing the _cleanup_list at this point,
+  // so it's not necessary to take any locks
+  while (!_cleanup_list.is_empty()) {
+    HeapRegion* hr = _cleanup_list.remove_head();
+    assert(hr != NULL, "the list was not empty");
+    hr->rem_set()->clear();
+    local_free_list.add_as_tail(hr);
+
+    // Instead of adding one region at a time to the secondary_free_list,
+    // we accumulate them in the local list and move them a few at a
+    // time. This also cuts down on the number of notify_all() calls
+    // we do during this process. We'll also append the local list when
+    // _cleanup_list is empty (which means we just removed the last
+    // region from the _cleanup_list).
+    if ((local_free_list.length() % G1SecondaryFreeListAppendLength == 0) ||
+        _cleanup_list.is_empty()) {
+      if (G1ConcRegionFreeingVerbose) {
+        gclog_or_tty->print_cr("G1ConcRegionFreeing [complete cleanup] : "
+                               "appending "SIZE_FORMAT" entries to the "
+                               "secondary_free_list, clean list still has "
+                               SIZE_FORMAT" entries",
+                               local_free_list.length(),
+                               _cleanup_list.length());
       }
-      hd = list->hd();
-      assert(hd == next_hd, "how not?");
+
+      {
+        MutexLockerEx x(SecondaryFreeList_lock, Mutex::_no_safepoint_check_flag);
+        g1h->secondary_free_list_add_as_tail(&local_free_list);
+        SecondaryFreeList_lock->notify_all();
+      }
+
+      if (G1StressConcRegionFreeing) {
+        for (uintx i = 0; i < G1StressConcRegionFreeingDelayMillis; ++i) {
+          os::sleep(Thread::current(), (jlong) 1, false);
+        }
+      }
     }
   }
+  assert(local_free_list.is_empty(), "post-condition");
 }
 
 bool G1CMIsAliveClosure::do_object_b(oop obj) {
@@ -2894,9 +2922,9 @@
   virtual void do_oop(      oop* p) { do_oop_work(p); }
 
   template <class T> void do_oop_work(T* p) {
-    assert(_g1h->is_in_g1_reserved((HeapWord*) p), "invariant");
-    assert(!_g1h->heap_region_containing((HeapWord*) p)->is_on_free_list(),
-           "invariant");
+    assert( _g1h->is_in_g1_reserved((HeapWord*) p), "invariant");
+    assert(!_g1h->is_on_free_list(
+                    _g1h->heap_region_containing((HeapWord*) p)), "invariant");
 
     oop obj = oopDesc::load_decode_heap_oop(p);
     if (_cm->verbose_high())
@@ -3116,8 +3144,8 @@
 void CMTask::push(oop obj) {
   HeapWord* objAddr = (HeapWord*) obj;
   assert(_g1h->is_in_g1_reserved(objAddr), "invariant");
-  assert(!_g1h->heap_region_containing(objAddr)->is_on_free_list(),
-         "invariant");
+  assert(!_g1h->is_on_free_list(
+              _g1h->heap_region_containing((HeapWord*) objAddr)), "invariant");
   assert(!_g1h->is_obj_ill(obj), "invariant");
   assert(_nextMarkBitMap->isMarked(objAddr), "invariant");
 
@@ -3362,8 +3390,8 @@
                                (void*) obj);
 
       assert(_g1h->is_in_g1_reserved((HeapWord*) obj), "invariant" );
-      assert(!_g1h->heap_region_containing(obj)->is_on_free_list(),
-             "invariant");
+      assert(!_g1h->is_on_free_list(
+                  _g1h->heap_region_containing((HeapWord*) obj)), "invariant");
 
       scan_object(obj);
 
--- a/src/share/vm/gc_implementation/g1/concurrentMark.hpp	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/share/vm/gc_implementation/g1/concurrentMark.hpp	Tue Feb 01 18:57:53 2011 -0800
@@ -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
@@ -25,7 +25,7 @@
 #ifndef SHARE_VM_GC_IMPLEMENTATION_G1_CONCURRENTMARK_HPP
 #define SHARE_VM_GC_IMPLEMENTATION_G1_CONCURRENTMARK_HPP
 
-#include "gc_implementation/g1/heapRegion.hpp"
+#include "gc_implementation/g1/heapRegionSets.hpp"
 #include "utilities/taskqueue.hpp"
 
 class G1CollectedHeap;
@@ -369,13 +369,7 @@
   double                _cleanup_sleep_factor;
   double                _cleanup_task_overhead;
 
-  // Stuff related to age cohort processing.
-  struct ParCleanupThreadState {
-    char _pre[64];
-    UncleanRegionList list;
-    char _post[64];
-  };
-  ParCleanupThreadState** _par_cleanup_thread_state;
+  FreeRegionList        _cleanup_list;
 
   // CMS marking support structures
   CMBitMap                _markBitMap1;
@@ -484,6 +478,10 @@
   // prints all gathered CM-related statistics
   void print_stats();
 
+  bool cleanup_list_is_empty() {
+    return _cleanup_list.is_empty();
+  }
+
   // accessor methods
   size_t parallel_marking_threads() { return _parallel_marking_threads; }
   double sleep_factor()             { return _sleep_factor; }
--- a/src/share/vm/gc_implementation/g1/concurrentMarkThread.cpp	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/share/vm/gc_implementation/g1/concurrentMarkThread.cpp	Tue Feb 01 18:57:53 2011 -0800
@@ -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
@@ -95,8 +95,8 @@
   _vtime_start = os::elapsedVTime();
   wait_for_universe_init();
 
-  G1CollectedHeap* g1 = G1CollectedHeap::heap();
-  G1CollectorPolicy* g1_policy = g1->g1_policy();
+  G1CollectedHeap* g1h = G1CollectedHeap::heap();
+  G1CollectorPolicy* g1_policy = g1h->g1_policy();
   G1MMUTracker *mmu_tracker = g1_policy->mmu_tracker();
   Thread *current_thread = Thread::current();
 
@@ -119,7 +119,7 @@
       if (!g1_policy->in_young_gc_mode()) {
         // this ensures the flag is not set if we bail out of the marking
         // cycle; normally the flag is cleared immediately after cleanup
-        g1->set_marking_complete();
+        g1h->set_marking_complete();
 
         if (g1_policy->adaptive_young_list_length()) {
           double now = os::elapsedTime();
@@ -228,10 +228,20 @@
         VM_CGC_Operation op(&cl_cl, verbose_str);
         VMThread::execute(&op);
       } else {
-        G1CollectedHeap::heap()->set_marking_complete();
+        g1h->set_marking_complete();
       }
 
-      if (!cm()->has_aborted()) {
+      // Check if cleanup set the free_regions_coming flag. If it
+      // hasn't, we can just skip the next step.
+      if (g1h->free_regions_coming()) {
+        // The following will finish freeing up any regions that we
+        // found to be empty during cleanup. We'll do this part
+        // without joining the suspendible set. If an evacuation pause
+        // takes places, then we would carry on freeing regions in
+        // case they are needed by the pause. If a Full GC takes
+        // places, it would wait for us to process the regions
+        // reclaimed by cleanup.
+
         double cleanup_start_sec = os::elapsedTime();
         if (PrintGC) {
           gclog_or_tty->date_stamp(PrintGCDateStamps);
@@ -240,23 +250,22 @@
         }
 
         // Now do the remainder of the cleanup operation.
-        _sts.join();
         _cm->completeCleanup();
-        if (!cm()->has_aborted()) {
-          g1_policy->record_concurrent_mark_cleanup_completed();
+        g1_policy->record_concurrent_mark_cleanup_completed();
 
-          double cleanup_end_sec = os::elapsedTime();
-          if (PrintGC) {
-            gclog_or_tty->date_stamp(PrintGCDateStamps);
-            gclog_or_tty->stamp(PrintGCTimeStamps);
-            gclog_or_tty->print_cr("[GC concurrent-cleanup-end, %1.7lf]",
-                                   cleanup_end_sec - cleanup_start_sec);
-          }
+        double cleanup_end_sec = os::elapsedTime();
+        if (PrintGC) {
+          gclog_or_tty->date_stamp(PrintGCDateStamps);
+          gclog_or_tty->stamp(PrintGCTimeStamps);
+          gclog_or_tty->print_cr("[GC concurrent-cleanup-end, %1.7lf]",
+                                 cleanup_end_sec - cleanup_start_sec);
         }
-        _sts.leave();
+
+        // We're done: no more free regions coming.
+        g1h->reset_free_regions_coming();
       }
-      // We're done: no more unclean regions coming.
-      G1CollectedHeap::heap()->set_unclean_regions_coming(false);
+      guarantee(cm()->cleanup_list_is_empty(),
+                "at this point there should be no regions on the cleanup list");
 
       if (cm()->has_aborted()) {
         if (PrintGC) {
@@ -278,7 +287,7 @@
     // Java thread is waiting for a full GC to happen (e.g., it
     // called System.gc() with +ExplicitGCInvokesConcurrent).
     _sts.join();
-    g1->increment_full_collections_completed(true /* concurrent */);
+    g1h->increment_full_collections_completed(true /* concurrent */);
     _sts.leave();
   }
   assert(_should_terminate, "just checking");
--- a/src/share/vm/gc_implementation/g1/concurrentZFThread.cpp	Thu Jan 27 17:28:34 2011 -0800
+++ /dev/null	Thu Jan 01 00:00:00 1970 +0000
@@ -1,194 +0,0 @@
-/*
- * Copyright (c) 2001, 2010, 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
- * under the terms of the GNU General Public License version 2 only, as
- * published by the Free Software Foundation.
- *
- * This code is distributed in the hope that it will be useful, but WITHOUT
- * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
- * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
- * version 2 for more details (a copy is included in the LICENSE file that
- * accompanied this code).
- *
- * You should have received a copy of the GNU General Public License version
- * 2 along with this work; if not, write to the Free Software Foundation,
- * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
- *
- * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
- * or visit www.oracle.com if you need additional information or have any
- * questions.
- *
- */
-
-#include "precompiled.hpp"
-#include "gc_implementation/g1/concurrentZFThread.hpp"
-#include "gc_implementation/g1/g1CollectedHeap.inline.hpp"
-#include "gc_implementation/g1/heapRegion.hpp"
-#include "memory/space.inline.hpp"
-#include "runtime/mutexLocker.hpp"
-#include "utilities/copy.hpp"
-
-// ======= Concurrent Zero-Fill Thread ========
-
-// The CM thread is created when the G1 garbage collector is used
-
-int ConcurrentZFThread::_region_allocs = 0;
-int ConcurrentZFThread::_sync_zfs = 0;
-int ConcurrentZFThread::_zf_waits = 0;
-int ConcurrentZFThread::_regions_filled = 0;
-
-ConcurrentZFThread::ConcurrentZFThread() :
-  ConcurrentGCThread()
-{
-  create_and_start();
-}
-
-void ConcurrentZFThread::wait_for_ZF_completed(HeapRegion* hr) {
-  assert(ZF_mon->owned_by_self(), "Precondition.");
-  note_zf_wait();
-  while (hr->zero_fill_state() == HeapRegion::ZeroFilling) {
-    ZF_mon->wait(Mutex::_no_safepoint_check_flag);
-  }
-}
-
-void ConcurrentZFThread::processHeapRegion(HeapRegion* hr) {
-  assert(!Universe::heap()->is_gc_active(),
-         "This should not happen during GC.");
-  assert(hr != NULL, "Precondition");
-  // These are unlocked reads, but if this test is successful, then no
-  // other thread will attempt this zero filling.  Only a GC thread can
-  // modify the ZF state of a region whose state is zero-filling, and this
-  // should only happen while the ZF thread is locking out GC.
-  if (hr->zero_fill_state() == HeapRegion::ZeroFilling
-      && hr->zero_filler() == Thread::current()) {
-    assert(hr->top() == hr->bottom(), "better be empty!");
-    assert(!hr->isHumongous(), "Only free regions on unclean list.");
-    Copy::fill_to_words(hr->bottom(), hr->capacity()/HeapWordSize);
-    note_region_filled();
-  }
-}
-
-void ConcurrentZFThread::run() {
-  initialize_in_thread();
-  Thread* thr_self = Thread::current();
-  _vtime_start = os::elapsedVTime();
-  wait_for_universe_init();
-
-  G1CollectedHeap* g1 = G1CollectedHeap::heap();
-  _sts.join();
-  while (!_should_terminate) {
-    _sts.leave();
-
-    {
-      MutexLockerEx x(ZF_mon, Mutex::_no_safepoint_check_flag);
-
-      // This local variable will hold a region being zero-filled.  This
-      // region will neither be on the unclean or zero-filled lists, and
-      // will not be available for allocation; thus, we might have an
-      // allocation fail, causing a full GC, because of this, but this is a
-      // price we will pay.  (In future, we might want to make the fact
-      // that there's a region being zero-filled apparent to the G1 heap,
-      // which could then wait for it in this extreme case...)
-      HeapRegion* to_fill;
-
-      while (!g1->should_zf()
-             || (to_fill = g1->pop_unclean_region_list_locked()) == NULL)
-        ZF_mon->wait(Mutex::_no_safepoint_check_flag);
-      while (to_fill->zero_fill_state() == HeapRegion::ZeroFilling)
-        ZF_mon->wait(Mutex::_no_safepoint_check_flag);
-
-      // So now to_fill is non-NULL and is not ZeroFilling.  It might be
-      // Allocated or ZeroFilled.  (The latter could happen if this thread
-      // starts the zero-filling of a region, but a GC intervenes and
-      // pushes new regions needing on the front of the filling on the
-      // front of the list.)
-
-      switch (to_fill->zero_fill_state()) {
-      case HeapRegion::Allocated:
-        to_fill = NULL;
-        break;
-
-      case HeapRegion::NotZeroFilled:
-        to_fill->set_zero_fill_in_progress(thr_self);
-
-        ZF_mon->unlock();
-        _sts.join();
-        processHeapRegion(to_fill);
-        _sts.leave();
-        ZF_mon->lock_without_safepoint_check();
-
-        if (to_fill->zero_fill_state() == HeapRegion::ZeroFilling
-            && to_fill->zero_filler() == thr_self) {
-          to_fill->set_zero_fill_complete();
-          (void)g1->put_free_region_on_list_locked(to_fill);
-        }
-        break;
-
-      case HeapRegion::ZeroFilled:
-        (void)g1->put_free_region_on_list_locked(to_fill);
-        break;
-
-      case HeapRegion::ZeroFilling:
-        ShouldNotReachHere();
-        break;
-      }
-    }
-    _vtime_accum = (os::elapsedVTime() - _vtime_start);
-    _sts.join();
-  }
-  _sts.leave();
-
-  assert(_should_terminate, "just checking");
-  terminate();
-}
-
-bool ConcurrentZFThread::offer_yield() {
-  if (_sts.should_yield()) {
-    _sts.yield("Concurrent ZF");
-    return true;
-  } else {
-    return false;
-  }
-}
-
-void ConcurrentZFThread::stop() {
-  // it is ok to take late safepoints here, if needed
-  MutexLockerEx mu(Terminator_lock);
-  _should_terminate = true;
-  while (!_has_terminated) {
-    Terminator_lock->wait();
-  }
-}
-
-void ConcurrentZFThread::print() const {
-  print_on(tty);
-}
-
-void ConcurrentZFThread::print_on(outputStream* st) const {
-  st->print("\"G1 Concurrent Zero-Fill Thread\" ");
-  Thread::print_on(st);
-  st->cr();
-}
-
-
-double ConcurrentZFThread::_vtime_accum;
-
-void ConcurrentZFThread::print_summary_info() {
-  gclog_or_tty->print("\nConcurrent Zero-Filling:\n");
-  gclog_or_tty->print("  Filled %d regions, used %5.2fs.\n",
-                      _regions_filled,
-                      vtime_accum());
-  gclog_or_tty->print("  Of %d region allocs, %d (%5.2f%%) required sync ZF,\n",
-                      _region_allocs, _sync_zfs,
-                      (_region_allocs > 0 ?
-                       (float)_sync_zfs/(float)_region_allocs*100.0 :
-                       0.0));
-  gclog_or_tty->print("     and %d (%5.2f%%) required a ZF wait.\n",
-                      _zf_waits,
-                      (_region_allocs > 0 ?
-                       (float)_zf_waits/(float)_region_allocs*100.0 :
-                       0.0));
-
-}
--- a/src/share/vm/gc_implementation/g1/concurrentZFThread.hpp	Thu Jan 27 17:28:34 2011 -0800
+++ /dev/null	Thu Jan 01 00:00:00 1970 +0000
@@ -1,91 +0,0 @@
-/*
- * Copyright (c) 2001, 2010, 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
- * under the terms of the GNU General Public License version 2 only, as
- * published by the Free Software Foundation.
- *
- * This code is distributed in the hope that it will be useful, but WITHOUT
- * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
- * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
- * version 2 for more details (a copy is included in the LICENSE file that
- * accompanied this code).
- *
- * You should have received a copy of the GNU General Public License version
- * 2 along with this work; if not, write to the Free Software Foundation,
- * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
- *
- * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
- * or visit www.oracle.com if you need additional information or have any
- * questions.
- *
- */
-
-#ifndef SHARE_VM_GC_IMPLEMENTATION_G1_CONCURRENTZFTHREAD_HPP
-#define SHARE_VM_GC_IMPLEMENTATION_G1_CONCURRENTZFTHREAD_HPP
-
-#include "gc_implementation/shared/concurrentGCThread.hpp"
-
-// The Concurrent ZF Thread.  Performs concurrent zero-filling.
-
-class ConcurrentZFThread: public ConcurrentGCThread {
-  friend class VMStructs;
-  friend class ZeroFillRegionClosure;
-
- private:
-
-  // Zero fill the heap region.
-  void processHeapRegion(HeapRegion* r);
-
-  // Stats
-  //   Allocation (protected by heap lock).
-  static int _region_allocs;  // Number of regions allocated
-  static int _sync_zfs;       //   Synchronous zero-fills +
-  static int _zf_waits;      //   Wait for conc zero-fill completion.
-
-  // Number of regions CFZ thread fills.
-  static int _regions_filled;
-
-  double _vtime_start;  // Initial virtual time.
-
-  // These are static because the "print_summary_info" method is, and
-  // it currently assumes there is only one ZF thread.  We'll change when
-  // we need to.
-  static double _vtime_accum;  // Initial virtual time.
-  static double vtime_accum() { return _vtime_accum; }
-
-  // Offer yield for GC.  Returns true if yield occurred.
-  bool offer_yield();
-
- public:
-  // Constructor
-  ConcurrentZFThread();
-
-  // Main loop.
-  virtual void run();
-
-  // Printing
-  void print_on(outputStream* st) const;
-  void print() const;
-
-  // Waits until "r" has been zero-filled.  Requires caller to hold the
-  // ZF_mon.
-  static void wait_for_ZF_completed(HeapRegion* r);
-
-  // Get or clear the current unclean region.  Should be done
-  // while holding the ZF_needed_mon lock.
-
-  // shutdown
-  void stop();
-
-  // Stats
-  static void note_region_alloc() {_region_allocs++; }
-  static void note_sync_zfs() { _sync_zfs++; }
-  static void note_zf_wait() { _zf_waits++; }
-  static void note_region_filled() { _regions_filled++; }
-
-  static void print_summary_info();
-};
-
-#endif // SHARE_VM_GC_IMPLEMENTATION_G1_CONCURRENTZFTHREAD_HPP
--- a/src/share/vm/gc_implementation/g1/g1BlockOffsetTable.cpp	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/share/vm/gc_implementation/g1/g1BlockOffsetTable.cpp	Tue Feb 01 18:57:53 2011 -0800
@@ -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
@@ -222,7 +222,7 @@
 
 // Action_mark - update the BOT for the block [blk_start, blk_end).
 //               Current typical use is for splitting a block.
-// Action_single - udpate the BOT for an allocation.
+// Action_single - update the BOT for an allocation.
 // Action_verify - BOT verification.
 void G1BlockOffsetArray::do_block_internal(HeapWord* blk_start,
                                            HeapWord* blk_end,
@@ -331,47 +331,6 @@
   do_block_internal(blk_start, blk_end, Action_mark);
 }
 
-void G1BlockOffsetArray::join_blocks(HeapWord* blk1, HeapWord* blk2) {
-  HeapWord* blk1_start = Universe::heap()->block_start(blk1);
-  HeapWord* blk2_start = Universe::heap()->block_start(blk2);
-  assert(blk1 == blk1_start && blk2 == blk2_start,
-         "Must be block starts.");
-  assert(blk1 + _sp->block_size(blk1) == blk2, "Must be contiguous.");
-  size_t blk1_start_index = _array->index_for(blk1);
-  size_t blk2_start_index = _array->index_for(blk2);
-  assert(blk1_start_index <= blk2_start_index, "sanity");
-  HeapWord* blk2_card_start = _array->address_for_index(blk2_start_index);
-  if (blk2 == blk2_card_start) {
-    // blk2 starts a card.  Does blk1 start on the prevous card, or futher
-    // back?
-    assert(blk1_start_index < blk2_start_index, "must be lower card.");
-    if (blk1_start_index + 1 == blk2_start_index) {
-      // previous card; new value for blk2 card is size of blk1.
-      _array->set_offset_array(blk2_start_index, (u_char) _sp->block_size(blk1));
-    } else {
-      // Earlier card; go back a card.
-      _array->set_offset_array(blk2_start_index, N_words);
-    }
-  } else {
-    // blk2 does not start a card.  Does it cross a card?  If not, nothing
-    // to do.
-    size_t blk2_end_index =
-      _array->index_for(blk2 + _sp->block_size(blk2) - 1);
-    assert(blk2_end_index >= blk2_start_index, "sanity");
-    if (blk2_end_index > blk2_start_index) {
-      // Yes, it crosses a card.  The value for the next card must change.
-      if (blk1_start_index + 1 == blk2_start_index) {
-        // previous card; new value for second blk2 card is size of blk1.
-        _array->set_offset_array(blk2_start_index + 1,
-                                 (u_char) _sp->block_size(blk1));
-      } else {
-        // Earlier card; go back a card.
-        _array->set_offset_array(blk2_start_index + 1, N_words);
-      }
-    }
-  }
-}
-
 HeapWord* G1BlockOffsetArray::block_start_unsafe(const void* addr) {
   assert(_bottom <= addr && addr < _end,
          "addr must be covered by this Array");
@@ -580,15 +539,50 @@
 #endif
 }
 
+bool
+G1BlockOffsetArray::verify_for_object(HeapWord* obj_start,
+                                      size_t word_size) const {
+  size_t first_card = _array->index_for(obj_start);
+  size_t last_card = _array->index_for(obj_start + word_size - 1);
+  if (!_array->is_card_boundary(obj_start)) {
+    // If the object is not on a card boundary the BOT entry of the
+    // first card should point to another object so we should not
+    // check that one.
+    first_card += 1;
+  }
+  for (size_t card = first_card; card <= last_card; card += 1) {
+    HeapWord* card_addr = _array->address_for_index(card);
+    HeapWord* block_start = block_start_const(card_addr);
+    if (block_start != obj_start) {
+      gclog_or_tty->print_cr("block start: "PTR_FORMAT" is incorrect - "
+                             "card index: "SIZE_FORMAT" "
+                             "card addr: "PTR_FORMAT" BOT entry: %u "
+                             "obj: "PTR_FORMAT" word size: "SIZE_FORMAT" "
+                             "cards: ["SIZE_FORMAT","SIZE_FORMAT"]",
+                             block_start, card, card_addr,
+                             _array->offset_array(card),
+                             obj_start, word_size, first_card, last_card);
+      return false;
+    }
+  }
+  return true;
+}
+
+#ifndef PRODUCT
 void
-G1BlockOffsetArray::set_for_starts_humongous(HeapWord* new_end) {
-  assert(_end ==  new_end, "_end should have already been updated");
-
-  // The first BOT entry should have offset 0.
-  _array->set_offset_array(_array->index_for(_bottom), 0);
-  // The rest should point to the first one.
-  set_remainder_to_point_to_start(_bottom + N_words, new_end);
+G1BlockOffsetArray::print_on(outputStream* out) {
+  size_t from_index = _array->index_for(_bottom);
+  size_t to_index = _array->index_for(_end);
+  out->print_cr(">> BOT for area ["PTR_FORMAT","PTR_FORMAT") "
+                "cards ["SIZE_FORMAT","SIZE_FORMAT")",
+                _bottom, _end, from_index, to_index);
+  for (size_t i = from_index; i < to_index; ++i) {
+    out->print_cr("  entry "SIZE_FORMAT_W(8)" | "PTR_FORMAT" : %3u",
+                  i, _array->address_for_index(i),
+                  (uint) _array->offset_array(i));
+  }
 }
+#endif // !PRODUCT
 
 //////////////////////////////////////////////////////////////////////
 // G1BlockOffsetArrayContigSpace
@@ -641,10 +635,20 @@
 }
 
 void
-G1BlockOffsetArrayContigSpace::set_for_starts_humongous(HeapWord* new_end) {
-  G1BlockOffsetArray::set_for_starts_humongous(new_end);
+G1BlockOffsetArrayContigSpace::set_for_starts_humongous(HeapWord* new_top) {
+  assert(new_top <= _end, "_end should have already been updated");
 
-  // Make sure _next_offset_threshold and _next_offset_index point to new_end.
-  _next_offset_threshold = new_end;
-  _next_offset_index     = _array->index_for(new_end);
+  // The first BOT entry should have offset 0.
+  zero_bottom_entry();
+  initialize_threshold();
+  alloc_block(_bottom, new_top);
+ }
+
+#ifndef PRODUCT
+void
+G1BlockOffsetArrayContigSpace::print_on(outputStream* out) {
+  G1BlockOffsetArray::print_on(out);
+  out->print_cr("  next offset threshold: "PTR_FORMAT, _next_offset_threshold);
+  out->print_cr("  next offset index:     "SIZE_FORMAT, _next_offset_index);
 }
+#endif // !PRODUCT
--- a/src/share/vm/gc_implementation/g1/g1BlockOffsetTable.hpp	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/share/vm/gc_implementation/g1/g1BlockOffsetTable.hpp	Tue Feb 01 18:57:53 2011 -0800
@@ -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
@@ -352,11 +352,6 @@
   // The following methods are useful and optimized for a
   // general, non-contiguous space.
 
-  // The given arguments are required to be the starts of adjacent ("blk1"
-  // before "blk2") well-formed blocks covered by "this".  After this call,
-  // they should be considered to form one block.
-  virtual void join_blocks(HeapWord* blk1, HeapWord* blk2);
-
   // Given a block [blk_start, blk_start + full_blk_size), and
   // a left_blk_size < full_blk_size, adjust the BOT to show two
   // blocks [blk_start, blk_start + left_blk_size) and
@@ -429,6 +424,12 @@
     verify_single_block(blk, blk + size);
   }
 
+  // Used by region verification. Checks that the contents of the
+  // BOT reflect that there's a single object that spans the address
+  // range [obj_start, obj_start + word_size); returns true if this is
+  // the case, returns false if it's not.
+  bool verify_for_object(HeapWord* obj_start, size_t word_size) const;
+
   // Verify that the given block is before _unallocated_block
   inline void verify_not_unallocated(HeapWord* blk_start,
                                      HeapWord* blk_end) const {
@@ -444,7 +445,7 @@
 
   void check_all_cards(size_t left_card, size_t right_card) const;
 
-  virtual void set_for_starts_humongous(HeapWord* new_end);
+  virtual void print_on(outputStream* out) PRODUCT_RETURN;
 };
 
 // A subtype of BlockOffsetArray that takes advantage of the fact
@@ -494,7 +495,9 @@
   HeapWord* block_start_unsafe(const void* addr);
   HeapWord* block_start_unsafe_const(const void* addr) const;
 
-  virtual void set_for_starts_humongous(HeapWord* new_end);
+  void set_for_starts_humongous(HeapWord* new_top);
+
+  virtual void print_on(outputStream* out) PRODUCT_RETURN;
 };
 
 #endif // SHARE_VM_GC_IMPLEMENTATION_G1_G1BLOCKOFFSETTABLE_HPP
--- a/src/share/vm/gc_implementation/g1/g1CollectedHeap.cpp	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/share/vm/gc_implementation/g1/g1CollectedHeap.cpp	Tue Feb 01 18:57:53 2011 -0800
@@ -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
@@ -28,7 +28,6 @@
 #include "gc_implementation/g1/concurrentG1Refine.hpp"
 #include "gc_implementation/g1/concurrentG1RefineThread.hpp"
 #include "gc_implementation/g1/concurrentMarkThread.inline.hpp"
-#include "gc_implementation/g1/concurrentZFThread.hpp"
 #include "gc_implementation/g1/g1CollectedHeap.inline.hpp"
 #include "gc_implementation/g1/g1CollectorPolicy.hpp"
 #include "gc_implementation/g1/g1MarkSweep.hpp"
@@ -425,11 +424,9 @@
 
 void G1CollectedHeap::stop_conc_gc_threads() {
   _cg1r->stop();
-  _czft->stop();
   _cmThread->stop();
 }
 
-
 void G1CollectedHeap::check_ct_logs_at_safepoint() {
   DirtyCardQueueSet& dcqs = JavaThread::dirty_card_queue_set();
   CardTableModRefBS* ct_bs = (CardTableModRefBS*)barrier_set();
@@ -481,49 +478,92 @@
 
 // Private methods.
 
-// Finds a HeapRegion that can be used to allocate a given size of block.
-
-
-HeapRegion* G1CollectedHeap::newAllocRegion_work(size_t word_size,
-                                                 bool do_expand,
-                                                 bool zero_filled) {
-  ConcurrentZFThread::note_region_alloc();
-  HeapRegion* res = alloc_free_region_from_lists(zero_filled);
+HeapRegion*
+G1CollectedHeap::new_region_try_secondary_free_list(size_t word_size) {
+  MutexLockerEx x(SecondaryFreeList_lock, Mutex::_no_safepoint_check_flag);
+  while (!_secondary_free_list.is_empty() || free_regions_coming()) {
+    if (!_secondary_free_list.is_empty()) {
+      if (G1ConcRegionFreeingVerbose) {
+        gclog_or_tty->print_cr("G1ConcRegionFreeing [region alloc] : "
+                               "secondary_free_list has "SIZE_FORMAT" entries",
+                               _secondary_free_list.length());
+      }
+      // It looks as if there are free regions available on the
+      // secondary_free_list. Let's move them to the free_list and try
+      // again to allocate from it.
+      append_secondary_free_list();
+
+      assert(!_free_list.is_empty(), "if the secondary_free_list was not "
+             "empty we should have moved at least one entry to the free_list");
+      HeapRegion* res = _free_list.remove_head();
+      if (G1ConcRegionFreeingVerbose) {
+        gclog_or_tty->print_cr("G1ConcRegionFreeing [region alloc] : "
+                               "allocated "HR_FORMAT" from secondary_free_list",
+                               HR_FORMAT_PARAMS(res));
+      }
+      return res;
+    }
+
+    // Wait here until we get notifed either when (a) there are no
+    // more free regions coming or (b) some regions have been moved on
+    // the secondary_free_list.
+    SecondaryFreeList_lock->wait(Mutex::_no_safepoint_check_flag);
+  }
+
+  if (G1ConcRegionFreeingVerbose) {
+    gclog_or_tty->print_cr("G1ConcRegionFreeing [region alloc] : "
+                           "could not allocate from secondary_free_list");
+  }
+  return NULL;
+}
+
+HeapRegion* G1CollectedHeap::new_region_work(size_t word_size,
+                                             bool do_expand) {
+  assert(!isHumongous(word_size) ||
+                                  word_size <= (size_t) HeapRegion::GrainWords,
+         "the only time we use this to allocate a humongous region is "
+         "when we are allocating a single humongous region");
+
+  HeapRegion* res;
+  if (G1StressConcRegionFreeing) {
+    if (!_secondary_free_list.is_empty()) {
+      if (G1ConcRegionFreeingVerbose) {
+        gclog_or_tty->print_cr("G1ConcRegionFreeing [region alloc] : "
+                               "forced to look at the secondary_free_list");
+      }
+      res = new_region_try_secondary_free_list(word_size);
+      if (res != NULL) {
+        return res;
+      }
+    }
+  }
+  res = _free_list.remove_head_or_null();
+  if (res == NULL) {
+    if (G1ConcRegionFreeingVerbose) {
+      gclog_or_tty->print_cr("G1ConcRegionFreeing [region alloc] : "
+                             "res == NULL, trying the secondary_free_list");
+    }
+    res = new_region_try_secondary_free_list(word_size);
+  }
   if (res == NULL && do_expand) {
     expand(word_size * HeapWordSize);
-    res = alloc_free_region_from_lists(zero_filled);
-    assert(res == NULL ||
-           (!res->isHumongous() &&
-            (!zero_filled ||
-             res->zero_fill_state() == HeapRegion::Allocated)),
-           "Alloc Regions must be zero filled (and non-H)");
+    res = _free_list.remove_head_or_null();
   }
   if (res != NULL) {
-    if (res->is_empty()) {
-      _free_regions--;
-    }
-    assert(!res->isHumongous() &&
-           (!zero_filled || res->zero_fill_state() == HeapRegion::Allocated),
-           err_msg("Non-young alloc Regions must be zero filled (and non-H):"
-                   " res->isHumongous()=%d, zero_filled=%d, res->zero_fill_state()=%d",
-                   res->isHumongous(), zero_filled, res->zero_fill_state()));
-    assert(!res->is_on_unclean_list(),
-           "Alloc Regions must not be on the unclean list");
     if (G1PrintHeapRegions) {
-      gclog_or_tty->print_cr("new alloc region %d:["PTR_FORMAT", "PTR_FORMAT"], "
-                             "top "PTR_FORMAT,
-                             res->hrs_index(), res->bottom(), res->end(), res->top());
+      gclog_or_tty->print_cr("new alloc region %d:["PTR_FORMAT","PTR_FORMAT"], "
+                             "top "PTR_FORMAT, res->hrs_index(),
+                             res->bottom(), res->end(), res->top());
     }
   }
   return res;
 }
 
-HeapRegion* G1CollectedHeap::newAllocRegionWithExpansion(int purpose,
-                                                         size_t word_size,
-                                                         bool zero_filled) {
+HeapRegion* G1CollectedHeap::new_gc_alloc_region(int purpose,
+                                                 size_t word_size) {
   HeapRegion* alloc_region = NULL;
   if (_gc_alloc_region_counts[purpose] < g1_policy()->max_regions(purpose)) {
-    alloc_region = newAllocRegion_work(word_size, true, zero_filled);
+    alloc_region = new_region_work(word_size, true /* do_expand */);
     if (purpose == GCAllocForSurvived && alloc_region != NULL) {
       alloc_region->set_survivor();
     }
@@ -534,81 +574,220 @@
   return alloc_region;
 }
 
+int G1CollectedHeap::humongous_obj_allocate_find_first(size_t num_regions,
+                                                       size_t word_size) {
+  int first = -1;
+  if (num_regions == 1) {
+    // Only one region to allocate, no need to go through the slower
+    // path. The caller will attempt the expasion if this fails, so
+    // let's not try to expand here too.
+    HeapRegion* hr = new_region_work(word_size, false /* do_expand */);
+    if (hr != NULL) {
+      first = hr->hrs_index();
+    } else {
+      first = -1;
+    }
+  } else {
+    // We can't allocate humongous regions while cleanupComplete() is
+    // running, since some of the regions we find to be empty might not
+    // yet be added to the free list and it is not straightforward to
+    // know which list they are on so that we can remove them. Note
+    // that we only need to do this if we need to allocate more than
+    // one region to satisfy the current humongous allocation
+    // request. If we are only allocating one region we use the common
+    // region allocation code (see above).
+    wait_while_free_regions_coming();
+    append_secondary_free_list_if_not_empty();
+
+    if (free_regions() >= num_regions) {
+      first = _hrs->find_contiguous(num_regions);
+      if (first != -1) {
+        for (int i = first; i < first + (int) num_regions; ++i) {
+          HeapRegion* hr = _hrs->at(i);
+          assert(hr->is_empty(), "sanity");
+          assert(is_on_free_list(hr), "sanity");
+          hr->set_pending_removal(true);
+        }
+        _free_list.remove_all_pending(num_regions);
+      }
+    }
+  }
+  return first;
+}
+
 // If could fit into free regions w/o expansion, try.
 // Otherwise, if can expand, do so.
 // Otherwise, if using ex regions might help, try with ex given back.
 HeapWord* G1CollectedHeap::humongous_obj_allocate(size_t word_size) {
-  assert_heap_locked_or_at_safepoint();
-  assert(regions_accounted_for(), "Region leakage!");
-
-  // We can't allocate humongous regions while cleanupComplete is
-  // running, since some of the regions we find to be empty might not
-  // yet be added to the unclean list. If we're already at a
-  // safepoint, this call is unnecessary, not to mention wrong.
-  if (!SafepointSynchronize::is_at_safepoint()) {
-    wait_for_cleanup_complete();
-  }
+  assert_heap_locked_or_at_safepoint(true /* should_be_vm_thread */);
+
+  verify_region_sets_optional();
 
   size_t num_regions =
          round_to(word_size, HeapRegion::GrainWords) / HeapRegion::GrainWords;
-
-  // Special case if < one region???
-
-  // Remember the ft size.
   size_t x_size = expansion_regions();
-
-  HeapWord* res = NULL;
-  bool eliminated_allocated_from_lists = false;
-
-  // Can the allocation potentially fit in the free regions?
-  if (free_regions() >= num_regions) {
-    res = _hrs->obj_allocate(word_size);
-  }
-  if (res == NULL) {
-    // Try expansion.
-    size_t fs = _hrs->free_suffix();
+  size_t fs = _hrs->free_suffix();
+  int first = humongous_obj_allocate_find_first(num_regions, word_size);
+  if (first == -1) {
+    // The only thing we can do now is attempt expansion.
     if (fs + x_size >= num_regions) {
       expand((num_regions - fs) * HeapRegion::GrainBytes);
-      res = _hrs->obj_allocate(word_size);
-      assert(res != NULL, "This should have worked.");
-    } else {
-      // Expansion won't help.  Are there enough free regions if we get rid
-      // of reservations?
-      size_t avail = free_regions();
-      if (avail >= num_regions) {
-        res = _hrs->obj_allocate(word_size);
-        if (res != NULL) {
-          remove_allocated_regions_from_lists();
-          eliminated_allocated_from_lists = true;
-        }
+      first = humongous_obj_allocate_find_first(num_regions, word_size);
+      assert(first != -1, "this should have worked");
+    }
+  }
+
+  if (first != -1) {
+    // Index of last region in the series + 1.
+    int last = first + (int) num_regions;
+
+    // We need to initialize the region(s) we just discovered. This is
+    // a bit tricky given that it can happen concurrently with
+    // refinement threads refining cards on these regions and
+    // potentially wanting to refine the BOT as they are scanning
+    // those cards (this can happen shortly after a cleanup; see CR
+    // 6991377). So we have to set up the region(s) carefully and in
+    // a specific order.
+
+    // The word size sum of all the regions we will allocate.
+    size_t word_size_sum = num_regions * HeapRegion::GrainWords;
+    assert(word_size <= word_size_sum, "sanity");
+
+    // This will be the "starts humongous" region.
+    HeapRegion* first_hr = _hrs->at(first);
+    // The header of the new object will be placed at the bottom of
+    // the first region.
+    HeapWord* new_obj = first_hr->bottom();
+    // This will be the new end of the first region in the series that
+    // should also match the end of the last region in the seriers.
+    HeapWord* new_end = new_obj + word_size_sum;
+    // This will be the new top of the first region that will reflect
+    // this allocation.
+    HeapWord* new_top = new_obj + word_size;
+
+    // First, we need to zero the header of the space that we will be
+    // allocating. When we update top further down, some refinement
+    // threads might try to scan the region. By zeroing the header we
+    // ensure that any thread that will try to scan the region will
+    // come across the zero klass word and bail out.
+    //
+    // NOTE: It would not have been correct to have used
+    // CollectedHeap::fill_with_object() and make the space look like
+    // an int array. The thread that is doing the allocation will
+    // later update the object header to a potentially different array
+    // type and, for a very short period of time, the klass and length
+    // fields will be inconsistent. This could cause a refinement
+    // thread to calculate the object size incorrectly.
+    Copy::fill_to_words(new_obj, oopDesc::header_size(), 0);
+
+    // We will set up the first region as "starts humongous". This
+    // will also update the BOT covering all the regions to reflect
+    // that there is a single object that starts at the bottom of the
+    // first region.
+    first_hr->set_startsHumongous(new_top, new_end);
+
+    // Then, if there are any, we will set up the "continues
+    // humongous" regions.
+    HeapRegion* hr = NULL;
+    for (int i = first + 1; i < last; ++i) {
+      hr = _hrs->at(i);
+      hr->set_continuesHumongous(first_hr);
+    }
+    // If we have "continues humongous" regions (hr != NULL), then the
+    // end of the last one should match new_end.
+    assert(hr == NULL || hr->end() == new_end, "sanity");
+
+    // Up to this point no concurrent thread would have been able to
+    // do any scanning on any region in this series. All the top
+    // fields still point to bottom, so the intersection between
+    // [bottom,top] and [card_start,card_end] will be empty. Before we
+    // update the top fields, we'll do a storestore to make sure that
+    // no thread sees the update to top before the zeroing of the
+    // object header and the BOT initialization.
+    OrderAccess::storestore();
+
+    // Now that the BOT and the object header have been initialized,
+    // we can update top of the "starts humongous" region.
+    assert(first_hr->bottom() < new_top && new_top <= first_hr->end(),
+           "new_top should be in this region");
+    first_hr->set_top(new_top);
+
+    // Now, we will update the top fields of the "continues humongous"
+    // regions. The reason we need to do this is that, otherwise,
+    // these regions would look empty and this will confuse parts of
+    // G1. For example, the code that looks for a consecutive number
+    // of empty regions will consider them empty and try to
+    // re-allocate them. We can extend is_empty() to also include
+    // !continuesHumongous(), but it is easier to just update the top
+    // fields here. The way we set top for all regions (i.e., top ==
+    // end for all regions but the last one, top == new_top for the
+    // last one) is actually used when we will free up the humongous
+    // region in free_humongous_region().
+    hr = NULL;
+    for (int i = first + 1; i < last; ++i) {
+      hr = _hrs->at(i);
+      if ((i + 1) == last) {
+        // last continues humongous region
+        assert(hr->bottom() < new_top && new_top <= hr->end(),
+               "new_top should fall on this region");
+        hr->set_top(new_top);
+      } else {
+        // not last one
+        assert(new_top > hr->end(), "new_top should be above this region");
+        hr->set_top(hr->end());
       }
     }
-  }
-  if (res != NULL) {
-    // Increment by the number of regions allocated.
-    // FIXME: Assumes regions all of size GrainBytes.
-#ifndef PRODUCT
-    mr_bs()->verify_clean_region(MemRegion(res, res + num_regions *
-                                           HeapRegion::GrainWords));
-#endif
-    if (!eliminated_allocated_from_lists)
-      remove_allocated_regions_from_lists();
-    _summary_bytes_used += word_size * HeapWordSize;
-    _free_regions -= num_regions;
-    _num_humongous_regions += (int) num_regions;
-  }
-  assert(regions_accounted_for(), "Region Leakage");
-  return res;
+    // If we have continues humongous regions (hr != NULL), then the
+    // end of the last one should match new_end and its top should
+    // match new_top.
+    assert(hr == NULL ||
+           (hr->end() == new_end && hr->top() == new_top), "sanity");
+
+    assert(first_hr->used() == word_size * HeapWordSize, "invariant");
+    _summary_bytes_used += first_hr->used();
+    _humongous_set.add(first_hr);
+
+    return new_obj;
+  }
+
+  verify_region_sets_optional();
+  return NULL;
 }
 
 void
 G1CollectedHeap::retire_cur_alloc_region(HeapRegion* cur_alloc_region) {
-  // The cleanup operation might update _summary_bytes_used
-  // concurrently with this method. So, right now, if we don't wait
-  // for it to complete, updates to _summary_bytes_used might get
-  // lost. This will be resolved in the near future when the operation
-  // of the free region list is revamped as part of CR 6977804.
-  wait_for_cleanup_complete();
+  // Other threads might still be trying to allocate using CASes out
+  // of the region we are retiring, as they can do so without holding
+  // the Heap_lock. So we first have to make sure that noone else can
+  // allocate in it by doing a maximal allocation. Even if our CAS
+  // attempt fails a few times, we'll succeed sooner or later given
+  // that a failed CAS attempt mean that the region is getting closed
+  // to being full (someone else succeeded in allocating into it).
+  size_t free_word_size = cur_alloc_region->free() / HeapWordSize;
+
+  // This is the minimum free chunk we can turn into a dummy
+  // object. If the free space falls below this, then noone can
+  // allocate in this region anyway (all allocation requests will be
+  // of a size larger than this) so we won't have to perform the dummy
+  // allocation.
+  size_t min_word_size_to_fill = CollectedHeap::min_fill_size();
+
+  while (free_word_size >= min_word_size_to_fill) {
+    HeapWord* dummy =
+      cur_alloc_region->par_allocate_no_bot_updates(free_word_size);
+    if (dummy != NULL) {
+      // If the allocation was successful we should fill in the space.
+      CollectedHeap::fill_with_object(dummy, free_word_size);
+      break;
+    }
+
+    free_word_size = cur_alloc_region->free() / HeapWordSize;
+    // It's also possible that someone else beats us to the
+    // allocation and they fill up the region. In that case, we can
+    // just get out of the loop
+  }
+  assert(cur_alloc_region->free() / HeapWordSize < min_word_size_to_fill,
+         "sanity");
 
   retire_cur_alloc_region_common(cur_alloc_region);
   assert(_cur_alloc_region == NULL, "post-condition");
@@ -621,7 +800,7 @@
                                                        bool at_safepoint,
                                                        bool do_dirtying,
                                                        bool can_expand) {
-  assert_heap_locked_or_at_safepoint();
+  assert_heap_locked_or_at_safepoint(true /* should_be_vm_thread */);
   assert(_cur_alloc_region == NULL,
          "replace_cur_alloc_region_and_allocate() should only be called "
          "after retiring the previous current alloc region");
@@ -632,25 +811,12 @@
          "we are not allowed to expand the young gen");
 
   if (can_expand || !g1_policy()->is_young_list_full()) {
-    if (!at_safepoint) {
-      // The cleanup operation might update _summary_bytes_used
-      // concurrently with this method. So, right now, if we don't
-      // wait for it to complete, updates to _summary_bytes_used might
-      // get lost. This will be resolved in the near future when the
-      // operation of the free region list is revamped as part of
-      // CR 6977804. If we're already at a safepoint, this call is
-      // unnecessary, not to mention wrong.
-      wait_for_cleanup_complete();
-    }
-
-    HeapRegion* new_cur_alloc_region = newAllocRegion(word_size,
-                                                      false /* zero_filled */);
+    HeapRegion* new_cur_alloc_region = new_alloc_region(word_size);
     if (new_cur_alloc_region != NULL) {
       assert(new_cur_alloc_region->is_empty(),
              "the newly-allocated region should be empty, "
              "as right now we only allocate new regions out of the free list");
       g1_policy()->update_region_num(true /* next_is_young */);
-      _summary_bytes_used -= new_cur_alloc_region->used();
       set_region_short_lived_locked(new_cur_alloc_region);
 
       assert(!new_cur_alloc_region->isHumongous(),
@@ -661,27 +827,29 @@
       // young type.
       OrderAccess::storestore();
 
-      // Now allocate out of the new current alloc region. We could
-      // have re-used allocate_from_cur_alloc_region() but its
-      // operation is slightly different to what we need here. First,
-      // allocate_from_cur_alloc_region() is only called outside a
-      // safepoint and will always unlock the Heap_lock if it returns
-      // a non-NULL result. Second, it assumes that the current alloc
-      // region is what's already assigned in _cur_alloc_region. What
-      // we want here is to actually do the allocation first before we
-      // assign the new region to _cur_alloc_region. This ordering is
-      // not currently important, but it will be essential when we
-      // change the code to support CAS allocation in the future (see
-      // CR 6994297).
-      //
-      // This allocate method does BOT updates and we don't need them in
-      // the young generation. This will be fixed in the near future by
-      // CR 6994297.
-      HeapWord* result = new_cur_alloc_region->allocate(word_size);
+      // Now, perform the allocation out of the region we just
+      // allocated. Note that noone else can access that region at
+      // this point (as _cur_alloc_region has not been updated yet),
+      // so we can just go ahead and do the allocation without any
+      // atomics (and we expect this allocation attempt to
+      // suceeded). Given that other threads can attempt an allocation
+      // with a CAS and without needing the Heap_lock, if we assigned
+      // the new region to _cur_alloc_region before first allocating
+      // into it other threads might have filled up the new region
+      // before we got a chance to do the allocation ourselves. In
+      // that case, we would have needed to retire the region, grab a
+      // new one, and go through all this again. Allocating out of the
+      // new region before assigning it to _cur_alloc_region avoids
+      // all this.
+      HeapWord* result =
+                     new_cur_alloc_region->allocate_no_bot_updates(word_size);
       assert(result != NULL, "we just allocate out of an empty region "
              "so allocation should have been successful");
       assert(is_in(result), "result should be in the heap");
 
+      // Now make sure that the store to _cur_alloc_region does not
+      // float above the store to top.
+      OrderAccess::storestore();
       _cur_alloc_region = new_cur_alloc_region;
 
       if (!at_safepoint) {
@@ -698,7 +866,7 @@
 
   assert(_cur_alloc_region == NULL, "we failed to allocate a new current "
          "alloc region, it should still be NULL");
-  assert_heap_locked_or_at_safepoint();
+  assert_heap_locked_or_at_safepoint(true /* should_be_vm_thread */);
   return NULL;
 }
 
@@ -710,6 +878,10 @@
   assert(!isHumongous(word_size), "attempt_allocation_slow() should not be "
          "used for humongous allocations");
 
+  // We should only reach here when we were unable to allocate
+  // otherwise. So, we should have not active current alloc region.
+  assert(_cur_alloc_region == NULL, "current alloc region should be NULL");
+
   // We will loop while succeeded is false, which means that we tried
   // to do a collection, but the VM op did not succeed. So, when we
   // exit the loop, either one of the allocation attempts was
@@ -718,28 +890,8 @@
   for (int try_count = 1; /* we'll return or break */; try_count += 1) {
     bool succeeded = true;
 
-    {
-      // We may have concurrent cleanup working at the time. Wait for
-      // it to complete. In the future we would probably want to make
-      // the concurrent cleanup truly concurrent by decoupling it from
-      // the allocation. This will happen in the near future as part
-      // of CR 6977804 which will revamp the operation of the free
-      // region list. The fact that wait_for_cleanup_complete() will
-      // do a wait() means that we'll give up the Heap_lock. So, it's
-      // possible that when we exit wait_for_cleanup_complete() we
-      // might be able to allocate successfully (since somebody else
-      // might have done a collection meanwhile). So, we'll attempt to
-      // allocate again, just in case. When we make cleanup truly
-      // concurrent with allocation, we should remove this allocation
-      // attempt as it's redundant (we only reach here after an
-      // allocation attempt has been unsuccessful).
-      wait_for_cleanup_complete();
-      HeapWord* result = attempt_allocation(word_size);
-      if (result != NULL) {
-        assert_heap_not_locked();
-        return result;
-      }
-    }
+    // Every time we go round the loop we should be holding the Heap_lock.
+    assert_heap_locked();
 
     if (GC_locker::is_active_and_needs_gc()) {
       // We are locked out of GC because of the GC locker. We can
@@ -748,7 +900,6 @@
       if (g1_policy()->can_expand_young_list()) {
         // Yes, we are allowed to expand the young gen. Let's try to
         // allocate a new current alloc region.
-
         HeapWord* result =
           replace_cur_alloc_region_and_allocate(word_size,
                                                 false, /* at_safepoint */
@@ -771,20 +922,23 @@
       // rather than causing more, now probably unnecessary, GC attempts.
       JavaThread* jthr = JavaThread::current();
       assert(jthr != NULL, "sanity");
-      if (!jthr->in_critical()) {
-        MutexUnlocker mul(Heap_lock);
-        GC_locker::stall_until_clear();
-
-        // We'll then fall off the end of the ("if GC locker active")
-        // if-statement and retry the allocation further down in the
-        // loop.
-      } else {
+      if (jthr->in_critical()) {
         if (CheckJNICalls) {
           fatal("Possible deadlock due to allocating while"
                 " in jni critical section");
         }
+        // We are returning NULL so the protocol is that we're still
+        // holding the Heap_lock.
+        assert_heap_locked();
         return NULL;
       }
+
+      Heap_lock->unlock();
+      GC_locker::stall_until_clear();
+
+      // No need to relock the Heap_lock. We'll fall off to the code
+      // below the else-statement which assumes that we are not
+      // holding the Heap_lock.
     } else {
       // We are not locked out. So, let's try to do a GC. The VM op
       // will retry the allocation before it completes.
@@ -805,11 +959,10 @@
         dirty_young_block(result, word_size);
         return result;
       }
-
-      Heap_lock->lock();
     }
 
-    assert_heap_locked();
+    // Both paths that get us here from above unlock the Heap_lock.
+    assert_heap_not_locked();
 
     // We can reach here when we were unsuccessful in doing a GC,
     // because another thread beat us to it, or because we were locked
@@ -854,7 +1007,7 @@
   // allocation paths that attempt to allocate a humongous object
   // should eventually reach here. Currently, the only paths are from
   // mem_allocate() and attempt_allocation_at_safepoint().
-  assert_heap_locked_or_at_safepoint();
+  assert_heap_locked_or_at_safepoint(true /* should_be_vm_thread */);
   assert(isHumongous(word_size), "attempt_allocation_humongous() "
          "should only be used for humongous allocations");
   assert(SafepointSynchronize::is_at_safepoint() == at_safepoint,
@@ -931,13 +1084,13 @@
     }
   }
 
-  assert_heap_locked_or_at_safepoint();
+  assert_heap_locked_or_at_safepoint(true /* should_be_vm_thread */);
   return NULL;
 }
 
 HeapWord* G1CollectedHeap::attempt_allocation_at_safepoint(size_t word_size,
                                            bool expect_null_cur_alloc_region) {
-  assert_at_safepoint();
+  assert_at_safepoint(true /* should_be_vm_thread */);
   assert(_cur_alloc_region == NULL || !expect_null_cur_alloc_region,
          err_msg("the current alloc region was unexpectedly found "
                  "to be non-NULL, cur alloc region: "PTR_FORMAT" "
@@ -948,10 +1101,8 @@
     if (!expect_null_cur_alloc_region) {
       HeapRegion* cur_alloc_region = _cur_alloc_region;
       if (cur_alloc_region != NULL) {
-        // This allocate method does BOT updates and we don't need them in
-        // the young generation. This will be fixed in the near future by
-        // CR 6994297.
-        HeapWord* result = cur_alloc_region->allocate(word_size);
+        // We are at a safepoint so no reason to use the MT-safe version.
+        HeapWord* result = cur_alloc_region->allocate_no_bot_updates(word_size);
         if (result != NULL) {
           assert(is_in(result), "result should be in the heap");
 
@@ -983,20 +1134,17 @@
   assert_heap_not_locked_and_not_at_safepoint();
   assert(!isHumongous(word_size), "we do not allow TLABs of humongous size");
 
-  Heap_lock->lock();
-
-  // First attempt: try allocating out of the current alloc region or
-  // after replacing the current alloc region.
+  // First attempt: Try allocating out of the current alloc region
+  // using a CAS. If that fails, take the Heap_lock and retry the
+  // allocation, potentially replacing the current alloc region.
   HeapWord* result = attempt_allocation(word_size);
   if (result != NULL) {
     assert_heap_not_locked();
     return result;
   }
 
-  assert_heap_locked();
-
-  // Second attempt: go into the even slower path where we might
-  // try to schedule a collection.
+  // Second attempt: Go to the slower path where we might try to
+  // schedule a collection.
   result = attempt_allocation_slow(word_size);
   if (result != NULL) {
     assert_heap_not_locked();
@@ -1004,6 +1152,7 @@
   }
 
   assert_heap_locked();
+  // Need to unlock the Heap_lock before returning.
   Heap_lock->unlock();
   return NULL;
 }
@@ -1022,11 +1171,10 @@
   for (int try_count = 1; /* we'll return */; try_count += 1) {
     unsigned int gc_count_before;
     {
-      Heap_lock->lock();
-
       if (!isHumongous(word_size)) {
-        // First attempt: try allocating out of the current alloc
-        // region or after replacing the current alloc region.
+        // First attempt: Try allocating out of the current alloc region
+        // using a CAS. If that fails, take the Heap_lock and retry the
+        // allocation, potentially replacing the current alloc region.
         HeapWord* result = attempt_allocation(word_size);
         if (result != NULL) {
           assert_heap_not_locked();
@@ -1035,14 +1183,17 @@
 
         assert_heap_locked();
 
-        // Second attempt: go into the even slower path where we might
-        // try to schedule a collection.
+        // Second attempt: Go to the slower path where we might try to
+        // schedule a collection.
         result = attempt_allocation_slow(word_size);
         if (result != NULL) {
           assert_heap_not_locked();
           return result;
         }
       } else {
+        // attempt_allocation_humongous() requires the Heap_lock to be held.
+        Heap_lock->lock();
+
         HeapWord* result = attempt_allocation_humongous(word_size,
                                                      false /* at_safepoint */);
         if (result != NULL) {
@@ -1054,7 +1205,8 @@
       assert_heap_locked();
       // Read the gc count while the heap lock is held.
       gc_count_before = SharedHeap::heap()->total_collections();
-      // We cannot be at a safepoint, so it is safe to unlock the Heap_lock
+
+      // Release the Heap_lock before attempting the collection.
       Heap_lock->unlock();
     }
 
@@ -1092,22 +1244,18 @@
 }
 
 void G1CollectedHeap::abandon_cur_alloc_region() {
-  if (_cur_alloc_region != NULL) {
-    // We're finished with the _cur_alloc_region.
-    if (_cur_alloc_region->is_empty()) {
-      _free_regions++;
-      free_region(_cur_alloc_region);
-    } else {
-      // As we're builing (at least the young portion) of the collection
-      // set incrementally we'll add the current allocation region to
-      // the collection set here.
-      if (_cur_alloc_region->is_young()) {
-        g1_policy()->add_region_to_incremental_cset_lhs(_cur_alloc_region);
-      }
-      _summary_bytes_used += _cur_alloc_region->used();
-    }
-    _cur_alloc_region = NULL;
-  }
+  assert_at_safepoint(true /* should_be_vm_thread */);
+
+  HeapRegion* cur_alloc_region = _cur_alloc_region;
+  if (cur_alloc_region != NULL) {
+    assert(!cur_alloc_region->is_empty(),
+           "the current alloc region can never be empty");
+    assert(cur_alloc_region->is_young(),
+           "the current alloc region should be young");
+
+    retire_cur_alloc_region_common(cur_alloc_region);
+  }
+  assert(_cur_alloc_region == NULL, "post-condition");
 }
 
 void G1CollectedHeap::abandon_gc_alloc_regions() {
@@ -1188,19 +1336,20 @@
 bool G1CollectedHeap::do_collection(bool explicit_gc,
                                     bool clear_all_soft_refs,
                                     size_t word_size) {
+  assert_at_safepoint(true /* should_be_vm_thread */);
+
   if (GC_locker::check_active_before_gc()) {
     return false;
   }
 
-  DTraceGCProbeMarker gc_probe_marker(true /* full */);
+  SvcGCMarker sgcm(SvcGCMarker::FULL);
   ResourceMark rm;
 
   if (PrintHeapAtGC) {
     Universe::print_heap_before_gc();
   }
 
-  assert(SafepointSynchronize::is_at_safepoint(), "should be at safepoint");
-  assert(Thread::current() == VMThread::vm_thread(), "should be in vm thread");
+  verify_region_sets_optional();
 
   const bool do_clear_all_soft_refs = clear_all_soft_refs ||
                            collector_policy()->should_clear_all_soft_refs();
@@ -1223,6 +1372,9 @@
     double start = os::elapsedTime();
     g1_policy()->record_full_collection_start();
 
+    wait_while_free_regions_coming();
+    append_secondary_free_list_if_not_empty();
+
     gc_prologue(true);
     increment_total_collections(true /* full gc */);
 
@@ -1235,7 +1387,6 @@
       gclog_or_tty->print(" VerifyBeforeGC:");
       Universe::verify(true);
     }
-    assert(regions_accounted_for(), "Region leakage!");
 
     COMPILER2_PRESENT(DerivedPointerTable::clear());
 
@@ -1259,7 +1410,6 @@
     assert(_cur_alloc_region == NULL, "Invariant.");
     g1_rem_set()->cleanupHRRS();
     tear_down_region_lists();
-    set_used_regions_to_need_zero_fill();
 
     // We may have added regions to the current incremental collection
     // set between the last GC or pause and now. We need to clear the
@@ -1294,9 +1444,7 @@
       HandleMark hm;  // Discard invalid handles created during gc
       G1MarkSweep::invoke_at_safepoint(ref_processor(), do_clear_all_soft_refs);
     }
-    // Because freeing humongous regions may have added some unclean
-    // regions, it is necessary to tear down again before rebuilding.
-    tear_down_region_lists();
+    assert(free_regions() == 0, "we should not have added any free regions");
     rebuild_region_lists();
 
     _summary_bytes_used = recalculate_used();
@@ -1378,7 +1526,6 @@
     JavaThread::dirty_card_queue_set().abandon_logs();
     assert(!G1DeferredRSUpdate
            || (G1DeferredRSUpdate && (dirty_card_queue_set().completed_buffers_num() == 0)), "Should not be any");
-    assert(regions_accounted_for(), "Region leakage!");
   }
 
   if (g1_policy()->in_young_gc_mode()) {
@@ -1392,6 +1539,8 @@
   // Update the number of full collections that have been completed.
   increment_full_collections_completed(false /* concurrent */);
 
+  verify_region_sets_optional();
+
   if (PrintHeapAtGC) {
     Universe::print_heap_after_gc();
   }
@@ -1532,10 +1681,7 @@
 HeapWord*
 G1CollectedHeap::satisfy_failed_allocation(size_t word_size,
                                            bool* succeeded) {
-  assert(SafepointSynchronize::is_at_safepoint(),
-         "satisfy_failed_allocation() should only be called at a safepoint");
-  assert(Thread::current()->is_VM_thread(),
-         "satisfy_failed_allocation() should only be called by the VM thread");
+  assert_at_safepoint(true /* should_be_vm_thread */);
 
   *succeeded = true;
   // Let's attempt the allocation first.
@@ -1607,53 +1753,22 @@
 // allocated block, or else "NULL".
 
 HeapWord* G1CollectedHeap::expand_and_allocate(size_t word_size) {
-  assert(SafepointSynchronize::is_at_safepoint(),
-         "expand_and_allocate() should only be called at a safepoint");
-  assert(Thread::current()->is_VM_thread(),
-         "expand_and_allocate() should only be called by the VM thread");
+  assert_at_safepoint(true /* should_be_vm_thread */);
+
+  verify_region_sets_optional();
 
   size_t expand_bytes = word_size * HeapWordSize;
   if (expand_bytes < MinHeapDeltaBytes) {
     expand_bytes = MinHeapDeltaBytes;
   }
   expand(expand_bytes);
-  assert(regions_accounted_for(), "Region leakage!");
+
+  verify_region_sets_optional();
 
   return attempt_allocation_at_safepoint(word_size,
                                      false /* expect_null_cur_alloc_region */);
 }
 
-size_t G1CollectedHeap::free_region_if_totally_empty(HeapRegion* hr) {
-  size_t pre_used = 0;
-  size_t cleared_h_regions = 0;
-  size_t freed_regions = 0;
-  UncleanRegionList local_list;
-  free_region_if_totally_empty_work(hr, pre_used, cleared_h_regions,
-                                    freed_regions, &local_list);
-
-  finish_free_region_work(pre_used, cleared_h_regions, freed_regions,
-                          &local_list);
-  return pre_used;
-}
-
-void
-G1CollectedHeap::free_region_if_totally_empty_work(HeapRegion* hr,
-                                                   size_t& pre_used,
-                                                   size_t& cleared_h,
-                                                   size_t& freed_regions,
-                                                   UncleanRegionList* list,
-                                                   bool par) {
-  assert(!hr->continuesHumongous(), "should have filtered these out");
-  size_t res = 0;
-  if (hr->used() > 0 && hr->garbage_bytes() == hr->used() &&
-      !hr->is_young()) {
-    if (G1PolicyVerbose > 0)
-      gclog_or_tty->print_cr("Freeing empty region "PTR_FORMAT "(" SIZE_FORMAT " bytes)"
-                                                                               " during cleanup", hr, hr->used());
-    free_region_work(hr, pre_used, cleared_h, freed_regions, list, par);
-  }
-}
-
 // FIXME: both this and shrink could probably be more efficient by
 // doing one "VirtualSpace::expand_by" call rather than several.
 void G1CollectedHeap::expand(size_t expand_bytes) {
@@ -1686,19 +1801,7 @@
 
       // Add it to the HeapRegionSeq.
       _hrs->insert(hr);
-      // Set the zero-fill state, according to whether it's already
-      // zeroed.
-      {
-        MutexLockerEx x(ZF_mon, Mutex::_no_safepoint_check_flag);
-        if (is_zeroed) {
-          hr->set_zero_fill_complete();
-          put_free_region_on_list_locked(hr);
-        } else {
-          hr->set_zero_fill_needed();
-          put_region_on_unclean_list_locked(hr);
-        }
-      }
-      _free_regions++;
+      _free_list.add_as_tail(hr);
       // And we used up an expansion region to create it.
       _expansion_regions--;
       // Tell the cardtable about it.
@@ -1707,6 +1810,7 @@
       _bot_shared->resize(_g1_committed.word_size());
     }
   }
+
   if (Verbose && PrintGC) {
     size_t new_mem_size = _g1_storage.committed_size();
     gclog_or_tty->print_cr("Expanding garbage-first heap from %ldK by %ldK to %ldK",
@@ -1731,7 +1835,6 @@
   assert(mr.start() == (HeapWord*)_g1_storage.high(), "Bad shrink!");
 
   _g1_committed.set_end(mr.start());
-  _free_regions -= num_regions_deleted;
   _expansion_regions += num_regions_deleted;
 
   // Tell the cardtable about it.
@@ -1751,10 +1854,17 @@
 }
 
 void G1CollectedHeap::shrink(size_t shrink_bytes) {
+  verify_region_sets_optional();
+
   release_gc_alloc_regions(true /* totally */);
+  // Instead of tearing down / rebuilding the free lists here, we
+  // could instead use the remove_all_pending() method on free_list to
+  // remove only the ones that we need to remove.
   tear_down_region_lists();  // We will rebuild them in a moment.
   shrink_helper(shrink_bytes);
   rebuild_region_lists();
+
+  verify_region_sets_optional();
 }
 
 // Public methods.
@@ -1773,18 +1883,17 @@
   _ref_processor(NULL),
   _process_strong_tasks(new SubTasksDone(G1H_PS_NumElements)),
   _bot_shared(NULL),
-  _par_alloc_during_gc_lock(Mutex::leaf, "par alloc during GC lock"),
   _objs_with_preserved_marks(NULL), _preserved_marks_of_objs(NULL),
   _evac_failure_scan_stack(NULL) ,
   _mark_in_progress(false),
-  _cg1r(NULL), _czft(NULL), _summary_bytes_used(0),
+  _cg1r(NULL), _summary_bytes_used(0),
   _cur_alloc_region(NULL),
   _refine_cte_cl(NULL),
-  _free_region_list(NULL), _free_region_list_size(0),
-  _free_regions(0),
   _full_collection(false),
-  _unclean_region_list(),
-  _unclean_regions_coming(false),
+  _free_list("Master Free List"),
+  _secondary_free_list("Secondary Free List"),
+  _humongous_set("Master Humongous Set"),
+  _free_regions_coming(false),
   _young_list(new YoungList(this)),
   _gc_time_stamp(0),
   _surviving_young_words(NULL),
@@ -1868,7 +1977,7 @@
 
   ReservedSpace heap_rs(max_byte_size + pgs->max_size(),
                         HeapRegion::GrainBytes,
-                        false /*ism*/, addr);
+                        UseLargePages, addr);
 
   if (UseCompressedOops) {
     if (addr != NULL && !heap_rs.is_reserved()) {
@@ -1877,13 +1986,13 @@
       // Try again to reserver heap higher.
       addr = Universe::preferred_heap_base(total_reserved, Universe::ZeroBasedNarrowOop);
       ReservedSpace heap_rs0(total_reserved, HeapRegion::GrainBytes,
-                             false /*ism*/, addr);
+                             UseLargePages, addr);
       if (addr != NULL && !heap_rs0.is_reserved()) {
         // Failed to reserve at specified address again - give up.
         addr = Universe::preferred_heap_base(total_reserved, Universe::HeapBasedNarrowOop);
         assert(addr == NULL, "");
         ReservedSpace heap_rs1(total_reserved, HeapRegion::GrainBytes,
-                               false /*ism*/, addr);
+                               UseLargePages, addr);
         heap_rs = heap_rs1;
       } else {
         heap_rs = heap_rs0;
@@ -1905,8 +2014,6 @@
 
   _expansion_regions = max_byte_size/HeapRegion::GrainBytes;
 
-  _num_humongous_regions = 0;
-
   // Create the gen rem set (and barrier set) for the entire reserved region.
   _rem_set = collector_policy()->create_rem_set(_reserved, 2);
   set_barrier_set(rem_set()->bs());
@@ -1951,6 +2058,8 @@
   guarantee((size_t) HeapRegion::CardsPerRegion < max_cards_per_region,
             "too many cards per region");
 
+  HeapRegionSet::set_unrealistically_long_length(max_regions() + 1);
+
   _bot_shared = new G1BlockOffsetSharedArray(_reserved,
                                              heap_word_size(init_byte_size));
 
@@ -1975,11 +2084,6 @@
   _cm       = new ConcurrentMark(heap_rs, (int) max_regions());
   _cmThread = _cm->cmThread();
 
-  // ...and the concurrent zero-fill thread, if necessary.
-  if (G1ConcZeroFill) {
-    _czft = new ConcurrentZFThread();
-  }
-
   // Initialize the from_card cache structure of HeapRegionRemSet.
   HeapRegionRemSet::init_heap(max_regions());
 
@@ -2153,7 +2257,7 @@
 #endif // PRODUCT
 
 size_t G1CollectedHeap::unsafe_max_alloc() {
-  if (_free_regions > 0) return HeapRegion::GrainBytes;
+  if (free_regions() > 0) return HeapRegion::GrainBytes;
   // otherwise, is there space in the current allocation region?
 
   // We need to store the current allocation region in a local variable
@@ -2233,8 +2337,7 @@
 }
 
 void G1CollectedHeap::collect_as_vm_thread(GCCause::Cause cause) {
-  assert(Thread::current()->is_VM_thread(), "Precondition#1");
-  assert(Heap_lock->is_locked(), "Precondition#2");
+  assert_at_safepoint(true /* should_be_vm_thread */);
   GCCauseSetter gcs(this, cause);
   switch (cause) {
     case GCCause::_heap_inspection:
@@ -2257,12 +2360,6 @@
   {
     MutexLocker ml(Heap_lock);
 
-    // Don't want to do a GC until cleanup is completed. This
-    // limitation will be removed in the near future when the
-    // operation of the free region list is revamped as part of
-    // CR 6977804.
-    wait_for_cleanup_complete();
-
     // Read the GC count while holding the Heap_lock
     gc_count_before = SharedHeap::heap()->total_collections();
     full_gc_count_before = SharedHeap::heap()->total_full_collections();
@@ -2641,10 +2738,6 @@
   }
 }
 
-bool G1CollectedHeap::allocs_are_zero_filled() {
-  return false;
-}
-
 size_t G1CollectedHeap::large_typearray_limit() {
   // FIXME
   return HeapRegion::GrainBytes/HeapWordSize;
@@ -2659,7 +2752,6 @@
   return 0;
 }
 
-
 void G1CollectedHeap::prepare_for_verify() {
   if (SafepointSynchronize::is_at_safepoint() || ! UseTLAB) {
     ensure_parsability(false);
@@ -2870,7 +2962,9 @@
                          &rootsCl);
     bool failures = rootsCl.failures();
     rem_set()->invalidate(perm_gen()->used_region(), false);
-    if (!silent) { gclog_or_tty->print("heapRegions "); }
+    if (!silent) { gclog_or_tty->print("HeapRegionSets "); }
+    verify_region_sets();
+    if (!silent) { gclog_or_tty->print("HeapRegions "); }
     if (GCParallelVerificationEnabled && ParallelGCThreads > 1) {
       assert(check_heap_region_claim_values(HeapRegion::InitialClaimValue),
              "sanity check");
@@ -2898,7 +2992,7 @@
         failures = true;
       }
     }
-    if (!silent) gclog_or_tty->print("remset ");
+    if (!silent) gclog_or_tty->print("RemSet ");
     rem_set()->verify();
 
     if (failures) {
@@ -2969,15 +3063,10 @@
   if (G1CollectedHeap::use_parallel_gc_threads()) {
     workers()->print_worker_threads_on(st);
   }
-
   _cmThread->print_on(st);
   st->cr();
-
   _cm->print_worker_threads_on(st);
-
   _cg1r->print_worker_threads_on(st);
-
-  _czft->print_on(st);
   st->cr();
 }
 
@@ -2987,7 +3076,6 @@
   }
   tc->do_thread(_cmThread);
   _cg1r->threads_do(tc);
-  tc->do_thread(_czft);
 }
 
 void G1CollectedHeap::print_tracing_info() const {
@@ -3003,15 +3091,10 @@
   if (G1SummarizeConcMark) {
     concurrent_mark()->print_summary_info();
   }
-  if (G1SummarizeZFStats) {
-    ConcurrentZFThread::print_summary_info();
-  }
   g1_policy()->print_yg_surv_rate_info();
-
   SpecializationStats::print();
 }
 
-
 int G1CollectedHeap::addr_to_arena_id(void* addr) const {
   HeapRegion* hr = heap_region_containing(addr);
   if (hr == NULL) {
@@ -3210,17 +3293,22 @@
 
 bool
 G1CollectedHeap::do_collection_pause_at_safepoint(double target_pause_time_ms) {
+  assert_at_safepoint(true /* should_be_vm_thread */);
+  guarantee(!is_gc_active(), "collection is not reentrant");
+
   if (GC_locker::check_active_before_gc()) {
     return false;
   }
 
-  DTraceGCProbeMarker gc_probe_marker(false /* full */);
+  SvcGCMarker sgcm(SvcGCMarker::MINOR);
   ResourceMark rm;
 
   if (PrintHeapAtGC) {
     Universe::print_heap_before_gc();
   }
 
+  verify_region_sets_optional();
+
   {
     // This call will decide whether this pause is an initial-mark
     // pause. If it is, during_initial_mark_pause() will return true
@@ -3251,10 +3339,16 @@
 
     TraceMemoryManagerStats tms(false /* fullGC */);
 
-    assert(SafepointSynchronize::is_at_safepoint(), "should be at safepoint");
-    assert(Thread::current() == VMThread::vm_thread(), "should be in vm thread");
-    guarantee(!is_gc_active(), "collection is not reentrant");
-    assert(regions_accounted_for(), "Region leakage!");
+    // If there are any free regions available on the secondary_free_list
+    // make sure we append them to the free_list. However, we don't
+    // have to wait for the rest of the cleanup operation to
+    // finish. If it's still going on that's OK. If we run out of
+    // regions, the region allocation code will check the
+    // secondary_free_list and potentially wait if more free regions
+    // are coming (see new_region_try_secondary_free_list()).
+    if (!G1StressConcRegionFreeing) {
+      append_secondary_free_list_if_not_empty();
+    }
 
     increment_gc_time_stamp();
 
@@ -3334,8 +3428,6 @@
       // progress, this will be zero.
       _cm->set_oops_do_bound();
 
-      assert(regions_accounted_for(), "Region leakage.");
-
       if (mark_in_progress())
         concurrent_mark()->newCSet();
 
@@ -3431,8 +3523,6 @@
       g1_policy()->record_pause_time_ms(pause_time_ms);
       g1_policy()->record_collection_pause_end();
 
-      assert(regions_accounted_for(), "Region leakage.");
-
       MemoryService::track_memory_usage();
 
       if (VerifyAfterGC && total_collections() >= VerifyGCStartAt) {
@@ -3463,8 +3553,6 @@
       gc_epilogue(false);
     }
 
-    assert(verify_region_lists(), "Bad region lists.");
-
     if (ExitAfterGCNum > 0 && total_collections() == ExitAfterGCNum) {
       gclog_or_tty->print_cr("Stopping after GC #%d", ExitAfterGCNum);
       print_tracing_info();
@@ -3472,6 +3560,8 @@
     }
   }
 
+  verify_region_sets_optional();
+
   TASKQUEUE_STATS_ONLY(if (ParallelGCVerbose) print_taskqueue_stats());
   TASKQUEUE_STATS_ONLY(reset_taskqueue_stats());
 
@@ -3578,7 +3668,7 @@
 
 void G1CollectedHeap::push_gc_alloc_region(HeapRegion* hr) {
   assert(Thread::current()->is_VM_thread() ||
-         par_alloc_during_gc_lock()->owned_by_self(), "Precondition");
+         FreeList_lock->owned_by_self(), "Precondition");
   assert(!hr->is_gc_alloc_region() && !hr->in_collection_set(),
          "Precondition.");
   hr->set_is_gc_alloc_region(true);
@@ -3600,7 +3690,7 @@
 #endif // G1_DEBUG
 
 void G1CollectedHeap::forget_alloc_region_list() {
-  assert(Thread::current()->is_VM_thread(), "Precondition");
+  assert_at_safepoint(true /* should_be_vm_thread */);
   while (_gc_alloc_region_list != NULL) {
     HeapRegion* r = _gc_alloc_region_list;
     assert(r->is_gc_alloc_region(), "Invariant.");
@@ -3620,9 +3710,6 @@
         _young_list->add_survivor_region(r);
       }
     }
-    if (r->is_empty()) {
-      ++_free_regions;
-    }
   }
 #ifdef G1_DEBUG
   FindGCAllocRegion fa;
@@ -3675,7 +3762,7 @@
 
     if (alloc_region == NULL) {
       // we will get a new GC alloc region
-      alloc_region = newAllocRegionWithExpansion(ap, 0);
+      alloc_region = new_gc_alloc_region(ap, 0);
     } else {
       // the region was retained from the last collection
       ++_gc_alloc_region_counts[ap];
@@ -3730,11 +3817,9 @@
       set_gc_alloc_region(ap, NULL);
 
       if (r->is_empty()) {
-        // we didn't actually allocate anything in it; let's just put
-        // it on the free list
-        MutexLockerEx x(ZF_mon, Mutex::_no_safepoint_check_flag);
-        r->set_zero_fill_complete();
-        put_free_region_on_list_locked(r);
+        // We didn't actually allocate anything in it; let's just put
+        // it back on the free list.
+        _free_list.add_as_tail(r);
       } else if (_retain_gc_alloc_region[ap] && !totally) {
         // retain it so that we can use it at the beginning of the next GC
         _retained_gc_alloc_regions[ap] = r;
@@ -3856,13 +3941,15 @@
   size_t _next_marked_bytes;
   OopsInHeapRegionClosure *_cl;
 public:
-  RemoveSelfPointerClosure(G1CollectedHeap* g1, OopsInHeapRegionClosure* cl) :
-    _g1(g1), _cm(_g1->concurrent_mark()),  _prev_marked_bytes(0),
+  RemoveSelfPointerClosure(G1CollectedHeap* g1, HeapRegion* hr,
+                           OopsInHeapRegionClosure* cl) :
+    _g1(g1), _hr(hr), _cm(_g1->concurrent_mark()),  _prev_marked_bytes(0),
     _next_marked_bytes(0), _cl(cl) {}
 
   size_t prev_marked_bytes() { return _prev_marked_bytes; }
   size_t next_marked_bytes() { return _next_marked_bytes; }
 
+  // <original comment>
   // The original idea here was to coalesce evacuated and dead objects.
   // However that caused complications with the block offset table (BOT).
   // In particular if there were two TLABs, one of them partially refined.
@@ -3871,15 +3958,24 @@
   // of TLAB_2. If the last object of the TLAB_1 and the first object
   // of TLAB_2 are coalesced, then the cards of the unrefined part
   // would point into middle of the filler object.
+  // The current approach is to not coalesce and leave the BOT contents intact.
+  // </original comment>
   //
-  // The current approach is to not coalesce and leave the BOT contents intact.
+  // We now reset the BOT when we start the object iteration over the
+  // region and refine its entries for every object we come across. So
+  // the above comment is not really relevant and we should be able
+  // to coalesce dead objects if we want to.
   void do_object(oop obj) {
+    HeapWord* obj_addr = (HeapWord*) obj;
+    assert(_hr->is_in(obj_addr), "sanity");
+    size_t obj_size = obj->size();
+    _hr->update_bot_for_object(obj_addr, obj_size);
     if (obj->is_forwarded() && obj->forwardee() == obj) {
       // The object failed to move.
       assert(!_g1->is_obj_dead(obj), "We should not be preserving dead objs.");
       _cm->markPrev(obj);
       assert(_cm->isPrevMarked(obj), "Should be marked!");
-      _prev_marked_bytes += (obj->size() * HeapWordSize);
+      _prev_marked_bytes += (obj_size * HeapWordSize);
       if (_g1->mark_in_progress() && !_g1->is_obj_ill(obj)) {
         _cm->markAndGrayObjectIfNecessary(obj);
       }
@@ -3901,7 +3997,7 @@
     } else {
       // The object has been either evacuated or is dead. Fill it with a
       // dummy object.
-      MemRegion mr((HeapWord*)obj, obj->size());
+      MemRegion mr((HeapWord*)obj, obj_size);
       CollectedHeap::fill_with_object(mr);
       _cm->clearRangeBothMaps(mr);
     }
@@ -3921,10 +4017,13 @@
   HeapRegion* cur = g1_policy()->collection_set();
   while (cur != NULL) {
     assert(g1_policy()->assertMarkedBytesDataOK(), "Should be!");
-
-    RemoveSelfPointerClosure rspc(_g1h, cl);
+    assert(!cur->isHumongous(), "sanity");
+
     if (cur->evacuation_failed()) {
       assert(cur->in_collection_set(), "bad CS");
+      RemoveSelfPointerClosure rspc(_g1h, cur, cl);
+
+      cur->reset_bot();
       cl->set_region(cur);
       cur->object_iterate(&rspc);
 
@@ -3989,15 +4088,6 @@
   }
 }
 
-void G1CollectedHeap::handle_evacuation_failure(oop old) {
-  markOop m = old->mark();
-  // forward to self
-  assert(!old->is_forwarded(), "precondition");
-
-  old->forward_to(old);
-  handle_evacuation_failure_common(old, m);
-}
-
 oop
 G1CollectedHeap::handle_evacuation_failure_par(OopsInHeapRegionClosure* cl,
                                                oop old) {
@@ -4084,8 +4174,6 @@
 
   HeapWord* block = alloc_region->par_allocate(word_size);
   if (block == NULL) {
-    MutexLockerEx x(par_alloc_during_gc_lock(),
-                    Mutex::_no_safepoint_check_flag);
     block = allocate_during_gc_slow(purpose, alloc_region, true, word_size);
   }
   return block;
@@ -4114,6 +4202,12 @@
          err_msg("we should not be seeing humongous allocation requests "
                  "during GC, word_size = "SIZE_FORMAT, word_size));
 
+  // We need to make sure we serialize calls to this method. Given
+  // that the FreeList_lock guards accesses to the free_list anyway,
+  // and we need to potentially remove a region from it, we'll use it
+  // to protect the whole call.
+  MutexLockerEx x(FreeList_lock, Mutex::_no_safepoint_check_flag);
+
   HeapWord* block = NULL;
   // In the parallel case, a previous thread to obtain the lock may have
   // already assigned a new gc_alloc_region.
@@ -4159,7 +4253,7 @@
   }
 
   // Now allocate a new region for allocation.
-  alloc_region = newAllocRegionWithExpansion(purpose, word_size, false /*zero_filled*/);
+  alloc_region = new_gc_alloc_region(purpose, word_size);
 
   // let the caller handle alloc failure
   if (alloc_region != NULL) {
@@ -4167,9 +4261,6 @@
     assert(check_gc_alloc_regions(), "alloc regions messed up");
     assert(alloc_region->saved_mark_at_top(),
            "Mark should have been saved already.");
-    // We used to assert that the region was zero-filled here, but no
-    // longer.
-
     // This must be done last: once it's installed, other regions may
     // allocate in it (without holding the lock.)
     set_gc_alloc_region(purpose, alloc_region);
@@ -4834,91 +4925,91 @@
   COMPILER2_PRESENT(DerivedPointerTable::update_pointers());
 }
 
-void G1CollectedHeap::free_region(HeapRegion* hr) {
-  size_t pre_used = 0;
-  size_t cleared_h_regions = 0;
-  size_t freed_regions = 0;
-  UncleanRegionList local_list;
-
-  HeapWord* start = hr->bottom();
-  HeapWord* end   = hr->prev_top_at_mark_start();
-  size_t used_bytes = hr->used();
-  size_t live_bytes = hr->max_live_bytes();
-  if (used_bytes > 0) {
-    guarantee( live_bytes <= used_bytes, "invariant" );
-  } else {
-    guarantee( live_bytes == 0, "invariant" );
-  }
-
-  size_t garbage_bytes = used_bytes - live_bytes;
-  if (garbage_bytes > 0)
-    g1_policy()->decrease_known_garbage_bytes(garbage_bytes);
-
-  free_region_work(hr, pre_used, cleared_h_regions, freed_regions,
-                   &local_list);
-  finish_free_region_work(pre_used, cleared_h_regions, freed_regions,
-                          &local_list);
+void G1CollectedHeap::free_region_if_totally_empty(HeapRegion* hr,
+                                     size_t* pre_used,
+                                     FreeRegionList* free_list,
+                                     HumongousRegionSet* humongous_proxy_set,
+                                     bool par) {
+  if (hr->used() > 0 && hr->max_live_bytes() == 0 && !hr->is_young()) {
+    if (hr->isHumongous()) {
+      assert(hr->startsHumongous(), "we should only see starts humongous");
+      free_humongous_region(hr, pre_used, free_list, humongous_proxy_set, par);
+    } else {
+      free_region(hr, pre_used, free_list, par);
+    }
+  }
 }
 
-void
-G1CollectedHeap::free_region_work(HeapRegion* hr,
-                                  size_t& pre_used,
-                                  size_t& cleared_h_regions,
-                                  size_t& freed_regions,
-                                  UncleanRegionList* list,
+void G1CollectedHeap::free_region(HeapRegion* hr,
+                                  size_t* pre_used,
+                                  FreeRegionList* free_list,
                                   bool par) {
-  pre_used += hr->used();
-  if (hr->isHumongous()) {
-    assert(hr->startsHumongous(),
-           "Only the start of a humongous region should be freed.");
-    int ind = _hrs->find(hr);
-    assert(ind != -1, "Should have an index.");
-    // Clear the start region.
-    hr->hr_clear(par, true /*clear_space*/);
-    list->insert_before_head(hr);
-    cleared_h_regions++;
-    freed_regions++;
-    // Clear any continued regions.
-    ind++;
-    while ((size_t)ind < n_regions()) {
-      HeapRegion* hrc = _hrs->at(ind);
-      if (!hrc->continuesHumongous()) break;
-      // Otherwise, does continue the H region.
-      assert(hrc->humongous_start_region() == hr, "Huh?");
-      hrc->hr_clear(par, true /*clear_space*/);
-      cleared_h_regions++;
-      freed_regions++;
-      list->insert_before_head(hrc);
-      ind++;
+  assert(!hr->isHumongous(), "this is only for non-humongous regions");
+  assert(!hr->is_empty(), "the region should not be empty");
+  assert(free_list != NULL, "pre-condition");
+
+  *pre_used += hr->used();
+  hr->hr_clear(par, true /* clear_space */);
+  free_list->add_as_tail(hr);
+}
+
+void G1CollectedHeap::free_humongous_region(HeapRegion* hr,
+                                     size_t* pre_used,
+                                     FreeRegionList* free_list,
+                                     HumongousRegionSet* humongous_proxy_set,
+                                     bool par) {
+  assert(hr->startsHumongous(), "this is only for starts humongous regions");
+  assert(free_list != NULL, "pre-condition");
+  assert(humongous_proxy_set != NULL, "pre-condition");
+
+  size_t hr_used = hr->used();
+  size_t hr_capacity = hr->capacity();
+  size_t hr_pre_used = 0;
+  _humongous_set.remove_with_proxy(hr, humongous_proxy_set);
+  hr->set_notHumongous();
+  free_region(hr, &hr_pre_used, free_list, par);
+
+  int i = hr->hrs_index() + 1;
+  size_t num = 1;
+  while ((size_t) i < n_regions()) {
+    HeapRegion* curr_hr = _hrs->at(i);
+    if (!curr_hr->continuesHumongous()) {
+      break;
     }
-  } else {
-    hr->hr_clear(par, true /*clear_space*/);
-    list->insert_before_head(hr);
-    freed_regions++;
-    // If we're using clear2, this should not be enabled.
-    // assert(!hr->in_cohort(), "Can't be both free and in a cohort.");
-  }
+    curr_hr->set_notHumongous();
+    free_region(curr_hr, &hr_pre_used, free_list, par);
+    num += 1;
+    i += 1;
+  }
+  assert(hr_pre_used == hr_used,
+         err_msg("hr_pre_used: "SIZE_FORMAT" and hr_used: "SIZE_FORMAT" "
+                 "should be the same", hr_pre_used, hr_used));
+  *pre_used += hr_pre_used;
 }
 
-void G1CollectedHeap::finish_free_region_work(size_t pre_used,
-                                              size_t cleared_h_regions,
-                                              size_t freed_regions,
-                                              UncleanRegionList* list) {
-  if (list != NULL && list->sz() > 0) {
-    prepend_region_list_on_unclean_list(list);
-  }
-  // Acquire a lock, if we're parallel, to update possibly-shared
-  // variables.
-  Mutex* lock = (n_par_threads() > 0) ? ParGCRareEvent_lock : NULL;
-  {
+void G1CollectedHeap::update_sets_after_freeing_regions(size_t pre_used,
+                                       FreeRegionList* free_list,
+                                       HumongousRegionSet* humongous_proxy_set,
+                                       bool par) {
+  if (pre_used > 0) {
+    Mutex* lock = (par) ? ParGCRareEvent_lock : NULL;
     MutexLockerEx x(lock, Mutex::_no_safepoint_check_flag);
+    assert(_summary_bytes_used >= pre_used,
+           err_msg("invariant: _summary_bytes_used: "SIZE_FORMAT" "
+                   "should be >= pre_used: "SIZE_FORMAT,
+                   _summary_bytes_used, pre_used));
     _summary_bytes_used -= pre_used;
-    _num_humongous_regions -= (int) cleared_h_regions;
-    _free_regions += freed_regions;
+  }
+  if (free_list != NULL && !free_list->is_empty()) {
+    MutexLockerEx x(FreeList_lock, Mutex::_no_safepoint_check_flag);
+    _free_list.add_as_tail(free_list);
+  }
+  if (humongous_proxy_set != NULL && !humongous_proxy_set->is_empty()) {
+    MutexLockerEx x(OldSets_lock, Mutex::_no_safepoint_check_flag);
+    _humongous_set.update_from_proxy(humongous_proxy_set);
   }
 }
 
-
 void G1CollectedHeap::dirtyCardsForYoungRegions(CardTableModRefBS* ct_bs, HeapRegion* list) {
   while (list != NULL) {
     guarantee( list->is_young(), "invariant" );
@@ -5041,6 +5132,9 @@
 }
 
 void G1CollectedHeap::free_collection_set(HeapRegion* cs_head) {
+  size_t pre_used = 0;
+  FreeRegionList local_free_list("Local List for CSet Freeing");
+
   double young_time_ms     = 0.0;
   double non_young_time_ms = 0.0;
 
@@ -5059,6 +5153,8 @@
   size_t rs_lengths = 0;
 
   while (cur != NULL) {
+    assert(!is_on_free_list(cur), "sanity");
+
     if (non_young) {
       if (cur->is_young()) {
         double end_sec = os::elapsedTime();
@@ -5069,14 +5165,12 @@
         non_young = false;
       }
     } else {
-      if (!cur->is_on_free_list()) {
-        double end_sec = os::elapsedTime();
-        double elapsed_ms = (end_sec - start_sec) * 1000.0;
-        young_time_ms += elapsed_ms;
-
-        start_sec = os::elapsedTime();
-        non_young = true;
-      }
+      double end_sec = os::elapsedTime();
+      double elapsed_ms = (end_sec - start_sec) * 1000.0;
+      young_time_ms += elapsed_ms;
+
+      start_sec = os::elapsedTime();
+      non_young = true;
     }
 
     rs_lengths += cur->rem_set()->occupied();
@@ -5109,9 +5203,8 @@
 
     if (!cur->evacuation_failed()) {
       // And the region is empty.
-      assert(!cur->is_empty(),
-             "Should not have empty regions in a CS.");
-      free_region(cur);
+      assert(!cur->is_empty(), "Should not have empty regions in a CS.");
+      free_region(cur, &pre_used, &local_free_list, false /* par */);
     } else {
       cur->uninstall_surv_rate_group();
       if (cur->is_young())
@@ -5132,6 +5225,9 @@
   else
     young_time_ms += elapsed_ms;
 
+  update_sets_after_freeing_regions(pre_used, &local_free_list,
+                                    NULL /* humongous_proxy_set */,
+                                    false /* par */);
   policy->record_young_free_cset_time_ms(young_time_ms);
   policy->record_non_young_free_cset_time_ms(non_young_time_ms);
 }
@@ -5157,291 +5253,53 @@
   }
 }
 
-HeapRegion*
-G1CollectedHeap::alloc_region_from_unclean_list_locked(bool zero_filled) {
-  assert(ZF_mon->owned_by_self(), "Precondition");
-  HeapRegion* res = pop_unclean_region_list_locked();
-  if (res != NULL) {
-    assert(!res->continuesHumongous() &&
-           res->zero_fill_state() != HeapRegion::Allocated,
-           "Only free regions on unclean list.");
-    if (zero_filled) {
-      res->ensure_zero_filled_locked();
-      res->set_zero_fill_allocated();
+void G1CollectedHeap::set_free_regions_coming() {
+  if (G1ConcRegionFreeingVerbose) {
+    gclog_or_tty->print_cr("G1ConcRegionFreeing [cm thread] : "
+                           "setting free regions coming");
+  }
+
+  assert(!free_regions_coming(), "pre-condition");
+  _free_regions_coming = true;
+}
+
+void G1CollectedHeap::reset_free_regions_coming() {
+  {
+    assert(free_regions_coming(), "pre-condition");
+    MutexLockerEx x(SecondaryFreeList_lock, Mutex::_no_safepoint_check_flag);
+    _free_regions_coming = false;
+    SecondaryFreeList_lock->notify_all();
+  }
+
+  if (G1ConcRegionFreeingVerbose) {
+    gclog_or_tty->print_cr("G1ConcRegionFreeing [cm thread] : "
+                           "reset free regions coming");
+  }
+}
+
+void G1CollectedHeap::wait_while_free_regions_coming() {
+  // Most of the time we won't have to wait, so let's do a quick test
+  // first before we take the lock.
+  if (!free_regions_coming()) {
+    return;
+  }
+
+  if (G1ConcRegionFreeingVerbose) {
+    gclog_or_tty->print_cr("G1ConcRegionFreeing [other] : "
+                           "waiting for free regions");
+  }
+
+  {
+    MutexLockerEx x(SecondaryFreeList_lock, Mutex::_no_safepoint_check_flag);
+    while (free_regions_coming()) {
+      SecondaryFreeList_lock->wait(Mutex::_no_safepoint_check_flag);
     }
   }
-  return res;
-}
-
-HeapRegion* G1CollectedHeap::alloc_region_from_unclean_list(bool zero_filled) {
-  MutexLockerEx zx(ZF_mon, Mutex::_no_safepoint_check_flag);
-  return alloc_region_from_unclean_list_locked(zero_filled);
-}
-
-void G1CollectedHeap::put_region_on_unclean_list(HeapRegion* r) {
-  MutexLockerEx x(ZF_mon, Mutex::_no_safepoint_check_flag);
-  put_region_on_unclean_list_locked(r);
-  if (should_zf()) ZF_mon->notify_all(); // Wake up ZF thread.
-}
-
-void G1CollectedHeap::set_unclean_regions_coming(bool b) {
-  MutexLockerEx x(Cleanup_mon);
-  set_unclean_regions_coming_locked(b);
-}
-
-void G1CollectedHeap::set_unclean_regions_coming_locked(bool b) {
-  assert(Cleanup_mon->owned_by_self(), "Precondition");
-  _unclean_regions_coming = b;
-  // Wake up mutator threads that might be waiting for completeCleanup to
-  // finish.
-  if (!b) Cleanup_mon->notify_all();
-}
-
-void G1CollectedHeap::wait_for_cleanup_complete() {
-  assert_not_at_safepoint();
-  MutexLockerEx x(Cleanup_mon);
-  wait_for_cleanup_complete_locked();
-}
-
-void G1CollectedHeap::wait_for_cleanup_complete_locked() {
-  assert(Cleanup_mon->owned_by_self(), "precondition");
-  while (_unclean_regions_coming) {
-    Cleanup_mon->wait();
-  }
-}
-
-void
-G1CollectedHeap::put_region_on_unclean_list_locked(HeapRegion* r) {
-  assert(ZF_mon->owned_by_self(), "precondition.");
-#ifdef ASSERT
-  if (r->is_gc_alloc_region()) {
-    ResourceMark rm;
-    stringStream region_str;
-    print_on(&region_str);
-    assert(!r->is_gc_alloc_region(), err_msg("Unexpected GC allocation region: %s",
-                                             region_str.as_string()));
-  }
-#endif
-  _unclean_region_list.insert_before_head(r);
-}
-
-void
-G1CollectedHeap::prepend_region_list_on_unclean_list(UncleanRegionList* list) {
-  MutexLockerEx x(ZF_mon, Mutex::_no_safepoint_check_flag);
-  prepend_region_list_on_unclean_list_locked(list);
-  if (should_zf()) ZF_mon->notify_all(); // Wake up ZF thread.
-}
-
-void
-G1CollectedHeap::
-prepend_region_list_on_unclean_list_locked(UncleanRegionList* list) {
-  assert(ZF_mon->owned_by_self(), "precondition.");
-  _unclean_region_list.prepend_list(list);
-}
-
-HeapRegion* G1CollectedHeap::pop_unclean_region_list_locked() {
-  assert(ZF_mon->owned_by_self(), "precondition.");
-  HeapRegion* res = _unclean_region_list.pop();
-  if (res != NULL) {
-    // Inform ZF thread that there's a new unclean head.
-    if (_unclean_region_list.hd() != NULL && should_zf())
-      ZF_mon->notify_all();
-  }
-  return res;
-}
-
-HeapRegion* G1CollectedHeap::peek_unclean_region_list_locked() {
-  assert(ZF_mon->owned_by_self(), "precondition.");
-  return _unclean_region_list.hd();
-}
-
-
-bool G1CollectedHeap::move_cleaned_region_to_free_list_locked() {
-  assert(ZF_mon->owned_by_self(), "Precondition");
-  HeapRegion* r = peek_unclean_region_list_locked();
-  if (r != NULL && r->zero_fill_state() == HeapRegion::ZeroFilled) {
-    // Result of below must be equal to "r", since we hold the lock.
-    (void)pop_unclean_region_list_locked();
-    put_free_region_on_list_locked(r);
-    return true;
-  } else {
-    return false;
-  }
-}
-
-bool G1CollectedHeap::move_cleaned_region_to_free_list() {
-  MutexLockerEx x(ZF_mon, Mutex::_no_safepoint_check_flag);
-  return move_cleaned_region_to_free_list_locked();
-}
-
-
-void G1CollectedHeap::put_free_region_on_list_locked(HeapRegion* r) {
-  assert(ZF_mon->owned_by_self(), "precondition.");
-  assert(_free_region_list_size == free_region_list_length(), "Inv");
-  assert(r->zero_fill_state() == HeapRegion::ZeroFilled,
-        "Regions on free list must be zero filled");
-  assert(!r->isHumongous(), "Must not be humongous.");
-  assert(r->is_empty(), "Better be empty");
-  assert(!r->is_on_free_list(),
-         "Better not already be on free list");
-  assert(!r->is_on_unclean_list(),
-         "Better not already be on unclean list");
-  r->set_on_free_list(true);
-  r->set_next_on_free_list(_free_region_list);
-  _free_region_list = r;
-  _free_region_list_size++;
-  assert(_free_region_list_size == free_region_list_length(), "Inv");
-}
-
-void G1CollectedHeap::put_free_region_on_list(HeapRegion* r) {
-  MutexLockerEx x(ZF_mon, Mutex::_no_safepoint_check_flag);
-  put_free_region_on_list_locked(r);
-}
-
-HeapRegion* G1CollectedHeap::pop_free_region_list_locked() {
-  assert(ZF_mon->owned_by_self(), "precondition.");
-  assert(_free_region_list_size == free_region_list_length(), "Inv");
-  HeapRegion* res = _free_region_list;
-  if (res != NULL) {
-    _free_region_list = res->next_from_free_list();
-    _free_region_list_size--;
-    res->set_on_free_list(false);
-    res->set_next_on_free_list(NULL);
-    assert(_free_region_list_size == free_region_list_length(), "Inv");
-  }
-  return res;
-}
-
-
-HeapRegion* G1CollectedHeap::alloc_free_region_from_lists(bool zero_filled) {
-  // By self, or on behalf of self.
-  assert(Heap_lock->is_locked(), "Precondition");
-  HeapRegion* res = NULL;
-  bool first = true;
-  while (res == NULL) {
-    if (zero_filled || !first) {
-      MutexLockerEx x(ZF_mon, Mutex::_no_safepoint_check_flag);
-      res = pop_free_region_list_locked();
-      if (res != NULL) {
-        assert(!res->zero_fill_is_allocated(),
-               "No allocated regions on free list.");
-        res->set_zero_fill_allocated();
-      } else if (!first) {
-        break;  // We tried both, time to return NULL.
-      }
-    }
-
-    if (res == NULL) {
-      res = alloc_region_from_unclean_list(zero_filled);
-    }
-    assert(res == NULL ||
-           !zero_filled ||
-           res->zero_fill_is_allocated(),
-           "We must have allocated the region we're returning");
-    first = false;
-  }
-  return res;
-}
-
-void G1CollectedHeap::remove_allocated_regions_from_lists() {
-  MutexLockerEx x(ZF_mon, Mutex::_no_safepoint_check_flag);
-  {
-    HeapRegion* prev = NULL;
-    HeapRegion* cur = _unclean_region_list.hd();
-    while (cur != NULL) {
-      HeapRegion* next = cur->next_from_unclean_list();
-      if (cur->zero_fill_is_allocated()) {
-        // Remove from the list.
-        if (prev == NULL) {
-          (void)_unclean_region_list.pop();
-        } else {
-          _unclean_region_list.delete_after(prev);
-        }
-        cur->set_on_unclean_list(false);
-        cur->set_next_on_unclean_list(NULL);
-      } else {
-        prev = cur;
-      }
-      cur = next;
-    }
-    assert(_unclean_region_list.sz() == unclean_region_list_length(),
-           "Inv");
-  }
-
-  {
-    HeapRegion* prev = NULL;
-    HeapRegion* cur = _free_region_list;
-    while (cur != NULL) {
-      HeapRegion* next = cur->next_from_free_list();
-      if (cur->zero_fill_is_allocated()) {
-        // Remove from the list.
-        if (prev == NULL) {
-          _free_region_list = cur->next_from_free_list();
-        } else {
-          prev->set_next_on_free_list(cur->next_from_free_list());
-        }
-        cur->set_on_free_list(false);
-        cur->set_next_on_free_list(NULL);
-        _free_region_list_size--;
-      } else {
-        prev = cur;
-      }
-      cur = next;
-    }
-    assert(_free_region_list_size == free_region_list_length(), "Inv");
-  }
-}
-
-bool G1CollectedHeap::verify_region_lists() {
-  MutexLockerEx x(ZF_mon, Mutex::_no_safepoint_check_flag);
-  return verify_region_lists_locked();
-}
-
-bool G1CollectedHeap::verify_region_lists_locked() {
-  HeapRegion* unclean = _unclean_region_list.hd();
-  while (unclean != NULL) {
-    guarantee(unclean->is_on_unclean_list(), "Well, it is!");
-    guarantee(!unclean->is_on_free_list(), "Well, it shouldn't be!");
-    guarantee(unclean->zero_fill_state() != HeapRegion::Allocated,
-              "Everything else is possible.");
-    unclean = unclean->next_from_unclean_list();
-  }
-  guarantee(_unclean_region_list.sz() == unclean_region_list_length(), "Inv");
-
-  HeapRegion* free_r = _free_region_list;
-  while (free_r != NULL) {
-    assert(free_r->is_on_free_list(), "Well, it is!");
-    assert(!free_r->is_on_unclean_list(), "Well, it shouldn't be!");
-    switch (free_r->zero_fill_state()) {
-    case HeapRegion::NotZeroFilled:
-    case HeapRegion::ZeroFilling:
-      guarantee(false, "Should not be on free list.");
-      break;
-    default:
-      // Everything else is possible.
-      break;
-    }
-    free_r = free_r->next_from_free_list();
-  }
-  guarantee(_free_region_list_size == free_region_list_length(), "Inv");
-  // If we didn't do an assertion...
-  return true;
-}
-
-size_t G1CollectedHeap::free_region_list_length() {
-  assert(ZF_mon->owned_by_self(), "precondition.");
-  size_t len = 0;
-  HeapRegion* cur = _free_region_list;
-  while (cur != NULL) {
-    len++;
-    cur = cur->next_from_free_list();
-  }
-  return len;
-}
-
-size_t G1CollectedHeap::unclean_region_list_length() {
-  assert(ZF_mon->owned_by_self(), "precondition.");
-  return _unclean_region_list.length();
+
+  if (G1ConcRegionFreeingVerbose) {
+    gclog_or_tty->print_cr("G1ConcRegionFreeing [other] : "
+                           "done waiting for free regions");
+  }
 }
 
 size_t G1CollectedHeap::n_regions() {
@@ -5454,55 +5312,6 @@
     HeapRegion::GrainBytes;
 }
 
-size_t G1CollectedHeap::free_regions() {
-  /* Possibly-expensive assert.
-  assert(_free_regions == count_free_regions(),
-         "_free_regions is off.");
-  */
-  return _free_regions;
-}
-
-bool G1CollectedHeap::should_zf() {
-  return _free_region_list_size < (size_t) G1ConcZFMaxRegions;
-}
-
-class RegionCounter: public HeapRegionClosure {
-  size_t _n;
-public:
-  RegionCounter() : _n(0) {}
-  bool doHeapRegion(HeapRegion* r) {
-    if (r->is_empty()) {
-      assert(!r->isHumongous(), "H regions should not be empty.");
-      _n++;
-    }
-    return false;
-  }
-  int res() { return (int) _n; }
-};
-
-size_t G1CollectedHeap::count_free_regions() {
-  RegionCounter rc;
-  heap_region_iterate(&rc);
-  size_t n = rc.res();
-  if (_cur_alloc_region != NULL && _cur_alloc_region->is_empty())
-    n--;
-  return n;
-}
-
-size_t G1CollectedHeap::count_free_regions_list() {
-  size_t n = 0;
-  size_t o = 0;
-  ZF_mon->lock_without_safepoint_check();
-  HeapRegion* cur = _free_region_list;
-  while (cur != NULL) {
-    cur = cur->next_from_free_list();
-    n++;
-  }
-  size_t m = unclean_region_list_length();
-  ZF_mon->unlock();
-  return n + m;
-}
-
 void G1CollectedHeap::set_region_short_lived_locked(HeapRegion* hr) {
   assert(heap_lock_held_for_gc(),
               "the heap lock should already be held by or for this thread");
@@ -5574,28 +5383,19 @@
   }
 }
 
-
 // Done at the start of full GC.
 void G1CollectedHeap::tear_down_region_lists() {
-  MutexLockerEx x(ZF_mon, Mutex::_no_safepoint_check_flag);
-  while (pop_unclean_region_list_locked() != NULL) ;
-  assert(_unclean_region_list.hd() == NULL && _unclean_region_list.sz() == 0,
-         "Postconditions of loop.");
-  while (pop_free_region_list_locked() != NULL) ;
-  assert(_free_region_list == NULL, "Postcondition of loop.");
-  if (_free_region_list_size != 0) {
-    gclog_or_tty->print_cr("Size is %d.", _free_region_list_size);
-    print_on(gclog_or_tty, true /* extended */);
-  }
-  assert(_free_region_list_size == 0, "Postconditions of loop.");
+  _free_list.remove_all();
 }
 
-
 class RegionResetter: public HeapRegionClosure {
-  G1CollectedHeap* _g1;
-  int _n;
+  G1CollectedHeap* _g1h;
+  FreeRegionList _local_free_list;
+
 public:
-  RegionResetter() : _g1(G1CollectedHeap::heap()), _n(0) {}
+  RegionResetter() : _g1h(G1CollectedHeap::heap()),
+                     _local_free_list("Local Free List for RegionResetter") { }
+
   bool doHeapRegion(HeapRegion* r) {
     if (r->continuesHumongous()) return false;
     if (r->top() > r->bottom()) {
@@ -5603,152 +5403,32 @@
         Copy::fill_to_words(r->top(),
                           pointer_delta(r->end(), r->top()));
       }
-      r->set_zero_fill_allocated();
     } else {
       assert(r->is_empty(), "tautology");
-      _n++;
-      switch (r->zero_fill_state()) {
-        case HeapRegion::NotZeroFilled:
-        case HeapRegion::ZeroFilling:
-          _g1->put_region_on_unclean_list_locked(r);
-          break;
-        case HeapRegion::Allocated:
-          r->set_zero_fill_complete();
-          // no break; go on to put on free list.
-        case HeapRegion::ZeroFilled:
-          _g1->put_free_region_on_list_locked(r);
-          break;
-      }
+      _local_free_list.add_as_tail(r);
     }
     return false;
   }
 
-  int getFreeRegionCount() {return _n;}
+  void update_free_lists() {
+    _g1h->update_sets_after_freeing_regions(0, &_local_free_list, NULL,
+                                            false /* par */);
+  }
 };
 
 // Done at the end of full GC.
 void G1CollectedHeap::rebuild_region_lists() {
-  MutexLockerEx x(ZF_mon, Mutex::_no_safepoint_check_flag);
   // This needs to go at the end of the full GC.
   RegionResetter rs;
   heap_region_iterate(&rs);
-  _free_regions = rs.getFreeRegionCount();
-  // Tell the ZF thread it may have work to do.
-  if (should_zf()) ZF_mon->notify_all();
-}
-
-class UsedRegionsNeedZeroFillSetter: public HeapRegionClosure {
-  G1CollectedHeap* _g1;
-  int _n;
-public:
-  UsedRegionsNeedZeroFillSetter() : _g1(G1CollectedHeap::heap()), _n(0) {}
-  bool doHeapRegion(HeapRegion* r) {
-    if (r->continuesHumongous()) return false;
-    if (r->top() > r->bottom()) {
-      // There are assertions in "set_zero_fill_needed()" below that
-      // require top() == bottom(), so this is technically illegal.
-      // We'll skirt the law here, by making that true temporarily.
-      DEBUG_ONLY(HeapWord* save_top = r->top();
-                 r->set_top(r->bottom()));
-      r->set_zero_fill_needed();
-      DEBUG_ONLY(r->set_top(save_top));
-    }
-    return false;
-  }
-};
-
-// Done at the start of full GC.
-void G1CollectedHeap::set_used_regions_to_need_zero_fill() {
-  MutexLockerEx x(ZF_mon, Mutex::_no_safepoint_check_flag);
-  // This needs to go at the end of the full GC.
-  UsedRegionsNeedZeroFillSetter rs;
-  heap_region_iterate(&rs);
+  rs.update_free_lists();
 }
 
 void G1CollectedHeap::set_refine_cte_cl_concurrency(bool concurrent) {
   _refine_cte_cl->set_concurrent(concurrent);
 }
 
-#ifndef PRODUCT
-
-class PrintHeapRegionClosure: public HeapRegionClosure {
-public:
-  bool doHeapRegion(HeapRegion *r) {
-    gclog_or_tty->print("Region: "PTR_FORMAT":", r);
-    if (r != NULL) {
-      if (r->is_on_free_list())
-        gclog_or_tty->print("Free ");
-      if (r->is_young())
-        gclog_or_tty->print("Young ");
-      if (r->isHumongous())
-        gclog_or_tty->print("Is Humongous ");
-      r->print();
-    }
-    return false;
-  }
-};
-
-class SortHeapRegionClosure : public HeapRegionClosure {
-  size_t young_regions,free_regions, unclean_regions;
-  size_t hum_regions, count;
-  size_t unaccounted, cur_unclean, cur_alloc;
-  size_t total_free;
-  HeapRegion* cur;
-public:
-  SortHeapRegionClosure(HeapRegion *_cur) : cur(_cur), young_regions(0),
-    free_regions(0), unclean_regions(0),
-    hum_regions(0),
-    count(0), unaccounted(0),
-    cur_alloc(0), total_free(0)
-  {}
-  bool doHeapRegion(HeapRegion *r) {
-    count++;
-    if (r->is_on_free_list()) free_regions++;
-    else if (r->is_on_unclean_list()) unclean_regions++;
-    else if (r->isHumongous())  hum_regions++;
-    else if (r->is_young()) young_regions++;
-    else if (r == cur) cur_alloc++;
-    else unaccounted++;
-    return false;
-  }
-  void print() {
-    total_free = free_regions + unclean_regions;
-    gclog_or_tty->print("%d regions\n", count);
-    gclog_or_tty->print("%d free: free_list = %d unclean = %d\n",
-                        total_free, free_regions, unclean_regions);
-    gclog_or_tty->print("%d humongous %d young\n",
-                        hum_regions, young_regions);
-    gclog_or_tty->print("%d cur_alloc\n", cur_alloc);
-    gclog_or_tty->print("UHOH unaccounted = %d\n", unaccounted);
-  }
-};
-
-void G1CollectedHeap::print_region_counts() {
-  SortHeapRegionClosure sc(_cur_alloc_region);
-  PrintHeapRegionClosure cl;
-  heap_region_iterate(&cl);
-  heap_region_iterate(&sc);
-  sc.print();
-  print_region_accounting_info();
-};
-
-bool G1CollectedHeap::regions_accounted_for() {
-  // TODO: regions accounting for young/survivor/tenured
-  return true;
-}
-
-bool G1CollectedHeap::print_region_accounting_info() {
-  gclog_or_tty->print_cr("Free regions: %d (count: %d count list %d) (clean: %d unclean: %d).",
-                         free_regions(),
-                         count_free_regions(), count_free_regions_list(),
-                         _free_region_list_size, _unclean_region_list.sz());
-  gclog_or_tty->print_cr("cur_alloc: %d.",
-                         (_cur_alloc_region == NULL ? 0 : 1));
-  gclog_or_tty->print_cr("H regions: %d.", _num_humongous_regions);
-
-  // TODO: check regions accounting for young/survivor/tenured
-  return true;
-}
+#ifdef ASSERT
 
 bool G1CollectedHeap::is_in_closed_subset(const void* p) const {
   HeapRegion* hr = heap_region_containing(p);
@@ -5758,8 +5438,84 @@
     return hr->is_in(p);
   }
 }
-#endif // !PRODUCT
-
-void G1CollectedHeap::g1_unimplemented() {
-  // Unimplemented();
+#endif // ASSERT
+
+class VerifyRegionListsClosure : public HeapRegionClosure {
+private:
+  HumongousRegionSet* _humongous_set;
+  FreeRegionList*     _free_list;
+  size_t              _region_count;
+
+public:
+  VerifyRegionListsClosure(HumongousRegionSet* humongous_set,
+                           FreeRegionList* free_list) :
+    _humongous_set(humongous_set), _free_list(free_list),
+    _region_count(0) { }
+
+  size_t region_count()      { return _region_count;      }
+
+  bool doHeapRegion(HeapRegion* hr) {
+    _region_count += 1;
+
+    if (hr->continuesHumongous()) {
+      return false;
+    }
+
+    if (hr->is_young()) {
+      // TODO
+    } else if (hr->startsHumongous()) {
+      _humongous_set->verify_next_region(hr);
+    } else if (hr->is_empty()) {
+      _free_list->verify_next_region(hr);
+    }
+    return false;
+  }
+};
+
+void G1CollectedHeap::verify_region_sets() {
+  assert_heap_locked_or_at_safepoint(true /* should_be_vm_thread */);
+
+  // First, check the explicit lists.
+  _free_list.verify();
+  {
+    // Given that a concurrent operation might be adding regions to
+    // the secondary free list we have to take the lock before
+    // verifying it.
+    MutexLockerEx x(SecondaryFreeList_lock, Mutex::_no_safepoint_check_flag);
+    _secondary_free_list.verify();
+  }
+  _humongous_set.verify();
+
+  // If a concurrent region freeing operation is in progress it will
+  // be difficult to correctly attributed any free regions we come
+  // across to the correct free list given that they might belong to
+  // one of several (free_list, secondary_free_list, any local lists,
+  // etc.). So, if that's the case we will skip the rest of the
+  // verification operation. Alternatively, waiting for the concurrent
+  // operation to complete will have a non-trivial effect on the GC's
+  // operation (no concurrent operation will last longer than the
+  // interval between two calls to verification) and it might hide
+  // any issues that we would like to catch during testing.
+  if (free_regions_coming()) {
+    return;
+  }
+
+  {
+    MutexLockerEx x(SecondaryFreeList_lock, Mutex::_no_safepoint_check_flag);
+    // Make sure we append the secondary_free_list on the free_list so
+    // that all free regions we will come across can be safely
+    // attributed to the free_list.
+    append_secondary_free_list();
+  }
+
+  // Finally, make sure that the region accounting in the lists is
+  // consistent with what we see in the heap.
+  _humongous_set.verify_start();
+  _free_list.verify_start();
+
+  VerifyRegionListsClosure cl(&_humongous_set, &_free_list);
+  heap_region_iterate(&cl);
+
+  _humongous_set.verify_end();
+  _free_list.verify_end();
 }
--- a/src/share/vm/gc_implementation/g1/g1CollectedHeap.hpp	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/share/vm/gc_implementation/g1/g1CollectedHeap.hpp	Tue Feb 01 18:57:53 2011 -0800
@@ -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
@@ -27,7 +27,7 @@
 
 #include "gc_implementation/g1/concurrentMark.hpp"
 #include "gc_implementation/g1/g1RemSet.hpp"
-#include "gc_implementation/g1/heapRegion.hpp"
+#include "gc_implementation/g1/heapRegionSets.hpp"
 #include "gc_implementation/parNew/parGCAllocBuffer.hpp"
 #include "memory/barrierSet.hpp"
 #include "memory/memRegion.hpp"
@@ -66,8 +66,7 @@
 enum G1GCThreadGroups {
   G1CRGroup = 0,
   G1ZFGroup = 1,
-  G1CMGroup = 2,
-  G1CLGroup = 3
+  G1CMGroup = 2
 };
 
 enum GCAllocPurpose {
@@ -155,6 +154,7 @@
   friend class RefineCardTableEntryClosure;
   friend class G1PrepareCompactClosure;
   friend class RegionSorter;
+  friend class RegionResetter;
   friend class CountRCClosure;
   friend class EvacPopObjClosure;
   friend class G1ParCleanupCTTask;
@@ -178,17 +178,20 @@
   // The maximum part of _g1_storage that has ever been committed.
   MemRegion _g1_max_committed;
 
-  // The number of regions that are completely free.
-  size_t _free_regions;
+  // The master free list. It will satisfy all new region allocations.
+  MasterFreeRegionList      _free_list;
+
+  // The secondary free list which contains regions that have been
+  // freed up during the cleanup process. This will be appended to the
+  // master free list when appropriate.
+  SecondaryFreeRegionList   _secondary_free_list;
+
+  // It keeps track of the humongous regions.
+  MasterHumongousRegionSet  _humongous_set;
 
   // The number of regions we could create by expansion.
   size_t _expansion_regions;
 
-  // Return the number of free regions in the heap (by direct counting.)
-  size_t count_free_regions();
-  // Return the number of free regions on the free and unclean lists.
-  size_t count_free_regions_list();
-
   // The block offset table for the G1 heap.
   G1BlockOffsetSharedArray* _bot_shared;
 
@@ -196,9 +199,6 @@
   // lists, before and after full GC.
   void tear_down_region_lists();
   void rebuild_region_lists();
-  // This sets all non-empty regions to need zero-fill (which they will if
-  // they are empty after full collection.)
-  void set_used_regions_to_need_zero_fill();
 
   // The sequence of all heap regions in the heap.
   HeapRegionSeq* _hrs;
@@ -231,7 +231,7 @@
   // Determines PLAB size for a particular allocation purpose.
   static size_t desired_plab_sz(GCAllocPurpose purpose);
 
-  // When called by par thread, require par_alloc_during_gc_lock() to be held.
+  // When called by par thread, requires the FreeList_lock to be held.
   void push_gc_alloc_region(HeapRegion* hr);
 
   // This should only be called single-threaded.  Undeclares all GC alloc
@@ -294,10 +294,11 @@
   // line number, file, etc.
 
 #define heap_locking_asserts_err_msg(__extra_message)                         \
-  err_msg("%s : Heap_lock %slocked, %sat a safepoint",                        \
+  err_msg("%s : Heap_lock locked: %s, at safepoint: %s, is VM thread: %s",    \
           (__extra_message),                                                  \
-          (!Heap_lock->owned_by_self()) ? "NOT " : "",                        \
-          (!SafepointSynchronize::is_at_safepoint()) ? "NOT " : "")
+          BOOL_TO_STR(Heap_lock->owned_by_self()),                            \
+          BOOL_TO_STR(SafepointSynchronize::is_at_safepoint()),               \
+          BOOL_TO_STR(Thread::current()->is_VM_thread()))
 
 #define assert_heap_locked()                                                  \
   do {                                                                        \
@@ -305,10 +306,11 @@
            heap_locking_asserts_err_msg("should be holding the Heap_lock"));  \
   } while (0)
 
-#define assert_heap_locked_or_at_safepoint()                                  \
+#define assert_heap_locked_or_at_safepoint(__should_be_vm_thread)             \
   do {                                                                        \
     assert(Heap_lock->owned_by_self() ||                                      \
-                                     SafepointSynchronize::is_at_safepoint(), \
+           (SafepointSynchronize::is_at_safepoint() &&                        \
+             ((__should_be_vm_thread) == Thread::current()->is_VM_thread())), \
            heap_locking_asserts_err_msg("should be holding the Heap_lock or " \
                                         "should be at a safepoint"));         \
   } while (0)
@@ -335,9 +337,10 @@
                                    "should not be at a safepoint"));          \
   } while (0)
 
-#define assert_at_safepoint()                                                 \
+#define assert_at_safepoint(__should_be_vm_thread)                            \
   do {                                                                        \
-    assert(SafepointSynchronize::is_at_safepoint(),                           \
+    assert(SafepointSynchronize::is_at_safepoint() &&                         \
+              ((__should_be_vm_thread) == Thread::current()->is_VM_thread()), \
            heap_locking_asserts_err_msg("should be at a safepoint"));         \
   } while (0)
 
@@ -362,31 +365,41 @@
   // The current policy object for the collector.
   G1CollectorPolicy* _g1_policy;
 
-  // Parallel allocation lock to protect the current allocation region.
-  Mutex  _par_alloc_during_gc_lock;
-  Mutex* par_alloc_during_gc_lock() { return &_par_alloc_during_gc_lock; }
+  // This is the second level of trying to allocate a new region. If
+  // new_region_work didn't find a region in the free_list, this call
+  // will check whether there's anything available in the
+  // secondary_free_list and/or wait for more regions to appear in that
+  // list, if _free_regions_coming is set.
+  HeapRegion* new_region_try_secondary_free_list(size_t word_size);
 
-  // If possible/desirable, allocate a new HeapRegion for normal object
-  // allocation sufficient for an allocation of the given "word_size".
-  // If "do_expand" is true, will attempt to expand the heap if necessary
-  // to to satisfy the request.  If "zero_filled" is true, requires a
-  // zero-filled region.
-  // (Returning NULL will trigger a GC.)
-  virtual HeapRegion* newAllocRegion_work(size_t word_size,
-                                          bool do_expand,
-                                          bool zero_filled);
+  // It will try to allocate a single non-humongous HeapRegion
+  // sufficient for an allocation of the given word_size.  If
+  // do_expand is true, it will attempt to expand the heap if
+  // necessary to satisfy the allocation request. Note that word_size
+  // is only used to make sure that we expand sufficiently but, given
+  // that the allocation request is assumed not to be humongous,
+  // having word_size is not strictly necessary (expanding by a single
+  // region will always be sufficient). But let's keep that parameter
+  // in case we need it in the future.
+  HeapRegion* new_region_work(size_t word_size, bool do_expand);
 
-  virtual HeapRegion* newAllocRegion(size_t word_size,
-                                     bool zero_filled = true) {
-    return newAllocRegion_work(word_size, false, zero_filled);
+  // It will try to allocate a new region to be used for allocation by
+  // mutator threads. It will not try to expand the heap if not region
+  // is available.
+  HeapRegion* new_alloc_region(size_t word_size) {
+    return new_region_work(word_size, false /* do_expand */);
   }
-  virtual HeapRegion* newAllocRegionWithExpansion(int purpose,
-                                                  size_t word_size,
-                                                  bool zero_filled = true);
+
+  // It will try to allocate a new region to be used for allocation by
+  // a GC thread. It will try to expand the heap if no region is
+  // available.
+  HeapRegion* new_gc_alloc_region(int purpose, size_t word_size);
+
+  int humongous_obj_allocate_find_first(size_t num_regions, size_t word_size);
 
   // Attempt to allocate an object of the given (very large) "word_size".
   // Returns "NULL" on failure.
-  virtual HeapWord* humongous_obj_allocate(size_t word_size);
+  HeapWord* humongous_obj_allocate(size_t word_size);
 
   // The following two methods, allocate_new_tlab() and
   // mem_allocate(), are the two main entry points from the runtime
@@ -430,7 +443,8 @@
                                  bool*  gc_overhead_limit_was_exceeded);
 
   // The following methods, allocate_from_cur_allocation_region(),
-  // attempt_allocation(), replace_cur_alloc_region_and_allocate(),
+  // attempt_allocation(), attempt_allocation_locked(),
+  // replace_cur_alloc_region_and_allocate(),
   // attempt_allocation_slow(), and attempt_allocation_humongous()
   // have very awkward pre- and post-conditions with respect to
   // locking:
@@ -481,20 +495,30 @@
   // successfully manage to allocate it, or NULL.
 
   // It tries to satisfy an allocation request out of the current
-  // allocating region, which is passed as a parameter. It assumes
-  // that the caller has checked that the current allocating region is
-  // not NULL. Given that the caller has to check the current
-  // allocating region for at least NULL, it might as well pass it as
-  // the first parameter so that the method doesn't have to read it
-  // from the _cur_alloc_region field again.
+  // alloc region, which is passed as a parameter. It assumes that the
+  // caller has checked that the current alloc region is not NULL.
+  // Given that the caller has to check the current alloc region for
+  // at least NULL, it might as well pass it as the first parameter so
+  // that the method doesn't have to read it from the
+  // _cur_alloc_region field again. It is called from both
+  // attempt_allocation() and attempt_allocation_locked() and the
+  // with_heap_lock parameter indicates whether the caller was holding
+  // the heap lock when it called it or not.
   inline HeapWord* allocate_from_cur_alloc_region(HeapRegion* cur_alloc_region,
-                                                  size_t word_size);
+                                                  size_t word_size,
+                                                  bool with_heap_lock);
 
-  // It attempts to allocate out of the current alloc region. If that
-  // fails, it retires the current alloc region (if there is one),
-  // tries to get a new one and retries the allocation.
+  // First-level of allocation slow path: it attempts to allocate out
+  // of the current alloc region in a lock-free manner using a CAS. If
+  // that fails it takes the Heap_lock and calls
+  // attempt_allocation_locked() for the second-level slow path.
   inline HeapWord* attempt_allocation(size_t word_size);
 
+  // Second-level of allocation slow path: while holding the Heap_lock
+  // it tries to allocate out of the current alloc region and, if that
+  // fails, tries to allocate out of a new current alloc region.
+  inline HeapWord* attempt_allocation_locked(size_t word_size);
+
   // It assumes that the current alloc region has been retired and
   // tries to allocate a new one. If it's successful, it performs the
   // allocation out of the new current alloc region and updates
@@ -506,11 +530,11 @@
                                                   bool do_dirtying,
                                                   bool can_expand);
 
-  // The slow path when we are unable to allocate a new current alloc
-  // region to satisfy an allocation request (i.e., when
-  // attempt_allocation() fails). It will try to do an evacuation
-  // pause, which might stall due to the GC locker, and retry the
-  // allocation attempt when appropriate.
+  // Third-level of allocation slow path: when we are unable to
+  // allocate a new current alloc region to satisfy an allocation
+  // request (i.e., when attempt_allocation_locked() fails). It will
+  // try to do an evacuation pause, which might stall due to the GC
+  // locker, and retry the allocation attempt when appropriate.
   HeapWord* attempt_allocation_slow(size_t word_size);
 
   // The method that tries to satisfy a humongous allocation
@@ -749,20 +773,29 @@
   // Invoke "save_marks" on all heap regions.
   void save_marks();
 
-  // Free a heap region.
-  void free_region(HeapRegion* hr);
-  // A component of "free_region", exposed for 'batching'.
-  // All the params after "hr" are out params: the used bytes of the freed
-  // region(s), the number of H regions cleared, the number of regions
-  // freed, and pointers to the head and tail of a list of freed contig
-  // regions, linked throught the "next_on_unclean_list" field.
-  void free_region_work(HeapRegion* hr,
-                        size_t& pre_used,
-                        size_t& cleared_h,
-                        size_t& freed_regions,
-                        UncleanRegionList* list,
-                        bool par = false);
+  // It frees a non-humongous region by initializing its contents and
+  // adding it to the free list that's passed as a parameter (this is
+  // usually a local list which will be appended to the master free
+  // list later). The used bytes of freed regions are accumulated in
+  // pre_used. If par is true, the region's RSet will not be freed
+  // up. The assumption is that this will be done later.
+  void free_region(HeapRegion* hr,
+                   size_t* pre_used,
+                   FreeRegionList* free_list,
+                   bool par);
 
+  // It frees a humongous region by collapsing it into individual
+  // regions and calling free_region() for each of them. The freed
+  // regions will be added to the free list that's passed as a parameter
+  // (this is usually a local list which will be appended to the
+  // master free list later). The used bytes of freed regions are
+  // accumulated in pre_used. If par is true, the region's RSet will
+  // not be freed up. The assumption is that this will be done later.
+  void free_humongous_region(HeapRegion* hr,
+                             size_t* pre_used,
+                             FreeRegionList* free_list,
+                             HumongousRegionSet* humongous_proxy_set,
+                             bool par);
 
   // The concurrent marker (and the thread it runs in.)
   ConcurrentMark* _cm;
@@ -772,9 +805,6 @@
   // The concurrent refiner.
   ConcurrentG1Refine* _cg1r;
 
-  // The concurrent zero-fill thread.
-  ConcurrentZFThread* _czft;
-
   // The parallel task queues
   RefToScanQueueSet *_task_queues;
 
@@ -826,7 +856,6 @@
   void finalize_for_evac_failure();
 
   // An attempt to evacuate "obj" has failed; take necessary steps.
-  void handle_evacuation_failure(oop obj);
   oop handle_evacuation_failure_par(OopsInHeapRegionClosure* cl, oop obj);
   void handle_evacuation_failure_common(oop obj, markOop m);
 
@@ -867,9 +896,7 @@
 
   SubTasksDone* _process_strong_tasks;
 
-  // List of regions which require zero filling.
-  UncleanRegionList _unclean_region_list;
-  bool _unclean_regions_coming;
+  volatile bool _free_regions_coming;
 
 public:
 
@@ -992,71 +1019,64 @@
   size_t max_regions();
 
   // The number of regions that are completely free.
-  size_t free_regions();
+  size_t free_regions() {
+    return _free_list.length();
+  }
 
   // The number of regions that are not completely free.
   size_t used_regions() { return n_regions() - free_regions(); }
 
-  // True iff the ZF thread should run.
-  bool should_zf();
-
   // The number of regions available for "regular" expansion.
   size_t expansion_regions() { return _expansion_regions; }
 
-#ifndef PRODUCT
-  bool regions_accounted_for();
-  bool print_region_accounting_info();
-  void print_region_counts();
-#endif
+  // verify_region_sets() performs verification over the region
+  // lists. It will be compiled in the product code to be used when
+  // necessary (i.e., during heap verification).
+  void verify_region_sets();
 
-  HeapRegion* alloc_region_from_unclean_list(bool zero_filled);
-  HeapRegion* alloc_region_from_unclean_list_locked(bool zero_filled);
+  // verify_region_sets_optional() is planted in the code for
+  // list verification in non-product builds (and it can be enabled in
+  // product builds by definning HEAP_REGION_SET_FORCE_VERIFY to be 1).
+#if HEAP_REGION_SET_FORCE_VERIFY
+  void verify_region_sets_optional() {
+    verify_region_sets();
+  }
+#else // HEAP_REGION_SET_FORCE_VERIFY
+  void verify_region_sets_optional() { }
+#endif // HEAP_REGION_SET_FORCE_VERIFY
 
-  void put_region_on_unclean_list(HeapRegion* r);
-  void put_region_on_unclean_list_locked(HeapRegion* r);
+#ifdef ASSERT
+  bool is_on_free_list(HeapRegion* hr) {
+    return hr->containing_set() == &_free_list;
+  }
 
-  void prepend_region_list_on_unclean_list(UncleanRegionList* list);
-  void prepend_region_list_on_unclean_list_locked(UncleanRegionList* list);
+  bool is_on_humongous_set(HeapRegion* hr) {
+    return hr->containing_set() == &_humongous_set;
+}
+#endif // ASSERT
 
-  void set_unclean_regions_coming(bool b);
-  void set_unclean_regions_coming_locked(bool b);
-  // Wait for cleanup to be complete.
-  void wait_for_cleanup_complete();
-  // Like above, but assumes that the calling thread owns the Heap_lock.
-  void wait_for_cleanup_complete_locked();
+  // Wrapper for the region list operations that can be called from
+  // methods outside this class.
 
-  // Return the head of the unclean list.
-  HeapRegion* peek_unclean_region_list_locked();
-  // Remove and return the head of the unclean list.
-  HeapRegion* pop_unclean_region_list_locked();
+  void secondary_free_list_add_as_tail(FreeRegionList* list) {
+    _secondary_free_list.add_as_tail(list);
+  }
 
-  // List of regions which are zero filled and ready for allocation.
-  HeapRegion* _free_region_list;
-  // Number of elements on the free list.
-  size_t _free_region_list_size;
+  void append_secondary_free_list() {
+    _free_list.add_as_tail(&_secondary_free_list);
+  }
 
-  // If the head of the unclean list is ZeroFilled, move it to the free
-  // list.
-  bool move_cleaned_region_to_free_list_locked();
-  bool move_cleaned_region_to_free_list();
+  void append_secondary_free_list_if_not_empty() {
+    if (!_secondary_free_list.is_empty()) {
+      MutexLockerEx x(SecondaryFreeList_lock, Mutex::_no_safepoint_check_flag);
+      append_secondary_free_list();
+    }
+  }
 
-  void put_free_region_on_list_locked(HeapRegion* r);
-  void put_free_region_on_list(HeapRegion* r);
-
-  // Remove and return the head element of the free list.
-  HeapRegion* pop_free_region_list_locked();
-
-  // If "zero_filled" is true, we first try the free list, then we try the
-  // unclean list, zero-filling the result.  If "zero_filled" is false, we
-  // first try the unclean list, then the zero-filled list.
-  HeapRegion* alloc_free_region_from_lists(bool zero_filled);
-
-  // Verify the integrity of the region lists.
-  void remove_allocated_regions_from_lists();
-  bool verify_region_lists();
-  bool verify_region_lists_locked();
-  size_t unclean_region_list_length();
-  size_t free_region_list_length();
+  void set_free_regions_coming();
+  void reset_free_regions_coming();
+  bool free_regions_coming() { return _free_regions_coming; }
+  void wait_while_free_regions_coming();
 
   // Perform a collection of the heap; intended for use in implementing
   // "System.gc".  This probably implies as full a collection as the
@@ -1075,23 +1095,24 @@
   // True iff a evacuation has failed in the most-recent collection.
   bool evacuation_failed() { return _evacuation_failed; }
 
-  // Free a region if it is totally full of garbage.  Returns the number of
-  // bytes freed (0 ==> didn't free it).
-  size_t free_region_if_totally_empty(HeapRegion *hr);
-  void free_region_if_totally_empty_work(HeapRegion *hr,
-                                         size_t& pre_used,
-                                         size_t& cleared_h_regions,
-                                         size_t& freed_regions,
-                                         UncleanRegionList* list,
-                                         bool par = false);
+  // It will free a region if it has allocated objects in it that are
+  // all dead. It calls either free_region() or
+  // free_humongous_region() depending on the type of the region that
+  // is passed to it.
+  void free_region_if_totally_empty(HeapRegion* hr,
+                                    size_t* pre_used,
+                                    FreeRegionList* free_list,
+                                    HumongousRegionSet* humongous_proxy_set,
+                                    bool par);
 
-  // If we've done free region work that yields the given changes, update
-  // the relevant global variables.
-  void finish_free_region_work(size_t pre_used,
-                               size_t cleared_h_regions,
-                               size_t freed_regions,
-                               UncleanRegionList* list);
-
+  // It appends the free list to the master free list and updates the
+  // master humongous list according to the contents of the proxy
+  // list. It also adjusts the total used bytes according to pre_used
+  // (if par is true, it will do so by taking the ParGCRareEvent_lock).
+  void update_sets_after_freeing_regions(size_t pre_used,
+                                       FreeRegionList* free_list,
+                                       HumongousRegionSet* humongous_proxy_set,
+                                       bool par);
 
   // Returns "TRUE" iff "p" points into the allocated area of the heap.
   virtual bool is_in(const void* p) const;
@@ -1304,8 +1325,6 @@
     return true;
   }
 
-  virtual bool allocs_are_zero_filled();
-
   // The boundary between a "large" and "small" array of primitives, in
   // words.
   virtual size_t large_typearray_limit();
@@ -1536,13 +1555,6 @@
 
 protected:
   size_t _max_heap_capacity;
-
-public:
-  // Temporary: call to mark things unimplemented for the G1 heap (e.g.,
-  // MemoryService).  In productization, we can make this assert false
-  // to catch such places (as well as searching for calls to this...)
-  static void g1_unimplemented();
-
 };
 
 #define use_local_bitmaps         1
--- a/src/share/vm/gc_implementation/g1/g1CollectedHeap.inline.hpp	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/share/vm/gc_implementation/g1/g1CollectedHeap.inline.hpp	Tue Feb 01 18:57:53 2011 -0800
@@ -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
@@ -28,7 +28,7 @@
 #include "gc_implementation/g1/concurrentMark.hpp"
 #include "gc_implementation/g1/g1CollectedHeap.hpp"
 #include "gc_implementation/g1/g1CollectorPolicy.hpp"
-#include "gc_implementation/g1/heapRegionSeq.hpp"
+#include "gc_implementation/g1/heapRegionSeq.inline.hpp"
 #include "utilities/taskqueue.hpp"
 
 // Inline functions for G1CollectedHeap
@@ -63,10 +63,12 @@
 // assumptions of this method (and other related ones).
 inline HeapWord*
 G1CollectedHeap::allocate_from_cur_alloc_region(HeapRegion* cur_alloc_region,
-                                                size_t word_size) {
-  assert_heap_locked_and_not_at_safepoint();
+                                                size_t word_size,
+                                                bool with_heap_lock) {
+  assert_not_at_safepoint();
+  assert(with_heap_lock == Heap_lock->owned_by_self(),
+         "with_heap_lock and Heap_lock->owned_by_self() should be a tautology");
   assert(cur_alloc_region != NULL, "pre-condition of the method");
-  assert(cur_alloc_region == _cur_alloc_region, "pre-condition of the method");
   assert(cur_alloc_region->is_young(),
          "we only support young current alloc regions");
   assert(!isHumongous(word_size), "allocate_from_cur_alloc_region() "
@@ -76,47 +78,100 @@
   assert(!cur_alloc_region->is_empty(),
          err_msg("region ["PTR_FORMAT","PTR_FORMAT"] should not be empty",
                  cur_alloc_region->bottom(), cur_alloc_region->end()));
-  // This allocate method does BOT updates and we don't need them in
-  // the young generation. This will be fixed in the near future by
-  // CR 6994297.
-  HeapWord* result = cur_alloc_region->allocate(word_size);
+  HeapWord* result = cur_alloc_region->par_allocate_no_bot_updates(word_size);
   if (result != NULL) {
     assert(is_in(result), "result should be in the heap");
-    Heap_lock->unlock();
 
+    if (with_heap_lock) {
+      Heap_lock->unlock();
+    }
+    assert_heap_not_locked();
     // Do the dirtying after we release the Heap_lock.
     dirty_young_block(result, word_size);
     return result;
   }
 
+  if (with_heap_lock) {
+    assert_heap_locked();
+  } else {
+    assert_heap_not_locked();
+  }
+  return NULL;
+}
+
+// See the comment in the .hpp file about the locking protocol and
+// assumptions of this method (and other related ones).
+inline HeapWord*
+G1CollectedHeap::attempt_allocation(size_t word_size) {
+  assert_heap_not_locked_and_not_at_safepoint();
+  assert(!isHumongous(word_size), "attempt_allocation() should not be called "
+         "for humongous allocation requests");
+
+  HeapRegion* cur_alloc_region = _cur_alloc_region;
+  if (cur_alloc_region != NULL) {
+    HeapWord* result = allocate_from_cur_alloc_region(cur_alloc_region,
+                                                   word_size,
+                                                   false /* with_heap_lock */);
+    assert_heap_not_locked();
+    if (result != NULL) {
+      return result;
+    }
+  }
+
+  // Our attempt to allocate lock-free failed as the current
+  // allocation region is either NULL or full. So, we'll now take the
+  // Heap_lock and retry.
+  Heap_lock->lock();
+
+  HeapWord* result = attempt_allocation_locked(word_size);
+  if (result != NULL) {
+    assert_heap_not_locked();
+    return result;
+  }
+
   assert_heap_locked();
   return NULL;
 }
 
-// See the comment in the .hpp file about the locking protocol and
-// assumptions of this method (and other related ones).
+inline void
+G1CollectedHeap::retire_cur_alloc_region_common(HeapRegion* cur_alloc_region) {
+  assert_heap_locked_or_at_safepoint(true /* should_be_vm_thread */);
+  assert(cur_alloc_region != NULL && cur_alloc_region == _cur_alloc_region,
+         "pre-condition of the call");
+  assert(cur_alloc_region->is_young(),
+         "we only support young current alloc regions");
+
+  // The region is guaranteed to be young
+  g1_policy()->add_region_to_incremental_cset_lhs(cur_alloc_region);
+  _summary_bytes_used += cur_alloc_region->used();
+  _cur_alloc_region = NULL;
+}
+
 inline HeapWord*
-G1CollectedHeap::attempt_allocation(size_t word_size) {
+G1CollectedHeap::attempt_allocation_locked(size_t word_size) {
   assert_heap_locked_and_not_at_safepoint();
-  assert(!isHumongous(word_size), "attempt_allocation() should not be called "
-         "for humongous allocation requests");
+  assert(!isHumongous(word_size), "attempt_allocation_locked() "
+         "should not be called for humongous allocation requests");
 
+  // First, reread the current alloc region and retry the allocation
+  // in case somebody replaced it while we were waiting to get the
+  // Heap_lock.
   HeapRegion* cur_alloc_region = _cur_alloc_region;
   if (cur_alloc_region != NULL) {
-    HeapWord* result = allocate_from_cur_alloc_region(cur_alloc_region,
-                                                      word_size);
+    HeapWord* result = allocate_from_cur_alloc_region(
+                                                  cur_alloc_region, word_size,
+                                                  true /* with_heap_lock */);
     if (result != NULL) {
       assert_heap_not_locked();
       return result;
     }
 
-    assert_heap_locked();
-
-    // Since we couldn't successfully allocate into it, retire the
-    // current alloc region.
+    // We failed to allocate out of the current alloc region, so let's
+    // retire it before getting a new one.
     retire_cur_alloc_region(cur_alloc_region);
   }
 
+  assert_heap_locked();
   // Try to get a new region and allocate out of it
   HeapWord* result = replace_cur_alloc_region_and_allocate(word_size,
                                                      false, /* at_safepoint */
@@ -131,20 +186,6 @@
   return NULL;
 }
 
-inline void
-G1CollectedHeap::retire_cur_alloc_region_common(HeapRegion* cur_alloc_region) {
-  assert_heap_locked_or_at_safepoint();
-  assert(cur_alloc_region != NULL && cur_alloc_region == _cur_alloc_region,
-         "pre-condition of the call");
-  assert(cur_alloc_region->is_young(),
-         "we only support young current alloc regions");
-
-  // The region is guaranteed to be young
-  g1_policy()->add_region_to_incremental_cset_lhs(cur_alloc_region);
-  _summary_bytes_used += cur_alloc_region->used();
-  _cur_alloc_region = NULL;
-}
-
 // It dirties the cards that cover the block so that so that the post
 // write barrier never queues anything when updating objects on this
 // block. It is assumed (and in fact we assert) that the block
--- a/src/share/vm/gc_implementation/g1/g1CollectorPolicy.cpp	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/share/vm/gc_implementation/g1/g1CollectorPolicy.cpp	Tue Feb 01 18:57:53 2011 -0800
@@ -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
@@ -2875,8 +2875,6 @@
   // Adjust for expansion and slop.
   max_live_bytes = max_live_bytes + expansion_bytes;
 
-  assert(_g1->regions_accounted_for(), "Region leakage!");
-
   HeapRegion* hr;
   if (in_young_gc_mode()) {
     double young_start_time_sec = os::elapsedTime();
--- a/src/share/vm/gc_implementation/g1/g1MarkSweep.cpp	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/share/vm/gc_implementation/g1/g1MarkSweep.cpp	Tue Feb 01 18:57:53 2011 -0800
@@ -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
@@ -113,6 +113,7 @@
 
   Threads::gc_epilogue();
   CodeCache::gc_epilogue();
+  JvmtiExport::gc_epilogue();
 
   // refs processing: clean slate
   GenMarkSweep::_ref_processor = NULL;
@@ -180,26 +181,46 @@
 }
 
 class G1PrepareCompactClosure: public HeapRegionClosure {
+  G1CollectedHeap* _g1h;
   ModRefBarrierSet* _mrbs;
   CompactPoint _cp;
+  size_t _pre_used;
+  FreeRegionList _free_list;
+  HumongousRegionSet _humongous_proxy_set;
 
   void free_humongous_region(HeapRegion* hr) {
-    HeapWord* bot = hr->bottom();
     HeapWord* end = hr->end();
     assert(hr->startsHumongous(),
            "Only the start of a humongous region should be freed.");
-    G1CollectedHeap::heap()->free_region(hr);
+    _g1h->free_humongous_region(hr, &_pre_used, &_free_list,
+                                &_humongous_proxy_set, false /* par */);
+    // Do we also need to do this for the continues humongous regions
+    // we just collapsed?
     hr->prepare_for_compaction(&_cp);
     // Also clear the part of the card table that will be unused after
     // compaction.
-    _mrbs->clear(MemRegion(hr->compaction_top(), hr->end()));
+    _mrbs->clear(MemRegion(hr->compaction_top(), end));
   }
 
 public:
-  G1PrepareCompactClosure(CompactibleSpace* cs) :
+  G1PrepareCompactClosure(CompactibleSpace* cs)
+  : _g1h(G1CollectedHeap::heap()),
+    _mrbs(G1CollectedHeap::heap()->mr_bs()),
     _cp(NULL, cs, cs->initialize_threshold()),
-    _mrbs(G1CollectedHeap::heap()->mr_bs())
-  {}
+    _pre_used(0),
+    _free_list("Local Free List for G1MarkSweep"),
+    _humongous_proxy_set("G1MarkSweep Humongous Proxy Set") { }
+
+  void update_sets() {
+    // We'll recalculate total used bytes and recreate the free list
+    // at the end of the GC, so no point in updating those values here.
+    _g1h->update_sets_after_freeing_regions(0, /* pre_used */
+                                            NULL, /* free_list */
+                                            &_humongous_proxy_set,
+                                            false /* par */);
+    _free_list.remove_all();
+  }
+
   bool doHeapRegion(HeapRegion* hr) {
     if (hr->isHumongous()) {
       if (hr->startsHumongous()) {
@@ -265,6 +286,7 @@
 
   G1PrepareCompactClosure blk(sp);
   g1h->heap_region_iterate(&blk);
+  blk.update_sets();
 
   CompactPoint perm_cp(pg, NULL, NULL);
   pg->prepare_for_compaction(&perm_cp);
--- a/src/share/vm/gc_implementation/g1/g1_globals.hpp	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/share/vm/gc_implementation/g1/g1_globals.hpp	Tue Feb 01 18:57:53 2011 -0800
@@ -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
@@ -75,21 +75,12 @@
           "(0 means do not periodically generate this info); "              \
           "it also requires -XX:+G1SummarizeRSetStats")                     \
                                                                             \
-  diagnostic(bool, G1SummarizeZFStats, false,                               \
-          "Summarize zero-filling info")                                    \
-                                                                            \
   diagnostic(bool, G1TraceConcRefinement, false,                            \
           "Trace G1 concurrent refinement")                                 \
                                                                             \
   product(intx, G1MarkRegionStackSize, 1024 * 1024,                         \
           "Size of the region stack for concurrent marking.")               \
                                                                             \
-  develop(bool, G1ConcZeroFill, true,                                       \
-          "If true, run concurrent zero-filling thread")                    \
-                                                                            \
-  develop(intx, G1ConcZFMaxRegions, 1,                                      \
-          "Stop zero-filling when # of zf'd regions reaches")               \
-                                                                            \
   develop(bool, G1SATBBarrierPrintNullPreVals, false,                       \
           "If true, count frac of ptr writes with null pre-vals.")          \
                                                                             \
@@ -99,6 +90,13 @@
   develop(intx, G1SATBProcessCompletedThreshold, 20,                        \
           "Number of completed buffers that triggers log processing.")      \
                                                                             \
+  product(uintx, G1SATBBufferEnqueueingThresholdPercent, 60,                \
+          "Before enqueueing them, each mutator thread tries to do some "   \
+          "filtering on the SATB buffers it generates. If post-filtering "  \
+          "the percentage of retained entries is over this threshold "      \
+          "the buffer will be enqueued for processing. A value of 0 "       \
+          "specifies that mutator threads should not do such filtering.")   \
+                                                                            \
   develop(intx, G1ExtraRegionSurvRate, 33,                                  \
           "If the young survival rate is S, and there's room left in "      \
           "to-space, we will allow regions whose survival rate is up to "   \
@@ -282,7 +280,20 @@
           "Size of a work unit of cards claimed by a worker thread"         \
           "during RSet scanning.")                                          \
                                                                             \
-  develop(bool, ReduceInitialCardMarksForG1, false,                         \
+  develop(uintx, G1SecondaryFreeListAppendLength, 5,                        \
+          "The number of regions we will add to the secondary free list "   \
+          "at every append operation")                                      \
+                                                                            \
+  develop(bool, G1ConcRegionFreeingVerbose, false,                          \
+          "Enables verboseness during concurrent region freeing")           \
+                                                                            \
+  develop(bool, G1StressConcRegionFreeing, false,                           \
+          "It stresses the concurrent region freeing operation")            \
+                                                                            \
+  develop(uintx, G1StressConcRegionFreeingDelayMillis, 0,                   \
+          "Artificial delay during concurrent region freeing")              \
+                                                                            \
+   develop(bool, ReduceInitialCardMarksForG1, false,                        \
           "When ReduceInitialCardMarks is true, this flag setting "         \
           " controls whether G1 allows the RICM optimization")
 
--- a/src/share/vm/gc_implementation/g1/heapRegion.cpp	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/share/vm/gc_implementation/g1/heapRegion.cpp	Tue Feb 01 18:57:53 2011 -0800
@@ -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
@@ -23,7 +23,6 @@
  */
 
 #include "precompiled.hpp"
-#include "gc_implementation/g1/concurrentZFThread.hpp"
 #include "gc_implementation/g1/g1BlockOffsetTable.inline.hpp"
 #include "gc_implementation/g1/g1CollectedHeap.inline.hpp"
 #include "gc_implementation/g1/g1OopClosures.inline.hpp"
@@ -348,22 +347,20 @@
 }
 
 void HeapRegion::hr_clear(bool par, bool clear_space) {
-  _humongous_type = NotHumongous;
-  _humongous_start_region = NULL;
+  assert(_humongous_type == NotHumongous,
+         "we should have already filtered out humongous regions");
+  assert(_humongous_start_region == NULL,
+         "we should have already filtered out humongous regions");
+  assert(_end == _orig_end,
+         "we should have already filtered out humongous regions");
+
   _in_collection_set = false;
   _is_gc_alloc_region = false;
 
-  // Age stuff (if parallel, this will be done separately, since it needs
-  // to be sequential).
-  G1CollectedHeap* g1h = G1CollectedHeap::heap();
-
   set_young_index_in_cset(-1);
   uninstall_surv_rate_group();
   set_young_type(NotYoung);
 
-  // In case it had been the start of a humongous sequence, reset its end.
-  set_end(_orig_end);
-
   if (!par) {
     // If this is parallel, this will be done later.
     HeapRegionRemSet* hrrs = rem_set();
@@ -386,26 +383,49 @@
 }
 // </PREDICTION>
 
-void HeapRegion::set_startsHumongous(HeapWord* new_end) {
+void HeapRegion::set_startsHumongous(HeapWord* new_top, HeapWord* new_end) {
+  assert(!isHumongous(), "sanity / pre-condition");
   assert(end() == _orig_end,
          "Should be normal before the humongous object allocation");
   assert(top() == bottom(), "should be empty");
+  assert(bottom() <= new_top && new_top <= new_end, "pre-condition");
 
   _humongous_type = StartsHumongous;
   _humongous_start_region = this;
 
   set_end(new_end);
-  _offsets.set_for_starts_humongous(new_end);
+  _offsets.set_for_starts_humongous(new_top);
 }
 
-void HeapRegion::set_continuesHumongous(HeapRegion* start) {
+void HeapRegion::set_continuesHumongous(HeapRegion* first_hr) {
+  assert(!isHumongous(), "sanity / pre-condition");
   assert(end() == _orig_end,
          "Should be normal before the humongous object allocation");
   assert(top() == bottom(), "should be empty");
-  assert(start->startsHumongous(), "pre-condition");
+  assert(first_hr->startsHumongous(), "pre-condition");
 
   _humongous_type = ContinuesHumongous;
-  _humongous_start_region = start;
+  _humongous_start_region = first_hr;
+}
+
+void HeapRegion::set_notHumongous() {
+  assert(isHumongous(), "pre-condition");
+
+  if (startsHumongous()) {
+    assert(top() <= end(), "pre-condition");
+    set_end(_orig_end);
+    if (top() > end()) {
+      // at least one "continues humongous" region after it
+      set_top(end());
+    }
+  } else {
+    // continues humongous
+    assert(end() == _orig_end, "sanity");
+  }
+
+  assert(capacity() == (size_t) HeapRegion::GrainBytes, "pre-condition");
+  _humongous_type = NotHumongous;
+  _humongous_start_region = NULL;
 }
 
 bool HeapRegion::claimHeapRegion(jint claimValue) {
@@ -442,15 +462,6 @@
   return low;
 }
 
-void HeapRegion::set_next_on_unclean_list(HeapRegion* r) {
-  assert(r == NULL || r->is_on_unclean_list(), "Malformed unclean list.");
-  _next_in_special_set = r;
-}
-
-void HeapRegion::set_on_unclean_list(bool b) {
-  _is_on_unclean_list = b;
-}
-
 void HeapRegion::initialize(MemRegion mr, bool clear_space, bool mangle_space) {
   G1OffsetTableContigSpace::initialize(mr, false, mangle_space);
   hr_clear(false/*par*/, clear_space);
@@ -468,15 +479,16 @@
     _hrs_index(-1),
     _humongous_type(NotHumongous), _humongous_start_region(NULL),
     _in_collection_set(false), _is_gc_alloc_region(false),
-    _is_on_free_list(false), _is_on_unclean_list(false),
     _next_in_special_set(NULL), _orig_end(NULL),
     _claimed(InitialClaimValue), _evacuation_failed(false),
     _prev_marked_bytes(0), _next_marked_bytes(0), _sort_index(-1),
     _young_type(NotYoung), _next_young_region(NULL),
-    _next_dirty_cards_region(NULL),
-    _young_index_in_cset(-1), _surv_rate_group(NULL), _age_index(-1),
-    _rem_set(NULL), _zfs(NotZeroFilled),
-    _recorded_rs_length(0), _predicted_elapsed_time_ms(0),
+    _next_dirty_cards_region(NULL), _next(NULL), _pending_removal(false),
+#ifdef ASSERT
+    _containing_set(NULL),
+#endif // ASSERT
+     _young_index_in_cset(-1), _surv_rate_group(NULL), _age_index(-1),
+    _rem_set(NULL), _recorded_rs_length(0), _predicted_elapsed_time_ms(0),
     _predicted_bytes_to_copy(0)
 {
   _orig_end = mr.end();
@@ -551,86 +563,6 @@
   oops_in_mr_iterate(MemRegion(bottom(), saved_mark_word()), cl);
 }
 
-#ifdef DEBUG
-HeapWord* HeapRegion::allocate(size_t size) {
-  jint state = zero_fill_state();
-  assert(!G1CollectedHeap::heap()->allocs_are_zero_filled() ||
-         zero_fill_is_allocated(),
-         "When ZF is on, only alloc in ZF'd regions");
-  return G1OffsetTableContigSpace::allocate(size);
-}
-#endif
-
-void HeapRegion::set_zero_fill_state_work(ZeroFillState zfs) {
-  assert(ZF_mon->owned_by_self() ||
-         Universe::heap()->is_gc_active(),
-         "Must hold the lock or be a full GC to modify.");
-#ifdef ASSERT
-  if (top() != bottom() && zfs != Allocated) {
-    ResourceMark rm;
-    stringStream region_str;
-    print_on(&region_str);
-    assert(top() == bottom() || zfs == Allocated,
-           err_msg("Region must be empty, or we must be setting it to allocated. "
-                   "_zfs=%d, zfs=%d, region: %s", _zfs, zfs, region_str.as_string()));
-  }
-#endif
-  _zfs = zfs;
-}
-
-void HeapRegion::set_zero_fill_complete() {
-  set_zero_fill_state_work(ZeroFilled);
-  if (ZF_mon->owned_by_self()) {
-    ZF_mon->notify_all();
-  }
-}
-
-
-void HeapRegion::ensure_zero_filled() {
-  MutexLockerEx x(ZF_mon, Mutex::_no_safepoint_check_flag);
-  ensure_zero_filled_locked();
-}
-
-void HeapRegion::ensure_zero_filled_locked() {
-  assert(ZF_mon->owned_by_self(), "Precondition");
-  bool should_ignore_zf = SafepointSynchronize::is_at_safepoint();
-  assert(should_ignore_zf || Heap_lock->is_locked(),
-         "Either we're in a GC or we're allocating a region.");
-  switch (zero_fill_state()) {
-  case HeapRegion::NotZeroFilled:
-    set_zero_fill_in_progress(Thread::current());
-    {
-      ZF_mon->unlock();
-      Copy::fill_to_words(bottom(), capacity()/HeapWordSize);
-      ZF_mon->lock_without_safepoint_check();
-    }
-    // A trap.
-    guarantee(zero_fill_state() == HeapRegion::ZeroFilling
-              && zero_filler() == Thread::current(),
-              "AHA!  Tell Dave D if you see this...");
-    set_zero_fill_complete();
-    // gclog_or_tty->print_cr("Did sync ZF.");
-    ConcurrentZFThread::note_sync_zfs();
-    break;
-  case HeapRegion::ZeroFilling:
-    if (should_ignore_zf) {
-      // We can "break" the lock and take over the work.
-      Copy::fill_to_words(bottom(), capacity()/HeapWordSize);
-      set_zero_fill_complete();
-      ConcurrentZFThread::note_sync_zfs();
-      break;
-    } else {
-      ConcurrentZFThread::wait_for_ZF_completed(this);
-    }
-  case HeapRegion::ZeroFilled:
-    // Nothing to do.
-    break;
-  case HeapRegion::Allocated:
-    guarantee(false, "Should not call on allocated regions.");
-  }
-  assert(zero_fill_state() == HeapRegion::ZeroFilled, "Post");
-}
-
 HeapWord*
 HeapRegion::object_iterate_mem_careful(MemRegion mr,
                                                  ObjectClosure* cl) {
@@ -782,9 +714,6 @@
   verify(allow_dirty, /* use_prev_marking */ true, /* failures */ &dummy);
 }
 
-#define OBJ_SAMPLE_INTERVAL 0
-#define BLOCK_SAMPLE_INTERVAL 100
-
 // This really ought to be commoned up into OffsetTableContigSpace somehow.
 // We would need a mechanism to make that code skip dead objects.
 
@@ -795,83 +724,125 @@
   *failures = false;
   HeapWord* p = bottom();
   HeapWord* prev_p = NULL;
-  int objs = 0;
-  int blocks = 0;
   VerifyLiveClosure vl_cl(g1, use_prev_marking);
   bool is_humongous = isHumongous();
+  bool do_bot_verify = !is_young();
   size_t object_num = 0;
   while (p < top()) {
-    size_t size = oop(p)->size();
-    if (is_humongous != g1->isHumongous(size)) {
+    oop obj = oop(p);
+    size_t obj_size = obj->size();
+    object_num += 1;
+
+    if (is_humongous != g1->isHumongous(obj_size)) {
       gclog_or_tty->print_cr("obj "PTR_FORMAT" is of %shumongous size ("
                              SIZE_FORMAT" words) in a %shumongous region",
-                             p, g1->isHumongous(size) ? "" : "non-",
-                             size, is_humongous ? "" : "non-");
+                             p, g1->isHumongous(obj_size) ? "" : "non-",
+                             obj_size, is_humongous ? "" : "non-");
        *failures = true;
+       return;
     }
-    object_num += 1;
-    if (blocks == BLOCK_SAMPLE_INTERVAL) {
-      HeapWord* res = block_start_const(p + (size/2));
-      if (p != res) {
-        gclog_or_tty->print_cr("offset computation 1 for "PTR_FORMAT" and "
-                               SIZE_FORMAT" returned "PTR_FORMAT,
-                               p, size, res);
+
+    // If it returns false, verify_for_object() will output the
+    // appropriate messasge.
+    if (do_bot_verify && !_offsets.verify_for_object(p, obj_size)) {
+      *failures = true;
+      return;
+    }
+
+    if (!g1->is_obj_dead_cond(obj, this, use_prev_marking)) {
+      if (obj->is_oop()) {
+        klassOop klass = obj->klass();
+        if (!klass->is_perm()) {
+          gclog_or_tty->print_cr("klass "PTR_FORMAT" of object "PTR_FORMAT" "
+                                 "not in perm", klass, obj);
+          *failures = true;
+          return;
+        } else if (!klass->is_klass()) {
+          gclog_or_tty->print_cr("klass "PTR_FORMAT" of object "PTR_FORMAT" "
+                                 "not a klass", klass, obj);
+          *failures = true;
+          return;
+        } else {
+          vl_cl.set_containing_obj(obj);
+          obj->oop_iterate(&vl_cl);
+          if (vl_cl.failures()) {
+            *failures = true;
+          }
+          if (G1MaxVerifyFailures >= 0 &&
+              vl_cl.n_failures() >= G1MaxVerifyFailures) {
+            return;
+          }
+        }
+      } else {
+        gclog_or_tty->print_cr(PTR_FORMAT" no an oop", obj);
         *failures = true;
         return;
       }
-      blocks = 0;
-    } else {
-      blocks++;
-    }
-    if (objs == OBJ_SAMPLE_INTERVAL) {
-      oop obj = oop(p);
-      if (!g1->is_obj_dead_cond(obj, this, use_prev_marking)) {
-        if (obj->is_oop()) {
-          klassOop klass = obj->klass();
-          if (!klass->is_perm()) {
-            gclog_or_tty->print_cr("klass "PTR_FORMAT" of object "PTR_FORMAT" "
-                                   "not in perm", klass, obj);
-            *failures = true;
-            return;
-          } else if (!klass->is_klass()) {
-            gclog_or_tty->print_cr("klass "PTR_FORMAT" of object "PTR_FORMAT" "
-                                   "not a klass", klass, obj);
-            *failures = true;
-            return;
-          } else {
-            vl_cl.set_containing_obj(obj);
-            obj->oop_iterate(&vl_cl);
-            if (vl_cl.failures()) {
-              *failures = true;
-            }
-            if (G1MaxVerifyFailures >= 0 &&
-                vl_cl.n_failures() >= G1MaxVerifyFailures) {
-              return;
-            }
-          }
-        } else {
-          gclog_or_tty->print_cr(PTR_FORMAT" no an oop", obj);
-          *failures = true;
-          return;
-        }
-      }
-      objs = 0;
-    } else {
-      objs++;
     }
     prev_p = p;
-    p += size;
+    p += obj_size;
   }
-  HeapWord* rend = end();
-  HeapWord* rtop = top();
-  if (rtop < rend) {
-    HeapWord* res = block_start_const(rtop + (rend - rtop) / 2);
-    if (res != rtop) {
-        gclog_or_tty->print_cr("offset computation 2 for "PTR_FORMAT" and "
-                               PTR_FORMAT" returned "PTR_FORMAT,
-                               rtop, rend, res);
+
+  if (p != top()) {
+    gclog_or_tty->print_cr("end of last object "PTR_FORMAT" "
+                           "does not match top "PTR_FORMAT, p, top());
+    *failures = true;
+    return;
+  }
+
+  HeapWord* the_end = end();
+  assert(p == top(), "it should still hold");
+  // Do some extra BOT consistency checking for addresses in the
+  // range [top, end). BOT look-ups in this range should yield
+  // top. No point in doing that if top == end (there's nothing there).
+  if (p < the_end) {
+    // Look up top
+    HeapWord* addr_1 = p;
+    HeapWord* b_start_1 = _offsets.block_start_const(addr_1);
+    if (b_start_1 != p) {
+      gclog_or_tty->print_cr("BOT look up for top: "PTR_FORMAT" "
+                             " yielded "PTR_FORMAT", expecting "PTR_FORMAT,
+                             addr_1, b_start_1, p);
+      *failures = true;
+      return;
+    }
+
+    // Look up top + 1
+    HeapWord* addr_2 = p + 1;
+    if (addr_2 < the_end) {
+      HeapWord* b_start_2 = _offsets.block_start_const(addr_2);
+      if (b_start_2 != p) {
+        gclog_or_tty->print_cr("BOT look up for top + 1: "PTR_FORMAT" "
+                               " yielded "PTR_FORMAT", expecting "PTR_FORMAT,
+                               addr_2, b_start_2, p);
         *failures = true;
         return;
+      }
+    }
+
+    // Look up an address between top and end
+    size_t diff = pointer_delta(the_end, p) / 2;
+    HeapWord* addr_3 = p + diff;
+    if (addr_3 < the_end) {
+      HeapWord* b_start_3 = _offsets.block_start_const(addr_3);
+      if (b_start_3 != p) {
+        gclog_or_tty->print_cr("BOT look up for top + diff: "PTR_FORMAT" "
+                               " yielded "PTR_FORMAT", expecting "PTR_FORMAT,
+                               addr_3, b_start_3, p);
+        *failures = true;
+        return;
+      }
+    }
+
+    // Loook up end - 1
+    HeapWord* addr_4 = the_end - 1;
+    HeapWord* b_start_4 = _offsets.block_start_const(addr_4);
+    if (b_start_4 != p) {
+      gclog_or_tty->print_cr("BOT look up for end - 1: "PTR_FORMAT" "
+                             " yielded "PTR_FORMAT", expecting "PTR_FORMAT,
+                             addr_4, b_start_4, p);
+      *failures = true;
+      return;
     }
   }
 
@@ -880,12 +851,6 @@
                            "but has "SIZE_FORMAT", objects",
                            bottom(), end(), object_num);
     *failures = true;
-  }
-
-  if (p != top()) {
-    gclog_or_tty->print_cr("end of last object "PTR_FORMAT" "
-                           "does not match top "PTR_FORMAT, p, top());
-    *failures = true;
     return;
   }
 }
@@ -976,67 +941,3 @@
   _offsets.set_space(this);
   initialize(mr, !is_zeroed, SpaceDecorator::Mangle);
 }
-
-size_t RegionList::length() {
-  size_t len = 0;
-  HeapRegion* cur = hd();
-  DEBUG_ONLY(HeapRegion* last = NULL);
-  while (cur != NULL) {
-    len++;
-    DEBUG_ONLY(last = cur);
-    cur = get_next(cur);
-  }
-  assert(last == tl(), "Invariant");
-  return len;
-}
-
-void RegionList::insert_before_head(HeapRegion* r) {
-  assert(well_formed(), "Inv");
-  set_next(r, hd());
-  _hd = r;
-  _sz++;
-  if (tl() == NULL) _tl = r;
-  assert(well_formed(), "Inv");
-}
-
-void RegionList::prepend_list(RegionList* new_list) {
-  assert(well_formed(), "Precondition");
-  assert(new_list->well_formed(), "Precondition");
-  HeapRegion* new_tl = new_list->tl();
-  if (new_tl != NULL) {
-    set_next(new_tl, hd());
-    _hd = new_list->hd();
-    _sz += new_list->sz();
-    if (tl() == NULL) _tl = new_list->tl();
-  } else {
-    assert(new_list->hd() == NULL && new_list->sz() == 0, "Inv");
-  }
-  assert(well_formed(), "Inv");
-}
-
-void RegionList::delete_after(HeapRegion* r) {
-  assert(well_formed(), "Precondition");
-  HeapRegion* next = get_next(r);
-  assert(r != NULL, "Precondition");
-  HeapRegion* next_tl = get_next(next);
-  set_next(r, next_tl);
-  dec_sz();
-  if (next == tl()) {
-    assert(next_tl == NULL, "Inv");
-    _tl = r;
-  }
-  assert(well_formed(), "Inv");
-}
-
-HeapRegion* RegionList::pop() {
-  assert(well_formed(), "Inv");
-  HeapRegion* res = hd();
-  if (res != NULL) {
-    _hd = get_next(res);
-    _sz--;
-    set_next(res, NULL);
-    if (sz() == 0) _tl = NULL;
-  }
-  assert(well_formed(), "Inv");
-  return res;
-}
--- a/src/share/vm/gc_implementation/g1/heapRegion.hpp	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/share/vm/gc_implementation/g1/heapRegion.hpp	Tue Feb 01 18:57:53 2011 -0800
@@ -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
@@ -50,6 +50,11 @@
 class HeapRegionRemSet;
 class HeapRegionRemSetIterator;
 class HeapRegion;
+class HeapRegionSetBase;
+
+#define HR_FORMAT "%d:["PTR_FORMAT","PTR_FORMAT","PTR_FORMAT"]"
+#define HR_FORMAT_PARAMS(__hr) (__hr)->hrs_index(), (__hr)->bottom(), \
+                               (__hr)->top(), (__hr)->end()
 
 // A dirty card to oop closure for heap regions. It
 // knows how to get the G1 heap and how to use the bitmap
@@ -173,6 +178,19 @@
   virtual HeapWord* cross_threshold(HeapWord* start, HeapWord* end);
 
   virtual void print() const;
+
+  void reset_bot() {
+    _offsets.zero_bottom_entry();
+    _offsets.initialize_threshold();
+  }
+
+  void update_bot_for_object(HeapWord* start, size_t word_size) {
+    _offsets.alloc_block(start, word_size);
+  }
+
+  void print_bot_on(outputStream* out) {
+    _offsets.print_on(out);
+  }
 };
 
 class HeapRegion: public G1OffsetTableContigSpace {
@@ -214,12 +232,6 @@
   // True iff the region is in current collection_set.
   bool _in_collection_set;
 
-    // True iff the region is on the unclean list, waiting to be zero filled.
-  bool _is_on_unclean_list;
-
-  // True iff the region is on the free list, ready for allocation.
-  bool _is_on_free_list;
-
   // Is this or has it been an allocation region in the current collection
   // pause.
   bool _is_gc_alloc_region;
@@ -241,6 +253,13 @@
   // Next region whose cards need cleaning
   HeapRegion* _next_dirty_cards_region;
 
+  // Fields used by the HeapRegionSetBase class and subclasses.
+  HeapRegion* _next;
+#ifdef ASSERT
+  HeapRegionSetBase* _containing_set;
+#endif // ASSERT
+  bool _pending_removal;
+
   // For parallel heapRegion traversal.
   jint _claimed;
 
@@ -292,10 +311,6 @@
     _top_at_conc_mark_count = bot;
   }
 
-  jint _zfs;  // A member of ZeroFillState.  Protected by ZF_lock.
-  Thread* _zero_filler; // If _zfs is ZeroFilling, the thread that (last)
-                        // made it so.
-
   void set_young_type(YoungType new_type) {
     //assert(_young_type != new_type, "setting the same type" );
     // TODO: add more assertions here
@@ -349,15 +364,14 @@
     RebuildRSClaimValue   = 5
   };
 
-  // Concurrent refinement requires contiguous heap regions (in which TLABs
-  // might be allocated) to be zero-filled.  Each region therefore has a
-  // zero-fill-state.
-  enum ZeroFillState {
-    NotZeroFilled,
-    ZeroFilling,
-    ZeroFilled,
-    Allocated
-  };
+  inline HeapWord* par_allocate_no_bot_updates(size_t word_size) {
+    assert(is_young(), "we can only skip BOT updates on young regions");
+    return ContiguousSpace::par_allocate(word_size);
+  }
+  inline HeapWord* allocate_no_bot_updates(size_t word_size) {
+    assert(is_young(), "we can only skip BOT updates on young regions");
+    return ContiguousSpace::allocate(word_size);
+  }
 
   // If this region is a member of a HeapRegionSeq, the index in that
   // sequence, otherwise -1.
@@ -404,13 +418,38 @@
     return _humongous_start_region;
   }
 
-  // Causes the current region to represent a humongous object spanning "n"
-  // regions.
-  void set_startsHumongous(HeapWord* new_end);
+  // Makes the current region be a "starts humongous" region, i.e.,
+  // the first region in a series of one or more contiguous regions
+  // that will contain a single "humongous" object. The two parameters
+  // are as follows:
+  //
+  // new_top : The new value of the top field of this region which
+  // points to the end of the humongous object that's being
+  // allocated. If there is more than one region in the series, top
+  // will lie beyond this region's original end field and on the last
+  // region in the series.
+  //
+  // new_end : The new value of the end field of this region which
+  // points to the end of the last region in the series. If there is
+  // one region in the series (namely: this one) end will be the same
+  // as the original end of this region.
+  //
+  // Updating top and end as described above makes this region look as
+  // if it spans the entire space taken up by all the regions in the
+  // series and an single allocation moved its top to new_top. This
+  // ensures that the space (capacity / allocated) taken up by all
+  // humongous regions can be calculated by just looking at the
+  // "starts humongous" regions and by ignoring the "continues
+  // humongous" regions.
+  void set_startsHumongous(HeapWord* new_top, HeapWord* new_end);
 
-  // The regions that continue a humongous sequence should be added using
-  // this method, in increasing address order.
-  void set_continuesHumongous(HeapRegion* start);
+  // Makes the current region be a "continues humongous'
+  // region. first_hr is the "start humongous" region of the series
+  // which this region will be part of.
+  void set_continuesHumongous(HeapRegion* first_hr);
+
+  // Unsets the humongous-related fields on the region.
+  void set_notHumongous();
 
   // If the region has a remembered set, return a pointer to it.
   HeapRegionRemSet* rem_set() const {
@@ -458,45 +497,56 @@
     _next_in_special_set = r;
   }
 
-  bool is_on_free_list() {
-    return _is_on_free_list;
+  // Methods used by the HeapRegionSetBase class and subclasses.
+
+  // Getter and setter for the next field used to link regions into
+  // linked lists.
+  HeapRegion* next()              { return _next; }
+
+  void set_next(HeapRegion* next) { _next = next; }
+
+  // Every region added to a set is tagged with a reference to that
+  // set. This is used for doing consistency checking to make sure that
+  // the contents of a set are as they should be and it's only
+  // available in non-product builds.
+#ifdef ASSERT
+  void set_containing_set(HeapRegionSetBase* containing_set) {
+    assert((containing_set == NULL && _containing_set != NULL) ||
+           (containing_set != NULL && _containing_set == NULL),
+           err_msg("containing_set: "PTR_FORMAT" "
+                   "_containing_set: "PTR_FORMAT,
+                   containing_set, _containing_set));
+
+    _containing_set = containing_set;
+}
+
+  HeapRegionSetBase* containing_set() { return _containing_set; }
+#else // ASSERT
+  void set_containing_set(HeapRegionSetBase* containing_set) { }
+
+  // containing_set() is only used in asserts so there's not reason
+  // to provide a dummy version of it.
+#endif // ASSERT
+
+  // If we want to remove regions from a list in bulk we can simply tag
+  // them with the pending_removal tag and call the
+  // remove_all_pending() method on the list.
+
+  bool pending_removal() { return _pending_removal; }
+
+  void set_pending_removal(bool pending_removal) {
+    // We can only set pending_removal to true, if it's false and the
+    // region belongs to a set.
+    assert(!pending_removal ||
+           (!_pending_removal && containing_set() != NULL), "pre-condition");
+    // We can only set pending_removal to false, if it's true and the
+    // region does not belong to a set.
+    assert( pending_removal ||
+           ( _pending_removal && containing_set() == NULL), "pre-condition");
+
+    _pending_removal = pending_removal;
   }
 
-  void set_on_free_list(bool b) {
-    _is_on_free_list = b;
-  }
-
-  HeapRegion* next_from_free_list() {
-    assert(is_on_free_list(),
-           "Should only invoke on free space.");
-    assert(_next_in_special_set == NULL ||
-           _next_in_special_set->is_on_free_list(),
-           "Malformed Free List.");
-    return _next_in_special_set;
-  }
-
-  void set_next_on_free_list(HeapRegion* r) {
-    assert(r == NULL || r->is_on_free_list(), "Malformed free list.");
-    _next_in_special_set = r;
-  }
-
-  bool is_on_unclean_list() {
-    return _is_on_unclean_list;
-  }
-
-  void set_on_unclean_list(bool b);
-
-  HeapRegion* next_from_unclean_list() {
-    assert(is_on_unclean_list(),
-           "Should only invoke on unclean space.");
-    assert(_next_in_special_set == NULL ||
-           _next_in_special_set->is_on_unclean_list(),
-           "Malformed unclean List.");
-    return _next_in_special_set;
-  }
-
-  void set_next_on_unclean_list(HeapRegion* r);
-
   HeapRegion* get_next_young_region() { return _next_young_region; }
   void set_next_young_region(HeapRegion* hr) {
     _next_young_region = hr;
@@ -515,11 +565,6 @@
 
   void initialize(MemRegion mr, bool clear_space, bool mangle_space);
 
-  // Ensure that "this" is zero-filled.
-  void ensure_zero_filled();
-  // This one requires that the calling thread holds ZF_mon.
-  void ensure_zero_filled_locked();
-
   // Get the start of the unmarked area in this region.
   HeapWord* prev_top_at_mark_start() const { return _prev_top_at_mark_start; }
   HeapWord* next_top_at_mark_start() const { return _next_top_at_mark_start; }
@@ -754,36 +799,6 @@
   // "end" of the region if there is no such block.
   HeapWord* next_block_start_careful(HeapWord* addr);
 
-  // Returns the zero-fill-state of the current region.
-  ZeroFillState zero_fill_state() { return (ZeroFillState)_zfs; }
-  bool zero_fill_is_allocated() { return _zfs == Allocated; }
-  Thread* zero_filler() { return _zero_filler; }
-
-  // Indicate that the contents of the region are unknown, and therefore
-  // might require zero-filling.
-  void set_zero_fill_needed() {
-    set_zero_fill_state_work(NotZeroFilled);
-  }
-  void set_zero_fill_in_progress(Thread* t) {
-    set_zero_fill_state_work(ZeroFilling);
-    _zero_filler = t;
-  }
-  void set_zero_fill_complete();
-  void set_zero_fill_allocated() {
-    set_zero_fill_state_work(Allocated);
-  }
-
-  void set_zero_fill_state_work(ZeroFillState zfs);
-
-  // This is called when a full collection shrinks the heap.
-  // We want to set the heap region to a value which says
-  // it is no longer part of the heap.  For now, we'll let "NotZF" fill
-  // that role.
-  void reset_zero_fill() {
-    set_zero_fill_state_work(NotZeroFilled);
-    _zero_filler = NULL;
-  }
-
   size_t recorded_rs_length() const        { return _recorded_rs_length; }
   double predicted_elapsed_time_ms() const { return _predicted_elapsed_time_ms; }
   size_t predicted_bytes_to_copy() const   { return _predicted_bytes_to_copy; }
@@ -822,10 +837,6 @@
 
   // Override; it uses the "prev" marking information
   virtual void verify(bool allow_dirty) const;
-
-#ifdef DEBUG
-  HeapWord* allocate(size_t size);
-#endif
 };
 
 // HeapRegionClosure is used for iterating over regions.
@@ -848,113 +859,6 @@
   bool complete() { return _complete; }
 };
 
-// A linked lists of heap regions.  It leaves the "next" field
-// unspecified; that's up to subtypes.
-class RegionList VALUE_OBJ_CLASS_SPEC {
-protected:
-  virtual HeapRegion* get_next(HeapRegion* chr) = 0;
-  virtual void set_next(HeapRegion* chr,
-                        HeapRegion* new_next) = 0;
-
-  HeapRegion* _hd;
-  HeapRegion* _tl;
-  size_t _sz;
-
-  // Protected constructor because this type is only meaningful
-  // when the _get/_set next functions are defined.
-  RegionList() : _hd(NULL), _tl(NULL), _sz(0) {}
-public:
-  void reset() {
-    _hd = NULL;
-    _tl = NULL;
-    _sz = 0;
-  }
-  HeapRegion* hd() { return _hd; }
-  HeapRegion* tl() { return _tl; }
-  size_t sz() { return _sz; }
-  size_t length();
-
-  bool well_formed() {
-    return
-      ((hd() == NULL && tl() == NULL && sz() == 0)
-       || (hd() != NULL && tl() != NULL && sz() > 0))
-      && (sz() == length());
-  }
-  virtual void insert_before_head(HeapRegion* r);
-  void prepend_list(RegionList* new_list);
-  virtual HeapRegion* pop();
-  void dec_sz() { _sz--; }
-  // Requires that "r" is an element of the list, and is not the tail.
-  void delete_after(HeapRegion* r);
-};
-
-class EmptyNonHRegionList: public RegionList {
-protected:
-  // Protected constructor because this type is only meaningful
-  // when the _get/_set next functions are defined.
-  EmptyNonHRegionList() : RegionList() {}
-
-public:
-  void insert_before_head(HeapRegion* r) {
-    //    assert(r->is_empty(), "Better be empty");
-    assert(!r->isHumongous(), "Better not be humongous.");
-    RegionList::insert_before_head(r);
-  }
-  void prepend_list(EmptyNonHRegionList* new_list) {
-    //    assert(new_list->hd() == NULL || new_list->hd()->is_empty(),
-    //     "Better be empty");
-    assert(new_list->hd() == NULL || !new_list->hd()->isHumongous(),
-           "Better not be humongous.");
-    //    assert(new_list->tl() == NULL || new_list->tl()->is_empty(),
-    //     "Better be empty");
-    assert(new_list->tl() == NULL || !new_list->tl()->isHumongous(),
-           "Better not be humongous.");
-    RegionList::prepend_list(new_list);
-  }
-};
-
-class UncleanRegionList: public EmptyNonHRegionList {
-public:
-  HeapRegion* get_next(HeapRegion* hr) {
-    return hr->next_from_unclean_list();
-  }
-  void set_next(HeapRegion* hr, HeapRegion* new_next) {
-    hr->set_next_on_unclean_list(new_next);
-  }
-
-  UncleanRegionList() : EmptyNonHRegionList() {}
-
-  void insert_before_head(HeapRegion* r) {
-    assert(!r->is_on_free_list(),
-           "Better not already be on free list");
-    assert(!r->is_on_unclean_list(),
-           "Better not already be on unclean list");
-    r->set_zero_fill_needed();
-    r->set_on_unclean_list(true);
-    EmptyNonHRegionList::insert_before_head(r);
-  }
-  void prepend_list(UncleanRegionList* new_list) {
-    assert(new_list->tl() == NULL || !new_list->tl()->is_on_free_list(),
-           "Better not already be on free list");
-    assert(new_list->tl() == NULL || new_list->tl()->is_on_unclean_list(),
-           "Better already be marked as on unclean list");
-    assert(new_list->hd() == NULL || !new_list->hd()->is_on_free_list(),
-           "Better not already be on free list");
-    assert(new_list->hd() == NULL || new_list->hd()->is_on_unclean_list(),
-           "Better already be marked as on unclean list");
-    EmptyNonHRegionList::prepend_list(new_list);
-  }
-  HeapRegion* pop() {
-    HeapRegion* res = RegionList::pop();
-    if (res != NULL) res->set_on_unclean_list(false);
-    return res;
-  }
-};
-
-// Local Variables: ***
-// c-indentation-style: gnu ***
-// End: ***
-
 #endif // SERIALGC
 
 #endif // SHARE_VM_GC_IMPLEMENTATION_G1_HEAPREGION_HPP
--- a/src/share/vm/gc_implementation/g1/heapRegionSeq.cpp	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/share/vm/gc_implementation/g1/heapRegionSeq.cpp	Tue Feb 01 18:57:53 2011 -0800
@@ -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
@@ -65,152 +65,6 @@
 
 // Private methods.
 
-HeapWord*
-HeapRegionSeq::alloc_obj_from_region_index(int ind, size_t word_size) {
-  assert(G1CollectedHeap::isHumongous(word_size),
-         "Allocation size should be humongous");
-  int cur = ind;
-  int first = cur;
-  size_t sumSizes = 0;
-  while (cur < _regions.length() && sumSizes < word_size) {
-    // Loop invariant:
-    //  For all i in [first, cur):
-    //       _regions.at(i)->is_empty()
-    //    && _regions.at(i) is contiguous with its predecessor, if any
-    //  && sumSizes is the sum of the sizes of the regions in the interval
-    //       [first, cur)
-    HeapRegion* curhr = _regions.at(cur);
-    if (curhr->is_empty()
-        && (first == cur
-            || (_regions.at(cur-1)->end() ==
-                curhr->bottom()))) {
-      sumSizes += curhr->capacity() / HeapWordSize;
-    } else {
-      first = cur + 1;
-      sumSizes = 0;
-    }
-    cur++;
-  }
-  if (sumSizes >= word_size) {
-    _alloc_search_start = cur;
-
-    // We need to initialize the region(s) we just discovered. This is
-    // a bit tricky given that it can happen concurrently with
-    // refinement threads refining cards on these regions and
-    // potentially wanting to refine the BOT as they are scanning
-    // those cards (this can happen shortly after a cleanup; see CR
-    // 6991377). So we have to set up the region(s) carefully and in
-    // a specific order.
-
-    // Currently, allocs_are_zero_filled() returns false. The zero
-    // filling infrastructure will be going away soon (see CR 6977804).
-    // So no need to do anything else here.
-    bool zf = G1CollectedHeap::heap()->allocs_are_zero_filled();
-    assert(!zf, "not supported");
-
-    // This will be the "starts humongous" region.
-    HeapRegion* first_hr = _regions.at(first);
-    {
-      MutexLockerEx x(ZF_mon, Mutex::_no_safepoint_check_flag);
-      first_hr->set_zero_fill_allocated();
-    }
-    // The header of the new object will be placed at the bottom of
-    // the first region.
-    HeapWord* new_obj = first_hr->bottom();
-    // This will be the new end of the first region in the series that
-    // should also match the end of the last region in the seriers.
-    // (Note: sumSizes = "region size" x "number of regions we found").
-    HeapWord* new_end = new_obj + sumSizes;
-    // This will be the new top of the first region that will reflect
-    // this allocation.
-    HeapWord* new_top = new_obj + word_size;
-
-    // First, we need to zero the header of the space that we will be
-    // allocating. When we update top further down, some refinement
-    // threads might try to scan the region. By zeroing the header we
-    // ensure that any thread that will try to scan the region will
-    // come across the zero klass word and bail out.
-    //
-    // NOTE: It would not have been correct to have used
-    // CollectedHeap::fill_with_object() and make the space look like
-    // an int array. The thread that is doing the allocation will
-    // later update the object header to a potentially different array
-    // type and, for a very short period of time, the klass and length
-    // fields will be inconsistent. This could cause a refinement
-    // thread to calculate the object size incorrectly.
-    Copy::fill_to_words(new_obj, oopDesc::header_size(), 0);
-
-    // We will set up the first region as "starts humongous". This
-    // will also update the BOT covering all the regions to reflect
-    // that there is a single object that starts at the bottom of the
-    // first region.
-    first_hr->set_startsHumongous(new_end);
-
-    // Then, if there are any, we will set up the "continues
-    // humongous" regions.
-    HeapRegion* hr = NULL;
-    for (int i = first + 1; i < cur; ++i) {
-      hr = _regions.at(i);
-      {
-        MutexLockerEx x(ZF_mon, Mutex::_no_safepoint_check_flag);
-        hr->set_zero_fill_allocated();
-      }
-      hr->set_continuesHumongous(first_hr);
-    }
-    // If we have "continues humongous" regions (hr != NULL), then the
-    // end of the last one should match new_end.
-    assert(hr == NULL || hr->end() == new_end, "sanity");
-
-    // Up to this point no concurrent thread would have been able to
-    // do any scanning on any region in this series. All the top
-    // fields still point to bottom, so the intersection between
-    // [bottom,top] and [card_start,card_end] will be empty. Before we
-    // update the top fields, we'll do a storestore to make sure that
-    // no thread sees the update to top before the zeroing of the
-    // object header and the BOT initialization.
-    OrderAccess::storestore();
-
-    // Now that the BOT and the object header have been initialized,
-    // we can update top of the "starts humongous" region.
-    assert(first_hr->bottom() < new_top && new_top <= first_hr->end(),
-           "new_top should be in this region");
-    first_hr->set_top(new_top);
-
-    // Now, we will update the top fields of the "continues humongous"
-    // regions. The reason we need to do this is that, otherwise,
-    // these regions would look empty and this will confuse parts of
-    // G1. For example, the code that looks for a consecutive number
-    // of empty regions will consider them empty and try to
-    // re-allocate them. We can extend is_empty() to also include
-    // !continuesHumongous(), but it is easier to just update the top
-    // fields here.
-    hr = NULL;
-    for (int i = first + 1; i < cur; ++i) {
-      hr = _regions.at(i);
-      if ((i + 1) == cur) {
-        // last continues humongous region
-        assert(hr->bottom() < new_top && new_top <= hr->end(),
-               "new_top should fall on this region");
-        hr->set_top(new_top);
-      } else {
-        // not last one
-        assert(new_top > hr->end(), "new_top should be above this region");
-        hr->set_top(hr->end());
-      }
-    }
-    // If we have continues humongous regions (hr != NULL), then the
-    // end of the last one should match new_end and its top should
-    // match new_top.
-    assert(hr == NULL ||
-           (hr->end() == new_end && hr->top() == new_top), "sanity");
-
-    return new_obj;
-  } else {
-    // If we started from the beginning, we want to know why we can't alloc.
-    return NULL;
-  }
-}
-
 void HeapRegionSeq::print_empty_runs() {
   int empty_run = 0;
   int n_empty = 0;
@@ -284,13 +138,67 @@
   return res;
 }
 
-HeapWord* HeapRegionSeq::obj_allocate(size_t word_size) {
-  int cur = _alloc_search_start;
-  // Make sure "cur" is a valid index.
-  assert(cur >= 0, "Invariant.");
-  HeapWord* res = alloc_obj_from_region_index(cur, word_size);
-  if (res == NULL)
-    res = alloc_obj_from_region_index(0, word_size);
+int HeapRegionSeq::find_contiguous_from(int from, size_t num) {
+  assert(num > 1, "pre-condition");
+  assert(0 <= from && from <= _regions.length(),
+         err_msg("from: %d should be valid and <= than %d",
+                 from, _regions.length()));
+
+  int curr = from;
+  int first = -1;
+  size_t num_so_far = 0;
+  while (curr < _regions.length() && num_so_far < num) {
+    HeapRegion* curr_hr = _regions.at(curr);
+    if (curr_hr->is_empty()) {
+      if (first == -1) {
+        first = curr;
+        num_so_far = 1;
+      } else {
+        num_so_far += 1;
+      }
+    } else {
+      first = -1;
+      num_so_far = 0;
+    }
+    curr += 1;
+  }
+
+  assert(num_so_far <= num, "post-condition");
+  if (num_so_far == num) {
+    // we find enough space for the humongous object
+    assert(from <= first && first < _regions.length(), "post-condition");
+    assert(first < curr && (curr - first) == (int) num, "post-condition");
+    for (int i = first; i < first + (int) num; ++i) {
+      assert(_regions.at(i)->is_empty(), "post-condition");
+    }
+    return first;
+  } else {
+    // we failed to find enough space for the humongous object
+    return -1;
+  }
+}
+
+int HeapRegionSeq::find_contiguous(size_t num) {
+  assert(num > 1, "otherwise we should not be calling this");
+  assert(0 <= _alloc_search_start && _alloc_search_start <= _regions.length(),
+         err_msg("_alloc_search_start: %d should be valid and <= than %d",
+                 _alloc_search_start, _regions.length()));
+
+  int start = _alloc_search_start;
+  int res = find_contiguous_from(start, num);
+  if (res == -1 && start != 0) {
+    // Try starting from the beginning. If _alloc_search_start was 0,
+    // no point in doing this again.
+    res = find_contiguous_from(0, num);
+  }
+  if (res != -1) {
+    assert(0 <= res && res < _regions.length(),
+           err_msg("res: %d should be valid", res));
+    _alloc_search_start = res + (int) num;
+  }
+  assert(0 < _alloc_search_start && _alloc_search_start <= _regions.length(),
+         err_msg("_alloc_search_start: %d should be valid",
+                 _alloc_search_start));
   return res;
 }
 
@@ -376,6 +284,10 @@
 
 MemRegion HeapRegionSeq::shrink_by(size_t shrink_bytes,
                                    size_t& num_regions_deleted) {
+  // Reset this in case it's currently pointing into the regions that
+  // we just removed.
+  _alloc_search_start = 0;
+
   assert(shrink_bytes % os::vm_page_size() == 0, "unaligned");
   assert(shrink_bytes % HeapRegion::GrainBytes == 0, "unaligned");
 
@@ -395,7 +307,6 @@
     }
     assert(cur == _regions.top(), "Should be top");
     if (!cur->is_empty()) break;
-    cur->reset_zero_fill();
     shrink_bytes -= cur->capacity();
     num_regions_deleted++;
     _regions.pop();
@@ -410,7 +321,6 @@
   return MemRegion(last_start, end);
 }
 
-
 class PrintHeapRegionClosure : public  HeapRegionClosure {
 public:
   bool doHeapRegion(HeapRegion* r) {
--- a/src/share/vm/gc_implementation/g1/heapRegionSeq.hpp	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/share/vm/gc_implementation/g1/heapRegionSeq.hpp	Tue Feb 01 18:57:53 2011 -0800
@@ -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
@@ -41,9 +41,9 @@
   // (For efficiency only; private to obj_allocate after initialization.)
   int _alloc_search_start;
 
-  // Attempts to allocate a block of the (assumed humongous) word_size,
-  // starting at the region "ind".
-  HeapWord* alloc_obj_from_region_index(int ind, size_t word_size);
+  // Finds a contiguous set of empty regions of length num, starting
+  // from a given index.
+  int find_contiguous_from(int from, size_t num);
 
   // Currently, we're choosing collection sets in a round-robin fashion,
   // starting here.
@@ -76,11 +76,8 @@
   // that are available for allocation.
   size_t free_suffix();
 
-  // Requires "word_size" to be humongous (in the technical sense).  If
-  // possible, allocates a contiguous subsequence of the heap regions to
-  // satisfy the allocation, and returns the address of the beginning of
-  // that sequence, otherwise returns NULL.
-  HeapWord* obj_allocate(size_t word_size);
+  // Finds a contiguous set of empty regions of length num.
+  int find_contiguous(size_t num);
 
   // Apply the "doHeapRegion" method of "blk" to all regions in "this",
   // in address order, terminating the iteration early
@@ -106,7 +103,7 @@
 
   // If "addr" falls within a region in the sequence, return that region,
   // or else NULL.
-  HeapRegion* addr_to_region(const void* addr);
+  inline HeapRegion* addr_to_region(const void* addr);
 
   void print();
 
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/src/share/vm/gc_implementation/g1/heapRegionSet.cpp	Tue Feb 01 18:57:53 2011 -0800
@@ -0,0 +1,438 @@
+/*
+ * Copyright (c) 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
+ * under the terms of the GNU General Public License version 2 only, as
+ * published by the Free Software Foundation.
+ *
+ * This code is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
+ * version 2 for more details (a copy is included in the LICENSE file that
+ * accompanied this code).
+ *
+ * You should have received a copy of the GNU General Public License version
+ * 2 along with this work; if not, write to the Free Software Foundation,
+ * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
+ *
+ * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
+ * or visit www.oracle.com if you need additional information or have any
+ * questions.
+ *
+ */
+
+#include "precompiled.hpp"
+#include "gc_implementation/g1/heapRegionSet.inline.hpp"
+
+size_t HeapRegionSetBase::_unrealistically_long_length = 0;
+
+//////////////////// HeapRegionSetBase ////////////////////
+
+void HeapRegionSetBase::set_unrealistically_long_length(size_t len) {
+  guarantee(_unrealistically_long_length == 0, "should only be set once");
+  _unrealistically_long_length = len;
+}
+
+size_t HeapRegionSetBase::calculate_region_num(HeapRegion* hr) {
+  assert(hr->startsHumongous(), "pre-condition");
+  assert(hr->capacity() % HeapRegion::GrainBytes == 0, "invariant");
+  size_t region_num = hr->capacity() >> HeapRegion::LogOfHRGrainBytes;
+  assert(region_num > 0, "sanity");
+  return region_num;
+}
+
+void HeapRegionSetBase::fill_in_ext_msg(hrl_ext_msg* msg, const char* message) {
+  msg->append("[%s] %s "
+              "ln: "SIZE_FORMAT" rn: "SIZE_FORMAT" "
+              "cy: "SIZE_FORMAT" ud: "SIZE_FORMAT,
+              name(), message, length(), region_num(),
+              total_capacity_bytes(), total_used_bytes());
+  fill_in_ext_msg_extra(msg);
+}
+
+bool HeapRegionSetBase::verify_region(HeapRegion* hr,
+                                  HeapRegionSetBase* expected_containing_set) {
+  const char* error_message = NULL;
+
+  if (!regions_humongous()) {
+    if (hr->isHumongous()) {
+      error_message = "the region should not be humongous";
+    }
+  } else {
+    if (!hr->isHumongous() || !hr->startsHumongous()) {
+      error_message = "the region should be 'starts humongous'";
+    }
+  }
+
+  if (!regions_empty()) {
+    if (hr->is_empty()) {
+      error_message = "the region should not be empty";
+    }
+  } else {
+    if (!hr->is_empty()) {
+      error_message = "the region should be empty";
+    }
+  }
+
+#ifdef ASSERT
+  // The _containing_set field is only available when ASSERT is defined.
+  if (hr->containing_set() != expected_containing_set) {
+    error_message = "inconsistent containing set found";
+  }
+#endif // ASSERT
+
+  const char* extra_error_message = verify_region_extra(hr);
+  if (extra_error_message != NULL) {
+    error_message = extra_error_message;
+  }
+
+  if (error_message != NULL) {
+    outputStream* out = tty;
+    out->cr();
+    out->print_cr("## [%s] %s", name(), error_message);
+    out->print_cr("## Offending Region: "PTR_FORMAT, hr);
+    out->print_cr("   "HR_FORMAT, HR_FORMAT_PARAMS(hr));
+#ifdef ASSERT
+    out->print_cr("   containing set: "PTR_FORMAT, hr->containing_set());
+#endif // ASSERT
+    out->print_cr("## Offending Region Set: "PTR_FORMAT, this);
+    print_on(out);
+    return false;
+  } else {
+    return true;
+  }
+}
+
+void HeapRegionSetBase::verify() {
+  // It's important that we also observe the MT safety protocol even
+  // for the verification calls. If we do verification without the
+  // appropriate locks and the set changes underneath our feet
+  // verification might fail and send us on a wild goose chase.
+  hrl_assert_mt_safety_ok(this);
+
+  guarantee(( is_empty() && length() == 0 && region_num() == 0 &&
+              total_used_bytes() == 0 && total_capacity_bytes() == 0) ||
+            (!is_empty() && length() >= 0 && region_num() >= 0 &&
+              total_used_bytes() >= 0 && total_capacity_bytes() >= 0),
+            hrl_ext_msg(this, "invariant"));
+
+  guarantee((!regions_humongous() && region_num() == length()) ||
+            ( regions_humongous() && region_num() >= length()),
+            hrl_ext_msg(this, "invariant"));
+
+  guarantee(!regions_empty() || total_used_bytes() == 0,
+            hrl_ext_msg(this, "invariant"));
+
+  guarantee(total_used_bytes() <= total_capacity_bytes(),
+            hrl_ext_msg(this, "invariant"));
+}
+
+void HeapRegionSetBase::verify_start() {
+  // See comment in verify() about MT safety and verification.
+  hrl_assert_mt_safety_ok(this);
+  assert(!_verify_in_progress,
+         hrl_ext_msg(this, "verification should not be in progress"));
+
+  // Do the basic verification first before we do the checks over the regions.
+  HeapRegionSetBase::verify();
+
+  _calc_length               = 0;
+  _calc_region_num           = 0;
+  _calc_total_capacity_bytes = 0;
+  _calc_total_used_bytes     = 0;
+  _verify_in_progress        = true;
+}
+
+void HeapRegionSetBase::verify_next_region(HeapRegion* hr) {
+  // See comment in verify() about MT safety and verification.
+  hrl_assert_mt_safety_ok(this);
+  assert(_verify_in_progress,
+         hrl_ext_msg(this, "verification should be in progress"));
+
+  guarantee(verify_region(hr, this), hrl_ext_msg(this, "region verification"));
+
+  _calc_length               += 1;
+  if (!hr->isHumongous()) {
+    _calc_region_num         += 1;
+  } else {
+    _calc_region_num         += calculate_region_num(hr);
+  }
+  _calc_total_capacity_bytes += hr->capacity();
+  _calc_total_used_bytes     += hr->used();
+}
+
+void HeapRegionSetBase::verify_end() {
+  // See comment in verify() about MT safety and verification.
+  hrl_assert_mt_safety_ok(this);
+  assert(_verify_in_progress,
+         hrl_ext_msg(this, "verification should be in progress"));
+
+  guarantee(length() == _calc_length,
+            hrl_err_msg("[%s] length: "SIZE_FORMAT" should be == "
+                        "calc length: "SIZE_FORMAT,
+                        name(), length(), _calc_length));
+
+  guarantee(region_num() == _calc_region_num,
+            hrl_err_msg("[%s] region num: "SIZE_FORMAT" should be == "
+                        "calc region num: "SIZE_FORMAT,
+                        name(), region_num(), _calc_region_num));
+
+  guarantee(total_capacity_bytes() == _calc_total_capacity_bytes,
+            hrl_err_msg("[%s] capacity bytes: "SIZE_FORMAT" should be == "
+                        "calc capacity bytes: "SIZE_FORMAT,
+                        name(),
+                        total_capacity_bytes(), _calc_total_capacity_bytes));
+
+  guarantee(total_used_bytes() == _calc_total_used_bytes,
+            hrl_err_msg("[%s] used bytes: "SIZE_FORMAT" should be == "
+                        "calc used bytes: "SIZE_FORMAT,
+                        name(), total_used_bytes(), _calc_total_used_bytes));
+
+  _verify_in_progress = false;
+}
+
+void HeapRegionSetBase::print_on(outputStream* out, bool print_contents) {
+  out->cr();
+  out->print_cr("Set: %s ("PTR_FORMAT")", name(), this);
+  out->print_cr("  Region Assumptions");
+  out->print_cr("    humongous         : %s", BOOL_TO_STR(regions_humongous()));
+  out->print_cr("    empty             : %s", BOOL_TO_STR(regions_empty()));
+  out->print_cr("  Attributes");
+  out->print_cr("    length            : "SIZE_FORMAT_W(14), length());
+  out->print_cr("    region num        : "SIZE_FORMAT_W(14), region_num());
+  out->print_cr("    total capacity    : "SIZE_FORMAT_W(14)" bytes",
+                total_capacity_bytes());
+  out->print_cr("    total used        : "SIZE_FORMAT_W(14)" bytes",
+                total_used_bytes());
+}
+
+void HeapRegionSetBase::clear() {
+  _length           = 0;
+  _region_num       = 0;
+  _total_used_bytes = 0;
+}
+
+HeapRegionSetBase::HeapRegionSetBase(const char* name)
+  : _name(name), _verify_in_progress(false),
+    _calc_length(0), _calc_region_num(0),
+    _calc_total_capacity_bytes(0), _calc_total_used_bytes(0) { }
+
+//////////////////// HeapRegionSet ////////////////////
+
+void HeapRegionSet::update_from_proxy(HeapRegionSet* proxy_set) {
+  hrl_assert_mt_safety_ok(this);
+  hrl_assert_mt_safety_ok(proxy_set);
+  hrl_assert_sets_match(this, proxy_set);
+
+  verify_optional();
+  proxy_set->verify_optional();
+
+  if (proxy_set->is_empty()) return;
+
+  assert(proxy_set->length() <= _length,
+         hrl_err_msg("[%s] proxy set length: "SIZE_FORMAT" "
+                     "should be <= length: "SIZE_FORMAT,
+                     name(), proxy_set->length(), _length));
+  _length -= proxy_set->length();
+
+  assert(proxy_set->region_num() <= _region_num,
+         hrl_err_msg("[%s] proxy set region num: "SIZE_FORMAT" "
+                     "should be <= region num: "SIZE_FORMAT,
+                     name(), proxy_set->region_num(), _region_num));
+  _region_num -= proxy_set->region_num();
+
+  assert(proxy_set->total_used_bytes() <= _total_used_bytes,
+         hrl_err_msg("[%s] proxy set used bytes: "SIZE_FORMAT" "
+                     "should be <= used bytes: "SIZE_FORMAT,
+                     name(), proxy_set->total_used_bytes(),
+                     _total_used_bytes));
+  _total_used_bytes -= proxy_set->total_used_bytes();
+
+  proxy_set->clear();
+
+  verify_optional();
+  proxy_set->verify_optional();
+}
+
+//////////////////// HeapRegionLinkedList ////////////////////
+
+void HeapRegionLinkedList::fill_in_ext_msg_extra(hrl_ext_msg* msg) {
+  msg->append(" hd: "PTR_FORMAT" tl: "PTR_FORMAT, head(), tail());
+}
+
+void HeapRegionLinkedList::add_as_tail(HeapRegionLinkedList* from_list) {
+  hrl_assert_mt_safety_ok(this);
+  hrl_assert_mt_safety_ok(from_list);
+
+  verify_optional();
+  from_list->verify_optional();
+
+  if (from_list->is_empty()) return;
+
+#ifdef ASSERT
+  HeapRegionLinkedListIterator iter(from_list);
+  while (iter.more_available()) {
+    HeapRegion* hr = iter.get_next();
+    // In set_containing_set() we check that we either set the value
+    // from NULL to non-NULL or vice versa to catch bugs. So, we have
+    // to NULL it first before setting it to the value.
+    hr->set_containing_set(NULL);
+    hr->set_containing_set(this);
+  }
+#endif // ASSERT
+
+  if (_tail != NULL) {
+    assert(length() >  0 && _head != NULL, hrl_ext_msg(this, "invariant"));
+    _tail->set_next(from_list->_head);
+  } else {
+    assert(length() == 0 && _head == NULL, hrl_ext_msg(this, "invariant"));
+    _head = from_list->_head;
+  }
+  _tail = from_list->_tail;
+
+  _length           += from_list->length();
+  _region_num       += from_list->region_num();
+  _total_used_bytes += from_list->total_used_bytes();
+  from_list->clear();
+
+  verify_optional();
+  from_list->verify_optional();
+}
+
+void HeapRegionLinkedList::remove_all() {
+  hrl_assert_mt_safety_ok(this);
+  verify_optional();
+
+  HeapRegion* curr = _head;
+  while (curr != NULL) {
+    hrl_assert_region_ok(this, curr, this);
+
+    HeapRegion* next = curr->next();
+    curr->set_next(NULL);
+    curr->set_containing_set(NULL);
+    curr = next;
+  }
+  clear();
+
+  verify_optional();
+}
+
+void HeapRegionLinkedList::remove_all_pending(size_t target_count) {
+  hrl_assert_mt_safety_ok(this);
+  assert(target_count > 1, hrl_ext_msg(this, "pre-condition"));
+  assert(!is_empty(), hrl_ext_msg(this, "pre-condition"));
+
+  verify_optional();
+  DEBUG_ONLY(size_t old_length = length();)
+
+  HeapRegion* curr = _head;
+  HeapRegion* prev = NULL;
+  size_t count = 0;
+  while (curr != NULL) {
+    hrl_assert_region_ok(this, curr, this);
+    HeapRegion* next = curr->next();
+
+    if (curr->pending_removal()) {
+      assert(count < target_count,
+             hrl_err_msg("[%s] should not come across more regions "
+                         "pending for removal than target_count: "SIZE_FORMAT,
+                         name(), target_count));
+
+      if (prev == NULL) {
+        assert(_head == curr, hrl_ext_msg(this, "invariant"));
+        _head = next;
+      } else {
+        assert(_head != curr, hrl_ext_msg(this, "invariant"));
+        prev->set_next(next);
+      }
+      if (next == NULL) {
+        assert(_tail == curr, hrl_ext_msg(this, "invariant"));
+        _tail = prev;
+      } else {
+        assert(_tail != curr, hrl_ext_msg(this, "invariant"));
+      }
+
+      curr->set_next(NULL);
+      remove_internal(curr);
+      curr->set_pending_removal(false);
+
+      count += 1;
+
+      // If we have come across the target number of regions we can
+      // just bail out. However, for debugging purposes, we can just
+      // carry on iterating to make sure there are not more regions
+      // tagged with pending removal.
+      DEBUG_ONLY(if (count == target_count) break;)
+    } else {
+      prev = curr;
+    }
+    curr = next;
+  }
+
+  assert(count == target_count,
+         hrl_err_msg("[%s] count: "SIZE_FORMAT" should be == "
+                     "target_count: "SIZE_FORMAT, name(), count, target_count));
+  assert(length() + target_count == old_length,
+         hrl_err_msg("[%s] new length should be consistent "
+                     "new length: "SIZE_FORMAT" old length: "SIZE_FORMAT" "
+                     "target_count: "SIZE_FORMAT,
+                     name(), length(), old_length, target_count));
+
+  verify_optional();
+}
+
+void HeapRegionLinkedList::verify() {
+  // See comment in HeapRegionSetBase::verify() about MT safety and
+  // verification.
+  hrl_assert_mt_safety_ok(this);
+
+  // This will also do the basic verification too.
+  verify_start();
+
+  HeapRegion* curr  = _head;
+  HeapRegion* prev1 = NULL;
+  HeapRegion* prev0 = NULL;
+  size_t      count = 0;
+  while (curr != NULL) {
+    verify_next_region(curr);
+
+    count += 1;
+    guarantee(count < _unrealistically_long_length,
+              hrl_err_msg("[%s] the calculated length: "SIZE_FORMAT" "
+                          "seems very long, is there maybe a cycle? "
+                          "curr: "PTR_FORMAT" prev0: "PTR_FORMAT" "
+                          "prev1: "PTR_FORMAT" length: "SIZE_FORMAT,
+                          name(), count, curr, prev0, prev1, length()));
+
+    prev1 = prev0;
+    prev0 = curr;
+    curr  = curr->next();
+  }
+
+  guarantee(_tail == prev0, hrl_ext_msg(this, "post-condition"));
+
+  verify_end();
+}
+
+void HeapRegionLinkedList::clear() {
+  HeapRegionSetBase::clear();
+  _head = NULL;
+  _tail = NULL;
+}
+
+void HeapRegionLinkedList::print_on(outputStream* out, bool print_contents) {
+  HeapRegionSetBase::print_on(out, print_contents);
+  out->print_cr("  Linking");
+  out->print_cr("    head              : "PTR_FORMAT, _head);
+  out->print_cr("    tail              : "PTR_FORMAT, _tail);
+
+  if (print_contents) {
+    out->print_cr("  Contents");
+    HeapRegionLinkedListIterator iter(this);
+    while (iter.more_available()) {
+      HeapRegion* hr = iter.get_next();
+      hr->print_on(out);
+    }
+  }
+}
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/src/share/vm/gc_implementation/g1/heapRegionSet.hpp	Tue Feb 01 18:57:53 2011 -0800
@@ -0,0 +1,346 @@
+/*
+ * copyright (c) 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
+ * under the terms of the GNU General Public License version 2 only, as
+ * published by the Free Software Foundation.
+ *
+ * This code is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
+ * version 2 for more details (a copy is included in the LICENSE file that
+ * accompanied this code).
+ *
+ * You should have received a copy of the GNU General Public License version
+ * 2 along with this work; if not, write to the Free Software Foundation,
+ * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
+ *
+ * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
+ * or visit www.oracle.com if you need additional information or have any
+ * questions.
+ *
+ */
+
+#ifndef SHARE_VM_GC_IMPLEMENTATION_G1_HEAPREGIONSET_HPP
+#define SHARE_VM_GC_IMPLEMENTATION_G1_HEAPREGIONSET_HPP
+
+#include "gc_implementation/g1/heapRegion.hpp"
+
+// Large buffer for some cases where the output might be larger than normal.
+#define HRL_ERR_MSG_BUFSZ 512
+typedef FormatBuffer<HRL_ERR_MSG_BUFSZ> hrl_err_msg;
+
+// Set verification will be forced either if someone defines
+// HEAP_REGION_SET_FORCE_VERIFY to be 1, or in builds in which
+// asserts are compiled in.
+#ifndef HEAP_REGION_SET_FORCE_VERIFY
+#define HEAP_REGION_SET_FORCE_VERIFY defined(ASSERT)
+#endif // HEAP_REGION_SET_FORCE_VERIFY
+
+//////////////////// HeapRegionSetBase ////////////////////
+
+// Base class for all the classes that represent heap region sets. It
+// contains the basic attributes that each set needs to maintain
+// (e.g., length, region num, used bytes sum) plus any shared
+// functionality (e.g., verification).
+
+class hrl_ext_msg;
+
+class HeapRegionSetBase VALUE_OBJ_CLASS_SPEC {
+  friend class hrl_ext_msg;
+
+protected:
+  static size_t calculate_region_num(HeapRegion* hr);
+
+  static size_t _unrealistically_long_length;
+
+  // The number of regions added to the set. If the set contains
+  // only humongous regions, this reflects only 'starts humongous'
+  // regions and does not include 'continues humongous' ones.
+  size_t _length;
+
+  // The total number of regions represented by the set. If the set
+  // does not contain humongous regions, this should be the same as
+  // _length. If the set contains only humongous regions, this will
+  // include the 'continues humongous' regions.
+  size_t _region_num;
+
+  // We don't keep track of the total capacity explicitly, we instead
+  // recalculate it based on _region_num and the heap region size.
+
+  // The sum of used bytes in the all the regions in the set.
+  size_t _total_used_bytes;
+
+  const char* _name;
+
+  bool        _verify_in_progress;
+  size_t      _calc_length;
+  size_t      _calc_region_num;
+  size_t      _calc_total_capacity_bytes;
+  size_t      _calc_total_used_bytes;
+
+  // verify_region() is used to ensure that the contents of a region
+  // added to / removed from a set are consistent. Different sets
+  // make different assumptions about the regions added to them. So
+  // each set can override verify_region_extra(), which is called
+  // from verify_region(), and do any extra verification it needs to
+  // perform in that.
+  virtual const char* verify_region_extra(HeapRegion* hr) { return NULL; }
+  bool verify_region(HeapRegion* hr,
+                     HeapRegionSetBase* expected_containing_set);
+
+  // Indicates whether all regions in the set should be humongous or
+  // not. Only used during verification.
+  virtual bool regions_humongous() = 0;
+
+  // Indicates whether all regions in the set should be empty or
+  // not. Only used during verification.
+  virtual bool regions_empty() = 0;
+
+  // Subclasses can optionally override this to do MT safety protocol
+  // checks. It is called in an assert from all methods that perform
+  // updates on the set (and subclasses should also call it too).
+  virtual bool check_mt_safety() { return true; }
+
+  // fill_in_ext_msg() writes the the values of the set's attributes
+  // in the custom err_msg (hrl_ext_msg). fill_in_ext_msg_extra()
+  // allows subclasses to append further information.
+  virtual void fill_in_ext_msg_extra(hrl_ext_msg* msg) { }
+  void fill_in_ext_msg(hrl_ext_msg* msg, const char* message);
+
+  // It updates the fields of the set to reflect hr being added to
+  // the set.
+  inline void update_for_addition(HeapRegion* hr);
+
+  // It updates the fields of the set to reflect hr being added to
+  // the set and tags the region appropriately.
+  inline void add_internal(HeapRegion* hr);
+
+  // It updates the fields of the set to reflect hr being removed
+  // from the set.
+  inline void update_for_removal(HeapRegion* hr);
+
+  // It updates the fields of the set to reflect hr being removed
+  // from the set and tags the region appropriately.
+  inline void remove_internal(HeapRegion* hr);
+
+  // It clears all the fields of the sets. Note: it will not iterate
+  // over the set and remove regions from it. It assumes that the
+  // caller has already done so. It will literally just clear the fields.
+  virtual void clear();
+
+  HeapRegionSetBase(const char* name);
+
+public:
+  static void set_unrealistically_long_length(size_t len);
+
+  const char* name() { return _name; }
+
+  size_t length() { return _length; }
+
+  bool is_empty() { return _length == 0; }
+
+  size_t region_num() { return _region_num; }
+
+  size_t total_capacity_bytes() {
+    return region_num() << HeapRegion::LogOfHRGrainBytes;
+  }
+
+  size_t total_used_bytes() { return _total_used_bytes; }
+
+  virtual void verify();
+  void verify_start();
+  void verify_next_region(HeapRegion* hr);
+  void verify_end();
+
+#if HEAP_REGION_SET_FORCE_VERIFY
+  void verify_optional() {
+    verify();
+  }
+#else // HEAP_REGION_SET_FORCE_VERIFY
+  void verify_optional() { }
+#endif // HEAP_REGION_SET_FORCE_VERIFY
+
+  virtual void print_on(outputStream* out, bool print_contents = false);
+};
+
+// Customized err_msg for heap region sets. Apart from a
+// assert/guarantee-specific message it also prints out the values of
+// the fields of the associated set. This can be very helpful in
+// diagnosing failures.
+
+class hrl_ext_msg : public hrl_err_msg {
+public:
+  hrl_ext_msg(HeapRegionSetBase* set, const char* message) : hrl_err_msg("") {
+    set->fill_in_ext_msg(this, message);
+  }
+};
+
+// These two macros are provided for convenience, to keep the uses of
+// these two asserts a bit more concise.
+
+#define hrl_assert_mt_safety_ok(_set_)                                        \
+  do {                                                                        \
+    assert((_set_)->check_mt_safety(), hrl_ext_msg((_set_), "MT safety"));    \
+  } while (0)
+
+#define hrl_assert_region_ok(_set_, _hr_, _expected_)                         \
+  do {                                                                        \
+    assert((_set_)->verify_region((_hr_), (_expected_)),                      \
+           hrl_ext_msg((_set_), "region verification"));                      \
+  } while (0)
+
+//////////////////// HeapRegionSet ////////////////////
+
+#define hrl_assert_sets_match(_set1_, _set2_)                                 \
+  do {                                                                        \
+    assert(((_set1_)->regions_humongous() ==                                  \
+                                            (_set2_)->regions_humongous()) && \
+           ((_set1_)->regions_empty() == (_set2_)->regions_empty()),          \
+           hrl_err_msg("the contents of set %s and set %s should match",      \
+                       (_set1_)->name(), (_set2_)->name()));                  \
+  } while (0)
+
+// This class represents heap region sets whose members are not
+// explicitly tracked. It's helpful to group regions using such sets
+// so that we can reason about all the region groups in the heap using
+// the same interface (namely, the HeapRegionSetBase API).
+
+class HeapRegionSet : public HeapRegionSetBase {
+protected:
+  virtual const char* verify_region_extra(HeapRegion* hr) {
+    if (hr->next() != NULL) {
+      return "next() should always be NULL as we do not link the regions";
+    }
+
+    return HeapRegionSetBase::verify_region_extra(hr);
+  }
+
+  HeapRegionSet(const char* name) : HeapRegionSetBase(name) {
+    clear();
+  }
+
+public:
+  // It adds hr to the set. The region should not be a member of
+  // another set.
+  inline void add(HeapRegion* hr);
+
+  // It removes hr from the set. The region should be a member of
+  // this set.
+  inline void remove(HeapRegion* hr);
+
+  // It removes a region from the set. Instead of updating the fields
+  // of the set to reflect this removal, it accumulates the updates
+  // in proxy_set. The idea is that proxy_set is thread-local to
+  // avoid multiple threads updating the fields of the set
+  // concurrently and having to synchronize. The method
+  // update_from_proxy() will update the fields of the set from the
+  // proxy_set.
+  inline void remove_with_proxy(HeapRegion* hr, HeapRegionSet* proxy_set);
+
+  // After multiple calls to remove_with_proxy() the updates to the
+  // fields of the set are accumulated in proxy_set. This call
+  // updates the fields of the set from proxy_set.
+  void update_from_proxy(HeapRegionSet* proxy_set);
+};
+
+//////////////////// HeapRegionLinkedList ////////////////////
+
+// A set that links all the regions added to it in a singly-linked
+// list. We should try to avoid doing operations that iterate over
+// such lists in performance critical paths. Typically we should
+// add / remove one region at a time or concatenate two lists. All
+// those operations are done in constant time.
+
+class HeapRegionLinkedListIterator;
+
+class HeapRegionLinkedList : public HeapRegionSetBase {
+  friend class HeapRegionLinkedListIterator;
+
+private:
+  HeapRegion* _head;
+  HeapRegion* _tail;
+
+  // These are provided for use by the friend classes.
+  HeapRegion* head() { return _head; }
+  HeapRegion* tail() { return _tail; }
+
+protected:
+  virtual void fill_in_ext_msg_extra(hrl_ext_msg* msg);
+
+  // See the comment for HeapRegionSetBase::clear()
+  virtual void clear();
+
+  HeapRegionLinkedList(const char* name) : HeapRegionSetBase(name) {
+    clear();
+  }
+
+public:
+  // It adds hr to the list as the new tail. The region should not be
+  // a member of another set.
+  inline void add_as_tail(HeapRegion* hr);
+
+  // It removes and returns the head of the list. It assumes that the
+  // list is not empty so it will return a non-NULL value.
+  inline HeapRegion* remove_head();
+
+  // Convenience method.
+  inline HeapRegion* remove_head_or_null();
+
+  // It moves the regions from from_list to this list and empties
+  // from_list. The new regions will appear in the same order as they
+  // were in from_list and be linked in the end of this list.
+  void add_as_tail(HeapRegionLinkedList* from_list);
+
+  // It empties the list by removing all regions from it.
+  void remove_all();
+
+  // It removes all regions in the list that are pending for removal
+  // (i.e., they have been tagged with "pending_removal"). The list
+  // must not be empty, target_count should reflect the exact number
+  // of regions that are pending for removal in the list, and
+  // target_count should be > 1 (currently, we never need to remove a
+  // single region using this).
+  void remove_all_pending(size_t target_count);
+
+  virtual void verify();
+
+  virtual void print_on(outputStream* out, bool print_contents = false);
+};
+
+//////////////////// HeapRegionLinkedList ////////////////////
+
+// Iterator class that provides a convenient way to iterator over the
+// regions in a HeapRegionLinkedList instance.
+
+class HeapRegionLinkedListIterator : public StackObj {
+private:
+  HeapRegionLinkedList* _list;
+  HeapRegion*           _curr;
+
+public:
+  bool more_available() {
+    return _curr != NULL;
+  }
+
+  HeapRegion* get_next() {
+    assert(more_available(),
+           "get_next() should be called when more regions are available");
+
+    // If we are going to introduce a count in the iterator we should
+    // do the "cycle" check.
+
+    HeapRegion* hr = _curr;
+    assert(_list->verify_region(hr, _list), "region verification");
+    _curr = hr->next();
+    return hr;
+  }
+
+  HeapRegionLinkedListIterator(HeapRegionLinkedList* list)
+    : _curr(NULL), _list(list) {
+    _curr = list->head();
+  }
+};
+
+#endif // SHARE_VM_GC_IMPLEMENTATION_G1_HEAPREGIONSET_HPP
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/src/share/vm/gc_implementation/g1/heapRegionSet.inline.hpp	Tue Feb 01 18:57:53 2011 -0800
@@ -0,0 +1,159 @@
+/*
+ * copyright (c) 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
+ * under the terms of the GNU General Public License version 2 only, as
+ * published by the Free Software Foundation.
+ *
+ * This code is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
+ * version 2 for more details (a copy is included in the LICENSE file that
+ * accompanied this code).
+ *
+ * You should have received a copy of the GNU General Public License version
+ * 2 along with this work; if not, write to the Free Software Foundation,
+ * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
+ *
+ * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
+ * or visit www.oracle.com if you need additional information or have any
+ * questions.
+ *
+ */
+
+#ifndef SHARE_VM_GC_IMPLEMENTATION_G1_HEAPREGIONSET_INLINE_HPP
+#define SHARE_VM_GC_IMPLEMENTATION_G1_HEAPREGIONSET_INLINE_HPP
+
+#include "gc_implementation/g1/heapRegionSet.hpp"
+
+//////////////////// HeapRegionSetBase ////////////////////
+
+inline void HeapRegionSetBase::update_for_addition(HeapRegion* hr) {
+  // Assumes the caller has already verified the region.
+
+  _length           += 1;
+  if (!hr->isHumongous()) {
+    _region_num     += 1;
+  } else {
+    _region_num     += calculate_region_num(hr);
+  }
+  _total_used_bytes += hr->used();
+}
+
+inline void HeapRegionSetBase::add_internal(HeapRegion* hr) {
+  hrl_assert_region_ok(this, hr, NULL);
+  assert(hr->next() == NULL, hrl_ext_msg(this, "should not already be linked"));
+
+  update_for_addition(hr);
+  hr->set_containing_set(this);
+}
+
+inline void HeapRegionSetBase::update_for_removal(HeapRegion* hr) {
+  // Assumes the caller has already verified the region.
+  assert(_length > 0, hrl_ext_msg(this, "pre-condition"));
+  _length -= 1;
+
+  size_t region_num_diff;
+  if (!hr->isHumongous()) {
+    region_num_diff = 1;
+  } else {
+    region_num_diff = calculate_region_num(hr);
+  }
+  assert(region_num_diff <= _region_num,
+         hrl_err_msg("[%s] region's region num: "SIZE_FORMAT" "
+                     "should be <= region num: "SIZE_FORMAT,
+                     name(), region_num_diff, _region_num));
+  _region_num -= region_num_diff;
+
+  size_t used_bytes = hr->used();
+  assert(used_bytes <= _total_used_bytes,
+         hrl_err_msg("[%s] region's used bytes: "SIZE_FORMAT" "
+                     "should be <= used bytes: "SIZE_FORMAT,
+                     name(), used_bytes, _total_used_bytes));
+  _total_used_bytes -= used_bytes;
+}
+
+inline void HeapRegionSetBase::remove_internal(HeapRegion* hr) {
+  hrl_assert_region_ok(this, hr, this);
+  assert(hr->next() == NULL, hrl_ext_msg(this, "should already be unlinked"));
+
+  hr->set_containing_set(NULL);
+  update_for_removal(hr);
+}
+
+//////////////////// HeapRegionSet ////////////////////
+
+inline void HeapRegionSet::add(HeapRegion* hr) {
+  hrl_assert_mt_safety_ok(this);
+  // add_internal() will verify the region.
+  add_internal(hr);
+}
+
+inline void HeapRegionSet::remove(HeapRegion* hr) {
+  hrl_assert_mt_safety_ok(this);
+  // remove_internal() will verify the region.
+  remove_internal(hr);
+}
+
+inline void HeapRegionSet::remove_with_proxy(HeapRegion* hr,
+                                             HeapRegionSet* proxy_set) {
+  // No need to fo the MT safety check here given that this method
+  // does not update the contents of the set but instead accumulates
+  // the changes in proxy_set which is assumed to be thread-local.
+  hrl_assert_sets_match(this, proxy_set);
+  hrl_assert_region_ok(this, hr, this);
+
+  hr->set_containing_set(NULL);
+  proxy_set->update_for_addition(hr);
+}
+
+//////////////////// HeapRegionLinkedList ////////////////////
+
+inline void HeapRegionLinkedList::add_as_tail(HeapRegion* hr) {
+  hrl_assert_mt_safety_ok(this);
+  assert((length() == 0 && _head == NULL && _tail == NULL) ||
+         (length() >  0 && _head != NULL && _tail != NULL),
+         hrl_ext_msg(this, "invariant"));
+  // add_internal() will verify the region.
+  add_internal(hr);
+
+  // Now link the region.
+  if (_tail != NULL) {
+    _tail->set_next(hr);
+  } else {
+    _head = hr;
+  }
+  _tail = hr;
+}
+
+inline HeapRegion* HeapRegionLinkedList::remove_head() {
+  hrl_assert_mt_safety_ok(this);
+  assert(!is_empty(), hrl_ext_msg(this, "the list should not be empty"));
+  assert(length() > 0 && _head != NULL && _tail != NULL,
+         hrl_ext_msg(this, "invariant"));
+
+  // We need to unlink it first.
+  HeapRegion* hr = _head;
+  _head = hr->next();
+  if (_head == NULL) {
+    _tail = NULL;
+  }
+  hr->set_next(NULL);
+
+  // remove_internal() will verify the region.
+  remove_internal(hr);
+  return hr;
+}
+
+inline HeapRegion* HeapRegionLinkedList::remove_head_or_null() {
+  hrl_assert_mt_safety_ok(this);
+
+  if (!is_empty()) {
+    return remove_head();
+  } else {
+    return NULL;
+  }
+}
+
+#endif // SHARE_VM_GC_IMPLEMENTATION_G1_HEAPREGIONSET_INLINE_HPP
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/src/share/vm/gc_implementation/g1/heapRegionSets.cpp	Tue Feb 01 18:57:53 2011 -0800
@@ -0,0 +1,102 @@
+/*
+ * Copyright (c) 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
+ * under the terms of the GNU General Public License version 2 only, as
+ * published by the Free Software Foundation.
+ *
+ * This code is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
+ * version 2 for more details (a copy is included in the LICENSE file that
+ * accompanied this code).
+ *
+ * You should have received a copy of the GNU General Public License version
+ * 2 along with this work; if not, write to the Free Software Foundation,
+ * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
+ *
+ * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
+ * or visit www.oracle.com if you need additional information or have any
+ * questions.
+ *
+ */
+
+#include "precompiled.hpp"
+#include "gc_implementation/g1/heapRegionSets.hpp"
+
+//////////////////// FreeRegionList ////////////////////
+
+const char* FreeRegionList::verify_region_extra(HeapRegion* hr) {
+  if (hr->is_young()) {
+    return "the region should not be young";
+  }
+  // The superclass will check that the region is empty and
+  // not-humongous.
+  return HeapRegionLinkedList::verify_region_extra(hr);
+}
+
+//////////////////// MasterFreeRegionList ////////////////////
+
+bool MasterFreeRegionList::check_mt_safety() {
+  // Master Free List MT safety protocol:
+  // (a) If we're at a safepoint, operations on the master free list
+  // should be invoked by either the VM thread (which will serialize
+  // them) or by the GC workers while holding the
+  // FreeList_lock.
+  // (b) If we're not at a safepoint, operations on the master free
+  // list should be invoked while holding the Heap_lock.
+
+  guarantee((SafepointSynchronize::is_at_safepoint() &&
+               (Thread::current()->is_VM_thread() ||
+                                            FreeList_lock->owned_by_self())) ||
+            (!SafepointSynchronize::is_at_safepoint() &&
+                                                Heap_lock->owned_by_self()),
+            hrl_ext_msg(this, "master free list MT safety protocol"));
+
+  return FreeRegionList::check_mt_safety();
+}
+
+//////////////////// SecondaryFreeRegionList ////////////////////
+
+bool SecondaryFreeRegionList::check_mt_safety() {
+  // Secondary Free List MT safety protocol:
+  // Operations on the secondary free list should always be invoked
+  // while holding the SecondaryFreeList_lock.
+
+  guarantee(SecondaryFreeList_lock->owned_by_self(),
+            hrl_ext_msg(this, "secondary free list MT safety protocol"));
+
+  return FreeRegionList::check_mt_safety();
+}
+
+//////////////////// HumongousRegionSet ////////////////////
+
+const char* HumongousRegionSet::verify_region_extra(HeapRegion* hr) {
+  if (hr->is_young()) {
+    return "the region should not be young";
+  }
+  // The superclass will check that the region is not empty and
+  // humongous.
+  return HeapRegionSet::verify_region_extra(hr);
+}
+
+//////////////////// HumongousRegionSet ////////////////////
+
+bool MasterHumongousRegionSet::check_mt_safety() {
+  // Master Humongous Set MT safety protocol:
+  // (a) If we're at a safepoint, operations on the master humongous
+  // set should be invoked by either the VM thread (which will
+  // serialize them) or by the GC workers while holding the
+  // OldSets_lock.
+  // (b) If we're not at a safepoint, operations on the master
+  // humongous set should be invoked while holding the Heap_lock.
+
+  guarantee((SafepointSynchronize::is_at_safepoint() &&
+               (Thread::current()->is_VM_thread() ||
+                                             OldSets_lock->owned_by_self())) ||
+            (!SafepointSynchronize::is_at_safepoint() &&
+                                                 Heap_lock->owned_by_self()),
+            hrl_ext_msg(this, "master humongous set MT safety protocol"));
+  return HumongousRegionSet::check_mt_safety();
+}
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/src/share/vm/gc_implementation/g1/heapRegionSets.hpp	Tue Feb 01 18:57:53 2011 -0800
@@ -0,0 +1,86 @@
+/*
+ * copyright (c) 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
+ * under the terms of the GNU General Public License version 2 only, as
+ * published by the Free Software Foundation.
+ *
+ * This code is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
+ * version 2 for more details (a copy is included in the LICENSE file that
+ * accompanied this code).
+ *
+ * You should have received a copy of the GNU General Public License version
+ * 2 along with this work; if not, write to the Free Software Foundation,
+ * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
+ *
+ * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
+ * or visit www.oracle.com if you need additional information or have any
+ * questions.
+ *
+ */
+
+#ifndef SHARE_VM_GC_IMPLEMENTATION_G1_HEAPREGIONSETS_HPP
+#define SHARE_VM_GC_IMPLEMENTATION_G1_HEAPREGIONSETS_HPP
+
+#include "gc_implementation/g1/heapRegionSet.inline.hpp"
+
+//////////////////// FreeRegionList ////////////////////
+
+class FreeRegionList : public HeapRegionLinkedList {
+protected:
+  virtual const char* verify_region_extra(HeapRegion* hr);
+
+  virtual bool regions_humongous() { return false; }
+  virtual bool regions_empty()     { return true;  }
+
+public:
+  FreeRegionList(const char* name) : HeapRegionLinkedList(name) { }
+};
+
+//////////////////// MasterFreeRegionList ////////////////////
+
+class MasterFreeRegionList : public FreeRegionList {
+protected:
+  virtual bool check_mt_safety();
+
+public:
+  MasterFreeRegionList(const char* name) : FreeRegionList(name) { }
+};
+
+//////////////////// SecondaryFreeRegionList ////////////////////
+
+class SecondaryFreeRegionList : public FreeRegionList {
+protected:
+  virtual bool check_mt_safety();
+
+public:
+  SecondaryFreeRegionList(const char* name) : FreeRegionList(name) { }
+};
+
+//////////////////// HumongousRegionSet ////////////////////
+
+class HumongousRegionSet : public HeapRegionSet {
+protected:
+  virtual const char* verify_region_extra(HeapRegion* hr);
+
+  virtual bool regions_humongous() { return true;  }
+  virtual bool regions_empty()     { return false; }
+
+public:
+  HumongousRegionSet(const char* name) : HeapRegionSet(name) { }
+};
+
+//////////////////// MasterHumongousRegionSet ////////////////////
+
+class MasterHumongousRegionSet : public HumongousRegionSet {
+protected:
+  virtual bool check_mt_safety();
+
+public:
+  MasterHumongousRegionSet(const char* name) : HumongousRegionSet(name) { }
+};
+
+#endif // SHARE_VM_GC_IMPLEMENTATION_G1_HEAPREGIONSETS_HPP
--- a/src/share/vm/gc_implementation/g1/ptrQueue.cpp	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/share/vm/gc_implementation/g1/ptrQueue.cpp	Tue Feb 01 18:57:53 2011 -0800
@@ -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
@@ -38,8 +38,8 @@
 # include "thread_windows.inline.hpp"
 #endif
 
-PtrQueue::PtrQueue(PtrQueueSet* qset_, bool perm, bool active) :
-  _qset(qset_), _buf(NULL), _index(0), _active(active),
+PtrQueue::PtrQueue(PtrQueueSet* qset, bool perm, bool active) :
+  _qset(qset), _buf(NULL), _index(0), _active(active),
   _perm(perm), _lock(NULL)
 {}
 
@@ -153,10 +153,16 @@
 }
 
 void PtrQueue::handle_zero_index() {
-  assert(0 == _index, "Precondition.");
+  assert(_index == 0, "Precondition.");
+
   // This thread records the full buffer and allocates a new one (while
   // holding the lock if there is one).
   if (_buf != NULL) {
+    if (!should_enqueue_buffer()) {
+      assert(_index > 0, "the buffer can only be re-used if it's not full");
+      return;
+    }
+
     if (_lock) {
       assert(_lock->owned_by_self(), "Required.");
 
--- a/src/share/vm/gc_implementation/g1/ptrQueue.hpp	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/share/vm/gc_implementation/g1/ptrQueue.hpp	Tue Feb 01 18:57:53 2011 -0800
@@ -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
@@ -68,7 +68,7 @@
 public:
   // Initialize this queue to contain a null buffer, and be part of the
   // given PtrQueueSet.
-  PtrQueue(PtrQueueSet*, bool perm = false, bool active = false);
+  PtrQueue(PtrQueueSet* qset, bool perm = false, bool active = false);
   // Release any contained resources.
   void flush();
   // Calls flush() when destroyed.
@@ -85,6 +85,14 @@
     else enqueue_known_active(ptr);
   }
 
+  // This method is called when we're doing the zero index handling
+  // and gives a chance to the queues to do any pre-enqueueing
+  // processing they might want to do on the buffer. It should return
+  // true if the buffer should be enqueued, or false if enough
+  // entries were cleared from it so that it can be re-used. It should
+  // not return false if the buffer is still full (otherwise we can
+  // get into an infinite loop).
+  virtual bool should_enqueue_buffer() { return true; }
   void handle_zero_index();
   void locking_enqueue_completed_buffer(void** buf);
 
--- a/src/share/vm/gc_implementation/g1/satbQueue.cpp	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/share/vm/gc_implementation/g1/satbQueue.cpp	Tue Feb 01 18:57:53 2011 -0800
@@ -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
@@ -23,12 +23,98 @@
  */
 
 #include "precompiled.hpp"
+#include "gc_implementation/g1/g1CollectedHeap.inline.hpp"
 #include "gc_implementation/g1/satbQueue.hpp"
 #include "memory/allocation.inline.hpp"
 #include "memory/sharedHeap.hpp"
 #include "runtime/mutexLocker.hpp"
 #include "runtime/thread.hpp"
 
+// This method removes entries from an SATB buffer that will not be
+// useful to the concurrent marking threads. An entry is removed if it
+// satisfies one of the following conditions:
+//
+// * it points to an object outside the G1 heap (G1's concurrent
+//     marking only visits objects inside the G1 heap),
+// * it points to an object that has been allocated since marking
+//     started (according to SATB those objects do not need to be
+//     visited during marking), or
+// * it points to an object that has already been marked (no need to
+//     process it again).
+//
+// The rest of the entries will be retained and are compacted towards
+// the top of the buffer. If with this filtering we clear a large
+// enough chunk of the buffer we can re-use it (instead of enqueueing
+// it) and we can just allow the mutator to carry on executing.
+
+bool ObjPtrQueue::should_enqueue_buffer() {
+  assert(_lock == NULL || _lock->owned_by_self(),
+         "we should have taken the lock before calling this");
+
+  // A value of 0 means "don't filter SATB buffers".
+  if (G1SATBBufferEnqueueingThresholdPercent == 0) {
+    return true;
+  }
+
+  G1CollectedHeap* g1h = G1CollectedHeap::heap();
+
+  // This method should only be called if there is a non-NULL buffer
+  // that is full.
+  assert(_index == 0, "pre-condition");
+  assert(_buf != NULL, "pre-condition");
+
+  void** buf = _buf;
+  size_t sz = _sz;
+
+  // Used for sanity checking at the end of the loop.
+  debug_only(size_t entries = 0; size_t retained = 0;)
+
+  size_t i = sz;
+  size_t new_index = sz;
+
+  // Given that we are expecting _index == 0, we could have changed
+  // the loop condition to (i > 0). But we are using _index for
+  // generality.
+  while (i > _index) {
+    assert(i > 0, "we should have at least one more entry to process");
+    i -= oopSize;
+    debug_only(entries += 1;)
+    oop* p = (oop*) &buf[byte_index_to_index((int) i)];
+    oop obj = *p;
+    // NULL the entry so that unused parts of the buffer contain NULLs
+    // at the end. If we are going to retain it we will copy it to its
+    // final place. If we have retained all entries we have visited so
+    // far, we'll just end up copying it to the same place.
+    *p = NULL;
+
+    bool retain = g1h->is_obj_ill(obj);
+    if (retain) {
+      assert(new_index > 0, "we should not have already filled up the buffer");
+      new_index -= oopSize;
+      assert(new_index >= i,
+             "new_index should never be below i, as we alwaysr compact 'up'");
+      oop* new_p = (oop*) &buf[byte_index_to_index((int) new_index)];
+      assert(new_p >= p, "the destination location should never be below "
+             "the source as we always compact 'up'");
+      assert(*new_p == NULL,
+             "we should have already cleared the destination location");
+      *new_p = obj;
+      debug_only(retained += 1;)
+    }
+  }
+  size_t entries_calc = (sz - _index) / oopSize;
+  assert(entries == entries_calc, "the number of entries we counted "
+         "should match the number of entries we calculated");
+  size_t retained_calc = (sz - new_index) / oopSize;
+  assert(retained == retained_calc, "the number of retained entries we counted "
+         "should match the number of retained entries we calculated");
+  size_t perc = retained_calc * 100 / entries_calc;
+  bool should_enqueue = perc > (size_t) G1SATBBufferEnqueueingThresholdPercent;
+  _index = new_index;
+
+  return should_enqueue;
+}
+
 void ObjPtrQueue::apply_closure(ObjectClosure* cl) {
   if (_buf != NULL) {
     apply_closure_to_buffer(cl, _buf, _index, _sz);
--- a/src/share/vm/gc_implementation/g1/satbQueue.hpp	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/share/vm/gc_implementation/g1/satbQueue.hpp	Tue Feb 01 18:57:53 2011 -0800
@@ -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
@@ -33,13 +33,18 @@
 // A ptrQueue whose elements are "oops", pointers to object heads.
 class ObjPtrQueue: public PtrQueue {
 public:
-  ObjPtrQueue(PtrQueueSet* qset_, bool perm = false) :
+  ObjPtrQueue(PtrQueueSet* qset, bool perm = false) :
     // SATB queues are only active during marking cycles. We create
     // them with their active field set to false. If a thread is
     // created during a cycle and its SATB queue needs to be activated
     // before the thread starts running, we'll need to set its active
     // field to true. This is done in JavaThread::initialize_queues().
-    PtrQueue(qset_, perm, false /* active */) { }
+    PtrQueue(qset, perm, false /* active */) { }
+
+  // Overrides PtrQueue::should_enqueue_buffer(). See the method's
+  // definition for more information.
+  virtual bool should_enqueue_buffer();
+
   // Apply the closure to all elements, and reset the index to make the
   // buffer empty.
   void apply_closure(ObjectClosure* cl);
--- a/src/share/vm/gc_implementation/g1/vm_operations_g1.cpp	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/share/vm/gc_implementation/g1/vm_operations_g1.cpp	Tue Feb 01 18:57:53 2011 -0800
@@ -38,7 +38,6 @@
 }
 
 void VM_G1CollectForAllocation::doit() {
-  JvmtiGCForAllocationMarker jgcm;
   G1CollectedHeap* g1h = G1CollectedHeap::heap();
   _result = g1h->satisfy_failed_allocation(_word_size, &_pause_succeeded);
   assert(_result == NULL || _pause_succeeded,
@@ -46,7 +45,6 @@
 }
 
 void VM_G1CollectFull::doit() {
-  JvmtiGCFullMarker jgcm;
   G1CollectedHeap* g1h = G1CollectedHeap::heap();
   GCCauseSetter x(g1h, _gc_cause);
   g1h->do_full_collection(false /* clear_all_soft_refs */);
@@ -72,7 +70,6 @@
 }
 
 void VM_G1IncCollectionPause::doit() {
-  JvmtiGCForAllocationMarker jgcm;
   G1CollectedHeap* g1h = G1CollectedHeap::heap();
   assert(!_should_initiate_conc_mark ||
   ((_gc_cause == GCCause::_gc_locker && GCLockerInvokesConcurrent) ||
--- a/src/share/vm/gc_implementation/parallelScavenge/psMarkSweep.cpp	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/share/vm/gc_implementation/parallelScavenge/psMarkSweep.cpp	Tue Feb 01 18:57:53 2011 -0800
@@ -258,6 +258,7 @@
     BiasedLocking::restore_marks();
     Threads::gc_epilogue();
     CodeCache::gc_epilogue();
+    JvmtiExport::gc_epilogue();
 
     COMPILER2_PRESENT(DerivedPointerTable::update_pointers());
 
--- a/src/share/vm/gc_implementation/parallelScavenge/psParallelCompact.cpp	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/share/vm/gc_implementation/parallelScavenge/psParallelCompact.cpp	Tue Feb 01 18:57:53 2011 -0800
@@ -1054,6 +1054,7 @@
 
   Threads::gc_epilogue();
   CodeCache::gc_epilogue();
+  JvmtiExport::gc_epilogue();
 
   COMPILER2_PRESENT(DerivedPointerTable::update_pointers());
 
--- a/src/share/vm/gc_implementation/parallelScavenge/vmPSOperations.cpp	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/share/vm/gc_implementation/parallelScavenge/vmPSOperations.cpp	Tue Feb 01 18:57:53 2011 -0800
@@ -42,8 +42,7 @@
 }
 
 void VM_ParallelGCFailedAllocation::doit() {
-  JvmtiGCForAllocationMarker jgcm;
-  notify_gc_begin(false);
+  SvcGCMarker sgcm(SvcGCMarker::MINOR);
 
   ParallelScavengeHeap* heap = (ParallelScavengeHeap*)Universe::heap();
   assert(heap->kind() == CollectedHeap::ParallelScavengeHeap, "must be a ParallelScavengeHeap");
@@ -54,8 +53,6 @@
   if (_result == NULL && GC_locker::is_active_and_needs_gc()) {
     set_gc_locked();
   }
-
-  notify_gc_end();
 }
 
 VM_ParallelGCFailedPermanentAllocation::VM_ParallelGCFailedPermanentAllocation(size_t size,
@@ -67,8 +64,7 @@
 }
 
 void VM_ParallelGCFailedPermanentAllocation::doit() {
-  JvmtiGCFullMarker jgcm;
-  notify_gc_begin(true);
+  SvcGCMarker sgcm(SvcGCMarker::FULL);
 
   ParallelScavengeHeap* heap = (ParallelScavengeHeap*)Universe::heap();
   assert(heap->kind() == CollectedHeap::ParallelScavengeHeap, "must be a ParallelScavengeHeap");
@@ -78,7 +74,6 @@
   if (_result == NULL && GC_locker::is_active_and_needs_gc()) {
     set_gc_locked();
   }
-  notify_gc_end();
 }
 
 // Only used for System.gc() calls
@@ -91,8 +86,7 @@
 }
 
 void VM_ParallelGCSystemGC::doit() {
-  JvmtiGCFullMarker jgcm;
-  notify_gc_begin(true);
+  SvcGCMarker sgcm(SvcGCMarker::FULL);
 
   ParallelScavengeHeap* heap = (ParallelScavengeHeap*)Universe::heap();
   assert(heap->kind() == CollectedHeap::ParallelScavengeHeap,
@@ -106,5 +100,4 @@
   } else {
     heap->invoke_full_gc(false);
   }
-  notify_gc_end();
 }
--- a/src/share/vm/gc_implementation/shared/vmGCOperations.cpp	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/share/vm/gc_implementation/shared/vmGCOperations.cpp	Tue Feb 01 18:57:53 2011 -0800
@@ -31,7 +31,6 @@
 #include "memory/oopFactory.hpp"
 #include "oops/instanceKlass.hpp"
 #include "oops/instanceRefKlass.hpp"
-#include "prims/jvmtiExport.hpp"
 #include "runtime/handles.inline.hpp"
 #include "runtime/init.hpp"
 #include "runtime/interfaceSupport.hpp"
@@ -40,6 +39,7 @@
 #ifndef SERIALGC
 #include "gc_implementation/g1/g1CollectedHeap.inline.hpp"
 #endif
+
 HS_DTRACE_PROBE_DECL1(hotspot, gc__begin, bool);
 HS_DTRACE_PROBE_DECL(hotspot, gc__end);
 
@@ -158,8 +158,7 @@
 
 
 void VM_GenCollectForAllocation::doit() {
-  JvmtiGCForAllocationMarker jgcm;
-  notify_gc_begin(false);
+  SvcGCMarker sgcm(SvcGCMarker::MINOR);
 
   GenCollectedHeap* gch = GenCollectedHeap::heap();
   GCCauseSetter gccs(gch, _gc_cause);
@@ -169,22 +168,19 @@
   if (_res == NULL && GC_locker::is_active_and_needs_gc()) {
     set_gc_locked();
   }
-  notify_gc_end();
 }
 
 void VM_GenCollectFull::doit() {
-  JvmtiGCFullMarker jgcm;
-  notify_gc_begin(true);
+  SvcGCMarker sgcm(SvcGCMarker::FULL);
 
   GenCollectedHeap* gch = GenCollectedHeap::heap();
   GCCauseSetter gccs(gch, _gc_cause);
   gch->do_full_collection(gch->must_clear_all_soft_refs(), _max_level);
-  notify_gc_end();
 }
 
 void VM_GenCollectForPermanentAllocation::doit() {
-  JvmtiGCForAllocationMarker jgcm;
-  notify_gc_begin(true);
+  SvcGCMarker sgcm(SvcGCMarker::FULL);
+
   SharedHeap* heap = (SharedHeap*)Universe::heap();
   GCCauseSetter gccs(heap, _gc_cause);
   switch (heap->kind()) {
@@ -209,5 +205,4 @@
   if (_res == NULL && GC_locker::is_active_and_needs_gc()) {
     set_gc_locked();
   }
-  notify_gc_end();
 }
--- a/src/share/vm/gc_implementation/shared/vmGCOperations.hpp	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/share/vm/gc_implementation/shared/vmGCOperations.hpp	Tue Feb 01 18:57:53 2011 -0800
@@ -30,6 +30,7 @@
 #include "runtime/jniHandles.hpp"
 #include "runtime/synchronizer.hpp"
 #include "runtime/vm_operations.hpp"
+#include "prims/jvmtiExport.hpp"
 
 // The following class hierarchy represents
 // a set of operations (VM_Operation) related to GC.
@@ -209,13 +210,17 @@
   HeapWord* result() const       { return _res; }
 };
 
-class DTraceGCProbeMarker : public StackObj {
-public:
-  DTraceGCProbeMarker(bool full) {
-    VM_GC_Operation::notify_gc_begin(full);
+class SvcGCMarker : public StackObj {
+ private:
+  JvmtiGCMarker _jgcm;
+ public:
+  typedef enum { MINOR, FULL, OTHER } reason_type;
+
+  SvcGCMarker(reason_type reason ) {
+    VM_GC_Operation::notify_gc_begin(reason == FULL);
   }
 
-  ~DTraceGCProbeMarker() {
+  ~SvcGCMarker() {
     VM_GC_Operation::notify_gc_end();
   }
 };
--- a/src/share/vm/interpreter/bytecode.cpp	Thu Jan 27 17:28:34 2011 -0800
+++ b/src/share/vm/interpreter/bytecode.cpp	Tue Feb 01 18:57:53 2011 -0800
@@ -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
@@ -34,30 +34,6 @@
 
 // Implementation of Bytecode
 
-bool Bytecode::check_must_rewrite(Bytecodes::Code code) const {
-  assert(Bytecodes::can_rewrite(code), "post-check only");
-
-  // Some codes are conditionally rewriting.  Look closely at them.
-  switch (code) {
-  case Bytecodes::_aload_0:
-    // Even if RewriteFrequentPairs is turned on,
-    // the _aload_0 code might delay its rewrite until
-    // a following _getfield rewrites itself.
-    return false;
-
-  case Bytecodes::_lookupswitch:
-    return false;  // the rewrite is not done by the interpreter
-
-  case Bytecodes::_new:
-    // (Could actually look at the class here, but the profit would be small.)
-    return false;  // the rewrite is not always done
-  }
-
-  // No other special cases.
-  return true;
-}
-
-
 #ifdef ASSERT
 
 void Bytecode::assert_same_format_as(Bytecodes::Code testbc, bool is_wide) const {
@@ -188,17 +164,16 @@
   // Note:  Rewriter::rewrite changes the Java_u2 of an invokedynamic to a native_u4,
   // at the same time it allocates per-call-site CP cache entries.
   Bytecodes::Code rawc = code();
-  Bytecode* invoke = bytecode();
-  if (invoke->has_index_u4(rawc))
-    return invoke->get_index_u4(rawc);
+  if (has_index_u4(rawc))
+    return get_index_u4(rawc);
   else
-    return invoke->get_index_u2_cpcache(rawc);
+    return get_index_u2_cpcache(rawc);
 }
 
 int Bytecode_member_ref::pool_index() const {
   int index = this->index();
   DEBUG_ONLY({
-      if (!bytecode()->has_index_u4(code()))
+      if (!has_index_u4(code()))
         index -= constantPoolOopDesc::CPCACHE_INDEX_TAG;
     });
   return _method->constants()->cache()->entry_at(index)->constant_pool_index();
@@ -214,13 +189,12 @@
 // Implementation of Bytecode_loadconstant
 
 int Bytecode_loadconstant::raw_index() const {
-  Bytecode* bcp = bytecode();
-  Bytecodes::Code rawc = bcp->code();
+  Bytecodes::Code rawc = code();
   assert(rawc != Bytecodes::_wide, "verifier prevents this");
   if (Bytecodes::java_code(rawc) == Bytecodes::_ldc)
-    return bcp->get_index_u1(rawc);
+    return get_index_u1(rawc);
   else