--- a/src/cpu/ppc/vm/interp_masm_ppc_64.cpp	Tue Dec 30 13:19:30 2014 -0800
+++ b/src/cpu/ppc/vm/interp_masm_ppc_64.cpp	Thu Jan 01 16:11:49 2015 -0800
@@ -544,6 +544,9 @@
   cmplw(CCR0, Rindex, Rlength);
   sldi(RsxtIndex, RsxtIndex, index_shift);
   blt(CCR0, LnotOOR);
+  // Index should be in R17_tos, array should be in R4_ARG2.
+  mr(R17_tos, Rindex);
+  mr(R4_ARG2, Rarray);
   load_dispatch_table(Rtmp, (address*)Interpreter::_throw_ArrayIndexOutOfBoundsException_entry);
   mtctr(Rtmp);
   bctr();
@@ -1678,6 +1681,228 @@
   }
 }
 
+// Argument and return type profilig.
+// kills: tmp, tmp2, R0, CR0, CR1
+void InterpreterMacroAssembler::profile_obj_type(Register obj, Register mdo_addr_base,
+                                                 RegisterOrConstant mdo_addr_offs, Register tmp, Register tmp2) {
+  Label do_nothing, do_update;
+
+  // tmp2 = obj is allowed
+  assert_different_registers(obj, mdo_addr_base, tmp, R0);
+  assert_different_registers(tmp2, mdo_addr_base, tmp, R0);
+  const Register klass = tmp2;
+
+  verify_oop(obj);
+
+  ld(tmp, mdo_addr_offs, mdo_addr_base);
+
+  // Set null_seen if obj is 0.
+  cmpdi(CCR0, obj, 0);
+  ori(R0, tmp, TypeEntries::null_seen);
+  beq(CCR0, do_update);
+
+  load_klass(klass, obj);
+
+  clrrdi(R0, tmp, exact_log2(-TypeEntries::type_klass_mask));
+  // Basically same as andi(R0, tmp, TypeEntries::type_klass_mask);
+  cmpd(CCR1, R0, klass);
+  // Klass seen before, nothing to do (regardless of unknown bit).
+  //beq(CCR1, do_nothing);
+
+  andi_(R0, klass, TypeEntries::type_unknown);
+  // Already unknown. Nothing to do anymore.
+  //bne(CCR0, do_nothing);
+  crorc(/*CCR0 eq*/2, /*CCR1 eq*/4+2, /*CCR0 eq*/2); // cr0 eq = cr1 eq or cr0 ne
+  beq(CCR0, do_nothing);
+
+  clrrdi_(R0, tmp, exact_log2(-TypeEntries::type_mask));
+  orr(R0, klass, tmp); // Combine klass and null_seen bit (only used if (tmp & type_mask)==0).
+  beq(CCR0, do_update); // First time here. Set profile type.
+
+  // Different than before. Cannot keep accurate profile.
+  ori(R0, tmp, TypeEntries::type_unknown);
+
+  bind(do_update);
+  // update profile
+  std(R0, mdo_addr_offs, mdo_addr_base);
+
+  align(32, 12);
+  bind(do_nothing);
+}
+
+void InterpreterMacroAssembler::profile_arguments_type(Register callee, Register tmp1, Register tmp2, bool is_virtual) {
+  if (!ProfileInterpreter) {
+    return;
+  }
+
+  assert_different_registers(callee, tmp1, tmp2, R28_mdx);
+
+  if (MethodData::profile_arguments() || MethodData::profile_return()) {
+    Label profile_continue;
+
+    test_method_data_pointer(profile_continue);
+
+    int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size());
+
+    lbz(tmp1, in_bytes(DataLayout::tag_offset()) - off_to_start, R28_mdx);
+    cmpwi(CCR0, tmp1, is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag);
+    bne(CCR0, profile_continue);
+
+    if (MethodData::profile_arguments()) {
+      Label done;
+      int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset());
+      add(R28_mdx, off_to_args, R28_mdx);
+
+      for (int i = 0; i < TypeProfileArgsLimit; i++) {
+        if (i > 0 || MethodData::profile_return()) {
+          // If return value type is profiled we may have no argument to profile.
+          ld(tmp1, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args, R28_mdx);
+          cmpdi(CCR0, tmp1, (i+1)*TypeStackSlotEntries::per_arg_count());
+          addi(tmp1, tmp1, -i*TypeStackSlotEntries::per_arg_count());
+          blt(CCR0, done);
+        }
+        ld(tmp1, in_bytes(Method::const_offset()), callee);
+        lhz(tmp1, in_bytes(ConstMethod::size_of_parameters_offset()), tmp1);
+        // Stack offset o (zero based) from the start of the argument
+        // list, for n arguments translates into offset n - o - 1 from
+        // the end of the argument list. But there's an extra slot at
+        // the top of the stack. So the offset is n - o from Lesp.
+        ld(tmp2, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))-off_to_args, R28_mdx);
+        subf(tmp1, tmp2, tmp1);
+
+        sldi(tmp1, tmp1, Interpreter::logStackElementSize);
+        ldx(tmp1, tmp1, R15_esp);
+
+        profile_obj_type(tmp1, R28_mdx, in_bytes(TypeEntriesAtCall::argument_type_offset(i))-off_to_args, tmp2, tmp1);
+
+        int to_add = in_bytes(TypeStackSlotEntries::per_arg_size());
+        addi(R28_mdx, R28_mdx, to_add);
+        off_to_args += to_add;
+      }
+
+      if (MethodData::profile_return()) {
+        ld(tmp1, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args, R28_mdx);
+        addi(tmp1, tmp1, -TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count());
+      }
+
+      bind(done);
+
+      if (MethodData::profile_return()) {
+        // We're right after the type profile for the last
+        // argument. tmp1 is the number of cells left in the
+        // CallTypeData/VirtualCallTypeData to reach its end. Non null
+        // if there's a return to profile.
+        assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type");
+        sldi(tmp1, tmp1, exact_log2(DataLayout::cell_size));
+        add(R28_mdx, tmp1, R28_mdx);
+      }
+    } else {
+      assert(MethodData::profile_return(), "either profile call args or call ret");
+      update_mdp_by_constant(in_bytes(TypeEntriesAtCall::return_only_size()));
+    }
+
+    // Mdp points right after the end of the
+    // CallTypeData/VirtualCallTypeData, right after the cells for the
+    // return value type if there's one.
+    align(32, 12);
+    bind(profile_continue);
+  }
+}
+
+void InterpreterMacroAssembler::profile_return_type(Register ret, Register tmp1, Register tmp2) {
+  assert_different_registers(ret, tmp1, tmp2);
+  if (ProfileInterpreter && MethodData::profile_return()) {
+    Label profile_continue;
+
+    test_method_data_pointer(profile_continue);
+
+    if (MethodData::profile_return_jsr292_only()) {
+      // If we don't profile all invoke bytecodes we must make sure
+      // it's a bytecode we indeed profile. We can't go back to the
+      // begining of the ProfileData we intend to update to check its
+      // type because we're right after it and we don't known its
+      // length.
+      lbz(tmp1, 0, R14_bcp);
+      lbz(tmp2, Method::intrinsic_id_offset_in_bytes(), R19_method);
+      cmpwi(CCR0, tmp1, Bytecodes::_invokedynamic);
+      cmpwi(CCR1, tmp1, Bytecodes::_invokehandle);
+      cror(/*CR0 eq*/2, /*CR1 eq*/4+2, /*CR0 eq*/2);
+      cmpwi(CCR1, tmp2, vmIntrinsics::_compiledLambdaForm);
+      cror(/*CR0 eq*/2, /*CR1 eq*/4+2, /*CR0 eq*/2);
+      bne(CCR0, profile_continue);
+    }
+
+    profile_obj_type(ret, R28_mdx, -in_bytes(ReturnTypeEntry::size()), tmp1, tmp2);
+
+    align(32, 12);
+    bind(profile_continue);
+  }
+}
+
+void InterpreterMacroAssembler::profile_parameters_type(Register tmp1, Register tmp2, Register tmp3, Register tmp4) {
+  if (ProfileInterpreter && MethodData::profile_parameters()) {
+    Label profile_continue, done;
+
+    test_method_data_pointer(profile_continue);
+
+    // Load the offset of the area within the MDO used for
+    // parameters. If it's negative we're not profiling any parameters.
+    lwz(tmp1, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset()), R28_mdx);
+    cmpwi(CCR0, tmp1, 0);
+    blt(CCR0, profile_continue);
+
+    // Compute a pointer to the area for parameters from the offset
+    // and move the pointer to the slot for the last
+    // parameters. Collect profiling from last parameter down.
+    // mdo start + parameters offset + array length - 1
+
+    // Pointer to the parameter area in the MDO.
+    const Register mdp = tmp1;
+    add(mdp, tmp1, R28_mdx);
+
+    // Pffset of the current profile entry to update.
+    const Register entry_offset = tmp2;
+    // entry_offset = array len in number of cells
+    ld(entry_offset, in_bytes(ArrayData::array_len_offset()), mdp);
+
+    int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0));
+    assert(off_base % DataLayout::cell_size == 0, "should be a number of cells");
+
+    // entry_offset (number of cells)  = array len - size of 1 entry + offset of the stack slot field
+    addi(entry_offset, entry_offset, -TypeStackSlotEntries::per_arg_count() + (off_base / DataLayout::cell_size));
+    // entry_offset in bytes
+    sldi(entry_offset, entry_offset, exact_log2(DataLayout::cell_size));
+
+    Label loop;
+    align(32, 12);
+    bind(loop);
+
+    // Load offset on the stack from the slot for this parameter.
+    ld(tmp3, entry_offset, mdp);
+    sldi(tmp3, tmp3, Interpreter::logStackElementSize);
+    neg(tmp3, tmp3);
+    // Read the parameter from the local area.
+    ldx(tmp3, tmp3, R18_locals);
+
+    // Make entry_offset now point to the type field for this parameter.
+    int type_base = in_bytes(ParametersTypeData::type_offset(0));
+    assert(type_base > off_base, "unexpected");
+    addi(entry_offset, entry_offset, type_base - off_base);
+
+    // Profile the parameter.
+    profile_obj_type(tmp3, mdp, entry_offset, tmp4, tmp3);
+
+    // Go to next parameter.
+    int delta = TypeStackSlotEntries::per_arg_count() * DataLayout::cell_size + (type_base - off_base);
+    cmpdi(CCR0, entry_offset, off_base + delta);
+    addi(entry_offset, entry_offset, -delta);
+    bge(CCR0, loop);
+
+    align(32, 12);
+    bind(profile_continue);
+  }
+}
+
 // Add a InterpMonitorElem to stack (see frame_sparc.hpp).
 void InterpreterMacroAssembler::add_monitor_to_stack(bool stack_is_empty, Register Rtemp1, Register Rtemp2) {
 
@@ -2039,20 +2264,19 @@
   bne(CCR0, test);
 
   address fd = CAST_FROM_FN_PTR(address, verify_return_address);
-  unsigned int nbytes_save = 10*8; // 10 volatile gprs
-
-  save_LR_CR(Rtmp);
+  const int nbytes_save = 11*8; // volatile gprs except R0
+  save_volatile_gprs(R1_SP, -nbytes_save); // except R0
+  save_LR_CR(Rtmp); // Save in old frame.
   push_frame_reg_args(nbytes_save, Rtmp);
-  save_volatile_gprs(R1_SP, 112); // except R0
 
   load_const_optimized(Rtmp, fd, R0);
   mr_if_needed(R4_ARG2, reg);
   mr(R3_ARG1, R19_method);
   call_c(Rtmp); // call C
 
-  restore_volatile_gprs(R1_SP, 112); // except R0
   pop_frame();
   restore_LR_CR(Rtmp);
+  restore_volatile_gprs(R1_SP, -nbytes_save); // except R0
   b(skip);
 
   // Perform a more elaborate out-of-line call.

--- a/src/cpu/ppc/vm/interp_masm_ppc_64.hpp	Tue Dec 30 13:19:30 2014 -0800
+++ b/src/cpu/ppc/vm/interp_masm_ppc_64.hpp	Thu Jan 01 16:11:49 2015 -0800
@@ -255,6 +255,12 @@
   void record_klass_in_profile(Register receiver, Register scratch1, Register scratch2, bool is_virtual_call);
   void record_klass_in_profile_helper(Register receiver, Register scratch1, Register scratch2, int start_row, Label& done, bool is_virtual_call);
 
+  // Argument and return type profiling.
+  void profile_obj_type(Register obj, Register mdo_addr_base, RegisterOrConstant mdo_addr_offs, Register tmp, Register tmp2);
+  void profile_arguments_type(Register callee, Register tmp1, Register tmp2, bool is_virtual);
+  void profile_return_type(Register ret, Register tmp1, Register tmp2);
+  void profile_parameters_type(Register tmp1, Register tmp2, Register tmp3, Register tmp4);
+
 #endif // !CC_INTERP
 
   // Debugging

--- a/src/cpu/ppc/vm/macroAssembler_ppc.cpp	Tue Dec 30 13:19:30 2014 -0800
+++ b/src/cpu/ppc/vm/macroAssembler_ppc.cpp	Thu Jan 01 16:11:49 2015 -0800
@@ -807,6 +807,7 @@
 
 // For verify_oops.
 void MacroAssembler::save_volatile_gprs(Register dst, int offset) {
+  std(R2,  offset, dst);   offset += 8;
   std(R3,  offset, dst);   offset += 8;
   std(R4,  offset, dst);   offset += 8;
   std(R5,  offset, dst);   offset += 8;
@@ -821,6 +822,7 @@
 
 // For verify_oops.
 void MacroAssembler::restore_volatile_gprs(Register src, int offset) {
+  ld(R2,  offset, src);   offset += 8;
   ld(R3,  offset, src);   offset += 8;
   ld(R4,  offset, src);   offset += 8;
   ld(R5,  offset, src);   offset += 8;
@@ -1187,6 +1189,16 @@
   call_VM(oop_result, entry_point, check_exceptions);
 }
 
+void MacroAssembler::call_VM(Register oop_result, address entry_point, Register arg_1, Register arg_2, Register arg_3,
+                             bool check_exceptions) {
+  // R3_ARG1 is reserved for the thread
+  mr_if_needed(R4_ARG2, arg_1);
+  assert(arg_2 != R4_ARG2, "smashed argument");
+  mr_if_needed(R5_ARG3, arg_2);
+  mr_if_needed(R6_ARG4, arg_3);
+  call_VM(oop_result, entry_point, check_exceptions);
+}
+
 void MacroAssembler::call_VM_leaf(address entry_point) {
   call_VM_leaf_base(entry_point);
 }
@@ -3059,35 +3071,27 @@
   if (!VerifyOops) {
     return;
   }
-  // Will be preserved.
-  Register tmp = R11;
-  assert(oop != tmp, "precondition");
-  unsigned int nbytes_save = 10*8; // 10 volatile gprs
+
   address/* FunctionDescriptor** */fd = StubRoutines::verify_oop_subroutine_entry_address();
-  // save tmp
-  mr(R0, tmp);
-  // kill tmp
-  save_LR_CR(tmp);
+  const Register tmp = R11; // Will be preserved.
+  const int nbytes_save = 11*8; // Volatile gprs except R0.
+  save_volatile_gprs(R1_SP, -nbytes_save); // except R0
+
+  if (oop == tmp) mr(R4_ARG2, oop);
+  save_LR_CR(tmp); // save in old frame
   push_frame_reg_args(nbytes_save, tmp);
-  // restore tmp
-  mr(tmp, R0);
-  save_volatile_gprs(R1_SP, 112); // except R0
   // load FunctionDescriptor** / entry_address *
-  load_const(tmp, fd);
+  load_const_optimized(tmp, fd, R0);
   // load FunctionDescriptor* / entry_address
   ld(tmp, 0, tmp);
-  mr(R4_ARG2, oop);
-  load_const(R3_ARG1, (address)msg);
-  // call destination for its side effect
+  if (oop != tmp) mr_if_needed(R4_ARG2, oop);
+  load_const_optimized(R3_ARG1, (address)msg, R0);
+  // Call destination for its side effect.
   call_c(tmp);
-  restore_volatile_gprs(R1_SP, 112); // except R0
+
   pop_frame();
-  // save tmp
-  mr(R0, tmp);
-  // kill tmp
   restore_LR_CR(tmp);
-  // restore tmp
-  mr(tmp, R0);
+  restore_volatile_gprs(R1_SP, -nbytes_save); // except R0
 }
 
 const char* stop_types[] = {

--- a/src/cpu/ppc/vm/macroAssembler_ppc.hpp	Tue Dec 30 13:19:30 2014 -0800
+++ b/src/cpu/ppc/vm/macroAssembler_ppc.hpp	Thu Jan 01 16:11:49 2015 -0800
@@ -369,6 +369,7 @@
   void call_VM(Register oop_result, address entry_point, bool check_exceptions = true);
   void call_VM(Register oop_result, address entry_point, Register arg_1, bool check_exceptions = true);
   void call_VM(Register oop_result, address entry_point, Register arg_1, Register arg_2, bool check_exceptions = true);
+  void call_VM(Register oop_result, address entry_point, Register arg_1, Register arg_2, Register arg3, bool check_exceptions = true);
   void call_VM_leaf(address entry_point);
   void call_VM_leaf(address entry_point, Register arg_1);
   void call_VM_leaf(address entry_point, Register arg_1, Register arg_2);

--- a/src/cpu/ppc/vm/nativeInst_ppc.cpp	Tue Dec 30 13:19:30 2014 -0800
+++ b/src/cpu/ppc/vm/nativeInst_ppc.cpp	Thu Jan 01 16:11:49 2015 -0800
@@ -100,10 +100,7 @@
   MacroAssembler* a = new MacroAssembler(&cb);
 
   // Patch the call.
-  if (ReoptimizeCallSequences &&
-      a->is_within_range_of_b(dest, addr_call)) {
-    a->bl(dest);
-  } else {
+  if (!ReoptimizeCallSequences || !a->is_within_range_of_b(dest, addr_call)) {
     address trampoline_stub_addr = get_trampoline();
 
     // We did not find a trampoline stub because the current codeblob
@@ -115,9 +112,12 @@
 
     // Patch the constant in the call's trampoline stub.
     NativeCallTrampolineStub_at(trampoline_stub_addr)->set_destination(dest);
+    dest = trampoline_stub_addr;
+  }
 
-    a->bl(trampoline_stub_addr);
-  }
+  OrderAccess::release();
+  a->bl(dest);
+
   ICache::ppc64_flush_icache_bytes(addr_call, code_size);
 }
 

--- a/src/cpu/ppc/vm/ppc.ad	Tue Dec 30 13:19:30 2014 -0800
+++ b/src/cpu/ppc/vm/ppc.ad	Thu Jan 01 16:11:49 2015 -0800
@@ -1936,8 +1936,9 @@
   // --------------------------------------------------------------------
   // Check for hi bits still needing moving. Only happens for misaligned
   // arguments to native calls.
-  if (src_hi == dst_hi)
+  if (src_hi == dst_hi) {
     return ppc64Opcode_none;               // Self copy; no move.
+  }
 
   ShouldNotReachHere();
   return ppc64Opcode_undefined;
@@ -1959,14 +1960,15 @@
 }
 
 uint MachNopNode::size(PhaseRegAlloc *ra_) const {
-   return _count * 4;
+  return _count * 4;
 }
 
 #ifndef PRODUCT
 void BoxLockNode::format(PhaseRegAlloc *ra_, outputStream *st) const {
   int offset = ra_->reg2offset(in_RegMask(0).find_first_elem());
-  int reg = ra_->get_reg_first(this);
-  st->print("ADDI %s, SP, %d \t// box node", Matcher::regName[reg], offset);
+  char reg_str[128];
+  ra_->dump_register(this, reg_str);
+  st->print("ADDI    %s, SP, %d \t// box node", reg_str, offset);
 }
 #endif
 

--- a/src/cpu/ppc/vm/templateInterpreter_ppc.cpp	Tue Dec 30 13:19:30 2014 -0800
+++ b/src/cpu/ppc/vm/templateInterpreter_ppc.cpp	Thu Jan 01 16:11:49 2015 -0800
@@ -91,7 +91,7 @@
 
   // Thread will be loaded to R3_ARG1.
   // Target class oop is in register R5_ARG3 by convention!
-  __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ClassCastException_verbose, R17_tos, R5_ARG3));
+  __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ClassCastException_verbose), R17_tos, R5_ARG3);
   // Above call must not return here since exception pending.
   DEBUG_ONLY(__ should_not_reach_here();)
   return entry;
@@ -172,6 +172,10 @@
   // Compiled code destroys templateTableBase, reload.
   __ load_const_optimized(R25_templateTableBase, (address)Interpreter::dispatch_table((TosState)0), R12_scratch2);
 
+  if (state == atos) {
+    __ profile_return_type(R3_RET, R11_scratch1, R12_scratch2);
+  }
+
   const Register cache = R11_scratch1;
   const Register size  = R12_scratch2;
   __ get_cache_and_index_at_bcp(cache, 1, index_size);
@@ -1189,6 +1193,10 @@
       __ li(R0, 1);
       __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread);
     }
+
+    // Argument and return type profiling.
+    __ profile_parameters_type(R3_ARG1, R4_ARG2, R5_ARG3, R6_ARG4);
+
     // Increment invocation counter and check for overflow.
     if (inc_counter) {
       generate_counter_incr(&invocation_counter_overflow, &profile_method, &profile_method_continue);
@@ -1469,6 +1477,8 @@
     __ resize_frame_absolute(R12_scratch2, R11_scratch1, R0);
     if (ProfileInterpreter) {
       __ set_method_data_pointer_for_bcp();
+      __ ld(R11_scratch1, 0, R1_SP);
+      __ std(R28_mdx, _ijava_state_neg(mdx), R11_scratch1);
     }
 #if INCLUDE_JVMTI
     Label L_done;
@@ -1480,13 +1490,11 @@
     // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call.
     // Detect such a case in the InterpreterRuntime function and return the member name argument, or NULL.
     __ ld(R4_ARG2, 0, R18_locals);
-    __ call_VM(R11_scratch1, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null),
-               R4_ARG2, R19_method, R14_bcp);
-
-    __ cmpdi(CCR0, R11_scratch1, 0);
+    __ MacroAssembler::call_VM(R4_ARG2, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), R4_ARG2, R19_method, R14_bcp, false);
+    __ restore_interpreter_state(R11_scratch1, /*bcp_and_mdx_only*/ true);
+    __ cmpdi(CCR0, R4_ARG2, 0);
     __ beq(CCR0, L_done);
-
-    __ std(R11_scratch1, wordSize, R15_esp);
+    __ std(R4_ARG2, wordSize, R15_esp);
     __ bind(L_done);
 #endif // INCLUDE_JVMTI
     __ dispatch_next(vtos);

--- a/src/cpu/ppc/vm/templateTable_ppc_64.cpp	Tue Dec 30 13:19:30 2014 -0800
+++ b/src/cpu/ppc/vm/templateTable_ppc_64.cpp	Thu Jan 01 16:11:49 2015 -0800
@@ -3235,6 +3235,8 @@
   // Load target.
   __ addi(Rrecv_klass, Rrecv_klass, base + vtableEntry::method_offset_in_bytes());
   __ ldx(Rtarget_method, Rindex, Rrecv_klass);
+  // Argument and return type profiling.
+  __ profile_arguments_type(Rtarget_method, Rrecv_klass /* scratch1 */, Rtemp /* scratch2 */, true);
   __ call_from_interpreter(Rtarget_method, Rret, Rrecv_klass /* scratch1 */, Rtemp /* scratch2 */);
 }
 
@@ -3318,6 +3320,8 @@
   __ null_check_throw(Rrecv, -1, Rscratch1);
 
   __ profile_final_call(Rrecv, Rscratch1);
+  // Argument and return type profiling.
+  __ profile_arguments_type(Rmethod, Rscratch1, Rscratch2, true);
 
   // Do the call.
   __ call_from_interpreter(Rmethod, Rret_addr, Rscratch1, Rscratch2);
@@ -3339,6 +3343,8 @@
   __ null_check_throw(Rreceiver, -1, R11_scratch1);
 
   __ profile_call(R11_scratch1, R12_scratch2);
+  // Argument and return type profiling.
+  __ profile_arguments_type(Rmethod, R11_scratch1, R12_scratch2, false);
   __ call_from_interpreter(Rmethod, Rret_addr, R11_scratch1, R12_scratch2);
 }
 
@@ -3353,6 +3359,8 @@
   prepare_invoke(byte_no, R19_method, Rret_addr, noreg, noreg, Rflags, R11_scratch1);
 
   __ profile_call(R11_scratch1, R12_scratch2);
+  // Argument and return type profiling.
+  __ profile_arguments_type(R19_method, R11_scratch1, R12_scratch2, false);
   __ call_from_interpreter(R19_method, Rret_addr, R11_scratch1, R12_scratch2);
 }
 
@@ -3374,6 +3382,8 @@
 
   // Final call case.
   __ profile_final_call(Rtemp1, Rscratch);
+  // Argument and return type profiling.
+  __ profile_arguments_type(Rindex, Rscratch, Rrecv_klass /* scratch */, true);
   // Do the final call - the index (f2) contains the method.
   __ call_from_interpreter(Rindex, Rret, Rscratch, Rrecv_klass /* scratch */);
 
@@ -3425,6 +3435,8 @@
   __ cmpdi(CCR0, Rindex, 0);
   __ beq(CCR0, Lthrow_ame);
   // Found entry. Jump off!
+  // Argument and return type profiling.
+  __ profile_arguments_type(Rindex, Rscratch1, Rscratch2, true);
   __ call_from_interpreter(Rindex, Rret_addr, Rscratch1, Rscratch2);
 
   // Vtable entry was NULL => Throw abstract method error.
@@ -3468,6 +3480,8 @@
   // to be the callsite object the bootstrap method returned. This is passed to a
   // "link" method which does the dispatch (Most likely just grabs the MH stored
   // inside the callsite and does an invokehandle).
+  // Argument and return type profiling.
+  __ profile_arguments_type(Rmethod, Rscratch1, Rscratch2, false);
   __ call_from_interpreter(Rmethod, Rret_addr, Rscratch1 /* scratch1 */, Rscratch2 /* scratch2 */);
 }
 
@@ -3488,6 +3502,8 @@
   __ profile_final_call(Rrecv, Rscratch1);
 
   // Still no call from handle => We call the method handle interpreter here.
+  // Argument and return type profiling.
+  __ profile_arguments_type(Rmethod, Rscratch1, Rscratch2, true);
   __ call_from_interpreter(Rmethod, Rret_addr, Rscratch1 /* scratch1 */, Rscratch2 /* scratch2 */);
 }
 

--- a/src/cpu/ppc/vm/vm_version_ppc.cpp	Tue Dec 30 13:19:30 2014 -0800
+++ b/src/cpu/ppc/vm/vm_version_ppc.cpp	Thu Jan 01 16:11:49 2015 -0800
@@ -134,13 +134,44 @@
   }
 
   assert(AllocatePrefetchLines > 0, "invalid value");
-  if (AllocatePrefetchLines < 1) // Set valid value in product VM.
+  if (AllocatePrefetchLines < 1) { // Set valid value in product VM.
     AllocatePrefetchLines = 1; // Conservative value.
+  }
 
-  if (AllocatePrefetchStyle == 3 && AllocatePrefetchDistance < cache_line_size)
+  if (AllocatePrefetchStyle == 3 && AllocatePrefetchDistance < cache_line_size) {
     AllocatePrefetchStyle = 1; // Fall back if inappropriate.
+  }
 
   assert(AllocatePrefetchStyle >= 0, "AllocatePrefetchStyle should be positive");
+
+  if (UseCRC32Intrinsics) {
+    if (!FLAG_IS_DEFAULT(UseCRC32Intrinsics))
+      warning("CRC32 intrinsics  are not available on this CPU");
+    FLAG_SET_DEFAULT(UseCRC32Intrinsics, false);
+  }
+
+  // The AES intrinsic stubs require AES instruction support.
+  if (UseAES) {
+    warning("AES instructions are not available on this CPU");
+    FLAG_SET_DEFAULT(UseAES, false);
+  }
+  if (UseAESIntrinsics) {
+    if (!FLAG_IS_DEFAULT(UseAESIntrinsics))
+      warning("AES intrinsics are not available on this CPU");
+    FLAG_SET_DEFAULT(UseAESIntrinsics, false);
+  }
+
+  if (UseSHA) {
+    warning("SHA instructions are not available on this CPU");
+    FLAG_SET_DEFAULT(UseSHA, false);
+  }
+  if (UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics) {
+    warning("SHA intrinsics are not available on this CPU");
+    FLAG_SET_DEFAULT(UseSHA1Intrinsics, false);
+    FLAG_SET_DEFAULT(UseSHA256Intrinsics, false);
+    FLAG_SET_DEFAULT(UseSHA512Intrinsics, false);
+  }
+
 }
 
 void VM_Version::print_features() {

--- a/src/os/aix/vm/os_aix.cpp	Tue Dec 30 13:19:30 2014 -0800
+++ b/src/os/aix/vm/os_aix.cpp	Thu Jan 01 16:11:49 2015 -0800
@@ -124,12 +124,6 @@
 }
 #endif
 
-// Excerpts from systemcfg.h definitions newer than AIX 5.3
-#ifndef PV_7
-# define PV_7 0x200000          // Power PC 7
-# define PV_7_Compat 0x208000   // Power PC 7
-#endif
-
 #define MAX_PATH (2 * K)
 
 // for timer info max values which include all bits
@@ -140,17 +134,40 @@
 #define ERROR_MP_VMGETINFO_FAILED                    102
 #define ERROR_MP_VMGETINFO_CLAIMS_NO_SUPPORT_FOR_64K 103
 
-// the semantics in this file are thus that codeptr_t is a *real code ptr*
+// The semantics in this file are thus that codeptr_t is a *real code ptr*.
 // This means that any function taking codeptr_t as arguments will assume
 // a real codeptr and won't handle function descriptors (eg getFuncName),
 // whereas functions taking address as args will deal with function
-// descriptors (eg os::dll_address_to_library_name)
+// descriptors (eg os::dll_address_to_library_name).
 typedef unsigned int* codeptr_t;
 
-// typedefs for stackslots, stack pointers, pointers to op codes
+// Typedefs for stackslots, stack pointers, pointers to op codes.
 typedef unsigned long stackslot_t;
 typedef stackslot_t* stackptr_t;
 
+// Excerpts from systemcfg.h definitions newer than AIX 5.3.
+#ifndef PV_7
+#define PV_7 0x200000          /* Power PC 7 */
+#define PV_7_Compat 0x208000   /* Power PC 7 */
+#endif
+#ifndef PV_8
+#define PV_8 0x300000          /* Power PC 8 */
+#define PV_8_Compat 0x308000   /* Power PC 8 */
+#endif
+
+#define trcVerbose(fmt, ...) { /* PPC port */  \
+  if (Verbose) { \
+    fprintf(stderr, fmt, ##__VA_ARGS__); \
+    fputc('\n', stderr); fflush(stderr); \
+  } \
+}
+#define trc(fmt, ...)        /* PPC port */
+
+#define ERRBYE(s) { \
+    trcVerbose(s); \
+    return -1; \
+}
+
 // query dimensions of the stack of the calling thread
 static void query_stack_dimensions(address* p_stack_base, size_t* p_stack_size);
 
@@ -182,12 +199,12 @@
   return true;
 }
 
-// macro to check a given stack pointer against given stack limits and to die if test fails
+// Macro to check a given stack pointer against given stack limits and to die if test fails.
 #define CHECK_STACK_PTR(sp, stack_base, stack_size) { \
     guarantee(is_valid_stackpointer((stackptr_t)(sp), (stackptr_t)(stack_base), stack_size), "Stack Pointer Invalid"); \
 }
 
-// macro to check the current stack pointer against given stacklimits
+// Macro to check the current stack pointer against given stacklimits.
 #define CHECK_CURRENT_STACK_PTR(stack_base, stack_size) { \
   address sp; \
   sp = os::current_stack_pointer(); \
@@ -221,7 +238,7 @@
 static pid_t    _initial_pid       = 0;
 static int      SR_signum          = SIGUSR2; // Signal used to suspend/resume a thread (must be > SIGSEGV, see 4355769)
 static sigset_t SR_sigset;
-static pthread_mutex_t dl_mutex;           // Used to protect dlsym() calls */
+static pthread_mutex_t dl_mutex;              // Used to protect dlsym() calls.
 
 julong os::available_memory() {
   return Aix::available_memory();
@@ -253,7 +270,6 @@
   return false;
 }
 
-
 // Return true if user is running as root.
 
 bool os::have_special_privileges() {
@@ -284,8 +300,7 @@
 
   for (int i = 0; i < numFullDisclaimsNeeded; i ++) {
     if (::disclaim(p, maxDisclaimSize, DISCLAIM_ZEROMEM) != 0) {
-      //if (Verbose)
-      fprintf(stderr, "Cannot disclaim %p - %p (errno %d)\n", p, p + maxDisclaimSize, errno);
+      trc("Cannot disclaim %p - %p (errno %d)\n", p, p + maxDisclaimSize, errno);
       return false;
     }
     p += maxDisclaimSize;
@@ -293,8 +308,7 @@
 
   if (lastDisclaimSize > 0) {
     if (::disclaim(p, lastDisclaimSize, DISCLAIM_ZEROMEM) != 0) {
-      //if (Verbose)
-        fprintf(stderr, "Cannot disclaim %p - %p (errno %d)\n", p, p + lastDisclaimSize, errno);
+      trc("Cannot disclaim %p - %p (errno %d)\n", p, p + lastDisclaimSize, errno);
       return false;
     }
   }
@@ -334,11 +348,11 @@
 
 void os::Aix::initialize_system_info() {
 
-  // get the number of online(logical) cpus instead of configured
+  // Get the number of online(logical) cpus instead of configured.
   os::_processor_count = sysconf(_SC_NPROCESSORS_ONLN);
   assert(_processor_count > 0, "_processor_count must be > 0");
 
-  // retrieve total physical storage
+  // Retrieve total physical storage.
   os::Aix::meminfo_t mi;
   if (!os::Aix::get_meminfo(&mi)) {
     fprintf(stderr, "os::Aix::get_meminfo failed.\n"); fflush(stderr);
@@ -513,7 +527,6 @@
 
 } // end os::Aix::query_multipage_support()
 
-// The code for this method was initially derived from the version in os_linux.cpp.
 void os::init_system_properties_values() {
 
 #define DEFAULT_LIBPATH "/usr/lib:/lib"
@@ -603,10 +616,11 @@
   sigaction(sig, (struct sigaction*)NULL, &oact);
   void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction)
     : CAST_FROM_FN_PTR(void*, oact.sa_handler);
-  if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN))
+  if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) {
     return true;
-  else
+  } else {
     return false;
+  }
 }
 
 void os::Aix::signal_sets_init() {
@@ -780,6 +794,9 @@
 
   // get the processor version from _system_configuration
   switch (_system_configuration.version) {
+  case PV_8:
+    strcpy(pci->version, "Power PC 8");
+    break;
   case PV_7:
     strcpy(pci->version, "Power PC 7");
     break;
@@ -807,6 +824,9 @@
   case PV_7_Compat:
     strcpy(pci->version, "PV_7_Compat");
     break;
+  case PV_8_Compat:
+    strcpy(pci->version, "PV_8_Compat");
+    break;
   default:
     strcpy(pci->version, "unknown");
   }
@@ -942,7 +962,9 @@
 
   pthread_attr_destroy(&attr);
 
-  if (ret != 0) {
+  if (ret == 0) {
+    // PPC port traceOsMisc(("Created New Thread : pthread-id %u", tid));
+  } else {
     if (PrintMiscellaneous && (Verbose || WizardMode)) {
       perror("pthread_create()");
     }
@@ -1103,8 +1125,7 @@
   if (os::Aix::on_pase()) {
     Unimplemented();
     return 0;
-  }
-  else {
+  } else {
     // On AIX use the precision of processors real time clock
     // or time base registers.
     timebasestruct_t time;
@@ -1152,7 +1173,6 @@
   }
 }
 
-
 char * os::local_time_string(char *buf, size_t buflen) {
   struct tm t;
   time_t long_time;
@@ -1190,7 +1210,6 @@
   if (abort_hook != NULL) {
     abort_hook();
   }
-
 }
 
 // Note: os::abort() might be called very early during initialization, or
@@ -1222,8 +1241,7 @@
 // from src/solaris/hpi/src/system_md.c
 
 size_t os::lasterror(char *buf, size_t len) {
-
-  if (errno == 0)  return 0;
+  if (errno == 0) return 0;
 
   const char *s = ::strerror(errno);
   size_t n = ::strlen(s);
@@ -1236,6 +1254,7 @@
 }
 
 intx os::current_thread_id() { return (intx)pthread_self(); }
+
 int os::current_process_id() {
 
   // This implementation returns a unique pid, the pid of the
@@ -1372,9 +1391,9 @@
   if (offset) {
     *offset = -1;
   }
-  if (buf) {
-    buf[0] = '\0';
-  }
+  // Buf is not optional, but offset is optional.
+  assert(buf != NULL, "sanity check");
+  buf[0] = '\0';
 
   // Resolve function ptr literals first.
   addr = resolve_function_descriptor_to_code_pointer(addr);
@@ -1407,12 +1426,9 @@
     return 0;
   }
 
-  if (Verbose) {
-    fprintf(stderr, "pc outside any module");
-  }
+  trcVerbose("pc outside any module");
 
   return -1;
-
 }
 
 bool os::dll_address_to_library_name(address addr, char* buf,
@@ -1420,9 +1436,9 @@
   if (offset) {
     *offset = -1;
   }
-  if (buf) {
-      buf[0] = '\0';
-  }
+  // Buf is not optional, but offset is optional.
+  assert(buf != NULL, "sanity check");
+  buf[0] = '\0';
 
   // Resolve function ptr literals first.
   addr = resolve_function_descriptor_to_code_pointer(addr);
@@ -1437,7 +1453,7 @@
 }
 
 // Loads .dll/.so and in case of error it checks if .dll/.so was built
-// for the same architecture as Hotspot is running on
+// for the same architecture as Hotspot is running on.
 void *os::dll_load(const char *filename, char *ebuf, int ebuflen) {
 
   if (ebuf && ebuflen > 0) {
@@ -1600,7 +1616,6 @@
   st->cr();
 }
 
-
 static void print_signal_handler(outputStream* st, int sig,
                                  char* buf, size_t buflen);
 
@@ -1624,7 +1639,7 @@
 
 static char saved_jvm_path[MAXPATHLEN] = {0};
 
-// Find the full path to the current module, libjvm.so or libjvm_g.so
+// Find the full path to the current module, libjvm.so.
 void os::jvm_path(char *buf, jint buflen) {
   // Error checking.
   if (buflen < MAXPATHLEN) {
@@ -1695,7 +1710,7 @@
   // Do not block out synchronous signals in the signal handler.
   // Blocking synchronous signals only makes sense if you can really
   // be sure that those signals won't happen during signal handling,
-  // when the blocking applies.  Normal signal handlers are lean and
+  // when the blocking applies. Normal signal handlers are lean and
   // do not cause signals. But our signal handlers tend to be "risky"
   // - secondary SIGSEGV, SIGILL, SIGBUS' may and do happen.
   // On AIX, PASE there was a case where a SIGSEGV happened, followed
@@ -2861,13 +2876,9 @@
   param.sched_priority = newpri;
   int ret = pthread_setschedparam(thr, policy, &param);
 
-  if (Verbose) {
-    if (ret == 0) {
-      fprintf(stderr, "changed priority of thread %d to %d\n", (int)thr, newpri);
-    } else {
-      fprintf(stderr, "Could not changed priority for thread %d to %d (error %d, %s)\n",
-              (int)thr, newpri, ret, strerror(ret));
-    }
+  if (ret != 0) {
+    trcVerbose("Could not change priority for thread %d to %d (error %d, %s)",
+        (int)thr, newpri, ret, strerror(ret));
   }
   return (ret == 0) ? OS_OK : OS_ERR;
 }
@@ -2988,7 +2999,6 @@
   errno = old_errno;
 }
 
-
 static int SR_initialize() {
   struct sigaction act;
   char *s;
@@ -3187,7 +3197,6 @@
   JVM_handle_aix_signal(sig, info, uc, true);
 }
 
-
 // This boolean allows users to forward their own non-matching signals
 // to JVM_handle_aix_signal, harmlessly.
 bool os::Aix::signal_handlers_are_installed = false;
@@ -3381,7 +3390,7 @@
     set_signal_handler(SIGDANGER, true);
 
     if (libjsig_is_loaded) {
-      // Tell libjsig jvm finishes setting signal handlers
+      // Tell libjsig jvm finishes setting signal handlers.
       (*end_signal_setting)();
     }
 
@@ -3397,7 +3406,7 @@
         tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
         check_signals = false;
       }
-      // need to initialize check_signal_done
+      // Need to initialize check_signal_done.
       ::sigemptyset(&check_signal_done);
     }
   }
@@ -3471,7 +3480,6 @@
   st->cr();
 }
 
-
 #define DO_SIGNAL_CHECK(sig) \
   if (!sigismember(&check_signal_done, sig)) \
     os::Aix::check_signal_handler(sig)
@@ -3532,7 +3540,6 @@
     ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
     : CAST_FROM_FN_PTR(address, act.sa_handler);
 
-
   switch(sig) {
   case SIGSEGV:
   case SIGBUS:
@@ -3685,15 +3692,13 @@
   pthread_mutex_init(&dl_mutex, NULL);
 }
 
-// this is called _after_ the global arguments have been parsed
+// This is called _after_ the global arguments have been parsed.
 jint os::init_2(void) {
 
-  if (Verbose) {
-    fprintf(stderr, "processor count: %d\n", os::_processor_count);
-    fprintf(stderr, "physical memory: %lu\n", Aix::_physical_memory);
-  }
-
-  // initially build up the loaded dll map
+  trcVerbose("processor count: %d", os::_processor_count);
+  trcVerbose("physical memory: %lu", Aix::_physical_memory);
+
+  // Initially build up the loaded dll map.
   LoadedLibraries::reload();
 
   const int page_size = Aix::page_size();
@@ -3743,7 +3748,7 @@
       }
 
       if (map_address != (address) MAP_FAILED) {
-        // map succeeded, but polling_page is not at wished address, unmap and continue.
+        // Map succeeded, but polling_page is not at wished address, unmap and continue.
         ::munmap(map_address, map_size);
         map_address = (address) MAP_FAILED;
       }
@@ -3797,7 +3802,7 @@
 
   // Make the stack size a multiple of the page size so that
   // the yellow/red zones can be guarded.
-  // note that this can be 0, if no default stacksize was set
+  // Note that this can be 0, if no default stacksize was set.
   JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes, vm_page_size()));
 
   Aix::libpthread_init();
@@ -4088,7 +4093,6 @@
   return fd;
 }
 
-
 // create binary file, rewriting existing file if required
 int os::create_binary_file(const char* path, bool rewrite_existing) {
   int oflags = O_WRONLY | O_CREAT;
@@ -4169,7 +4173,6 @@
   return mapped_address;
 }
 
-
 // Remap a block of memory.
 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
                           char *addr, size_t bytes, bool read_only,
@@ -4217,14 +4220,14 @@
   jlong sys_time = 0;
   jlong user_time = 0;
 
-  // reimplemented using getthrds64().
+  // Reimplemented using getthrds64().
   //
-  // goes like this:
+  // Works like this:
   // For the thread in question, get the kernel thread id. Then get the
   // kernel thread statistics using that id.
   //
   // This only works of course when no pthread scheduling is used,
-  // ie there is a 1:1 relationship to kernel threads.
+  // i.e. there is a 1:1 relationship to kernel threads.
   // On AIX, see AIXTHREAD_SCOPE variable.
 
   pthread_t pthtid = thread->osthread()->pthread_id();
@@ -4371,14 +4374,12 @@
   memset(&uts, 0, sizeof(uts));
   strcpy(uts.sysname, "?");
   if (::uname(&uts) == -1) {
-    fprintf(stderr, "uname failed (%d)\n", errno);
+    trc("uname failed (%d)", errno);
     guarantee(0, "Could not determine whether we run on AIX or PASE");
   } else {
-    if (Verbose) {
-      fprintf(stderr,"uname says: sysname \"%s\" version \"%s\" release \"%s\" "
-              "node \"%s\" machine \"%s\"\n",
-              uts.sysname, uts.version, uts.release, uts.nodename, uts.machine);
-    }
+    trcVerbose("uname says: sysname \"%s\" version \"%s\" release \"%s\" "
+               "node \"%s\" machine \"%s\"\n",
+               uts.sysname, uts.version, uts.release, uts.nodename, uts.machine);
     const int major = atoi(uts.version);
     assert(major > 0, "invalid OS version");
     const int minor = atoi(uts.release);
@@ -4390,12 +4391,10 @@
       // We run on AIX. We do not support versions older than AIX 5.3.
       _on_pase = 0;
       if (_os_version < 0x0503) {
-        fprintf(stderr, "AIX release older than AIX 5.3 not supported.\n");
+        trc("AIX release older than AIX 5.3 not supported.");
         assert(false, "AIX release too old.");
       } else {
-        if (Verbose) {
-          fprintf(stderr, "We run on AIX %d.%d\n", major, minor);
-        }
+        trcVerbose("We run on AIX %d.%d\n", major, minor);
       }
     } else {
       assert(false, "unknown OS");
@@ -4403,7 +4402,6 @@
   }
 
   guarantee(_on_pase != -1 && _os_version, "Could not determine AIX/OS400 release");
-
 } // end: os::Aix::initialize_os_info()
 
 // Scan environment for important settings which might effect the VM.
@@ -4441,12 +4439,10 @@
   // Note: Setting XPG_SUS_ENV in the process is too late. Must be set earlier (before
   // exec() ? before loading the libjvm ? ....)
   p = ::getenv("XPG_SUS_ENV");
-  if (Verbose) {
-    fprintf(stderr, "XPG_SUS_ENV=%s.\n", p ? p : "<unset>");
-  }
+  trcVerbose("XPG_SUS_ENV=%s.", p ? p : "<unset>");
   if (p && strcmp(p, "ON") == 0) {
     _xpg_sus_mode = 1;
-    fprintf(stderr, "Unsupported setting: XPG_SUS_ENV=ON\n");
+    trc("Unsupported setting: XPG_SUS_ENV=ON");
     // This is not supported. Worst of all, it changes behaviour of mmap MAP_FIXED to
     // clobber address ranges. If we ever want to support that, we have to do some
     // testing first.
@@ -4458,10 +4454,7 @@
   // Switch off AIX internal (pthread) guard pages. This has
   // immediate effect for any pthread_create calls which follow.
   p = ::getenv("AIXTHREAD_GUARDPAGES");
-  if (Verbose) {
-    fprintf(stderr, "AIXTHREAD_GUARDPAGES=%s.\n", p ? p : "<unset>");
-    fprintf(stderr, "setting AIXTHREAD_GUARDPAGES=0.\n");
-  }
+  trcVerbose("AIXTHREAD_GUARDPAGES=%s.", p ? p : "<unset>");
   rc = ::putenv("AIXTHREAD_GUARDPAGES=0");
   guarantee(rc == 0, "");
 
@@ -4479,7 +4472,7 @@
   assert(os::Aix::on_aix(), "AIX only");
 
   if (!libperfstat::init()) {
-    fprintf(stderr, "libperfstat initialization failed.\n");
+    trc("libperfstat initialization failed.");
     assert(false, "libperfstat initialization failed");
   } else {
     if (Verbose) {
@@ -4651,7 +4644,6 @@
   return abstime;
 }
 
-
 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
 // Conceptually TryPark() should be equivalent to park(0).
 
@@ -4732,7 +4724,7 @@
   while (_Event < 0) {
     status = pthread_cond_timedwait(_cond, _mutex, &abst);
     assert_status(status == 0 || status == ETIMEDOUT,
-          status, "cond_timedwait");
+                  status, "cond_timedwait");
     if (!FilterSpuriousWakeups) break;         // previous semantics
     if (status == ETIMEDOUT) break;
     // We consume and ignore EINTR and spurious wakeups.
@@ -4866,9 +4858,9 @@
   // Optional fast-path check:
   // Return immediately if a permit is available.
   if (_counter > 0) {
-      _counter = 0;
-      OrderAccess::fence();
-      return;
+    _counter = 0;
+    OrderAccess::fence();
+    return;
   }
 
   Thread* thread = Thread::current();
@@ -4890,7 +4882,6 @@
     unpackTime(&absTime, isAbsolute, time);
   }
 
-
   // Enter safepoint region
   // Beware of deadlocks such as 6317397.
   // The per-thread Parker:: mutex is a classic leaf-lock.
@@ -4978,7 +4969,6 @@
   }
 }
 
-
 extern char** environ;
 
 // Run the specified command in a separate process. Return its exit value,
@@ -4997,44 +4987,43 @@
   } else if (pid == 0) {
     // child process
 
-    // try to be consistent with system(), which uses "/usr/bin/sh" on AIX
+    // Try to be consistent with system(), which uses "/usr/bin/sh" on AIX.
     execve("/usr/bin/sh", argv, environ);
 
     // execve failed
     _exit(-1);
 
-  } else  {
+  } else {
     // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
     // care about the actual exit code, for now.
 
     int status;
 
-    // Wait for the child process to exit.  This returns immediately if
+    // Wait for the child process to exit. This returns immediately if
     // the child has already exited. */
     while (waitpid(pid, &status, 0) < 0) {
-        switch (errno) {
+      switch (errno) {
         case ECHILD: return 0;
         case EINTR: break;
         default: return -1;
-        }
+      }
     }
 
     if (WIFEXITED(status)) {
-       // The child exited normally; get its exit code.
-       return WEXITSTATUS(status);
+      // The child exited normally; get its exit code.
+      return WEXITSTATUS(status);
     } else if (WIFSIGNALED(status)) {
-       // The child exited because of a signal
-       // The best value to return is 0x80 + signal number,
-       // because that is what all Unix shells do, and because
-       // it allows callers to distinguish between process exit and
-       // process death by signal.
-       return 0x80 + WTERMSIG(status);
+      // The child exited because of a signal.
+      // The best value to return is 0x80 + signal number,
+      // because that is what all Unix shells do, and because
+      // it allows callers to distinguish between process exit and
+      // process death by signal.
+      return 0x80 + WTERMSIG(status);
     } else {
-       // Unknown exit code; pass it through
-       return status;
+      // Unknown exit code; pass it through.
+      return status;
     }
   }
-  // Remove warning.
   return -1;
 }
 
@@ -5049,7 +5038,7 @@
   struct stat statbuf;
   char buf[MAXPATHLEN];
   char libmawtpath[MAXPATHLEN];
-  const char *xawtstr  = "/xawt/libmawt.so";
+  const char *xawtstr = "/xawt/libmawt.so";
   const char *new_xawtstr = "/libawt_xawt.so";
 
   char *p;

--- a/src/os/aix/vm/os_aix.hpp	Tue Dec 30 13:19:30 2014 -0800
+++ b/src/os/aix/vm/os_aix.hpp	Thu Jan 01 16:11:49 2015 -0800
@@ -209,7 +209,7 @@
     return _can_use_16M_pages == 1 ? true : false;
   }
 
-  static address   ucontext_get_pc(ucontext_t* uc);
+  static address   ucontext_get_pc(const ucontext_t* uc);
   static intptr_t* ucontext_get_sp(ucontext_t* uc);
   static intptr_t* ucontext_get_fp(ucontext_t* uc);
   // Set PC into context. Needed for continuation after signal.

--- a/src/os_cpu/aix_ppc/vm/os_aix_ppc.cpp	Tue Dec 30 13:19:30 2014 -0800
+++ b/src/os_cpu/aix_ppc/vm/os_aix_ppc.cpp	Thu Jan 01 16:11:49 2015 -0800
@@ -91,8 +91,9 @@
 
 // Frame information (pc, sp, fp) retrieved via ucontext
 // always looks like a C-frame according to the frame
-// conventions in frame_ppc64.hpp.
-address os::Aix::ucontext_get_pc(ucontext_t * uc) {
+// conventions in frame_ppc.hpp.
+
+address os::Aix::ucontext_get_pc(const ucontext_t * uc) {
   return (address)uc->uc_mcontext.jmp_context.iar;
 }
 
@@ -486,7 +487,7 @@
 ////////////////////////////////////////////////////////////////////////////////
 // thread stack
 
-size_t os::Aix::min_stack_allowed = 768*K;
+size_t os::Aix::min_stack_allowed = 128*K;
 
 // Aix is always in floating stack mode. The stack size for a new
 // thread can be set via pthread_attr_setstacksize().
@@ -499,7 +500,7 @@
   // because of the strange 'fallback logic' in os::create_thread().
   // Better set CompilerThreadStackSize in globals_<os_cpu>.hpp if you want to
   // specify a different stack size for compiler threads!
-  size_t s = (thr_type == os::compiler_thread ? 4 * M : 1024 * K);
+  size_t s = (thr_type == os::compiler_thread ? 4 * M : 1 * M);
   return s;
 }
 

--- a/src/os_cpu/aix_ppc/vm/os_aix_ppc.hpp	Tue Dec 30 13:19:30 2014 -0800
+++ b/src/os_cpu/aix_ppc/vm/os_aix_ppc.hpp	Thu Jan 01 16:11:49 2015 -0800
@@ -23,8 +23,8 @@
  *
  */
 
-#ifndef OS_CPU_AIX_OJDKPPC_VM_OS_AIX_PPC_HPP
-#define OS_CPU_AIX_OJDKPPC_VM_OS_AIX_PPC_HPP
+#ifndef OS_CPU_AIX_PPC_VM_OS_AIX_PPC_HPP
+#define OS_CPU_AIX_PPC_VM_OS_AIX_PPC_HPP
 
   static void setup_fpu() {}
 
@@ -32,4 +32,4 @@
   // Note: Currently only used in 64 bit Windows implementations
   static bool register_code_area(char *low, char *high) { return true; }
 
-#endif // OS_CPU_AIX_OJDKPPC_VM_OS_AIX_PPC_HPP
+#endif // OS_CPU_AIX_PPC_VM_OS_AIX_PPC_HPP

--- a/src/os_cpu/aix_ppc/vm/prefetch_aix_ppc.inline.hpp	Tue Dec 30 13:19:30 2014 -0800
+++ b/src/os_cpu/aix_ppc/vm/prefetch_aix_ppc.inline.hpp	Thu Jan 01 16:11:49 2015 -0800
@@ -23,8 +23,8 @@
  *
  */
 
-#ifndef OS_CPU_AIX_PPC_64_VM_PREFETCH_AIX_PPC_64_INLINE_HPP
-#define OS_CPU_AIX_PPC_64_VM_PREFETCH_AIX_PPC_64_INLINE_HPP
+#ifndef OS_CPU_AIX_PPC_VM_PREFETCH_AIX_PPC_INLINE_HPP
+#define OS_CPU_AIX_PPC_VM_PREFETCH_AIX_PPC_INLINE_HPP
 
 #include "runtime/prefetch.hpp"
 
@@ -55,4 +55,4 @@
 #endif
 }
 
-#endif // OS_CPU_AIX_PPC_64_VM_PREFETCH_AIX_PPC_64_INLINE_HPP
+#endif // OS_CPU_AIX_PPC_VM_PREFETCH_AIX_PPC_INLINE_HPP

--- a/src/os_cpu/aix_ppc/vm/threadLS_aix_ppc.hpp	Tue Dec 30 13:19:30 2014 -0800
+++ b/src/os_cpu/aix_ppc/vm/threadLS_aix_ppc.hpp	Thu Jan 01 16:11:49 2015 -0800
@@ -23,8 +23,8 @@
  *
  */
 
-#ifndef OS_CPU_AIX_OJDKPPC_VM_THREADLS_AIX_PPC_HPP
-#define OS_CPU_AIX_OJDKPPC_VM_THREADLS_AIX_PPC_HPP
+#ifndef OS_CPU_AIX_PPC_VM_THREADLS_AIX_PPC_HPP
+#define OS_CPU_AIX_PPC_VM_THREADLS_AIX_PPC_HPP
 
   // Processor dependent parts of ThreadLocalStorage
 
@@ -33,4 +33,4 @@
     return (Thread *) os::thread_local_storage_at(thread_index());
   }
 
-#endif // OS_CPU_AIX_OJDKPPC_VM_THREADLS_AIX_PPC_HPP
+#endif // OS_CPU_AIX_PPC_VM_THREADLS_AIX_PPC_HPP

--- a/src/os_cpu/aix_ppc/vm/thread_aix_ppc.hpp	Tue Dec 30 13:19:30 2014 -0800
+++ b/src/os_cpu/aix_ppc/vm/thread_aix_ppc.hpp	Thu Jan 01 16:11:49 2015 -0800
@@ -23,8 +23,8 @@
  *
  */
 
-#ifndef OS_CPU_AIX_OJDKPPC_VM_THREAD_AIX_PPC_HPP
-#define OS_CPU_AIX_OJDKPPC_VM_THREAD_AIX_PPC_HPP
+#ifndef OS_CPU_AIX_PPC_VM_THREAD_AIX_PPC_HPP
+#define OS_CPU_AIX_PPC_VM_THREAD_AIX_PPC_HPP
 
  private:
   void pd_initialize() {
@@ -76,4 +76,4 @@
 
   intptr_t* last_interpreter_fp() { return _last_interpreter_fp; }
 
-#endif // OS_CPU_AIX_OJDKPPC_VM_THREAD_AIX_PPC_HPP
+#endif // OS_CPU_AIX_PPC_VM_THREAD_AIX_PPC_HPP

--- a/src/os_cpu/linux_ppc/vm/os_linux_ppc.cpp	Tue Dec 30 13:19:30 2014 -0800
+++ b/src/os_cpu/linux_ppc/vm/os_linux_ppc.cpp	Thu Jan 01 16:11:49 2015 -0800
@@ -453,7 +453,7 @@
 ////////////////////////////////////////////////////////////////////////////////
 // thread stack
 
-size_t os::Linux::min_stack_allowed = 768*K;
+size_t os::Linux::min_stack_allowed = 128*K;
 
 bool os::Linux::supports_variable_stack_size() { return true; }
 

--- a/src/share/vm/gc_implementation/concurrentMarkSweep/compactibleFreeListSpace.cpp	Tue Dec 30 13:19:30 2014 -0800
+++ b/src/share/vm/gc_implementation/concurrentMarkSweep/compactibleFreeListSpace.cpp	Thu Jan 01 16:11:49 2015 -0800
@@ -2560,12 +2560,12 @@
      x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x,   \
      x }
 
-// Initialize with default setting of CMSParPromoteBlocksToClaim, _not_
-// OldPLABSize, whose static default is different; if overridden at the
+// Initialize with default setting for CMS, _not_
+// generic OldPLABSize, whose static default is different; if overridden at the
 // command-line, this will get reinitialized via a call to
 // modify_initialization() below.
 AdaptiveWeightedAverage CFLS_LAB::_blocks_to_claim[]    =
-  VECTOR_257(AdaptiveWeightedAverage(OldPLABWeight, (float)CMSParPromoteBlocksToClaim));
+  VECTOR_257(AdaptiveWeightedAverage(OldPLABWeight, (float)CFLS_LAB::_default_dynamic_old_plab_size));
 size_t CFLS_LAB::_global_num_blocks[]  = VECTOR_257(0);
 uint   CFLS_LAB::_global_num_workers[] = VECTOR_257(0);
 

--- a/src/share/vm/gc_implementation/concurrentMarkSweep/compactibleFreeListSpace.hpp	Tue Dec 30 13:19:30 2014 -0800
+++ b/src/share/vm/gc_implementation/concurrentMarkSweep/compactibleFreeListSpace.hpp	Thu Jan 01 16:11:49 2015 -0800
@@ -690,6 +690,9 @@
   void get_from_global_pool(size_t word_sz, AdaptiveFreeList<FreeChunk>* fl);
 
 public:
+  static const int _default_dynamic_old_plab_size = 16;
+  static const int _default_static_old_plab_size  = 50;
+
   CFLS_LAB(CompactibleFreeListSpace* cfls);
 
   // Allocate and return a block of the given size, or else return NULL.

--- a/src/share/vm/gc_implementation/concurrentMarkSweep/concurrentMarkSweepGeneration.cpp	Tue Dec 30 13:19:30 2014 -0800
+++ b/src/share/vm/gc_implementation/concurrentMarkSweep/concurrentMarkSweepGeneration.cpp	Thu Jan 01 16:11:49 2015 -0800
@@ -42,6 +42,7 @@
 #include "gc_implementation/shared/isGCActiveMark.hpp"
 #include "gc_interface/collectedHeap.inline.hpp"
 #include "memory/allocation.hpp"
+#include "memory/cardGeneration.inline.hpp"
 #include "memory/cardTableRS.hpp"
 #include "memory/collectorPolicy.hpp"
 #include "memory/gcLocker.inline.hpp"
@@ -1610,14 +1611,15 @@
 
   // If the collection is being acquired from the background
   // collector, there may be references on the discovered
-  // references lists that have NULL referents (being those
-  // that were concurrently cleared by a mutator) or
-  // that are no longer active (having been enqueued concurrently
-  // by the mutator).
-  // Scrub the list of those references because Mark-Sweep-Compact
-  // code assumes referents are not NULL and that all discovered
-  // Reference objects are active.
-  ref_processor()->clean_up_discovered_references();
+  // references lists.  Abandon those references, since some
+  // of them may have become unreachable after concurrent
+  // discovery; the STW compacting collector will redo discovery
+  // more precisely, without being subject to floating garbage.
+  // Leaving otherwise unreachable references in the discovered
+  // lists would require special handling.
+  ref_processor()->disable_discovery();
+  ref_processor()->abandon_partial_discovery();
+  ref_processor()->verify_no_references_recorded();
 
   if (first_state > Idling) {
     save_heap_summary();
@@ -1683,7 +1685,7 @@
   ReferenceProcessorMTDiscoveryMutator rp_mut_discovery(ref_processor(), false);
 
   ref_processor()->set_enqueuing_is_done(false);
-  ref_processor()->enable_discovery(false /*verify_disabled*/, false /*check_no_refs*/);
+  ref_processor()->enable_discovery();
   ref_processor()->setup_policy(clear_all_soft_refs);
   // If an asynchronous collection finishes, the _modUnionTable is
   // all clear.  If we are assuming the collection from an asynchronous
@@ -3000,7 +3002,7 @@
                     Mutex::_no_safepoint_check_flag);
     checkpointRootsInitialWork();
     // enable ("weak") refs discovery
-    rp->enable_discovery(true /*verify_disabled*/, true /*check_no_refs*/);
+    rp->enable_discovery();
     _collectorState = Marking;
   }
   SpecializationStats::print();

--- a/src/share/vm/gc_implementation/g1/concurrentMark.cpp	Tue Dec 30 13:19:30 2014 -0800
+++ b/src/share/vm/gc_implementation/g1/concurrentMark.cpp	Thu Jan 01 16:11:49 2015 -0800
@@ -971,7 +971,7 @@
   // Start Concurrent Marking weak-reference discovery.
   ReferenceProcessor* rp = g1h->ref_processor_cm();
   // enable ("weak") refs discovery
-  rp->enable_discovery(true /*verify_disabled*/, true /*verify_no_refs*/);
+  rp->enable_discovery();
   rp->setup_policy(false); // snapshot the soft ref policy to be used in this cycle
 
   SATBMarkQueueSet& satb_mq_set = JavaThread::satb_mark_queue_set();

--- a/src/share/vm/gc_implementation/g1/g1AllocRegion.cpp	Tue Dec 30 13:19:30 2014 -0800
+++ b/src/share/vm/gc_implementation/g1/g1AllocRegion.cpp	Thu Jan 01 16:11:49 2015 -0800
@@ -254,25 +254,23 @@
 HeapRegion* SurvivorGCAllocRegion::allocate_new_region(size_t word_size,
                                                        bool force) {
   assert(!force, "not supported for GC alloc regions");
-  return _g1h->new_gc_alloc_region(word_size, count(), GCAllocForSurvived);
+  return _g1h->new_gc_alloc_region(word_size, count(), InCSetState::Young);
 }
 
 void SurvivorGCAllocRegion::retire_region(HeapRegion* alloc_region,
                                           size_t allocated_bytes) {
-  _g1h->retire_gc_alloc_region(alloc_region, allocated_bytes,
-                               GCAllocForSurvived);
+  _g1h->retire_gc_alloc_region(alloc_region, allocated_bytes, InCSetState::Young);
 }
 
 HeapRegion* OldGCAllocRegion::allocate_new_region(size_t word_size,
                                                   bool force) {
   assert(!force, "not supported for GC alloc regions");
-  return _g1h->new_gc_alloc_region(word_size, count(), GCAllocForTenured);
+  return _g1h->new_gc_alloc_region(word_size, count(), InCSetState::Old);
 }
 
 void OldGCAllocRegion::retire_region(HeapRegion* alloc_region,
                                      size_t allocated_bytes) {
-  _g1h->retire_gc_alloc_region(alloc_region, allocated_bytes,
-                               GCAllocForTenured);
+  _g1h->retire_gc_alloc_region(alloc_region, allocated_bytes, InCSetState::Old);
 }
 
 HeapRegion* OldGCAllocRegion::release() {

--- a/src/share/vm/gc_implementation/g1/g1Allocator.cpp	Tue Dec 30 13:19:30 2014 -0800
+++ b/src/share/vm/gc_implementation/g1/g1Allocator.cpp	Thu Jan 01 16:11:49 2015 -0800
@@ -113,15 +113,16 @@
 G1ParGCAllocBuffer::G1ParGCAllocBuffer(size_t gclab_word_size) :
   ParGCAllocBuffer(gclab_word_size), _retired(true) { }
 
-HeapWord* G1ParGCAllocator::allocate_slow(GCAllocPurpose purpose, size_t word_sz, AllocationContext_t context) {
-  HeapWord* obj = NULL;
-  size_t gclab_word_size = _g1h->desired_plab_sz(purpose);
+HeapWord* G1ParGCAllocator::allocate_direct_or_new_plab(InCSetState dest,
+                                                        size_t word_sz,
+                                                        AllocationContext_t context) {
+  size_t gclab_word_size = _g1h->desired_plab_sz(dest);
   if (word_sz * 100 < gclab_word_size * ParallelGCBufferWastePct) {
-    G1ParGCAllocBuffer* alloc_buf = alloc_buffer(purpose, context);
+    G1ParGCAllocBuffer* alloc_buf = alloc_buffer(dest, context);
     add_to_alloc_buffer_waste(alloc_buf->words_remaining());
     alloc_buf->retire(false /* end_of_gc */, false /* retain */);
 
-    HeapWord* buf = _g1h->par_allocate_during_gc(purpose, gclab_word_size, context);
+    HeapWord* buf = _g1h->par_allocate_during_gc(dest, gclab_word_size, context);
     if (buf == NULL) {
       return NULL; // Let caller handle allocation failure.
     }
@@ -129,30 +130,33 @@
     alloc_buf->set_word_size(gclab_word_size);
     alloc_buf->set_buf(buf);
 
-    obj = alloc_buf->allocate(word_sz);
+    HeapWord* const obj = alloc_buf->allocate(word_sz);
     assert(obj != NULL, "buffer was definitely big enough...");
+    return obj;
   } else {
-    obj = _g1h->par_allocate_during_gc(purpose, word_sz, context);
+    return _g1h->par_allocate_during_gc(dest, word_sz, context);
   }
-  return obj;
 }
 
 G1DefaultParGCAllocator::G1DefaultParGCAllocator(G1CollectedHeap* g1h) :
-            G1ParGCAllocator(g1h),
-            _surviving_alloc_buffer(g1h->desired_plab_sz(GCAllocForSurvived)),
-            _tenured_alloc_buffer(g1h->desired_plab_sz(GCAllocForTenured)) {
-
-  _alloc_buffers[GCAllocForSurvived] = &_surviving_alloc_buffer;
-  _alloc_buffers[GCAllocForTenured]  = &_tenured_alloc_buffer;
-
+  G1ParGCAllocator(g1h),
+  _surviving_alloc_buffer(g1h->desired_plab_sz(InCSetState::Young)),
+  _tenured_alloc_buffer(g1h->desired_plab_sz(InCSetState::Old)) {
+  for (uint state = 0; state < InCSetState::Num; state++) {
+    _alloc_buffers[state] = NULL;
+  }
+  _alloc_buffers[InCSetState::Young] = &_surviving_alloc_buffer;
+  _alloc_buffers[InCSetState::Old]  = &_tenured_alloc_buffer;
 }
 
 void G1DefaultParGCAllocator::retire_alloc_buffers() {
-  for (int ap = 0; ap < GCAllocPurposeCount; ++ap) {
-    size_t waste = _alloc_buffers[ap]->words_remaining();
-    add_to_alloc_buffer_waste(waste);
-    _alloc_buffers[ap]->flush_stats_and_retire(_g1h->stats_for_purpose((GCAllocPurpose)ap),
-                                               true /* end_of_gc */,
-                                               false /* retain */);
+  for (uint state = 0; state < InCSetState::Num; state++) {
+    G1ParGCAllocBuffer* const buf = _alloc_buffers[state];
+    if (buf != NULL) {
+      add_to_alloc_buffer_waste(buf->words_remaining());
+      buf->flush_stats_and_retire(_g1h->alloc_buffer_stats(state),
+                                  true /* end_of_gc */,
+                                  false /* retain */);
+    }
   }
 }

--- a/src/share/vm/gc_implementation/g1/g1Allocator.hpp	Tue Dec 30 13:19:30 2014 -0800
+++ b/src/share/vm/gc_implementation/g1/g1Allocator.hpp	Thu Jan 01 16:11:49 2015 -0800
@@ -27,14 +27,9 @@
 
 #include "gc_implementation/g1/g1AllocationContext.hpp"
 #include "gc_implementation/g1/g1AllocRegion.hpp"
+#include "gc_implementation/g1/g1InCSetState.hpp"
 #include "gc_implementation/shared/parGCAllocBuffer.hpp"
 
-enum GCAllocPurpose {
-  GCAllocForTenured,
-  GCAllocForSurvived,
-  GCAllocPurposeCount
-};
-
 // Base class for G1 allocators.
 class G1Allocator : public CHeapObj<mtGC> {
   friend class VMStructs;
@@ -178,20 +173,40 @@
 protected:
   G1CollectedHeap* _g1h;
 
+  // The survivor alignment in effect in bytes.
+  // == 0 : don't align survivors
+  // != 0 : align survivors to that alignment
+  // These values were chosen to favor the non-alignment case since some
+  // architectures have a special compare against zero instructions.
+  const uint _survivor_alignment_bytes;
+
   size_t _alloc_buffer_waste;
   size_t _undo_waste;
 
   void add_to_alloc_buffer_waste(size_t waste) { _alloc_buffer_waste += waste; }
   void add_to_undo_waste(size_t waste)         { _undo_waste += waste; }
 
-  HeapWord* allocate_slow(GCAllocPurpose purpose, size_t word_sz, AllocationContext_t context);
+  virtual void retire_alloc_buffers() = 0;
+  virtual G1ParGCAllocBuffer* alloc_buffer(InCSetState dest, AllocationContext_t context) = 0;
 
-  virtual void retire_alloc_buffers() = 0;
-  virtual G1ParGCAllocBuffer* alloc_buffer(GCAllocPurpose purpose, AllocationContext_t context) = 0;
+  // Calculate the survivor space object alignment in bytes. Returns that or 0 if
+  // there are no restrictions on survivor alignment.
+  static uint calc_survivor_alignment_bytes() {
+    assert(SurvivorAlignmentInBytes >= ObjectAlignmentInBytes, "sanity");
+    if (SurvivorAlignmentInBytes == ObjectAlignmentInBytes) {
+      // No need to align objects in the survivors differently, return 0
+      // which means "survivor alignment is not used".
+      return 0;
+    } else {
+      assert(SurvivorAlignmentInBytes > 0, "sanity");
+      return SurvivorAlignmentInBytes;
+    }
+  }
 
 public:
   G1ParGCAllocator(G1CollectedHeap* g1h) :
-    _g1h(g1h), _alloc_buffer_waste(0), _undo_waste(0) {
+    _g1h(g1h), _survivor_alignment_bytes(calc_survivor_alignment_bytes()),
+    _alloc_buffer_waste(0), _undo_waste(0) {
   }
 
   static G1ParGCAllocator* create_allocator(G1CollectedHeap* g1h);
@@ -199,24 +214,40 @@
   size_t alloc_buffer_waste() { return _alloc_buffer_waste; }
   size_t undo_waste() {return _undo_waste; }
 
-  HeapWord* allocate(GCAllocPurpose purpose, size_t word_sz, AllocationContext_t context) {
-    HeapWord* obj = NULL;
-    if (purpose == GCAllocForSurvived) {
-      obj = alloc_buffer(purpose, context)->allocate_aligned(word_sz, SurvivorAlignmentInBytes);
+  // Allocate word_sz words in dest, either directly into the regions or by
+  // allocating a new PLAB. Returns the address of the allocated memory, NULL if
+  // not successful.
+  HeapWord* allocate_direct_or_new_plab(InCSetState dest,
+                                        size_t word_sz,
+                                        AllocationContext_t context);
+
+  // Allocate word_sz words in the PLAB of dest.  Returns the address of the
+  // allocated memory, NULL if not successful.
+  HeapWord* plab_allocate(InCSetState dest,
+                          size_t word_sz,
+                          AllocationContext_t context) {
+    G1ParGCAllocBuffer* buffer = alloc_buffer(dest, context);
+    if (_survivor_alignment_bytes == 0) {
+      return buffer->allocate(word_sz);
     } else {
-      obj = alloc_buffer(purpose, context)->allocate(word_sz);
+      return buffer->allocate_aligned(word_sz, _survivor_alignment_bytes);
     }
+  }
+
+  HeapWord* allocate(InCSetState dest, size_t word_sz,
+                     AllocationContext_t context) {
+    HeapWord* const obj = plab_allocate(dest, word_sz, context);
     if (obj != NULL) {
       return obj;
     }
-    return allocate_slow(purpose, word_sz, context);
+    return allocate_direct_or_new_plab(dest, word_sz, context);
   }
 
-  void undo_allocation(GCAllocPurpose purpose, HeapWord* obj, size_t word_sz, AllocationContext_t context) {
-    if (alloc_buffer(purpose, context)->contains(obj)) {
-      assert(alloc_buffer(purpose, context)->contains(obj + word_sz - 1),
+  void undo_allocation(InCSetState dest, HeapWord* obj, size_t word_sz, AllocationContext_t context) {
+    if (alloc_buffer(dest, context)->contains(obj)) {
+      assert(alloc_buffer(dest, context)->contains(obj + word_sz - 1),
              "should contain whole object");
-      alloc_buffer(purpose, context)->undo_allocation(obj, word_sz);
+      alloc_buffer(dest, context)->undo_allocation(obj, word_sz);
     } else {
       CollectedHeap::fill_with_object(obj, word_sz);
       add_to_undo_waste(word_sz);
@@ -227,13 +258,17 @@
 class G1DefaultParGCAllocator : public G1ParGCAllocator {
   G1ParGCAllocBuffer  _surviving_alloc_buffer;
   G1ParGCAllocBuffer  _tenured_alloc_buffer;
-  G1ParGCAllocBuffer* _alloc_buffers[GCAllocPurposeCount];
+  G1ParGCAllocBuffer* _alloc_buffers[InCSetState::Num];
 
 public:
   G1DefaultParGCAllocator(G1CollectedHeap* g1h);
 
-  virtual G1ParGCAllocBuffer* alloc_buffer(GCAllocPurpose purpose, AllocationContext_t context) {
-    return _alloc_buffers[purpose];
+  virtual G1ParGCAllocBuffer* alloc_buffer(InCSetState dest, AllocationContext_t context) {
+    assert(dest.is_valid(),
+           err_msg("Allocation buffer index out-of-bounds: " CSETSTATE_FORMAT, dest.value()));
+    assert(_alloc_buffers[dest.value()] != NULL,
+           err_msg("Allocation buffer is NULL: " CSETSTATE_FORMAT, dest.value()));
+    return _alloc_buffers[dest.value()];
   }
 
   virtual void retire_alloc_buffers() ;

--- a/src/share/vm/gc_implementation/g1/g1CollectedHeap.cpp	Tue Dec 30 13:19:30 2014 -0800
+++ b/src/share/vm/gc_implementation/g1/g1CollectedHeap.cpp	Thu Jan 01 16:11:49 2015 -0800
@@ -1301,7 +1301,7 @@
       // Temporarily clear the STW ref processor's _is_alive_non_header field.
       ReferenceProcessorIsAliveMutator stw_rp_is_alive_null(ref_processor_stw(), NULL);
 
-      ref_processor_stw()->enable_discovery(true /*verify_disabled*/, true /*verify_no_refs*/);
+      ref_processor_stw()->enable_discovery();
       ref_processor_stw()->setup_policy(do_clear_all_soft_refs);
 
       // Do collection work
@@ -1886,13 +1886,12 @@
 
   initialize_reserved_region((HeapWord*)heap_rs.base(), (HeapWord*)(heap_rs.base() + heap_rs.size()));
 
-  // Create the gen rem set (and barrier set) for the entire reserved region.
-  _rem_set = collector_policy()->create_rem_set(reserved_region());
-  set_barrier_set(rem_set()->bs());
-  if (!barrier_set()->is_a(BarrierSet::G1SATBCTLogging)) {
-    vm_exit_during_initialization("G1 requires a G1SATBLoggingCardTableModRefBS");
-    return JNI_ENOMEM;
-  }
+  // Create the barrier set for the entire reserved region.
+  G1SATBCardTableLoggingModRefBS* bs
+    = new G1SATBCardTableLoggingModRefBS(reserved_region());
+  bs->initialize();
+  assert(bs->is_a(BarrierSet::G1SATBCTLogging), "sanity");
+  set_barrier_set(bs);
 
   // Also create a G1 rem set.
   _g1_rem_set = new G1RemSet(this, g1_barrier_set());
@@ -3153,8 +3152,6 @@
         failures = true;
       }
     }
-    if (!silent) gclog_or_tty->print("RemSet ");
-    rem_set()->verify();
 
     if (G1StringDedup::is_enabled()) {
       if (!silent) gclog_or_tty->print("StrDedup ");
@@ -3750,8 +3747,7 @@
       // reference processing currently works in G1.
 
       // Enable discovery in the STW reference processor
-      ref_processor_stw()->enable_discovery(true /*verify_disabled*/,
-                                            true /*verify_no_refs*/);
+      ref_processor_stw()->enable_discovery();
 
       {
         // We want to temporarily turn off discovery by the
@@ -3819,6 +3815,8 @@
 
         register_humongous_regions_with_in_cset_fast_test();
 
+        assert(check_cset_fast_test(), "Inconsistency in the InCSetState table.");
+
         _cm->note_start_of_gc();
         // We should not verify the per-thread SATB buffers given that
         // we have not filtered them yet (we'll do so during the
@@ -4048,29 +4046,6 @@
   return true;
 }
 
-size_t G1CollectedHeap::desired_plab_sz(GCAllocPurpose purpose)
-{
-  size_t gclab_word_size;
-  switch (purpose) {
-    case GCAllocForSurvived:
-      gclab_word_size = _survivor_plab_stats.desired_plab_sz();
-      break;
-    case GCAllocForTenured:
-      gclab_word_size = _old_plab_stats.desired_plab_sz();
-      break;
-    default:
-      assert(false, "unknown GCAllocPurpose");
-      gclab_word_size = _old_plab_stats.desired_plab_sz();
-      break;
-  }
-
-  // Prevent humongous PLAB sizes for two reasons:
-  // * PLABs are allocated using a similar paths as oops, but should
-  //   never be in a humongous region
-  // * Allowing humongous PLABs needlessly churns the region free lists
-  return MIN2(_humongous_object_threshold_in_words, gclab_word_size);
-}
-
 void G1CollectedHeap::init_for_evac_failure(OopsInHeapRegionClosure* cl) {
   _drain_in_progress = false;
   set_evac_failure_closure(cl);
@@ -4196,35 +4171,6 @@
   }
 }
 
-HeapWord* G1CollectedHeap::par_allocate_during_gc(GCAllocPurpose purpose,
-                                                  size_t word_size,
-                                                  AllocationContext_t context) {
-  if (purpose == GCAllocForSurvived) {
-    HeapWord* result = survivor_attempt_allocation(word_size, context);
-    if (result != NULL) {
-      return result;
-    } else {
-      // Let's try to allocate in the old gen in case we can fit the
-      // object there.
-      return old_attempt_allocation(word_size, context);
-    }
-  } else {
-    assert(purpose ==  GCAllocForTenured, "sanity");
-    HeapWord* result = old_attempt_allocation(word_size, context);
-    if (result != NULL) {
-      return result;
-    } else {
-      // Let's try to allocate in the survivors in case we can fit the
-      // object there.
-      return survivor_attempt_allocation(word_size, context);
-    }
-  }
-
-  ShouldNotReachHere();
-  // Trying to keep some compilers happy.
-  return NULL;
-}
-
 void G1ParCopyHelper::mark_object(oop obj) {
   assert(!_g1->heap_region_containing(obj)->in_collection_set(), "should not mark objects in the CSet");
 
@@ -4267,15 +4213,14 @@
 
   assert(_worker_id == _par_scan_state->queue_num(), "sanity");
 
-  G1CollectedHeap::in_cset_state_t state = _g1->in_cset_state(obj);
-
-  if (state == G1CollectedHeap::InCSet) {
+  const InCSetState state = _g1->in_cset_state(obj);
+  if (state.is_in_cset()) {
     oop forwardee;
     markOop m = obj->mark();
     if (m->is_marked()) {
       forwardee = (oop) m->decode_pointer();
     } else {
-      forwardee = _par_scan_state->copy_to_survivor_space(obj, m);
+      forwardee = _par_scan_state->copy_to_survivor_space(state, obj, m);
     }
     assert(forwardee != NULL, "forwardee should not be NULL");
     oopDesc::encode_store_heap_oop(p, forwardee);
@@ -4289,7 +4234,7 @@
       do_klass_barrier(p, forwardee);
     }
   } else {
-    if (state == G1CollectedHeap::IsHumongous) {
+    if (state.is_humongous()) {
       _g1->set_humongous_is_live(obj);
     }
     // The object is not in collection set. If we're a root scanning
@@ -4609,7 +4554,7 @@
 G1CollectedHeap::
 g1_process_roots(OopClosure* scan_non_heap_roots,
                  OopClosure* scan_non_heap_weak_roots,
-                 OopsInHeapRegionClosure* scan_rs,
+                 G1ParPushHeapRSClosure* scan_rs,
                  CLDClosure* scan_strong_clds,
                  CLDClosure* scan_weak_clds,
                  CodeBlobClosure* scan_strong_code,
@@ -5145,17 +5090,17 @@
     oop obj = *p;
     assert(obj != NULL, "the caller should have filtered out NULL values");
 
-    G1CollectedHeap::in_cset_state_t cset_state = _g1->in_cset_state(obj);
-    if (cset_state == G1CollectedHeap::InNeither) {
+    const InCSetState cset_state = _g1->in_cset_state(obj);
+    if (!cset_state.is_in_cset_or_humongous()) {
       return;
     }
-    if (cset_state == G1CollectedHeap::InCSet) {
+    if (cset_state.is_in_cset()) {
       assert( obj->is_forwarded(), "invariant" );
       *p = obj->forwardee();
     } else {
       assert(!obj->is_forwarded(), "invariant" );
-      assert(cset_state == G1CollectedHeap::IsHumongous,
-             err_msg("Only allowed InCSet state is IsHumongous, but is %d", cset_state));
+      assert(cset_state.is_humongous(),
+             err_msg("Only allowed InCSet state is IsHumongous, but is %d", cset_state.value()));
       _g1->set_humongous_is_live(obj);
     }
   }
@@ -5640,8 +5585,6 @@
 
   init_for_evac_failure(NULL);
 
-  rem_set()->prepare_for_younger_refs_iterate(true);
-
   assert(dirty_card_queue_set().completed_buffers_num() == 0, "Should be empty");
   double start_par_time_sec = os::elapsedTime();
   double end_par_time_sec;
@@ -5951,6 +5894,70 @@
   heap_region_iterate(&cl);
   guarantee(!cl.failures(), "bitmap verification");
 }
+
+class G1CheckCSetFastTableClosure : public HeapRegionClosure {
+ private:
+  bool _failures;
+ public:
+  G1CheckCSetFastTableClosure() : HeapRegionClosure(), _failures(false) { }
+
+  virtual bool doHeapRegion(HeapRegion* hr) {
+    uint i = hr->hrm_index();
+    InCSetState cset_state = (InCSetState) G1CollectedHeap::heap()->_in_cset_fast_test.get_by_index(i);
+    if (hr->is_humongous()) {
+      if (hr->in_collection_set()) {
+        gclog_or_tty->print_cr("\n## humongous region %u in CSet", i);
+        _failures = true;
+        return true;
+      }
+      if (cset_state.is_in_cset()) {
+        gclog_or_tty->print_cr("\n## inconsistent cset state %d for humongous region %u", cset_state.value(), i);
+        _failures = true;
+        return true;
+      }
+      if (hr->is_continues_humongous() && cset_state.is_humongous()) {
+        gclog_or_tty->print_cr("\n## inconsistent cset state %d for continues humongous region %u", cset_state.value(), i);
+        _failures = true;
+        return true;
+      }
+    } else {
+      if (cset_state.is_humongous()) {
+        gclog_or_tty->print_cr("\n## inconsistent cset state %d for non-humongous region %u", cset_state.value(), i);
+        _failures = true;
+        return true;
+      }
+      if (hr->in_collection_set() != cset_state.is_in_cset()) {
+        gclog_or_tty->print_cr("\n## in CSet %d / cset state %d inconsistency for region %u",
+                               hr->in_collection_set(), cset_state.value(), i);
+        _failures = true;
+        return true;
+      }
+      if (cset_state.is_in_cset()) {
+        if (hr->is_young() != (cset_state.is_young())) {
+          gclog_or_tty->print_cr("\n## is_young %d / cset state %d inconsistency for region %u",
+                                 hr->is_young(), cset_state.value(), i);
+          _failures = true;
+          return true;
+        }
+        if (hr->is_old() != (cset_state.is_old())) {
+          gclog_or_tty->print_cr("\n## is_old %d / cset state %d inconsistency for region %u",
+                                 hr->is_old(), cset_state.value(), i);
+          _failures = true;
+          return true;
+        }
+      }
+    }
+    return false;
+  }
+
+  bool failures() const { return _failures; }
+};
+
+bool G1CollectedHeap::check_cset_fast_test() {
+  G1CheckCSetFastTableClosure cl;
+  _hrm.iterate(&cl);
+  return !cl.failures();
+}
 #endif // PRODUCT
 
 void G1CollectedHeap::cleanUpCardTable() {
@@ -6519,20 +6526,20 @@
 
 HeapRegion* G1CollectedHeap::new_gc_alloc_region(size_t word_size,
                                                  uint count,
-                                                 GCAllocPurpose ap) {
+                                                 InCSetState dest) {
   assert(FreeList_lock->owned_by_self(), "pre-condition");
 
-  if (count < g1_policy()->max_regions(ap)) {
-    bool survivor = (ap == GCAllocForSurvived);
+  if (count < g1_policy()->max_regions(dest)) {
+    const bool is_survivor = (dest.is_young());
     HeapRegion* new_alloc_region = new_region(word_size,
-                                              !survivor,
+                                              !is_survivor,
                                               true /* do_expand */);
     if (new_alloc_region != NULL) {
       // We really only need to do this for old regions given that we
       // should never scan survivors. But it doesn't hurt to do it
       // for survivors too.
       new_alloc_region->record_timestamp();
-      if (survivor) {
+      if (is_survivor) {
         new_alloc_region->set_survivor();
         _hr_printer.alloc(new_alloc_region, G1HRPrinter::Survivor);
         check_bitmaps("Survivor Region Allocation", new_alloc_region);
@@ -6544,8 +6551,6 @@
       bool during_im = g1_policy()->during_initial_mark_pause();
       new_alloc_region->note_start_of_copying(during_im);
       return new_alloc_region;
-    } else {
-      g1_policy()->note_alloc_region_limit_reached(ap);
     }
   }
   return NULL;
@@ -6553,11 +6558,11 @@
 
 void G1CollectedHeap::retire_gc_alloc_region(HeapRegion* alloc_region,
                                              size_t allocated_bytes,
-                                             GCAllocPurpose ap) {
+                                             InCSetState dest) {
   bool during_im = g1_policy()->during_initial_mark_pause();
   alloc_region->note_end_of_copying(during_im);
   g1_policy()->record_bytes_copied_during_gc(allocated_bytes);
-  if (ap == GCAllocForSurvived) {
+  if (dest.is_young()) {
     young_list()->add_survivor_region(alloc_region);
   } else {
     _old_set.add(alloc_region);

--- a/src/share/vm/gc_implementation/g1/g1CollectedHeap.hpp	Tue Dec 30 13:19:30 2014 -0800
+++ b/src/share/vm/gc_implementation/g1/g1CollectedHeap.hpp	Thu Jan 01 16:11:49 2015 -0800
@@ -32,6 +32,7 @@
 #include "gc_implementation/g1/g1AllocRegion.hpp"
 #include "gc_implementation/g1/g1BiasedArray.hpp"
 #include "gc_implementation/g1/g1HRPrinter.hpp"
+#include "gc_implementation/g1/g1InCSetState.hpp"
 #include "gc_implementation/g1/g1MonitoringSupport.hpp"
 #include "gc_implementation/g1/g1SATBCardTableModRefBS.hpp"
 #include "gc_implementation/g1/g1YCTypes.hpp"
@@ -213,6 +214,9 @@
   friend class G1MarkSweep;
   friend class HeapRegionClaimer;
 
+  // Testing classes.
+  friend class G1CheckCSetFastTableClosure;
+
 private:
   // The one and only G1CollectedHeap, so static functions can find it.
   static G1CollectedHeap* _g1h;
@@ -547,15 +551,9 @@
   // allocation region, either by picking one or expanding the
   // heap, and then allocate a block of the given size. The block
   // may not be a humongous - it must fit into a single heap region.
-  HeapWord* par_allocate_during_gc(GCAllocPurpose purpose,
-                                   size_t word_size,
-                                   AllocationContext_t context);
-
-  HeapWord* allocate_during_gc_slow(GCAllocPurpose purpose,
-                                    HeapRegion*    alloc_region,
-                                    bool           par,
-                                    size_t         word_size);
-
+  inline HeapWord* par_allocate_during_gc(InCSetState dest,
+                                          size_t word_size,
+                                          AllocationContext_t context);
   // Ensure that no further allocations can happen in "r", bearing in mind
   // that parallel threads might be attempting allocations.
   void par_allocate_remaining_space(HeapRegion* r);
@@ -577,9 +575,9 @@
 
   // For GC alloc regions.
   HeapRegion* new_gc_alloc_region(size_t word_size, uint count,
-                                  GCAllocPurpose ap);
+                                  InCSetState dest);
   void retire_gc_alloc_region(HeapRegion* alloc_region,
-                              size_t allocated_bytes, GCAllocPurpose ap);
+                              size_t allocated_bytes, InCSetState dest);
 
   // - if explicit_gc is true, the GC is for a System.gc() or a heap
   //   inspection request and should collect the entire heap
@@ -640,26 +638,11 @@
   // (Rounds up to a HeapRegion boundary.)
   bool expand(size_t expand_bytes);
 
-  // Returns the PLAB statistics given a purpose.
-  PLABStats* stats_for_purpose(GCAllocPurpose purpose) {
-    PLABStats* stats = NULL;
+  // Returns the PLAB statistics for a given destination.
+  inline PLABStats* alloc_buffer_stats(InCSetState dest);
 
-    switch (purpose) {
-    case GCAllocForSurvived:
-      stats = &_survivor_plab_stats;
-      break;
-    case GCAllocForTenured:
-      stats = &_old_plab_stats;
-      break;
-    default:
-      assert(false, "unrecognized GCAllocPurpose");
-    }
-
-    return stats;
-  }
-
-  // Determines PLAB size for a particular allocation purpose.
-  size_t desired_plab_sz(GCAllocPurpose purpose);
+  // Determines PLAB size for a given destination.
+  inline size_t desired_plab_sz(InCSetState dest);
 
   inline AllocationContextStats& allocation_context_stats();
 
@@ -683,8 +666,11 @@
   void register_humongous_regions_with_in_cset_fast_test();
   // We register a region with the fast "in collection set" test. We
   // simply set to true the array slot corresponding to this region.
-  void register_region_with_in_cset_fast_test(HeapRegion* r) {
-    _in_cset_fast_test.set_in_cset(r->hrm_index());
+  void register_young_region_with_in_cset_fast_test(HeapRegion* r) {
+    _in_cset_fast_test.set_in_young(r->hrm_index());
+  }
+  void register_old_region_with_in_cset_fast_test(HeapRegion* r) {
+    _in_cset_fast_test.set_in_old(r->hrm_index());
   }
 
   // This is a fast test on whether a reference points into the
@@ -821,7 +807,7 @@
   // In the sequential case this param will be ignored.
   void g1_process_roots(OopClosure* scan_non_heap_roots,
                         OopClosure* scan_non_heap_weak_roots,
-                        OopsInHeapRegionClosure* scan_rs,
+                        G1ParPushHeapRSClosure* scan_rs,
                         CLDClosure* scan_strong_clds,
                         CLDClosure* scan_weak_clds,
                         CodeBlobClosure* scan_strong_code,
@@ -1181,6 +1167,9 @@
   // appropriate error messages and crash.
   void check_bitmaps(const char* caller) PRODUCT_RETURN;
 
+  // Do sanity check on the contents of the in-cset fast test table.
+  bool check_cset_fast_test() PRODUCT_RETURN_( return true; );
+
   // 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).
@@ -1276,53 +1265,15 @@
 
   inline bool is_in_cset_or_humongous(const oop obj);
 
-  enum in_cset_state_t {
-   InNeither,           // neither in collection set nor humongous
-   InCSet,              // region is in collection set only
-   IsHumongous          // region is a humongous start region
-  };
  private:
-  // Instances of this class are used for quick tests on whether a reference points
-  // into the collection set or is a humongous object (points into a humongous
-  // object).
-  // Each of the array's elements denotes whether the corresponding region is in
-  // the collection set or a humongous region.
-  // We use this to quickly reclaim humongous objects: by making a humongous region
-  // succeed this test, we sort-of add it to the collection set. During the reference
-  // iteration closures, when we see a humongous region, we simply mark it as
-  // referenced, i.e. live.
-  class G1FastCSetBiasedMappedArray : public G1BiasedMappedArray<char> {
-   protected:
-    char default_value() const { return G1CollectedHeap::InNeither; }
-   public:
-    void set_humongous(uintptr_t index) {
-      assert(get_by_index(index) != InCSet, "Should not overwrite InCSet values");
-      set_by_index(index, G1CollectedHeap::IsHumongous);
-    }
-
-    void clear_humongous(uintptr_t index) {
-      set_by_index(index, G1CollectedHeap::InNeither);
-    }
-
-    void set_in_cset(uintptr_t index) {
-      assert(get_by_index(index) != G1CollectedHeap::IsHumongous, "Should not overwrite IsHumongous value");
-      set_by_index(index, G1CollectedHeap::InCSet);
-    }
-
-    bool is_in_cset_or_humongous(HeapWord* addr) const { return get_by_address(addr) != G1CollectedHeap::InNeither; }
-    bool is_in_cset(HeapWord* addr) const { return get_by_address(addr) == G1CollectedHeap::InCSet; }
-    G1CollectedHeap::in_cset_state_t at(HeapWord* addr) const { return (G1CollectedHeap::in_cset_state_t)get_by_address(addr); }
-    void clear() { G1BiasedMappedArray<char>::clear(); }
-  };
-
   // This array is used for a quick test on whether a reference points into
   // the collection set or not. Each of the array's elements denotes whether the
   // corresponding region is in the collection set or not.
-  G1FastCSetBiasedMappedArray _in_cset_fast_test;
+  G1InCSetStateFastTestBiasedMappedArray _in_cset_fast_test;
 
  public:
 
-  inline in_cset_state_t in_cset_state(const oop obj);
+  inline InCSetState in_cset_state(const oop obj);
 
   // Return "TRUE" iff the given object address is in the reserved
   // region of g1.

--- a/src/share/vm/gc_implementation/g1/g1CollectedHeap.inline.hpp	Tue Dec 30 13:19:30 2014 -0800
+++ b/src/share/vm/gc_implementation/g1/g1CollectedHeap.inline.hpp	Thu Jan 01 16:11:49 2015 -0800
@@ -35,6 +35,41 @@
 #include "runtime/orderAccess.inline.hpp"
 #include "utilities/taskqueue.hpp"
 
+PLABStats* G1CollectedHeap::alloc_buffer_stats(InCSetState dest) {
+  switch (dest.value()) {
+    case InCSetState::Young:
+      return &_survivor_plab_stats;
+    case InCSetState::Old:
+      return &_old_plab_stats;
+    default:
+      ShouldNotReachHere();
+      return NULL; // Keep some compilers happy
+  }
+}
+
+size_t G1CollectedHeap::desired_plab_sz(InCSetState dest) {
+  size_t gclab_word_size = alloc_buffer_stats(dest)->desired_plab_sz();
+  // Prevent humongous PLAB sizes for two reasons:
+  // * PLABs are allocated using a similar paths as oops, but should
+  //   never be in a humongous region
+  // * Allowing humongous PLABs needlessly churns the region free lists
+  return MIN2(_humongous_object_threshold_in_words, gclab_word_size);
+}
+
+HeapWord* G1CollectedHeap::par_allocate_during_gc(InCSetState dest,
+                                                  size_t word_size,
+                                                  AllocationContext_t context) {
+  switch (dest.value()) {
+    case InCSetState::Young:
+      return survivor_attempt_allocation(word_size, context);
+    case InCSetState::Old:
+      return old_attempt_allocation(word_size, context);
+    default:
+      ShouldNotReachHere();
+      return NULL; // Keep some compilers happy
+  }
+}
+
 // Inline functions for G1CollectedHeap
 
 inline AllocationContextStats& G1CollectedHeap::allocation_context_stats() {
@@ -203,7 +238,7 @@
   return _in_cset_fast_test.is_in_cset_or_humongous((HeapWord*)obj);
 }
 
-G1CollectedHeap::in_cset_state_t G1CollectedHeap::in_cset_state(const oop obj) {
+InCSetState G1CollectedHeap::in_cset_state(const oop obj) {
   return _in_cset_fast_test.at((HeapWord*)obj);
 }
 

--- a/src/share/vm/gc_implementation/g1/g1CollectorPolicy.cpp	Tue Dec 30 13:19:30 2014 -0800
+++ b/src/share/vm/gc_implementation/g1/g1CollectorPolicy.cpp	Thu Jan 01 16:11:49 2015 -0800
@@ -1437,18 +1437,6 @@
   return young_list_length < young_list_max_length;
 }
 
-uint G1CollectorPolicy::max_regions(int purpose) {
-  switch (purpose) {
-    case GCAllocForSurvived:
-      return _max_survivor_regions;
-    case GCAllocForTenured:
-      return REGIONS_UNLIMITED;
-    default:
-      ShouldNotReachHere();
-      return REGIONS_UNLIMITED;
-  };
-}
-
 void G1CollectorPolicy::update_max_gc_locker_expansion() {
   uint expansion_region_num = 0;
   if (GCLockerEdenExpansionPercent > 0) {
@@ -1634,7 +1622,7 @@
   hr->set_next_in_collection_set(_collection_set);
   _collection_set = hr;
   _collection_set_bytes_used_before += hr->used();
-  _g1->register_region_with_in_cset_fast_test(hr);
+  _g1->register_old_region_with_in_cset_fast_test(hr);
   size_t rs_length = hr->rem_set()->occupied();
   _recorded_rs_lengths += rs_length;
   _old_cset_region_length += 1;
@@ -1767,7 +1755,7 @@
   hr->set_in_collection_set(true);
   assert( hr->next_in_collection_set() == NULL, "invariant");
 
-  _g1->register_region_with_in_cset_fast_test(hr);
+  _g1->register_young_region_with_in_cset_fast_test(hr);
 }
 
 // Add the region at the RHS of the incremental cset

--- a/src/share/vm/gc_implementation/g1/g1CollectorPolicy.hpp	Tue Dec 30 13:19:30 2014 -0800
+++ b/src/share/vm/gc_implementation/g1/g1CollectorPolicy.hpp	Thu Jan 01 16:11:49 2015 -0800
@@ -881,28 +881,20 @@
 public:
   uint tenuring_threshold() const { return _tenuring_threshold; }
 
-  inline GCAllocPurpose
-    evacuation_destination(HeapRegion* src_region, uint age, size_t word_sz) {
-      if (age < _tenuring_threshold && src_region->is_young()) {
-        return GCAllocForSurvived;
-      } else {
-        return GCAllocForTenured;
-      }
-  }
-
-  inline bool track_object_age(GCAllocPurpose purpose) {
-    return purpose == GCAllocForSurvived;
-  }
-
   static const uint REGIONS_UNLIMITED = (uint) -1;
 
-  uint max_regions(int purpose);
-
-  // The limit on regions for a particular purpose is reached.
-  void note_alloc_region_limit_reached(int purpose) {
-    if (purpose == GCAllocForSurvived) {
-      _tenuring_threshold = 0;
+  uint max_regions(InCSetState dest) {
+    switch (dest.value()) {
+      case InCSetState::Young:
+        return _max_survivor_regions;
+      case InCSetState::Old:
+        return REGIONS_UNLIMITED;
+      default:
+        assert(false, err_msg("Unknown dest state: " CSETSTATE_FORMAT, dest.value()));
+        break;
     }
+    // keep some compilers happy
+    return 0;
   }
 
   void note_start_adding_survivor_regions() {

--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/src/share/vm/gc_implementation/g1/g1InCSetState.hpp	Thu Jan 01 16:11:49 2015 -0800
@@ -0,0 +1,132 @@
+/*
+ * Copyright (c) 2014, 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_G1INCSETSTATE_HPP
+#define SHARE_VM_GC_IMPLEMENTATION_G1_G1INCSETSTATE_HPP
+
+#include "gc_implementation/g1/g1BiasedArray.hpp"
+#include "memory/allocation.hpp"
+
+// Per-region state during garbage collection.
+struct InCSetState {
+ public:
+  // We use different types to represent the state value. Particularly SPARC puts
+  // values in structs from "left to right", i.e. MSB to LSB. This results in many
+  // unnecessary shift operations when loading and storing values of this type.
+  // This degrades performance significantly (>10%) on that platform.
+  // Other tested ABIs do not seem to have this problem, and actually tend to
+  // favor smaller types, so we use the smallest usable type there.
+#ifdef SPARC
+  #define CSETSTATE_FORMAT INTPTR_FORMAT
+  typedef intptr_t in_cset_state_t;
+#else
+  #define CSETSTATE_FORMAT "%d"
+  typedef int8_t in_cset_state_t;
+#endif
+ private:
+  in_cset_state_t _value;
+ public:
+  enum {
+    // Selection of the values were driven to micro-optimize the encoding and
+    // frequency of the checks.
+    // The most common check is whether the region is in the collection set or not.
+    // This encoding allows us to use an != 0 check which in some architectures
+    // (x86*) can be encoded slightly more efficently than a normal comparison
+    // against zero.
+    // The same situation occurs when checking whether the region is humongous
+    // or not, which is encoded by values < 0.
+    // The other values are simply encoded in increasing generation order, which
+    // makes getting the next generation fast by a simple increment.
+    Humongous    = -1,    // The region is humongous - note that actually any value < 0 would be possible here.
+    NotInCSet    =  0,    // The region is not in the collection set.
+    Young        =  1,    // The region is in the collection set and a young region.
+    Old          =  2,    // The region is in the collection set and an old region.
+    Num
+  };
+
+  InCSetState(in_cset_state_t value = NotInCSet) : _value(value) {
+    assert(is_valid(), err_msg("Invalid state %d", _value));
+  }
+
+  in_cset_state_t value() const        { return _value; }
+
+  void set_old()                       { _value = Old; }
+
+  bool is_in_cset_or_humongous() const { return _value != NotInCSet; }
+  bool is_in_cset() const              { return _value > NotInCSet; }
+  bool is_humongous() const            { return _value < NotInCSet; }
+  bool is_young() const                { return _value == Young; }
+  bool is_old() const                  { return _value == Old; }
+
+#ifdef ASSERT
+  bool is_default() const              { return !is_in_cset_or_humongous(); }
+  bool is_valid() const                { return (_value >= Humongous) && (_value < Num); }
+  bool is_valid_gen() const            { return (_value >= Young && _value <= Old); }
+#endif
+};
+
+// Instances of this class are used for quick tests on whether a reference points
+// into the collection set and into which generation or is a humongous object
+//
+// Each of the array's elements indicates whether the corresponding region is in
+// the collection set and if so in which generation, or a humongous region.
+//
+// We use this to speed up reference processing during young collection and
+// quickly reclaim humongous objects. For the latter, by making a humongous region
+// succeed this test, we sort-of add it to the collection set. During the reference
+// iteration closures, when we see a humongous region, we then simply mark it as
+// referenced, i.e. live.
+class G1InCSetStateFastTestBiasedMappedArray : public G1BiasedMappedArray<InCSetState> {
+ protected:
+  InCSetState default_value() const { return InCSetState::NotInCSet; }
+ public:
+  void set_humongous(uintptr_t index) {
+    assert(get_by_index(index).is_default(),
+           err_msg("State at index " INTPTR_FORMAT" should be default but is " CSETSTATE_FORMAT, index, get_by_index(index).value()));
+    set_by_index(index, InCSetState::Humongous);
+  }
+
+  void clear_humongous(uintptr_t index) {
+    set_by_index(index, InCSetState::NotInCSet);
+  }
+
+  void set_in_young(uintptr_t index) {
+    assert(get_by_index(index).is_default(),
+           err_msg("State at index " INTPTR_FORMAT" should be default but is " CSETSTATE_FORMAT, index, get_by_index(index).value()));
+    set_by_index(index, InCSetState::Young);
+  }
+
+  void set_in_old(uintptr_t index) {
+    assert(get_by_index(index).is_default(),
+           err_msg("State at index " INTPTR_FORMAT" should be default but is " CSETSTATE_FORMAT, index, get_by_index(index).value()));
+    set_by_index(index, InCSetState::Old);
+  }
+
+  bool is_in_cset_or_humongous(HeapWord* addr) const { return at(addr).is_in_cset_or_humongous(); }
+  bool is_in_cset(HeapWord* addr) const { return at(addr).is_in_cset(); }
+  InCSetState at(HeapWord* addr) const { return get_by_address(addr); }
+  void clear() { G1BiasedMappedArray<InCSetState>::clear(); }
+};
+
+#endif // SHARE_VM_GC_IMPLEMENTATION_G1_G1INCSETSTATE_HPP

--- a/src/share/vm/gc_implementation/g1/g1OopClosures.hpp	Tue Dec 30 13:19:30 2014 -0800
+++ b/src/share/vm/gc_implementation/g1/g1OopClosures.hpp	Thu Jan 01 16:11:49 2015 -0800
@@ -26,6 +26,7 @@
 #define SHARE_VM_GC_IMPLEMENTATION_G1_G1OOPCLOSURES_HPP
 
 #include "memory/iterator.hpp"
+#include "oops/markOop.hpp"
 
 class HeapRegion;
 class G1CollectedHeap;
@@ -239,14 +240,14 @@
   G1CollectedHeap* _g1;
   G1RemSet* _g1_rem_set;
   HeapRegion* _from;
-  OopsInHeapRegionClosure* _push_ref_cl;
+  G1ParPushHeapRSClosure* _push_ref_cl;
   bool _record_refs_into_cset;
   uint _worker_i;
 
 public:
   G1UpdateRSOrPushRefOopClosure(G1CollectedHeap* g1h,
                                 G1RemSet* rs,
-                                OopsInHeapRegionClosure* push_ref_cl,
+                                G1ParPushHeapRSClosure* push_ref_cl,
                                 bool record_refs_into_cset,
                                 uint worker_i = 0);
 
@@ -256,7 +257,8 @@
   }
 
   bool self_forwarded(oop obj) {
-    bool result = (obj->is_forwarded() && (obj->forwardee()== obj));
+    markOop m = obj->mark();
+    bool result = (m->is_marked() && ((oop)m->decode_pointer() == obj));
     return result;
   }
 

--- a/src/share/vm/gc_implementation/g1/g1OopClosures.inline.hpp	Tue Dec 30 13:19:30 2014 -0800
+++ b/src/share/vm/gc_implementation/g1/g1OopClosures.inline.hpp	Thu Jan 01 16:11:49 2015 -0800
@@ -67,8 +67,8 @@
 
   if (!oopDesc::is_null(heap_oop)) {
     oop obj = oopDesc::decode_heap_oop_not_null(heap_oop);
-    G1CollectedHeap::in_cset_state_t state = _g1->in_cset_state(obj);
-    if (state == G1CollectedHeap::InCSet) {
+    const InCSetState state = _g1->in_cset_state(obj);
+    if (state.is_in_cset()) {
       // We're not going to even bother checking whether the object is
       // already forwarded or not, as this usually causes an immediate
       // stall. We'll try to prefetch the object (for write, given that
@@ -87,7 +87,7 @@
 
       _par_scan_state->push_on_queue(p);
     } else {
-      if (state == G1CollectedHeap::IsHumongous) {
+      if (state.is_humongous()) {
         _g1->set_humongous_is_live(obj);
       }
       _par_scan_state->update_rs(_from, p, _worker_id);

--- a/src/share/vm/gc_implementation/g1/g1PageBasedVirtualSpace.cpp	Tue Dec 30 13:19:30 2014 -0800
+++ b/src/share/vm/gc_implementation/g1/g1PageBasedVirtualSpace.cpp	Thu Jan 01 16:11:49 2015 -0800
@@ -131,6 +131,9 @@
   _committed.set_range(start, start + size_in_pages);
 
   MemRegion result((HeapWord*)page_start(start), byte_size_for_pages(size_in_pages) / HeapWordSize);
+  if (AlwaysPreTouch) {
+    os::pretouch_memory((char*)result.start(), (char*)result.end());
+  }
   return result;
 }
 

--- a/src/share/vm/gc_implementation/g1/g1ParScanThreadState.cpp	Tue Dec 30 13:19:30 2014 -0800
+++ b/src/share/vm/gc_implementation/g1/g1ParScanThreadState.cpp	Thu Jan 01 16:11:49 2015 -0800
@@ -38,6 +38,7 @@
     _g1_rem(g1h->g1_rem_set()),
     _hash_seed(17), _queue_num(queue_num),
     _term_attempts(0),
+    _tenuring_threshold(g1h->g1_policy()->tenuring_threshold()),
     _age_table(false), _scanner(g1h, rp),
     _strong_roots_time(0), _term_time(0) {
   _scanner.set_par_scan_thread_state(this);
@@ -59,6 +60,12 @@
 
   _g1_par_allocator = G1ParGCAllocator::create_allocator(_g1h);
 
+  _dest[InCSetState::NotInCSet]    = InCSetState::NotInCSet;
+  // The dest for Young is used when the objects are aged enough to
+  // need to be moved to the next space.
+  _dest[InCSetState::Young]        = InCSetState::Old;
+  _dest[InCSetState::Old]          = InCSetState::Old;
+
   _start = os::elapsedTime();
 }
 
@@ -150,52 +157,94 @@
   } while (!_refs->is_empty());
 }
 
-oop G1ParScanThreadState::copy_to_survivor_space(oop const old,
+HeapWord* G1ParScanThreadState::allocate_in_next_plab(InCSetState const state,
+                                                      InCSetState* dest,
+                                                      size_t word_sz,
+                                                      AllocationContext_t const context) {
+  assert(state.is_in_cset_or_humongous(), err_msg("Unexpected state: " CSETSTATE_FORMAT, state.value()));
+  assert(dest->is_in_cset_or_humongous(), err_msg("Unexpected dest: " CSETSTATE_FORMAT, dest->value()));
+
+  // Right now we only have two types of regions (young / old) so
+  // let's keep the logic here simple. We can generalize it when necessary.
+  if (dest->is_young()) {
+    HeapWord* const obj_ptr = _g1_par_allocator->allocate(InCSetState::Old,
+                                                          word_sz, context);
+    if (obj_ptr == NULL) {
+      return NULL;
+    }
+    // Make sure that we won't attempt to copy any other objects out
+    // of a survivor region (given that apparently we cannot allocate
+    // any new ones) to avoid coming into this slow path.
+    _tenuring_threshold = 0;
+    dest->set_old();
+    return obj_ptr;
+  } else {
+    assert(dest->is_old(), err_msg("Unexpected dest: " CSETSTATE_FORMAT, dest->value()));
+    // no other space to try.
+    return NULL;
+  }
+}
+
+InCSetState G1ParScanThreadState::next_state(InCSetState const state, markOop const m, uint& age) {
+  if (state.is_young()) {
+    age = !m->has_displaced_mark_helper() ? m->age()
+                                          : m->displaced_mark_helper()->age();
+    if (age < _tenuring_threshold) {
+      return state;
+    }
+  }
+  return dest(state);
+}
+
+oop G1ParScanThreadState::copy_to_survivor_space(InCSetState const state,
+                                                 oop const old,
                                                  markOop const old_mark) {
-  size_t word_sz = old->size();
-  HeapRegion* from_region = _g1h->heap_region_containing_raw(old);
+  const size_t word_sz = old->size();
+  HeapRegion* const from_region = _g1h->heap_region_containing_raw(old);
   // +1 to make the -1 indexes valid...
-  int       young_index = from_region->young_index_in_cset()+1;
+  const int young_index = from_region->young_index_in_cset()+1;
   assert( (from_region->is_young() && young_index >  0) ||
          (!from_region->is_young() && young_index == 0), "invariant" );
-  G1CollectorPolicy* g1p = _g1h->g1_policy();
-  uint age = old_mark->has_displaced_mark_helper() ? old_mark->displaced_mark_helper()->age()
-                                                   : old_mark->age();
-  GCAllocPurpose alloc_purpose = g1p->evacuation_destination(from_region, age,
-                                                             word_sz);
-  AllocationContext_t context = from_region->allocation_context();
-  HeapWord* obj_ptr = _g1_par_allocator->allocate(alloc_purpose, word_sz, context);
+  const AllocationContext_t context = from_region->allocation_context();
+
+  uint age = 0;
+  InCSetState dest_state = next_state(state, old_mark, age);
+  HeapWord* obj_ptr = _g1_par_allocator->plab_allocate(dest_state, word_sz, context);
+
+  // PLAB allocations should succeed most of the time, so we'll
+  // normally check against NULL once and that's it.
+  if (obj_ptr == NULL) {
+    obj_ptr = _g1_par_allocator->allocate_direct_or_new_plab(dest_state, word_sz, context);
+    if (obj_ptr == NULL) {
+      obj_ptr = allocate_in_next_plab(state, &dest_state, word_sz, context);
+      if (obj_ptr == NULL) {
+        // This will either forward-to-self, or detect that someone else has
+        // installed a forwarding pointer.
+        return _g1h->handle_evacuation_failure_par(this, old);
+      }
+    }
+  }
+
+  assert(obj_ptr != NULL, "when we get here, allocation should have succeeded");
 #ifndef PRODUCT
   // Should this evacuation fail?
   if (_g1h->evacuation_should_fail()) {
-    if (obj_ptr != NULL) {
-      _g1_par_allocator->undo_allocation(alloc_purpose, obj_ptr, word_sz, context);
-      obj_ptr = NULL;
-    }
+    // Doing this after all the allocation attempts also tests the
+    // undo_allocation() method too.
+    _g1_par_allocator->undo_allocation(dest_state, obj_ptr, word_sz, context);
+    return _g1h->handle_evacuation_failure_par(this, old);
   }
 #endif // !PRODUCT
 
-  if (obj_ptr == NULL) {
-    // This will either forward-to-self, or detect that someone else has
-    // installed a forwarding pointer.
-    return _g1h->handle_evacuation_failure_par(this, old);
-  }
-
-  oop obj = oop(obj_ptr);
-
   // We're going to allocate linearly, so might as well prefetch ahead.
   Prefetch::write(obj_ptr, PrefetchCopyIntervalInBytes);
 
-  oop forward_ptr = old->forward_to_atomic(obj);
+  const oop obj = oop(obj_ptr);
+  const oop forward_ptr = old->forward_to_atomic(obj);
   if (forward_ptr == NULL) {
     Copy::aligned_disjoint_words((HeapWord*) old, obj_ptr, word_sz);
 
-    // alloc_purpose is just a hint to allocate() above, recheck the type of region
-    // we actually allocated from and update alloc_purpose accordingly
-    HeapRegion* to_region = _g1h->heap_region_containing_raw(obj_ptr);
-    alloc_purpose = to_region->is_young() ? GCAllocForSurvived : GCAllocForTenured;
-
-    if (g1p->track_object_age(alloc_purpose)) {
+    if (dest_state.is_young()) {
       if (age < markOopDesc::max_age) {
         age++;
       }
@@ -215,13 +264,19 @@
     }
 
     if (G1StringDedup::is_enabled()) {
-      G1StringDedup::enqueue_from_evacuation(from_region->is_young(),
-                                             to_region->is_young(),
+      const bool is_from_young = state.is_young();
+      const bool is_to_young = dest_state.is_young();
+      assert(is_from_young == _g1h->heap_region_containing_raw(old)->is_young(),
+             "sanity");
+      assert(is_to_young == _g1h->heap_region_containing_raw(obj)->is_young(),
+             "sanity");
+      G1StringDedup::enqueue_from_evacuation(is_from_young,
+                                             is_to_young,
                                              queue_num(),
                                              obj);
     }
 
-    size_t* surv_young_words = surviving_young_words();
+    size_t* const surv_young_words = surviving_young_words();
     surv_young_words[young_index] += word_sz;
 
     if (obj->is_objArray() && arrayOop(obj)->length() >= ParGCArrayScanChunk) {
@@ -232,14 +287,13 @@
       oop* old_p = set_partial_array_mask(old);
       push_on_queue(old_p);
     } else {
-      // No point in using the slower heap_region_containing() method,
-      // given that we know obj is in the heap.
-      _scanner.set_region(_g1h->heap_region_containing_raw(obj));
+      HeapRegion* const to_region = _g1h->heap_region_containing_raw(obj_ptr);
+      _scanner.set_region(to_region);
       obj->oop_iterate_backwards(&_scanner);
     }
+    return obj;
   } else {
-    _g1_par_allocator->undo_allocation(alloc_purpose, obj_ptr, word_sz, context);
-    obj = forward_ptr;
+    _g1_par_allocator->undo_allocation(dest_state, obj_ptr, word_sz, context);
+    return forward_ptr;
   }
-  return obj;
 }

--- a/src/share/vm/gc_implementation/g1/g1ParScanThreadState.hpp	Tue Dec 30 13:19:30 2014 -0800
+++ b/src/share/vm/gc_implementation/g1/g1ParScanThreadState.hpp	Thu Jan 01 16:11:49 2015 -0800
@@ -46,14 +46,16 @@
   G1SATBCardTableModRefBS* _ct_bs;
   G1RemSet* _g1_rem;
 
-  G1ParGCAllocator*   _g1_par_allocator;
-
-  ageTable            _age_table;
+  G1ParGCAllocator* _g1_par_allocator;
 
-  G1ParScanClosure    _scanner;
+  ageTable          _age_table;
+  InCSetState       _dest[InCSetState::Num];
+  // Local tenuring threshold.
+  uint              _tenuring_threshold;
+  G1ParScanClosure  _scanner;
 
-  size_t           _alloc_buffer_waste;
-  size_t           _undo_waste;
+  size_t            _alloc_buffer_waste;
+  size_t            _undo_waste;
 
   OopsInHeapRegionClosure*      _evac_failure_cl;
 
@@ -82,6 +84,14 @@
   DirtyCardQueue& dirty_card_queue()             { return _dcq;  }
   G1SATBCardTableModRefBS* ctbs()                { return _ct_bs; }
 
+  InCSetState dest(InCSetState original) const {
+    assert(original.is_valid(),
+           err_msg("Original state invalid: " CSETSTATE_FORMAT, original.value()));
+    assert(_dest[original.value()].is_valid_gen(),
+           err_msg("Dest state is invalid: " CSETSTATE_FORMAT, _dest[original.value()].value()));
+    return _dest[original.value()];
+  }
+
  public:
   G1ParScanThreadState(G1CollectedHeap* g1h, uint queue_num, ReferenceProcessor* rp);
   ~G1ParScanThreadState();
@@ -112,7 +122,6 @@
       }
     }
   }
- public:
 
   void set_evac_failure_closure(OopsInHeapRegionClosure* evac_failure_cl) {
     _evac_failure_cl = evac_failure_cl;
@@ -193,9 +202,20 @@
   template <class T> inline void deal_with_reference(T* ref_to_scan);
 
   inline void dispatch_reference(StarTask ref);
+
+  // Tries to allocate word_sz in the PLAB of the next "generation" after trying to
+  // allocate into dest. State is the original (source) cset state for the object
+  // that is allocated for.
+  // Returns a non-NULL pointer if successful, and updates dest if required.
+  HeapWord* allocate_in_next_plab(InCSetState const state,
+                                  InCSetState* dest,
+                                  size_t word_sz,
+                                  AllocationContext_t const context);
+
+  inline InCSetState next_state(InCSetState const state, markOop const m, uint& age);
  public:
 
-  oop copy_to_survivor_space(oop const obj, markOop const old_mark);
+  oop copy_to_survivor_space(InCSetState const state, oop const obj, markOop const old_mark);
 
   void trim_queue();
 

--- a/src/share/vm/gc_implementation/g1/g1ParScanThreadState.inline.hpp	Tue Dec 30 13:19:30 2014 -0800
+++ b/src/share/vm/gc_implementation/g1/g1ParScanThreadState.inline.hpp	Thu Jan 01 16:11:49 2015 -0800
@@ -38,21 +38,21 @@
   // set, due to (benign) races in the claim mechanism during RSet scanning more
   // than one thread might claim the same card. So the same card may be
   // processed multiple times. So redo this check.
-  G1CollectedHeap::in_cset_state_t in_cset_state = _g1h->in_cset_state(obj);
-  if (in_cset_state == G1CollectedHeap::InCSet) {
+  const InCSetState in_cset_state = _g1h->in_cset_state(obj);
+  if (in_cset_state.is_in_cset()) {
     oop forwardee;
     markOop m = obj->mark();
     if (m->is_marked()) {
       forwardee = (oop) m->decode_pointer();
     } else {
-      forwardee = copy_to_survivor_space(obj, m);
+      forwardee = copy_to_survivor_space(in_cset_state, obj, m);
     }
     oopDesc::encode_store_heap_oop(p, forwardee);
-  } else if (in_cset_state == G1CollectedHeap::IsHumongous) {
+  } else if (in_cset_state.is_humongous()) {
     _g1h->set_humongous_is_live(obj);
   } else {
-    assert(in_cset_state == G1CollectedHeap::InNeither,
-           err_msg("In_cset_state must be InNeither here, but is %d", in_cset_state));
+    assert(!in_cset_state.is_in_cset_or_humongous(),
+           err_msg("In_cset_state must be NotInCSet here, but is " CSETSTATE_FORMAT, in_cset_state.value()));
   }
 
   assert(obj != NULL, "Must be");

--- a/src/share/vm/gc_implementation/g1/g1RemSet.cpp	Tue Dec 30 13:19:30 2014 -0800
+++ b/src/share/vm/gc_implementation/g1/g1RemSet.cpp	Thu Jan 01 16:11:49 2015 -0800
@@ -80,7 +80,7 @@
     _prev_period_summary()
 {
   _seq_task = new SubTasksDone(NumSeqTasks);
-  _cset_rs_update_cl = NEW_C_HEAP_ARRAY(OopsInHeapRegionClosure*, n_workers(), mtGC);
+  _cset_rs_update_cl = NEW_C_HEAP_ARRAY(G1ParPushHeapRSClosure*, n_workers(), mtGC);
   for (uint i = 0; i < n_workers(); i++) {
     _cset_rs_update_cl[i] = NULL;
   }
@@ -94,14 +94,14 @@
   for (uint i = 0; i < n_workers(); i++) {
     assert(_cset_rs_update_cl[i] == NULL, "it should be");
   }
-  FREE_C_HEAP_ARRAY(OopsInHeapRegionClosure*, _cset_rs_update_cl);
+  FREE_C_HEAP_ARRAY(G1ParPushHeapRSClosure*, _cset_rs_update_cl);
 }
 
 class ScanRSClosure : public HeapRegionClosure {
   size_t _cards_done, _cards;
   G1CollectedHeap* _g1h;
 
-  OopsInHeapRegionClosure* _oc;
+  G1ParPushHeapRSClosure* _oc;
   CodeBlobClosure* _code_root_cl;
 
   G1BlockOffsetSharedArray* _bot_shared;
@@ -113,7 +113,7 @@
   bool   _try_claimed;
 
 public:
-  ScanRSClosure(OopsInHeapRegionClosure* oc,
+  ScanRSClosure(G1ParPushHeapRSClosure* oc,
                 CodeBlobClosure* code_root_cl,
                 uint worker_i) :
     _oc(oc),
@@ -135,8 +135,7 @@
   void scanCard(size_t index, HeapRegion *r) {
     // Stack allocate the DirtyCardToOopClosure instance
     HeapRegionDCTOC cl(_g1h, r, _oc,
-                       CardTableModRefBS::Precise,
-                       HeapRegionDCTOC::IntoCSFilterKind);
+                       CardTableModRefBS::Precise);
 
     // Set the "from" region in the closure.
     _oc->set_region(r);
@@ -231,7 +230,7 @@
   size_t cards_looked_up() { return _cards;}
 };
 
-void G1RemSet::scanRS(OopsInHeapRegionClosure* oc,
+void G1RemSet::scanRS(G1ParPushHeapRSClosure* oc,
                       CodeBlobClosure* code_root_cl,
                       uint worker_i) {
   double rs_time_start = os::elapsedTime();
@@ -301,7 +300,7 @@
   HeapRegionRemSet::cleanup();
 }
 
-void G1RemSet::oops_into_collection_set_do(OopsInHeapRegionClosure* oc,
+void G1RemSet::oops_into_collection_set_do(G1ParPushHeapRSClosure* oc,
                                            CodeBlobClosure* code_root_cl,
                                            uint worker_i) {
 #if CARD_REPEAT_HISTO
@@ -417,7 +416,7 @@
 G1UpdateRSOrPushRefOopClosure::
 G1UpdateRSOrPushRefOopClosure(G1CollectedHeap* g1h,
                               G1RemSet* rs,
-                              OopsInHeapRegionClosure* push_ref_cl,
+                              G1ParPushHeapRSClosure* push_ref_cl,
                               bool record_refs_into_cset,
                               uint worker_i) :
   _g1(g1h), _g1_rem_set(rs), _from(NULL),
@@ -518,7 +517,7 @@
   ct_freq_note_card(_ct_bs->index_for(start));
 #endif
 
-  OopsInHeapRegionClosure* oops_in_heap_closure = NULL;
+  G1ParPushHeapRSClosure* oops_in_heap_closure = NULL;
   if (check_for_refs_into_cset) {
     // ConcurrentG1RefineThreads have worker numbers larger than what
     // _cset_rs_update_cl[] is set up to handle. But those threads should