Mergehs24-b02

43 files changed, 877 insertions, 209 deletions

diff --git a/make/bsd/makefiles/buildtree.make b/make/bsd/makefiles/buildtree.make
index 960a382bb..ccc812775 100644
--- a/make/bsd/makefiles/buildtree.make
+++ b/make/bsd/makefiles/buildtree.make
@@ -58,6 +58,7 @@
 # needs to be set here since this Makefile doesn't include defs.make
 OS_VENDOR:=$(shell uname -s)
 
+-include $(SPEC)
 include $(GAMMADIR)/make/scm.make
 include $(GAMMADIR)/make/altsrc.make
 
@@ -247,6 +248,8 @@ flags.make: $(BUILDTREE_MAKE) ../shared_dirs.lst
 	    echo "HOTSPOT_EXTRA_SYSDEFS\$$(HOTSPOT_EXTRA_SYSDEFS) = $(HOTSPOT_EXTRA_SYSDEFS)" && \
 	    echo "SYSDEFS += \$$(HOTSPOT_EXTRA_SYSDEFS)"; \
 	echo; \
+	[ -n "$(SPEC)" ] && \
+	    echo "include $(SPEC)"; \
 	echo "include \$$(GAMMADIR)/make/$(OS_FAMILY)/makefiles/$(VARIANT).make"; \
 	echo "include \$$(GAMMADIR)/make/$(OS_FAMILY)/makefiles/$(COMPILER).make"; \
 	) > $@
diff --git a/make/bsd/makefiles/gcc.make b/make/bsd/makefiles/gcc.make
index d7a6422eb..a73397bf0 100644
--- a/make/bsd/makefiles/gcc.make
+++ b/make/bsd/makefiles/gcc.make
@@ -27,53 +27,57 @@ OS_VENDOR = $(shell uname -s)
 #------------------------------------------------------------------------
 # CC, CXX & AS
 
-# When cross-compiling the ALT_COMPILER_PATH points
-# to the cross-compilation toolset
-ifdef CROSS_COMPILE_ARCH
- CXX = $(ALT_COMPILER_PATH)/g++
- CC  = $(ALT_COMPILER_PATH)/gcc
- HOSTCXX = g++
- HOSTCC  = gcc
-else ifneq ($(OS_VENDOR), Darwin)
- CXX = g++
- CC  = gcc
- HOSTCXX = $(CXX)
- HOSTCC  = $(CC)
-endif
-
-# i486 hotspot requires -mstackrealign on Darwin.
-# llvm-gcc supports this in Xcode 3.2.6 and 4.0.
-# gcc-4.0 supports this on earlier versions.
-# Prefer llvm-gcc where available.
-ifeq ($(OS_VENDOR), Darwin)
-  ifeq ($(origin CXX), default)
-   CXX = llvm-g++
-  endif
-  ifeq ($(origin CC), default)
-   CC  = llvm-gcc
+# If a SPEC is not set already, then use these defaults.
+ifeq ($(SPEC),)
+  # When cross-compiling the ALT_COMPILER_PATH points
+  # to the cross-compilation toolset
+  ifdef CROSS_COMPILE_ARCH
+    CXX = $(ALT_COMPILER_PATH)/g++
+    CC  = $(ALT_COMPILER_PATH)/gcc
+    HOSTCXX = g++
+    HOSTCC  = gcc
+  else ifneq ($(OS_VENDOR), Darwin)
+    CXX = g++
+    CC  = gcc
+    HOSTCXX = $(CXX)
+    HOSTCC  = $(CC)
   endif
 
-  ifeq ($(ARCH), i486)
-  LLVM_SUPPORTS_STACKREALIGN := $(shell \
-   [ "0"`llvm-gcc -v 2>&1 | grep LLVM | sed -E "s/.*LLVM build ([0-9]+).*/\1/"` -gt "2333" ] \
-   && echo true || echo false)
-
-  ifeq ($(LLVM_SUPPORTS_STACKREALIGN), true)
-    CXX32 ?= llvm-g++
-    CC32  ?= llvm-gcc
-  else
-    CXX32 ?= g++-4.0
-    CC32  ?= gcc-4.0
-  endif
-  CXX = $(CXX32)
-  CC  = $(CC32)
+  # i486 hotspot requires -mstackrealign on Darwin.
+  # llvm-gcc supports this in Xcode 3.2.6 and 4.0.
+  # gcc-4.0 supports this on earlier versions.
+  # Prefer llvm-gcc where available.
+  ifeq ($(OS_VENDOR), Darwin)
+    ifeq ($(origin CXX), default)
+      CXX = llvm-g++
+    endif
+    ifeq ($(origin CC), default)
+      CC  = llvm-gcc
+    endif
+
+    ifeq ($(ARCH), i486)
+      LLVM_SUPPORTS_STACKREALIGN := $(shell \
+       [ "0"`llvm-gcc -v 2>&1 | grep LLVM | sed -E "s/.*LLVM build ([0-9]+).*/\1/"` -gt "2333" ] \
+       && echo true || echo false)
+
+      ifeq ($(LLVM_SUPPORTS_STACKREALIGN), true)
+        CXX32 ?= llvm-g++
+        CC32  ?= llvm-gcc
+      else
+        CXX32 ?= g++-4.0
+        CC32  ?= gcc-4.0
+      endif
+      CXX = $(CXX32)
+      CC  = $(CC32)
+    endif
+
+    HOSTCXX = $(CXX)
+    HOSTCC  = $(CC)
   endif
 
-  HOSTCXX = $(CXX)
-  HOSTCC  = $(CC)
+  AS   = $(CC) -c -x assembler-with-cpp
 endif
 
-AS   = $(CC) -c -x assembler-with-cpp
 
 # -dumpversion in gcc-2.91 shows "egcs-2.91.66". In later version, it only
 # prints the numbers (e.g. "2.95", "3.2.1")
diff --git a/make/bsd/makefiles/sparcWorks.make b/make/bsd/makefiles/sparcWorks.make
index 81de9c33b..c87f50440 100644
--- a/make/bsd/makefiles/sparcWorks.make
+++ b/make/bsd/makefiles/sparcWorks.make
@@ -25,12 +25,15 @@
 #------------------------------------------------------------------------
 # CC, CXX & AS
 
-CXX = CC
-CC  = cc
-AS  = $(CC) -c
-
-HOSTCXX = $(CXX)
-HOSTCC  = $(CC)
+# If a SPEC is not set already, then use these defaults.
+ifeq ($(SPEC),)
+  CXX = CC
+  CC  = cc
+  AS  = $(CC) -c
+
+  HOSTCXX = $(CXX)
+  HOSTCC  = $(CC)
+endif
 
 ARCHFLAG = $(ARCHFLAG/$(BUILDARCH))
 ARCHFLAG/i486    = -m32
diff --git a/make/defs.make b/make/defs.make
index a0aa0e5c8..ac7268a96 100644
--- a/make/defs.make
+++ b/make/defs.make
@@ -24,6 +24,11 @@
 
 # The common definitions for hotspot builds.
 
+# Optionally include SPEC file generated by configure.
+ifneq ($(SPEC),)
+  include $(SPEC)
+endif
+
 # Default to verbose build logs (show all compile lines):
 MAKE_VERBOSE=y
 
diff --git a/make/hotspot_version b/make/hotspot_version
index f67badd6d..6cc9716c9 100644
--- a/make/hotspot_version
+++ b/make/hotspot_version
@@ -35,7 +35,7 @@ HOTSPOT_VM_COPYRIGHT=Copyright 2011
 
 HS_MAJOR_VER=24
 HS_MINOR_VER=0
-HS_BUILD_NUMBER=01
+HS_BUILD_NUMBER=02
 
 JDK_MAJOR_VER=1
 JDK_MINOR_VER=8
diff --git a/make/linux/makefiles/buildtree.make b/make/linux/makefiles/buildtree.make
index 5bc29368e..6af5490da 100644
--- a/make/linux/makefiles/buildtree.make
+++ b/make/linux/makefiles/buildtree.make
@@ -55,6 +55,7 @@
 # The makefiles are split this way so that "make foo" will run faster by not
 # having to read the dependency files for the vm.
 
+-include $(SPEC)
 include $(GAMMADIR)/make/scm.make
 include $(GAMMADIR)/make/altsrc.make
 
@@ -244,6 +245,8 @@ flags.make: $(BUILDTREE_MAKE) ../shared_dirs.lst
 	    echo "HOTSPOT_EXTRA_SYSDEFS\$$(HOTSPOT_EXTRA_SYSDEFS) = $(HOTSPOT_EXTRA_SYSDEFS)" && \
 	    echo "SYSDEFS += \$$(HOTSPOT_EXTRA_SYSDEFS)"; \
 	echo; \
+	[ -n "$(SPEC)" ] && \
+	    echo "include $(SPEC)"; \
 	echo "include \$$(GAMMADIR)/make/$(OS_FAMILY)/makefiles/$(VARIANT).make"; \
 	echo "include \$$(GAMMADIR)/make/$(OS_FAMILY)/makefiles/$(COMPILER).make"; \
 	) > $@
diff --git a/make/linux/makefiles/gcc.make b/make/linux/makefiles/gcc.make
index 0188ed772..d918ef77b 100644
--- a/make/linux/makefiles/gcc.make
+++ b/make/linux/makefiles/gcc.make
@@ -25,21 +25,26 @@
 #------------------------------------------------------------------------
 # CC, CXX & AS
 
-# When cross-compiling the ALT_COMPILER_PATH points
-# to the cross-compilation toolset
-ifdef CROSS_COMPILE_ARCH
-CXX = $(ALT_COMPILER_PATH)/g++
-CC  = $(ALT_COMPILER_PATH)/gcc
-HOSTCXX = g++
-HOSTCC  = gcc
-else
-CXX = g++
-CC  = gcc
-HOSTCXX = $(CXX)
-HOSTCC  = $(CC)
+# If a SPEC is not set already, then use these defaults.
+ifeq ($(SPEC),)
+  # When cross-compiling the ALT_COMPILER_PATH points
+  # to the cross-compilation toolset
+  ifdef CROSS_COMPILE_ARCH
+    CXX = $(ALT_COMPILER_PATH)/g++
+    CC  = $(ALT_COMPILER_PATH)/gcc
+    HOSTCXX = g++
+    HOSTCC  = gcc
+    STRIP = $(ALT_COMPILER_PATH)/strip
+  else
+    CXX = g++
+    CC  = gcc
+    HOSTCXX = $(CXX)
+    HOSTCC  = $(CC)
+    STRIP = strip
+  endif
+  AS  = $(CC) -c
 endif
 
-AS  = $(CC) -c
 
 # -dumpversion in gcc-2.91 shows "egcs-2.91.66". In later version, it only
 # prints the numbers (e.g. "2.95", "3.2.1")
@@ -261,9 +266,3 @@ endif
 ifdef MINIMIZE_RAM_USAGE
 CFLAGS += -DMINIMIZE_RAM_USAGE
 endif
-
-ifdef CROSS_COMPILE_ARCH
-  STRIP = $(ALT_COMPILER_PATH)/strip
-else
-  STRIP = strip
-endif
diff --git a/make/linux/makefiles/sparcWorks.make b/make/linux/makefiles/sparcWorks.make
index 81de9c33b..c87f50440 100644
--- a/make/linux/makefiles/sparcWorks.make
+++ b/make/linux/makefiles/sparcWorks.make
@@ -25,12 +25,15 @@
 #------------------------------------------------------------------------
 # CC, CXX & AS
 
-CXX = CC
-CC  = cc
-AS  = $(CC) -c
-
-HOSTCXX = $(CXX)
-HOSTCC  = $(CC)
+# If a SPEC is not set already, then use these defaults.
+ifeq ($(SPEC),)
+  CXX = CC
+  CC  = cc
+  AS  = $(CC) -c
+
+  HOSTCXX = $(CXX)
+  HOSTCC  = $(CC)
+endif
 
 ARCHFLAG = $(ARCHFLAG/$(BUILDARCH))
 ARCHFLAG/i486    = -m32
diff --git a/make/solaris/makefiles/buildtree.make b/make/solaris/makefiles/buildtree.make
index 129a48494..901100fb7 100644
--- a/make/solaris/makefiles/buildtree.make
+++ b/make/solaris/makefiles/buildtree.make
@@ -55,6 +55,7 @@
 # The makefiles are split this way so that "make foo" will run faster by not
 # having to read the dependency files for the vm.
 
+-include $(SPEC)
 include $(GAMMADIR)/make/scm.make
 include $(GAMMADIR)/make/altsrc.make
 
@@ -237,6 +238,8 @@ flags.make: $(BUILDTREE_MAKE) ../shared_dirs.lst
 	    echo "HOTSPOT_EXTRA_SYSDEFS\$$(HOTSPOT_EXTRA_SYSDEFS) = $(HOTSPOT_EXTRA_SYSDEFS)" && \
 	    echo "SYSDEFS += \$$(HOTSPOT_EXTRA_SYSDEFS)"; \
 	echo; \
+	[ -n "$(SPEC)" ] && \
+	    echo "include $(SPEC)"; \
 	echo "include \$$(GAMMADIR)/make/$(OS_FAMILY)/makefiles/$(VARIANT).make"; \
 	echo "include \$$(GAMMADIR)/make/$(OS_FAMILY)/makefiles/$(COMPILER).make"; \
 	) > $@
diff --git a/make/solaris/makefiles/gcc.make b/make/solaris/makefiles/gcc.make
index d4c54abd2..4d0db9e7a 100644
--- a/make/solaris/makefiles/gcc.make
+++ b/make/solaris/makefiles/gcc.make
@@ -25,9 +25,13 @@
 #------------------------------------------------------------------------
 # CC, CXX & AS
 
-CXX = g++
-CC  = gcc
-AS  = $(CC) -c
+# If a SPEC is not set already, then use these defaults.
+ifeq ($(SPEC),)
+  CXX = g++
+  CC  = gcc
+  AS  = $(CC) -c
+  MCS = /usr/ccs/bin/mcs
+endif
 
 Compiler = gcc
 
@@ -193,5 +197,3 @@ DEBUG_CFLAGS += $(DEBUG_CFLAGS/$(BUILDARCH))
 ifeq ($(DEBUG_CFLAGS/$(BUILDARCH)),) 
 DEBUG_CFLAGS += -gstabs 
 endif 
-
-MCS = /usr/ccs/bin/mcs
diff --git a/make/solaris/makefiles/sparcWorks.make b/make/solaris/makefiles/sparcWorks.make
index 8baf3c1d4..8ef47c563 100644
--- a/make/solaris/makefiles/sparcWorks.make
+++ b/make/solaris/makefiles/sparcWorks.make
@@ -22,18 +22,22 @@
 #  
 #
 
-# Compiler-specific flags for sparcworks.
+# If a SPEC is not set already, then use these defaults.
+ifeq ($(SPEC),)
+  # Compiler-specific flags for sparcworks.
+  CC	= cc
+  CXX	= CC
 
-# tell make which C and C++ compilers to use
-CC	= cc
-CXX	= CC
+  # Note that this 'as' is an older version of the Sun Studio 'fbe', and will
+  #   use the older style options. The 'fbe' options will match 'cc' and 'CC'.
+  AS	= /usr/ccs/bin/as
 
-# Note that this 'as' is an older version of the Sun Studio 'fbe', and will
-#   use the older style options. The 'fbe' options will match 'cc' and 'CC'.
-AS	= /usr/ccs/bin/as
+  NM    = /usr/ccs/bin/nm
+  NAWK  = /bin/nawk
 
-NM	= /usr/ccs/bin/nm
-NAWK    = /bin/nawk
+  MCS	= /usr/ccs/bin/mcs
+  STRIP	= /usr/ccs/bin/strip
+endif
 
 REORDER_FLAG = -xF
 
@@ -557,9 +561,6 @@ else
 #LINK_INTO = LIBJVM
 endif
 
-MCS	= /usr/ccs/bin/mcs
-STRIP	= /usr/ccs/bin/strip
-
 # Solaris platforms collect lots of redundant file-ident lines,
 # to the point of wasting a significant percentage of file space.
 # (The text is stored in ELF .comment sections, contributed by
diff --git a/make/windows/build.make b/make/windows/build.make
index 8ec004d49..fff8a8530 100644
--- a/make/windows/build.make
+++ b/make/windows/build.make
@@ -297,6 +297,10 @@ $(variantDir)\local.make: checks
 	@ echo BUILDARCH=$(BUILDARCH)         			>> $@
 	@ echo Platform_arch=$(Platform_arch)        		>> $@
 	@ echo Platform_arch_model=$(Platform_arch_model)	>> $@
+	@ echo CXX=$(CXX)					>> $@
+	@ echo LD=$(LD)						>> $@
+	@ echo MT=$(MT)						>> $@
+	@ echo RC=$(RC)						>> $@
 	@ sh $(WorkSpace)/make/windows/get_msc_ver.sh		>> $@
 
 checks: checkVariant checkWorkSpace checkSA
diff --git a/make/windows/makefiles/compile.make b/make/windows/makefiles/compile.make
index 1c9855c15..e1848a4ac 100644
--- a/make/windows/makefiles/compile.make
+++ b/make/windows/makefiles/compile.make
@@ -23,7 +23,9 @@
 #
 
 # Generic compiler settings
+!if "x$(CXX)" == "x"
 CXX=cl.exe
+!endif
 
 # CXX Flags: (these vary slightly from VC6->VS2003->VS2005 compilers)
 #   /nologo   Supress copyright message at every cl.exe startup
@@ -183,8 +185,10 @@ BUFFEROVERFLOWLIB = bufferoverflowU.lib
 LD_FLAGS = /manifest $(LD_FLAGS) $(BUFFEROVERFLOWLIB)
 # Manifest Tool - used in VS2005 and later to adjust manifests stored
 # as resources inside build artifacts.
+!if "x$(MT)" == "x"
 MT=mt.exe
 !endif
+!endif
 
 !if "$(COMPILER_NAME)" == "VS2008"
 PRODUCT_OPT_OPTION   = /O2 /Oy-
@@ -194,8 +198,10 @@ GX_OPTION = /EHsc
 LD_FLAGS = /manifest $(LD_FLAGS)
 # Manifest Tool - used in VS2005 and later to adjust manifests stored
 # as resources inside build artifacts.
+!if "x$(MT)" == "x"
 MT=mt.exe
 !endif
+!endif
 
 !if "$(COMPILER_NAME)" == "VS2010"
 PRODUCT_OPT_OPTION   = /O2 /Oy-
@@ -205,7 +211,9 @@ GX_OPTION = /EHsc
 LD_FLAGS = /manifest $(LD_FLAGS)
 # Manifest Tool - used in VS2005 and later to adjust manifests stored
 # as resources inside build artifacts.
+!if "x$(MT)" == "x"
 MT=mt.exe
+!endif
 !if "$(BUILDARCH)" == "i486"
 LD_FLAGS = /SAFESEH $(LD_FLAGS)
 !endif
@@ -225,7 +233,9 @@ FASTDEBUG_OPT_OPTION = $(DEBUG_OPT_OPTION)
 !endif
 
 # Generic linker settings
+!if "x$(LD)" == "x"
 LD=link.exe
+!endif
 LD_FLAGS= $(LD_FLAGS) kernel32.lib user32.lib gdi32.lib winspool.lib \
  comdlg32.lib advapi32.lib shell32.lib ole32.lib oleaut32.lib \
  uuid.lib Wsock32.lib winmm.lib /nologo /machine:$(MACHINE) /opt:REF \
@@ -237,7 +247,9 @@ LD_FLAGS= $(LD_FLAGS) psapi.lib
 !endif
 
 # Resource compiler settings
+!if "x$(RC)" == "x"
 RC=rc.exe
+!endif
 RC_FLAGS=/D "HS_VER=$(HS_VER)" \
 	 /D "HS_DOTVER=$(HS_DOTVER)" \
 	 /D "HS_BUILD_ID=$(HS_BUILD_ID)" \
diff --git a/make/windows/makefiles/defs.make b/make/windows/makefiles/defs.make
index c1be7d2c8..993799e37 100644
--- a/make/windows/makefiles/defs.make
+++ b/make/windows/makefiles/defs.make
@@ -202,3 +202,19 @@ ifeq ($(BUILD_WIN_SA), 1)
   # Must pass this down to nmake.
   MAKE_ARGS += BUILD_WIN_SA=1
 endif
+
+# Propagate compiler and tools paths from configure to nmake. 
+# Need to make sure they contain \\ and not /.
+ifneq ($(SPEC),)
+  ifeq ($(USING_CYGWIN), true)
+    MAKE_ARGS += CXX="$(subst /,\\,$(shell /bin/cygpath -s -m -a $(CXX)))"
+    MAKE_ARGS += LD="$(subst /,\\,$(shell /bin/cygpath -s -m -a $(LD)))"
+    MAKE_ARGS += RC="$(subst /,\\,$(shell /bin/cygpath -s -m -a $(RC)))"
+    MAKE_ARGS += MT="$(subst /,\\,$(shell /bin/cygpath -s -m -a $(MT)))"
+  else
+    MAKE_ARGS += CXX="$(subst /,\\,$(CXX))"
+    MAKE_ARGS += LD="$(subst /,\\,$(LD))"
+    MAKE_ARGS += RC="$(subst /,\\,$(RC))"
+    MAKE_ARGS += MT="$(subst /,\\,$(MT))"
+  endif
+endif
diff --git a/src/cpu/x86/vm/c1_Runtime1_x86.cpp b/src/cpu/x86/vm/c1_Runtime1_x86.cpp
index 5f2cf3886..30df6087d 100644
--- a/src/cpu/x86/vm/c1_Runtime1_x86.cpp
+++ b/src/cpu/x86/vm/c1_Runtime1_x86.cpp
@@ -47,6 +47,12 @@ int StubAssembler::call_RT(Register oop_result1, Register oop_result2, address e
   assert(!(oop_result1->is_valid() || oop_result2->is_valid()) || oop_result1 != oop_result2, "registers must be different");
   assert(oop_result1 != thread && oop_result2 != thread, "registers must be different");
   assert(args_size >= 0, "illegal args_size");
+  bool align_stack = false;
+#ifdef _LP64
+  // At a method handle call, the stack may not be properly aligned
+  // when returning with an exception.
+  align_stack = (stub_id() == Runtime1::handle_exception_from_callee_id);
+#endif
 
 #ifdef _LP64
   mov(c_rarg0, thread);
@@ -59,11 +65,21 @@ int StubAssembler::call_RT(Register oop_result1, Register oop_result2, address e
   push(thread);
 #endif // _LP64
 
-  set_last_Java_frame(thread, noreg, rbp, NULL);
+  int call_offset;
+  if (!align_stack) {
+    set_last_Java_frame(thread, noreg, rbp, NULL);
+  } else {
+    address the_pc = pc();
+    call_offset = offset();
+    set_last_Java_frame(thread, noreg, rbp, the_pc);
+    andptr(rsp, -(StackAlignmentInBytes));    // Align stack
+  }
 
   // do the call
   call(RuntimeAddress(entry));
-  int call_offset = offset();
+  if (!align_stack) {
+    call_offset = offset();
+  }
   // verify callee-saved register
 #ifdef ASSERT
   guarantee(thread != rax, "change this code");
@@ -78,7 +94,7 @@ int StubAssembler::call_RT(Register oop_result1, Register oop_result2, address e
   }
   pop(rax);
 #endif
-  reset_last_Java_frame(thread, true, false);
+  reset_last_Java_frame(thread, true, align_stack);
 
   // discard thread and arguments
   NOT_LP64(addptr(rsp, num_rt_args()*BytesPerWord));
diff --git a/src/cpu/x86/vm/sharedRuntime_x86_64.cpp b/src/cpu/x86/vm/sharedRuntime_x86_64.cpp
index b7af4544d..68a23939d 100644
--- a/src/cpu/x86/vm/sharedRuntime_x86_64.cpp
+++ b/src/cpu/x86/vm/sharedRuntime_x86_64.cpp
@@ -1181,14 +1181,13 @@ static void save_or_restore_arguments(MacroAssembler* masm,
                                       BasicType* in_sig_bt) {
   // if map is non-NULL then the code should store the values,
   // otherwise it should load them.
-  int handle_index = 0;
+  int slot = arg_save_area;
   // Save down double word first
   for ( int i = 0; i < total_in_args; i++) {
     if (in_regs[i].first()->is_XMMRegister() && in_sig_bt[i] == T_DOUBLE) {
-      int slot = handle_index * VMRegImpl::slots_per_word + arg_save_area;
       int offset = slot * VMRegImpl::stack_slot_size;
-      handle_index += 2;
-      assert(handle_index <= stack_slots, "overflow");
+      slot += VMRegImpl::slots_per_word;
+      assert(slot <= stack_slots, "overflow");
       if (map != NULL) {
         __ movdbl(Address(rsp, offset), in_regs[i].first()->as_XMMRegister());
       } else {
@@ -1197,10 +1196,7 @@ static void save_or_restore_arguments(MacroAssembler* masm,
     }
     if (in_regs[i].first()->is_Register() &&
         (in_sig_bt[i] == T_LONG || in_sig_bt[i] == T_ARRAY)) {
-      int slot = handle_index * VMRegImpl::slots_per_word + arg_save_area;
       int offset = slot * VMRegImpl::stack_slot_size;
-      handle_index += 2;
-      assert(handle_index <= stack_slots, "overflow");
       if (map != NULL) {
         __ movq(Address(rsp, offset), in_regs[i].first()->as_Register());
         if (in_sig_bt[i] == T_ARRAY) {
@@ -1209,14 +1205,15 @@ static void save_or_restore_arguments(MacroAssembler* masm,
       } else {
         __ movq(in_regs[i].first()->as_Register(), Address(rsp, offset));
       }
+      slot += VMRegImpl::slots_per_word;
     }
   }
   // Save or restore single word registers
   for ( int i = 0; i < total_in_args; i++) {
     if (in_regs[i].first()->is_Register()) {
-      int slot = handle_index++ * VMRegImpl::slots_per_word + arg_save_area;
       int offset = slot * VMRegImpl::stack_slot_size;
-      assert(handle_index <= stack_slots, "overflow");
+      slot++;
+      assert(slot <= stack_slots, "overflow");
 
       // Value is in an input register pass we must flush it to the stack
       const Register reg = in_regs[i].first()->as_Register();
@@ -1241,9 +1238,9 @@ static void save_or_restore_arguments(MacroAssembler* masm,
       }
     } else if (in_regs[i].first()->is_XMMRegister()) {
       if (in_sig_bt[i] == T_FLOAT) {
-        int slot = handle_index++ * VMRegImpl::slots_per_word + arg_save_area;
         int offset = slot * VMRegImpl::stack_slot_size;
-        assert(handle_index <= stack_slots, "overflow");
+        slot++;
+        assert(slot <= stack_slots, "overflow");
         if (map != NULL) {
           __ movflt(Address(rsp, offset), in_regs[i].first()->as_XMMRegister());
         } else {
@@ -1368,6 +1365,174 @@ static void unpack_array_argument(MacroAssembler* masm, VMRegPair reg, BasicType
   __ bind(done);
 }
 
+
+class ComputeMoveOrder: public StackObj {
+  class MoveOperation: public ResourceObj {
+    friend class ComputeMoveOrder;
+   private:
+    VMRegPair        _src;
+    VMRegPair        _dst;
+    int              _src_index;
+    int              _dst_index;
+    bool             _processed;
+    MoveOperation*  _next;
+    MoveOperation*  _prev;
+
+    static int get_id(VMRegPair r) {
+      return r.first()->value();
+    }
+
+   public:
+    MoveOperation(int src_index, VMRegPair src, int dst_index, VMRegPair dst):
+      _src(src)
+    , _src_index(src_index)
+    , _dst(dst)
+    , _dst_index(dst_index)
+    , _next(NULL)
+    , _prev(NULL)
+    , _processed(false) {
+    }
+
+    VMRegPair src() const              { return _src; }
+    int src_id() const                 { return get_id(src()); }
+    int src_index() const              { return _src_index; }
+    VMRegPair dst() const              { return _dst; }
+    void set_dst(int i, VMRegPair dst) { _dst_index = i, _dst = dst; }
+    int dst_index() const              { return _dst_index; }
+    int dst_id() const                 { return get_id(dst()); }
+    MoveOperation* next() const       { return _next; }
+    MoveOperation* prev() const       { return _prev; }
+    void set_processed()               { _processed = true; }
+    bool is_processed() const          { return _processed; }
+
+    // insert
+    void break_cycle(VMRegPair temp_register) {
+      // create a new store following the last store
+      // to move from the temp_register to the original
+      MoveOperation* new_store = new MoveOperation(-1, temp_register, dst_index(), dst());
+
+      // break the cycle of links and insert new_store at the end
+      // break the reverse link.
+      MoveOperation* p = prev();
+      assert(p->next() == this, "must be");
+      _prev = NULL;
+      p->_next = new_store;
+      new_store->_prev = p;
+
+      // change the original store to save it's value in the temp.
+      set_dst(-1, temp_register);
+    }
+
+    void link(GrowableArray<MoveOperation*>& killer) {
+      // link this store in front the store that it depends on
+      MoveOperation* n = killer.at_grow(src_id(), NULL);
+      if (n != NULL) {
+        assert(_next == NULL && n->_prev == NULL, "shouldn't have been set yet");
+        _next = n;
+        n->_prev = this;
+      }
+    }
+  };
+
+ private:
+  GrowableArray<MoveOperation*> edges;
+
+ public:
+  ComputeMoveOrder(int total_in_args, VMRegPair* in_regs, int total_c_args, VMRegPair* out_regs,
+                    BasicType* in_sig_bt, GrowableArray<int>& arg_order, VMRegPair tmp_vmreg) {
+    // Move operations where the dest is the stack can all be
+    // scheduled first since they can't interfere with the other moves.
+    for (int i = total_in_args - 1, c_arg = total_c_args - 1; i >= 0; i--, c_arg--) {
+      if (in_sig_bt[i] == T_ARRAY) {
+        c_arg--;
+        if (out_regs[c_arg].first()->is_stack() &&
+            out_regs[c_arg + 1].first()->is_stack()) {
+          arg_order.push(i);
+          arg_order.push(c_arg);
+        } else {
+          if (out_regs[c_arg].first()->is_stack() ||
+              in_regs[i].first() == out_regs[c_arg].first()) {
+            add_edge(i, in_regs[i].first(), c_arg, out_regs[c_arg + 1]);
+          } else {
+            add_edge(i, in_regs[i].first(), c_arg, out_regs[c_arg]);
+          }
+        }
+      } else if (in_sig_bt[i] == T_VOID) {
+        arg_order.push(i);
+        arg_order.push(c_arg);
+      } else {
+        if (out_regs[c_arg].first()->is_stack() ||
+            in_regs[i].first() == out_regs[c_arg].first()) {
+          arg_order.push(i);
+          arg_order.push(c_arg);
+        } else {
+          add_edge(i, in_regs[i].first(), c_arg, out_regs[c_arg]);
+        }
+      }
+    }
+    // Break any cycles in the register moves and emit the in the
+    // proper order.
+    GrowableArray<MoveOperation*>* stores = get_store_order(tmp_vmreg);
+    for (int i = 0; i < stores->length(); i++) {
+      arg_order.push(stores->at(i)->src_index());
+      arg_order.push(stores->at(i)->dst_index());
+    }
+ }
+
+  // Collected all the move operations
+  void add_edge(int src_index, VMRegPair src, int dst_index, VMRegPair dst) {
+    if (src.first() == dst.first()) return;
+    edges.append(new MoveOperation(src_index, src, dst_index, dst));
+  }
+
+  // Walk the edges breaking cycles between moves.  The result list
+  // can be walked in order to produce the proper set of loads
+  GrowableArray<MoveOperation*>* get_store_order(VMRegPair temp_register) {
+    // Record which moves kill which values
+    GrowableArray<MoveOperation*> killer;
+    for (int i = 0; i < edges.length(); i++) {
+      MoveOperation* s = edges.at(i);
+      assert(killer.at_grow(s->dst_id(), NULL) == NULL, "only one killer");
+      killer.at_put_grow(s->dst_id(), s, NULL);
+    }
+    assert(killer.at_grow(MoveOperation::get_id(temp_register), NULL) == NULL,
+           "make sure temp isn't in the registers that are killed");
+
+    // create links between loads and stores
+    for (int i = 0; i < edges.length(); i++) {
+      edges.at(i)->link(killer);
+    }
+
+    // at this point, all the move operations are chained together
+    // in a doubly linked list.  Processing it backwards finds
+    // the beginning of the chain, forwards finds the end.  If there's
+    // a cycle it can be broken at any point,  so pick an edge and walk
+    // backward until the list ends or we end where we started.
+    GrowableArray<MoveOperation*>* stores = new GrowableArray<MoveOperation*>();
+    for (int e = 0; e < edges.length(); e++) {
+      MoveOperation* s = edges.at(e);
+      if (!s->is_processed()) {
+        MoveOperation* start = s;
+        // search for the beginning of the chain or cycle
+        while (start->prev() != NULL && start->prev() != s) {
+          start = start->prev();
+        }
+        if (start->prev() == s) {
+          start->break_cycle(temp_register);
+        }
+        // walk the chain forward inserting to store list
+        while (start != NULL) {
+          stores->append(start);
+          start->set_processed();
+          start = start->next();
+        }
+      }
+    }
+    return stores;
+  }
+};
+
+
 // ---------------------------------------------------------------------------
 // Generate a native wrapper for a given method.  The method takes arguments
 // in the Java compiled code convention, marshals them to the native
@@ -1488,12 +1653,12 @@ nmethod *SharedRuntime::generate_native_wrapper(MacroAssembler *masm,
       if (in_regs[i].first()->is_Register()) {
         const Register reg = in_regs[i].first()->as_Register();
         switch (in_sig_bt[i]) {
-          case T_ARRAY:
           case T_BOOLEAN:
           case T_BYTE:
           case T_SHORT:
           case T_CHAR:
           case T_INT:  single_slots++; break;
+          case T_ARRAY:
           case T_LONG: double_slots++; break;
           default:  ShouldNotReachHere();
         }
@@ -1690,36 +1855,43 @@ nmethod *SharedRuntime::generate_native_wrapper(MacroAssembler *masm,
 
 #endif /* ASSERT */
 
-  if (is_critical_native) {
-    // The mapping of Java and C arguments passed in registers are
-    // rotated by one, which helps when passing arguments to regular
-    // Java method but for critical natives that creates a cycle which
-    // can cause arguments to be killed before they are used.  Break
-    // the cycle by moving the first argument into a temporary
-    // register.
-    for (int i = 0; i < total_c_args; i++) {
-      if (in_regs[i].first()->is_Register() &&
-          in_regs[i].first()->as_Register() == rdi) {
-        __ mov(rbx, rdi);
-        in_regs[i].set1(rbx->as_VMReg());
-      }
-    }
-  }
-
   // This may iterate in two different directions depending on the
   // kind of native it is.  The reason is that for regular JNI natives
   // the incoming and outgoing registers are offset upwards and for
   // critical natives they are offset down.
-  int c_arg = total_c_args - 1;
-  int stride = -1;
-  int init = total_in_args - 1;
-  if (is_critical_native) {
-    // stride forwards
-    c_arg = 0;
-    stride = 1;
-    init = 0;
-  }
-  for (int i = init, count = 0; count < total_in_args; i += stride, c_arg += stride, count++ ) {
+  GrowableArray<int> arg_order(2 * total_in_args);
+  VMRegPair tmp_vmreg;
+  tmp_vmreg.set1(rbx->as_VMReg());
+
+  if (!is_critical_native) {
+    for (int i = total_in_args - 1, c_arg = total_c_args - 1; i >= 0; i--, c_arg--) {
+      arg_order.push(i);
+      arg_order.push(c_arg);
+    }
+  } else {
+    // Compute a valid move order, using tmp_vmreg to break any cycles
+    ComputeMoveOrder cmo(total_in_args, in_regs, total_c_args, out_regs, in_sig_bt, arg_order, tmp_vmreg);
+  }
+
+  int temploc = -1;
+  for (int ai = 0; ai < arg_order.length(); ai += 2) {
+    int i = arg_order.at(ai);
+    int c_arg = arg_order.at(ai + 1);
+    __ block_comment(err_msg("move %d -> %d", i, c_arg));
+    if (c_arg == -1) {
+      assert(is_critical_native, "should only be required for critical natives");
+      // This arg needs to be moved to a temporary
+      __ mov(tmp_vmreg.first()->as_Register(), in_regs[i].first()->as_Register());
+      in_regs[i] = tmp_vmreg;
+      temploc = i;
+      continue;
+    } else if (i == -1) {
+      assert(is_critical_native, "should only be required for critical natives");
+      // Read from the temporary location
+      assert(temploc != -1, "must be valid");
+      i = temploc;
+      temploc = -1;
+    }
 #ifdef ASSERT
     if (in_regs[i].first()->is_Register()) {
       assert(!reg_destroyed[in_regs[i].first()->as_Register()->encoding()], "destroyed reg!");
@@ -1779,7 +1951,7 @@ nmethod *SharedRuntime::generate_native_wrapper(MacroAssembler *masm,
 
   // point c_arg at the first arg that is already loaded in case we
   // need to spill before we call out
-  c_arg++;
+  int c_arg = total_c_args - total_in_args;
 
   // Pre-load a static method's oop into r14.  Used both by locking code and
   // the normal JNI call code.
@@ -3620,8 +3792,12 @@ void OptoRuntime::generate_exception_blob() {
   //
   // address OptoRuntime::handle_exception_C(JavaThread* thread)
 
-  __ set_last_Java_frame(noreg, noreg, NULL);
+  // At a method handle call, the stack may not be properly aligned
+  // when returning with an exception.
+  address the_pc = __ pc();
+  __ set_last_Java_frame(noreg, noreg, the_pc);
   __ mov(c_rarg0, r15_thread);
+  __ andptr(rsp, -(StackAlignmentInBytes));    // Align stack
   __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, OptoRuntime::handle_exception_C)));
 
   // Set an oopmap for the call site.  This oopmap will only be used if we
@@ -3632,9 +3808,9 @@ void OptoRuntime::generate_exception_blob() {
 
   OopMapSet* oop_maps = new OopMapSet();
 
-  oop_maps->add_gc_map( __ pc()-start, new OopMap(SimpleRuntimeFrame::framesize, 0));
+  oop_maps->add_gc_map(the_pc - start, new OopMap(SimpleRuntimeFrame::framesize, 0));
 
-  __ reset_last_Java_frame(false, false);
+  __ reset_last_Java_frame(false, true);
 
   // Restore callee-saved registers
 
diff --git a/src/cpu/x86/vm/stubGenerator_x86_64.cpp b/src/cpu/x86/vm/stubGenerator_x86_64.cpp
index d68bdac6d..9d9472200 100644
--- a/src/cpu/x86/vm/stubGenerator_x86_64.cpp
+++ b/src/cpu/x86/vm/stubGenerator_x86_64.cpp
@@ -710,6 +710,21 @@ class StubGenerator: public StubCodeGenerator {
     return start;
   }
 
+  // Support for intptr_t get_previous_sp()
+  //
+  // This routine is used to find the previous stack pointer for the
+  // caller.
+  address generate_get_previous_sp() {
+    StubCodeMark mark(this, "StubRoutines", "get_previous_sp");
+    address start = __ pc();
+
+    __ movptr(rax, rsp);
+    __ addptr(rax, 8); // return address is at the top of the stack.
+    __ ret(0);
+
+    return start;
+  }
+
   //----------------------------------------------------------------------------------------------------
   // Support for void verify_mxcsr()
   //
@@ -3060,6 +3075,7 @@ class StubGenerator: public StubCodeGenerator {
 
     // platform dependent
     StubRoutines::x86::_get_previous_fp_entry = generate_get_previous_fp();
+    StubRoutines::x86::_get_previous_sp_entry = generate_get_previous_sp();
 
     StubRoutines::x86::_verify_mxcsr_entry    = generate_verify_mxcsr();
 
diff --git a/src/cpu/x86/vm/stubRoutines_x86_64.cpp b/src/cpu/x86/vm/stubRoutines_x86_64.cpp
index 182872b88..782dc9063 100644
--- a/src/cpu/x86/vm/stubRoutines_x86_64.cpp
+++ b/src/cpu/x86/vm/stubRoutines_x86_64.cpp
@@ -43,6 +43,7 @@
 // a description of how to extend it, see the stubRoutines.hpp file.
 
 address StubRoutines::x86::_get_previous_fp_entry = NULL;
+address StubRoutines::x86::_get_previous_sp_entry = NULL;
 
 address StubRoutines::x86::_verify_mxcsr_entry = NULL;
 
diff --git a/src/cpu/x86/vm/stubRoutines_x86_64.hpp b/src/cpu/x86/vm/stubRoutines_x86_64.hpp
index 3d2a4fb9a..7737c4eee 100644
--- a/src/cpu/x86/vm/stubRoutines_x86_64.hpp
+++ b/src/cpu/x86/vm/stubRoutines_x86_64.hpp
@@ -41,6 +41,7 @@ class x86 {
 
  private:
   static address _get_previous_fp_entry;
+  static address _get_previous_sp_entry;
   static address _verify_mxcsr_entry;
 
   static address _f2i_fixup;
@@ -61,6 +62,11 @@ class x86 {
     return _get_previous_fp_entry;
   }
 
+  static address get_previous_sp_entry()
+  {
+    return _get_previous_sp_entry;
+  }
+
   static address verify_mxcsr_entry()
   {
     return _verify_mxcsr_entry;
diff --git a/src/os/linux/vm/os_linux.cpp b/src/os/linux/vm/os_linux.cpp
index 72f3fd07a..6a33c0e80 100644
--- a/src/os/linux/vm/os_linux.cpp
+++ b/src/os/linux/vm/os_linux.cpp
@@ -4690,14 +4690,12 @@ char* os::map_memory(int fd, const char* file_name, size_t file_offset,
                      char *addr, size_t bytes, bool read_only,
                      bool allow_exec) {
   int prot;
-  int flags;
+  int flags = MAP_PRIVATE;
 
   if (read_only) {
     prot = PROT_READ;
-    flags = MAP_SHARED;
   } else {
     prot = PROT_READ | PROT_WRITE;
-    flags = MAP_PRIVATE;
   }
 
   if (allow_exec) {
diff --git a/src/os/solaris/vm/os_solaris.cpp b/src/os/solaris/vm/os_solaris.cpp
index c47d37433..31bd41862 100644
--- a/src/os/solaris/vm/os_solaris.cpp
+++ b/src/os/solaris/vm/os_solaris.cpp
@@ -1013,15 +1013,6 @@ extern "C" void breakpoint() {
   // use debugger to set breakpoint here
 }
 
-// Returns an estimate of the current stack pointer. Result must be guaranteed to
-// point into the calling threads stack, and be no lower than the current stack
-// pointer.
-address os::current_stack_pointer() {
-  volatile int dummy;
-  address sp = (address)&dummy + 8;     // %%%% need to confirm if this is right
-  return sp;
-}
-
 static thread_t main_thread;
 
 // Thread start routine for all new Java threads
diff --git a/src/os/windows/vm/os_windows.cpp b/src/os/windows/vm/os_windows.cpp
index 970aab747..de38c6c12 100644
--- a/src/os/windows/vm/os_windows.cpp
+++ b/src/os/windows/vm/os_windows.cpp
@@ -324,16 +324,6 @@ extern "C" void breakpoint() {
   os::breakpoint();
 }
 
-// Returns an estimate of the current stack pointer. Result must be guaranteed
-// to point into the calling threads stack, and be no lower than the current
-// stack pointer.
-
-address os::current_stack_pointer() {
-  int dummy;
-  address sp = (address)&dummy;
-  return sp;
-}
-
 // os::current_stack_base()
 //
 //   Returns the base of the stack, which is the stack's
diff --git a/src/os_cpu/bsd_x86/vm/os_bsd_x86.cpp b/src/os_cpu/bsd_x86/vm/os_bsd_x86.cpp
index 06ba58075..12d6871ca 100644
--- a/src/os_cpu/bsd_x86/vm/os_bsd_x86.cpp
+++ b/src/os_cpu/bsd_x86/vm/os_bsd_x86.cpp
@@ -1126,3 +1126,8 @@ void os::setup_fpu() {
                       : "r" (fpu_cntrl) : "memory");
 #endif // !AMD64
 }
+
+#ifndef PRODUCT
+void os::verify_stack_alignment() {
+}
+#endif
diff --git a/src/os_cpu/bsd_zero/vm/os_bsd_zero.cpp b/src/os_cpu/bsd_zero/vm/os_bsd_zero.cpp
index ff5b32dbf..4bb5a8abb 100644
--- a/src/os_cpu/bsd_zero/vm/os_bsd_zero.cpp
+++ b/src/os_cpu/bsd_zero/vm/os_bsd_zero.cpp
@@ -562,3 +562,8 @@ extern "C" {
   }
 };
 #endif // !_LP64
+
+#ifndef PRODUCT
+void os::verify_stack_alignment() {
+}
+#endif
diff --git a/src/os_cpu/linux_sparc/vm/os_linux_sparc.cpp b/src/os_cpu/linux_sparc/vm/os_linux_sparc.cpp
index 62131ee6f..11c90e522 100644
--- a/src/os_cpu/linux_sparc/vm/os_linux_sparc.cpp
+++ b/src/os_cpu/linux_sparc/vm/os_linux_sparc.cpp
@@ -756,3 +756,8 @@ size_t os::Linux::default_guard_size(os::ThreadType thr_type) {
   // guard page, only enable glibc guard page for non-Java threads.
   return (thr_type == java_thread ? 0 : page_size());
 }
+
+#ifndef PRODUCT
+void os::verify_stack_alignment() {
+}
+#endif
diff --git a/src/os_cpu/linux_x86/vm/os_linux_x86.cpp b/src/os_cpu/linux_x86/vm/os_linux_x86.cpp
index ba484b9fd..93bb1f459 100644
--- a/src/os_cpu/linux_x86/vm/os_linux_x86.cpp
+++ b/src/os_cpu/linux_x86/vm/os_linux_x86.cpp
@@ -862,3 +862,11 @@ void os::setup_fpu() {
                       : "r" (fpu_cntrl) : "memory");
 #endif // !AMD64
 }
+
+#ifndef PRODUCT
+void os::verify_stack_alignment() {
+#ifdef AMD64
+  assert(((intptr_t)os::current_stack_pointer() & (StackAlignmentInBytes-1)) == 0, "incorrect stack alignment");
+#endif
+}
+#endif
diff --git a/src/os_cpu/linux_zero/vm/os_linux_zero.cpp b/src/os_cpu/linux_zero/vm/os_linux_zero.cpp
index 32c2d0a51..203090ce0 100644
--- a/src/os_cpu/linux_zero/vm/os_linux_zero.cpp
+++ b/src/os_cpu/linux_zero/vm/os_linux_zero.cpp
@@ -506,3 +506,8 @@ extern "C" {
   }
 };
 #endif // !_LP64
+
+#ifndef PRODUCT
+void os::verify_stack_alignment() {
+}
+#endif
diff --git a/src/os_cpu/solaris_sparc/vm/os_solaris_sparc.cpp b/src/os_cpu/solaris_sparc/vm/os_solaris_sparc.cpp
index 971f600fe..752daf7bf 100644
--- a/src/os_cpu/solaris_sparc/vm/os_solaris_sparc.cpp
+++ b/src/os_cpu/solaris_sparc/vm/os_solaris_sparc.cpp
@@ -251,6 +251,15 @@ frame os::get_sender_for_C_frame(frame* fr) {
   return frame(fr->sender_sp(), frame::unpatchable, fr->sender_pc());
 }
 
+// Returns an estimate of the current stack pointer. Result must be guaranteed to
+// point into the calling threads stack, and be no lower than the current stack
+// pointer.
+address os::current_stack_pointer() {
+  volatile int dummy;
+  address sp = (address)&dummy + 8;     // %%%% need to confirm if this is right
+  return sp;
+}
+
 frame os::current_frame() {
   intptr_t* sp = StubRoutines::Sparc::flush_callers_register_windows_func()();
   frame myframe(sp, frame::unpatchable,
@@ -815,3 +824,8 @@ add_func_t*          os::atomic_add_func          = os::atomic_add_bootstrap;
    __asm__ __volatile__ ("wr %%g0, 0, %%fprs \n\t" : : :);
   }
 #endif //defined(__sparc) && defined(COMPILER2)
+
+#ifndef PRODUCT
+void os::verify_stack_alignment() {
+}
+#endif
diff --git a/src/os_cpu/solaris_x86/vm/os_solaris_x86.cpp b/src/os_cpu/solaris_x86/vm/os_solaris_x86.cpp
index 78e93ea66..ec047c16b 100644
--- a/src/os_cpu/solaris_x86/vm/os_solaris_x86.cpp
+++ b/src/os_cpu/solaris_x86/vm/os_solaris_x86.cpp
@@ -237,6 +237,12 @@ frame os::get_sender_for_C_frame(frame* fr) {
   return frame(fr->sender_sp(), fr->link(), fr->sender_pc());
 }
 
+extern "C" intptr_t *_get_current_sp();  // in .il file
+
+address os::current_stack_pointer() {
+  return (address)_get_current_sp();
+}
+
 extern "C" intptr_t *_get_current_fp();  // in .il file
 
 frame os::current_frame() {
@@ -954,3 +960,11 @@ void os::setup_fpu() {
   _solaris_raw_setup_fpu(fpu_cntrl);
 }
 #endif // AMD64
+
+#ifndef PRODUCT
+void os::verify_stack_alignment() {
+#ifdef AMD64
+  assert(((intptr_t)os::current_stack_pointer() & (StackAlignmentInBytes-1)) == 0, "incorrect stack alignment");
+#endif
+}
+#endif
diff --git a/src/os_cpu/solaris_x86/vm/solaris_x86_32.il b/src/os_cpu/solaris_x86/vm/solaris_x86_32.il
index b635a8292..9b0f07da3 100644
--- a/src/os_cpu/solaris_x86/vm/solaris_x86_32.il
+++ b/src/os_cpu/solaris_x86/vm/solaris_x86_32.il
@@ -37,6 +37,12 @@
       movl     %gs:0, %eax 
       .end
 
+  // Get current sp
+      .inline _get_current_sp,0
+      .volatile
+      movl     %esp, %eax 
+      .end
+
   // Get current fp
       .inline _get_current_fp,0
       .volatile
diff --git a/src/os_cpu/solaris_x86/vm/solaris_x86_64.il b/src/os_cpu/solaris_x86/vm/solaris_x86_64.il
index fb7946b8c..89809bc54 100644
--- a/src/os_cpu/solaris_x86/vm/solaris_x86_64.il
+++ b/src/os_cpu/solaris_x86/vm/solaris_x86_64.il
@@ -30,6 +30,12 @@
       movq     %fs:0, %rax 
       .end
 
+  // Get current sp
+      .inline _get_current_sp,0
+      .volatile
+      movq     %rsp, %rax 
+      .end
+
   // Get current fp
       .inline _get_current_fp,0
       .volatile
diff --git a/src/os_cpu/windows_x86/vm/os_windows_x86.cpp b/src/os_cpu/windows_x86/vm/os_windows_x86.cpp
index 960ceab66..52fe243ee 100644
--- a/src/os_cpu/windows_x86/vm/os_windows_x86.cpp
+++ b/src/os_cpu/windows_x86/vm/os_windows_x86.cpp
@@ -370,6 +370,26 @@ frame os::get_sender_for_C_frame(frame* fr) {
   return frame(fr->sender_sp(), fr->link(), fr->sender_pc());
 }
 
+#ifndef AMD64
+// Returns an estimate of the current stack pointer. Result must be guaranteed
+// to point into the calling threads stack, and be no lower than the current
+// stack pointer.
+address os::current_stack_pointer() {
+  int dummy;
+  address sp = (address)&dummy;
+  return sp;
+}
+#else
+// Returns the current stack pointer. Accurate value needed for
+// os::verify_stack_alignment().
+address os::current_stack_pointer() {
+  typedef address get_sp_func();
+  get_sp_func* func = CAST_TO_FN_PTR(get_sp_func*,
+                                     StubRoutines::x86::get_previous_sp_entry());
+  return (*func)();
+}
+#endif
+
 
 #ifndef AMD64
 intptr_t* _get_previous_fp() {
@@ -546,3 +566,11 @@ void os::setup_fpu() {
   __asm fldcw fpu_cntrl_word;
 #endif // !AMD64
 }
+
+#ifndef PRODUCT
+void os::verify_stack_alignment() {
+#ifdef AMD64
+  assert(((intptr_t)os::current_stack_pointer() & (StackAlignmentInBytes-1)) == 0, "incorrect stack alignment");
+#endif
+}
+#endif
diff --git a/src/share/tools/LogCompilation/src/com/sun/hotspot/tools/compiler/Compilation.java b/src/share/tools/LogCompilation/src/com/sun/hotspot/tools/compiler/Compilation.java
index 73e59566e..861fe4437 100644
--- a/src/share/tools/LogCompilation/src/com/sun/hotspot/tools/compiler/Compilation.java
+++ b/src/share/tools/LogCompilation/src/com/sun/hotspot/tools/compiler/Compilation.java
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2009, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2009, 2012, Oracle and/or its affiliates. All rights reserved.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
  * This code is free software; you can redistribute it and/or modify it
@@ -33,6 +33,7 @@ public class Compilation implements LogEvent {
     private boolean osr;
     private Method method;
     private CallSite call = new CallSite();
+    private CallSite lateInlineCall = new CallSite();
     private int osrBci;
     private String icount;
     private String bcount;
@@ -80,6 +81,13 @@ public class Compilation implements LogEvent {
             sb.append(site);
             sb.append("\n");
         }
+        if (getLateInlineCall().getCalls() != null) {
+            sb.append("late inline:\n");
+            for (CallSite site : getLateInlineCall().getCalls()) {
+                sb.append(site);
+                sb.append("\n");
+            }
+        }
         return sb.toString();
     }
 
@@ -115,6 +123,12 @@ public class Compilation implements LogEvent {
                     site.print(stream, indent + 2);
                 }
             }
+            if (printInlining && lateInlineCall.getCalls() != null) {
+                stream.println("late inline:");
+                for (CallSite site : lateInlineCall.getCalls()) {
+                    site.print(stream, indent + 2);
+                }
+            }
         }
     }
 
@@ -215,7 +229,11 @@ public class Compilation implements LogEvent {
     }
 
     public void setMethod(Method method) {
-        this.method = method;
+        // Don't change method if it is already set to avoid changing
+        // it by post parse inlining info.
+        if (getMethod() == null) {
+            this.method = method;
+        }
     }
 
     public CallSite getCall() {
@@ -226,6 +244,10 @@ public class Compilation implements LogEvent {
         this.call = call;
     }
 
+    public CallSite getLateInlineCall() {
+        return lateInlineCall;
+    }
+
     public double getElapsedTime() {
         return end - start;
     }
diff --git a/src/share/tools/LogCompilation/src/com/sun/hotspot/tools/compiler/LogParser.java b/src/share/tools/LogCompilation/src/com/sun/hotspot/tools/compiler/LogParser.java
index d864db0ff..dc0b7ab34 100644
--- a/src/share/tools/LogCompilation/src/com/sun/hotspot/tools/compiler/LogParser.java
+++ b/src/share/tools/LogCompilation/src/com/sun/hotspot/tools/compiler/LogParser.java
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2009, 2011, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2009, 2012, 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
@@ -146,6 +146,7 @@ public class LogParser extends DefaultHandler implements ErrorHandler, Constants
     private CallSite site;
     private Stack<Phase> phaseStack = new Stack<Phase>();
     private UncommonTrapEvent currentTrap;
+    private Stack<CallSite> late_inline_scope;
 
     long parseLong(String l) {
         try {
@@ -302,6 +303,7 @@ public class LogParser extends DefaultHandler implements ErrorHandler, Constants
             }
             events.add(compile);
             compiles.put(makeId(atts), compile);
+            site = compile.getCall();
         } else if (qname.equals("type")) {
             type(search(atts, "id"), search(atts, "name"));
         } else if (qname.equals("bc")) {
@@ -360,12 +362,22 @@ public class LogParser extends DefaultHandler implements ErrorHandler, Constants
                 // uncommon trap inserted during parsing.
                 // ignore for now
             }
+        } else if (qname.equals("late_inline")) {
+            late_inline_scope = new Stack<CallSite>();
+            site = new CallSite(-999, method(search(atts, "method")));
+            late_inline_scope.push(site);
         } else if (qname.equals("jvms")) {
             // <jvms bci='4' method='java/io/DataInputStream readChar ()C' bytes='40' count='5815' iicount='20815'/>
             if (currentTrap != null) {
                 currentTrap.addJVMS(atts.getValue("method"), Integer.parseInt(atts.getValue("bci")));
+            } else if (late_inline_scope != null) {
+                bci = Integer.parseInt(search(atts, "bci"));
+                site = new CallSite(bci, method(search(atts, "method")));
+                late_inline_scope.push(site);
             } else {
-                // Ignore <eliminate_allocation type='667'> and <eliminate_lock lock='1'>
+                // Ignore <eliminate_allocation type='667'>,
+                //        <eliminate_lock lock='1'>,
+                //        <replace_string_concat arguments='2' string_alloc='0' multiple='0'>
             }
         } else if (qname.equals("nmethod")) {
             String id = makeId(atts);
@@ -379,7 +391,7 @@ public class LogParser extends DefaultHandler implements ErrorHandler, Constants
             Method m = method(search(atts, "method"));
             if (scopes.size() == 0) {
                 compile.setMethod(m);
-                scopes.push(compile.getCall());
+                scopes.push(site);
             } else {
                 if (site.getMethod() == m) {
                     scopes.push(site);
@@ -393,7 +405,7 @@ public class LogParser extends DefaultHandler implements ErrorHandler, Constants
             }
         } else if (qname.equals("parse_done")) {
             CallSite call = scopes.pop();
-            call.setEndNodes(Integer.parseInt(search(atts, "nodes")));
+            call.setEndNodes(Integer.parseInt(search(atts, "nodes", "1")));
             call.setTimeStamp(Double.parseDouble(search(atts, "stamp")));
             scopes.push(call);
         }
@@ -408,6 +420,43 @@ public class LogParser extends DefaultHandler implements ErrorHandler, Constants
             scopes.pop();
         } else if (qname.equals("uncommon_trap")) {
             currentTrap = null;
+        } else if (qname.equals("late_inline")) {
+            // Populate late inlining info.
+
+            // late_inline scopes are specified in reverse order:
+            // compiled method should be on top of stack.
+            CallSite caller = late_inline_scope.pop();
+            Method m = compile.getMethod();
+            if (m != caller.getMethod()) {
+                System.out.println(m);
+                System.out.println(caller.getMethod() + " bci: " + bci);
+                throw new InternalError("call site and late_inline info don't match");
+            }
+
+            // late_inline contains caller+bci info, convert it
+            // to bci+callee info used by LogCompilation.
+            site = compile.getLateInlineCall();
+            do {
+                bci = caller.getBci();
+                // Next inlined call.
+                caller = late_inline_scope.pop();
+                CallSite callee =  new CallSite(bci, caller.getMethod());
+                site.add(callee);
+                site = callee;
+            } while (!late_inline_scope.empty());
+
+            if (caller.getBci() != -999) {
+                System.out.println(caller.getMethod());
+                throw new InternalError("broken late_inline info");
+            }
+            if (site.getMethod() != caller.getMethod()) {
+                System.out.println(site.getMethod());
+                System.out.println(caller.getMethod());
+                throw new InternalError("call site and late_inline info don't match");
+            }
+            // late_inline is followed by parse with scopes.size() == 0,
+            // 'site' will be pushed to scopes.
+            late_inline_scope = null;
         } else if (qname.equals("task")) {
             types.clear();
             methods.clear();
diff --git a/src/share/vm/c1/c1_LinearScan.cpp b/src/share/vm/c1/c1_LinearScan.cpp
index aaae71d46..fede09fe8 100644
--- a/src/share/vm/c1/c1_LinearScan.cpp
+++ b/src/share/vm/c1/c1_LinearScan.cpp
@@ -1884,7 +1884,7 @@ void LinearScan::resolve_exception_entry(BlockBegin* block, MoveResolver &move_r
 
   if (move_resolver.has_mappings()) {
     // insert moves after first instruction
-    move_resolver.set_insert_position(block->lir(), 1);
+    move_resolver.set_insert_position(block->lir(), 0);
     move_resolver.resolve_and_append_moves();
   }
 }
diff --git a/src/share/vm/classfile/vmSymbols.hpp b/src/share/vm/classfile/vmSymbols.hpp
index d6ed1def5..92c06342d 100644
--- a/src/share/vm/classfile/vmSymbols.hpp
+++ b/src/share/vm/classfile/vmSymbols.hpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 1997, 2011, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1997, 2012, 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
@@ -509,6 +509,9 @@
   template(clear_name,                                 "clear")                                                   \
   template(trigger_method_signature,                   "(ILjava/lang/management/MemoryUsage;)V")                                                 \
   template(startAgent_name,                            "startAgent")                                              \
+  template(startRemoteAgent_name,                      "startRemoteManagementAgent")                              \
+  template(startLocalAgent_name,                       "startLocalManagementAgent")                               \
+  template(stopRemoteAgent_name,                       "stopRemoteManagementAgent")                               \
   template(java_lang_management_ThreadInfo_constructor_signature, "(Ljava/lang/Thread;ILjava/lang/Object;Ljava/lang/Thread;JJJJ[Ljava/lang/StackTraceElement;)V") \
   template(java_lang_management_ThreadInfo_with_locks_constructor_signature, "(Ljava/lang/Thread;ILjava/lang/Object;Ljava/lang/Thread;JJJJ[Ljava/lang/StackTraceElement;[Ljava/lang/Object;[I[Ljava/lang/Object;)V") \
   template(long_long_long_long_void_signature,         "(JJJJ)V")                                                 \
diff --git a/src/share/vm/oops/cpCacheOop.cpp b/src/share/vm/oops/cpCacheOop.cpp
index 678bc1327..8ebc99678 100644
--- a/src/share/vm/oops/cpCacheOop.cpp
+++ b/src/share/vm/oops/cpCacheOop.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 1998, 2011, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1998, 2012, 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
@@ -504,17 +504,17 @@ bool ConstantPoolCacheEntry::is_interesting_method_entry(klassOop k) {
 
 void ConstantPoolCacheEntry::print(outputStream* st, int index) const {
   // print separator
-  if (index == 0) tty->print_cr("                 -------------");
+  if (index == 0) st->print_cr("                 -------------");
   // print entry
-  tty->print("%3d  ("PTR_FORMAT")  ", index, (intptr_t)this);
+  st->print("%3d  ("PTR_FORMAT")  ", index, (intptr_t)this);
   if (is_secondary_entry())
-    tty->print_cr("[%5d|secondary]", main_entry_index());
+    st->print_cr("[%5d|secondary]", main_entry_index());
   else
-    tty->print_cr("[%02x|%02x|%5d]", bytecode_2(), bytecode_1(), constant_pool_index());
-  tty->print_cr("                 [   "PTR_FORMAT"]", (intptr_t)(oop)_f1);
-  tty->print_cr("                 [   "PTR_FORMAT"]", (intptr_t)_f2);
-  tty->print_cr("                 [   "PTR_FORMAT"]", (intptr_t)_flags);
-  tty->print_cr("                 -------------");
+    st->print_cr("[%02x|%02x|%5d]", bytecode_2(), bytecode_1(), constant_pool_index());
+  st->print_cr("                 [   "PTR_FORMAT"]", (intptr_t)(oop)_f1);
+  st->print_cr("                 [   "PTR_FORMAT"]", (intptr_t)_f2);
+  st->print_cr("                 [   "PTR_FORMAT"]", (intptr_t)_flags);
+  st->print_cr("                 -------------");
 }
 
 void ConstantPoolCacheEntry::verify(outputStream* st) const {
diff --git a/src/share/vm/opto/connode.cpp b/src/share/vm/opto/connode.cpp
index 9a3b65031..7e072ff44 100644
--- a/src/share/vm/opto/connode.cpp
+++ b/src/share/vm/opto/connode.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 1997, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1997, 2012, 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
@@ -1051,6 +1051,7 @@ Node *ConvL2INode::Ideal(PhaseGVN *phase, bool can_reshape) {
 //------------------------------Value------------------------------------------
 const Type *CastX2PNode::Value( PhaseTransform *phase ) const {
   const Type* t = phase->type(in(1));
+  if (t == Type::TOP) return Type::TOP;
   if (t->base() == Type_X && t->singleton()) {
     uintptr_t bits = (uintptr_t) t->is_intptr_t()->get_con();
     if (bits == 0)   return TypePtr::NULL_PTR;
@@ -1121,6 +1122,7 @@ Node *CastX2PNode::Identity( PhaseTransform *phase ) {
 //------------------------------Value------------------------------------------
 const Type *CastP2XNode::Value( PhaseTransform *phase ) const {
   const Type* t = phase->type(in(1));
+  if (t == Type::TOP) return Type::TOP;
   if (t->base() == Type::RawPtr && t->singleton()) {
     uintptr_t bits = (uintptr_t) t->is_rawptr()->get_con();
     return TypeX::make(bits);
diff --git a/src/share/vm/opto/escape.cpp b/src/share/vm/opto/escape.cpp
index 27fd8171a..d7c67e252 100644
--- a/src/share/vm/opto/escape.cpp
+++ b/src/share/vm/opto/escape.cpp
@@ -2035,40 +2035,14 @@ void ConnectionGraph::find_init_values(Node* alloc, VectorSet* visited, PhaseTra
           Node* store = ini->find_captured_store(offset, type2aelembytes(ft), phase);
           if (store != NULL && store->is_Store()) {
             value = store->in(MemNode::ValueIn);
-          } else if (ptn->edge_count() > 0) { // Are there oop stores?
-            // Check for a store which follows allocation without branches.
+          } else {
+            // There could be initializing stores which follow allocation.
             // For example, a volatile field store is not collected
-            // by Initialize node. TODO: it would be nice to use idom() here.
-            //
-            // Search all references to the same field which use different
-            // AddP nodes, for example, in the next case:
-            //
-            //    Point p[] = new Point[1];
-            //    if ( x ) { p[0] = new Point(); p[0].x = x; }
-            //    if ( p[0] != null ) { y = p[0].x; } // has CastPP
+            // by Initialize node.
             //
-            for (uint next = ei; (next < ae_cnt) && (value == NULL); next++) {
-              uint fpi = pta->edge_target(next); // Field (AddP)
-              PointsToNode *ptf = ptnode_adr(fpi);
-              if (ptf->offset() == offset) {
-                Node* nf = ptf->_node;
-                for (DUIterator_Fast imax, i = nf->fast_outs(imax); i < imax; i++) {
-                  store = nf->fast_out(i);
-                  if (store->is_Store() && store->in(0) != NULL) {
-                    Node* ctrl = store->in(0);
-                    while(!(ctrl == ini || ctrl == alloc || ctrl == NULL ||
-                            ctrl == C->root() || ctrl == C->top() || ctrl->is_Region() ||
-                            ctrl->is_IfTrue() || ctrl->is_IfFalse())) {
-                       ctrl = ctrl->in(0);
-                    }
-                    if (ctrl == ini || ctrl == alloc) {
-                      value = store->in(MemNode::ValueIn);
-                      break;
-                    }
-                  }
-                }
-              }
-            }
+            // Need to check for dependent loads to separate such stores from
+            // stores which follow loads. For now, add initial value NULL so
+            // that compare pointers optimization works correctly.
           }
         }
         if (value == NULL || value != ptnode_adr(value->_idx)->_node) {
diff --git a/src/share/vm/runtime/interfaceSupport.hpp b/src/share/vm/runtime/interfaceSupport.hpp
index e1001eea1..2875ee0eb 100644
--- a/src/share/vm/runtime/interfaceSupport.hpp
+++ b/src/share/vm/runtime/interfaceSupport.hpp
@@ -436,6 +436,7 @@ class RuntimeHistogramElement : public HistogramElement {
 #define VM_LEAF_BASE(result_type, header)                            \
   TRACE_CALL(result_type, header)                                    \
   debug_only(NoHandleMark __hm;)                                     \
+  os::verify_stack_alignment();                                      \
   /* begin of body */
 
 
@@ -445,6 +446,7 @@ class RuntimeHistogramElement : public HistogramElement {
   TRACE_CALL(result_type, header)                                    \
   HandleMarkCleaner __hm(thread);                                    \
   Thread* THREAD = thread;                                           \
+  os::verify_stack_alignment();                                      \
   /* begin of body */
 
 
@@ -454,6 +456,7 @@ class RuntimeHistogramElement : public HistogramElement {
   TRACE_CALL(result_type, header)                                    \
   debug_only(NoHandleMark __hm;)                                     \
   Thread* THREAD = thread;                                           \
+  os::verify_stack_alignment();                                      \
   /* begin of body */
 
 
diff --git a/src/share/vm/runtime/os.hpp b/src/share/vm/runtime/os.hpp
index e77f3ce91..dd163e9a9 100644
--- a/src/share/vm/runtime/os.hpp
+++ b/src/share/vm/runtime/os.hpp
@@ -404,6 +404,8 @@ class os: AllStatic {
   static address current_stack_base();
   static size_t current_stack_size();
 
+  static void verify_stack_alignment() PRODUCT_RETURN;
+
   static int message_box(const char* title, const char* message);
   static char* do_you_want_to_debug(const char* message);
 
diff --git a/src/share/vm/services/diagnosticCommand.cpp b/src/share/vm/services/diagnosticCommand.cpp
index 78f4c2239..107b1abe5 100644
--- a/src/share/vm/services/diagnosticCommand.cpp
+++ b/src/share/vm/services/diagnosticCommand.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2011, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2011, 2012, 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
@@ -49,6 +49,11 @@ void DCmdRegistrant::register_dcmds(){
   DCmdFactory::register_DCmdFactory(new DCmdFactoryImpl<ClassHistogramDCmd>(true, false));
   DCmdFactory::register_DCmdFactory(new DCmdFactoryImpl<ThreadDumpDCmd>(true, false));
 
+  //Enhanced JMX Agent Support
+  DCmdFactory::register_DCmdFactory(new DCmdFactoryImpl<JMXStartRemoteDCmd>(true,false));
+  DCmdFactory::register_DCmdFactory(new DCmdFactoryImpl<JMXStartLocalDCmd>(true,false));
+  DCmdFactory::register_DCmdFactory(new DCmdFactoryImpl<JMXStopRemoteDCmd>(true,false));
+
 }
 
 #ifndef HAVE_EXTRA_DCMD
@@ -344,3 +349,185 @@ int ThreadDumpDCmd::num_arguments() {
     return 0;
   }
 }
+
+// Enhanced JMX Agent support
+
+JMXStartRemoteDCmd::JMXStartRemoteDCmd(outputStream *output, bool heap_allocated) :
+
+  DCmdWithParser(output, heap_allocated),
+
+  _config_file
+  ("config.file",
+   "set com.sun.management.config.file", "STRING", false),
+
+  _jmxremote_port
+  ("jmxremote.port",
+   "set com.sun.management.jmxremote.port", "STRING", false),
+
+  _jmxremote_rmi_port
+  ("jmxremote.rmi.port",
+   "set com.sun.management.jmxremote.rmi.port", "STRING", false),
+
+  _jmxremote_ssl
+  ("jmxremote.ssl",
+   "set com.sun.management.jmxremote.ssl", "STRING", false),
+
+  _jmxremote_registry_ssl
+  ("jmxremote.registry.ssl",
+   "set com.sun.management.jmxremote.registry.ssl", "STRING", false),
+
+  _jmxremote_authenticate
+  ("jmxremote.authenticate",
+   "set com.sun.management.jmxremote.authenticate", "STRING", false),
+
+  _jmxremote_password_file
+  ("jmxremote.password.file",
+   "set com.sun.management.jmxremote.password.file", "STRING", false),
+
+  _jmxremote_access_file
+  ("jmxremote.access.file",
+   "set com.sun.management.jmxremote.access.file", "STRING", false),
+
+  _jmxremote_login_config
+  ("jmxremote.login.config",
+   "set com.sun.management.jmxremote.login.config", "STRING", false),
+
+  _jmxremote_ssl_enabled_cipher_suites
+  ("jmxremote.ssl.enabled.cipher.suites",
+   "set com.sun.management.jmxremote.ssl.enabled.cipher.suite", "STRING", false),
+
+  _jmxremote_ssl_enabled_protocols
+  ("jmxremote.ssl.enabled.protocols",
+   "set com.sun.management.jmxremote.ssl.enabled.protocols", "STRING", false),
+
+  _jmxremote_ssl_need_client_auth
+  ("jmxremote.ssl.need.client.auth",
+   "set com.sun.management.jmxremote.need.client.auth", "STRING", false),
+
+  _jmxremote_ssl_config_file
+  ("jmxremote.ssl.config.file",
+   "set com.sun.management.jmxremote.ssl_config_file", "STRING", false)
+
+  {
+    _dcmdparser.add_dcmd_option(&_config_file);
+    _dcmdparser.add_dcmd_option(&_jmxremote_port);
+    _dcmdparser.add_dcmd_option(&_jmxremote_rmi_port);
+    _dcmdparser.add_dcmd_option(&_jmxremote_ssl);
+    _dcmdparser.add_dcmd_option(&_jmxremote_registry_ssl);
+    _dcmdparser.add_dcmd_option(&_jmxremote_authenticate);
+    _dcmdparser.add_dcmd_option(&_jmxremote_password_file);
+    _dcmdparser.add_dcmd_option(&_jmxremote_access_file);
+    _dcmdparser.add_dcmd_option(&_jmxremote_login_config);
+    _dcmdparser.add_dcmd_option(&_jmxremote_ssl_enabled_cipher_suites);
+    _dcmdparser.add_dcmd_option(&_jmxremote_ssl_enabled_protocols);
+    _dcmdparser.add_dcmd_option(&_jmxremote_ssl_need_client_auth);
+    _dcmdparser.add_dcmd_option(&_jmxremote_ssl_config_file);
+}
+
+
+int JMXStartRemoteDCmd::num_arguments() {
+  ResourceMark rm;
+  JMXStartRemoteDCmd* dcmd = new JMXStartRemoteDCmd(NULL, false);
+  if (dcmd != NULL) {
+    DCmdMark mark(dcmd);
+    return dcmd->_dcmdparser.num_arguments();
+  } else {
+    return 0;
+  }
+}
+
+
+void JMXStartRemoteDCmd::execute(TRAPS) {
+    ResourceMark rm(THREAD);
+    HandleMark hm(THREAD);
+
+    // Load and initialize the sun.management.Agent class
+    // invoke startRemoteManagementAgent(string) method to start
+    // the remote management server.
+    // throw java.lang.NoSuchMethodError if the method doesn't exist
+
+    Handle loader = Handle(THREAD, SystemDictionary::java_system_loader());
+    klassOop k = SystemDictionary::resolve_or_fail(vmSymbols::sun_management_Agent(), loader, Handle(), true, CHECK);
+    instanceKlassHandle ik (THREAD, k);
+
+    JavaValue result(T_VOID);
+
+    // Pass all command line arguments to java as key=value,...
+    // All checks are done on java side
+
+    int len = 0;
+    stringStream options;
+    char comma[2] = {0,0};
+
+    // Leave default values on Agent.class side and pass only
+    // agruments explicitly set by user. All arguments passed
+    // to jcmd override properties with the same name set by
+    // command line with -D or by managmenent.properties
+    // file.
+#define PUT_OPTION(a) \
+    if ( (a).is_set() ){ \
+        options.print("%scom.sun.management.%s=%s", comma, (a).name(), (a).value()); \
+        comma[0] = ','; \
+    }
+
+    PUT_OPTION(_config_file);
+    PUT_OPTION(_jmxremote_port);
+    PUT_OPTION(_jmxremote_rmi_port);
+    PUT_OPTION(_jmxremote_ssl);
+    PUT_OPTION(_jmxremote_registry_ssl);
+    PUT_OPTION(_jmxremote_authenticate);
+    PUT_OPTION(_jmxremote_password_file);
+    PUT_OPTION(_jmxremote_access_file);
+    PUT_OPTION(_jmxremote_login_config);
+    PUT_OPTION(_jmxremote_ssl_enabled_cipher_suites);
+    PUT_OPTION(_jmxremote_ssl_enabled_protocols);
+    PUT_OPTION(_jmxremote_ssl_need_client_auth);
+    PUT_OPTION(_jmxremote_ssl_config_file);
+
+#undef PUT_OPTION
+
+    Handle str = java_lang_String::create_from_str(options.as_string(), CHECK);
+    JavaCalls::call_static(&result, ik, vmSymbols::startRemoteAgent_name(), vmSymbols::string_void_signature(), str, CHECK);
+}
+
+JMXStartLocalDCmd::JMXStartLocalDCmd(outputStream *output, bool heap_allocated) :
+  DCmd(output, heap_allocated)
+{
+  // do nothing
+}
+
+void JMXStartLocalDCmd::execute(TRAPS) {
+    ResourceMark rm(THREAD);
+    HandleMark hm(THREAD);
+
+    // Load and initialize the sun.management.Agent class
+    // invoke startLocalManagementAgent(void) method to start
+    // the local management server
+    // throw java.lang.NoSuchMethodError if method doesn't exist
+
+    Handle loader = Handle(THREAD, SystemDictionary::java_system_loader());
+    klassOop k = SystemDictionary::resolve_or_fail(vmSymbols::sun_management_Agent(), loader, Handle(), true, CHECK);
+    instanceKlassHandle ik (THREAD, k);
+
+    JavaValue result(T_VOID);
+    JavaCalls::call_static(&result, ik, vmSymbols::startLocalAgent_name(), vmSymbols::void_method_signature(), CHECK);
+}
+
+
+void JMXStopRemoteDCmd::execute(TRAPS) {
+    ResourceMark rm(THREAD);
+    HandleMark hm(THREAD);
+
+    // Load and initialize the sun.management.Agent class
+    // invoke stopRemoteManagementAgent method to stop the
+    // management server
+    // throw java.lang.NoSuchMethodError if method doesn't exist
+
+    Handle loader = Handle(THREAD, SystemDictionary::java_system_loader());
+    klassOop k = SystemDictionary::resolve_or_fail(vmSymbols::sun_management_Agent(), loader, Handle(), true, CHECK);
+    instanceKlassHandle ik (THREAD, k);
+
+    JavaValue result(T_VOID);
+    JavaCalls::call_static(&result, ik, vmSymbols::stopRemoteAgent_name(), vmSymbols::void_method_signature(), CHECK);
+}
+
diff --git a/src/share/vm/services/diagnosticCommand.hpp b/src/share/vm/services/diagnosticCommand.hpp
index c226a4ad6..d5c5cd172 100644
--- a/src/share/vm/services/diagnosticCommand.hpp
+++ b/src/share/vm/services/diagnosticCommand.hpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2011, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2011, 2012, 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
@@ -214,4 +214,82 @@ public:
   virtual void execute(TRAPS);
 };
 
+// Enhanced JMX Agent support
+
+class JMXStartRemoteDCmd : public DCmdWithParser {
+
+  // Explicitly list all properties that could be
+  // passed to Agent.startRemoteManagementAgent()
+  // com.sun.management is omitted
+
+  DCmdArgument<char *> _config_file;
+  DCmdArgument<char *> _jmxremote_port;
+  DCmdArgument<char *> _jmxremote_rmi_port;
+  DCmdArgument<char *> _jmxremote_ssl;
+  DCmdArgument<char *> _jmxremote_registry_ssl;
+  DCmdArgument<char *> _jmxremote_authenticate;
+  DCmdArgument<char *> _jmxremote_password_file;
+  DCmdArgument<char *> _jmxremote_access_file;
+  DCmdArgument<char *> _jmxremote_login_config;
+  DCmdArgument<char *> _jmxremote_ssl_enabled_cipher_suites;
+  DCmdArgument<char *> _jmxremote_ssl_enabled_protocols;
+  DCmdArgument<char *> _jmxremote_ssl_need_client_auth;
+  DCmdArgument<char *> _jmxremote_ssl_config_file;
+
+public:
+  JMXStartRemoteDCmd(outputStream *output, bool heap_allocated);
+
+  static const char *name() {
+    return "ManagementAgent.start";
+  }
+
+  static const char *description() {
+    return "Start remote management agent.";
+  }
+
+  static int num_arguments();
+
+  virtual void execute(TRAPS);
+
+};
+
+class JMXStartLocalDCmd : public DCmd {
+
+  // Explicitly request start of local agent,
+  // it will not be started by start dcmd
+
+
+public:
+  JMXStartLocalDCmd(outputStream *output, bool heap_allocated);
+
+  static const char *name() {
+    return "ManagementAgent.start_local";
+  }
+
+  static const char *description() {
+    return "Start local management agent.";
+  }
+
+  virtual void execute(TRAPS);
+
+};
+
+class JMXStopRemoteDCmd : public DCmd {
+public:
+  JMXStopRemoteDCmd(outputStream *output, bool heap_allocated) :
+  DCmd(output, heap_allocated) {
+    // Do Nothing
+  }
+
+  static const char *name() {
+    return "ManagementAgent.stop";
+  }
+
+  static const char *description() {
+    return "Stop remote management agent.";
+  }
+
+  virtual void execute(TRAPS);
+};
+
 #endif // SHARE_VM_SERVICES_DIAGNOSTICCOMMAND_HPP