6655638: dynamic languages need method handles

Summary: initial implementation, with known omissions (x86/64, sparc, compiler optim., c-oops, C++ interp.)
Reviewed-by: kvn, twisti, never

24 files changed, 1556 insertions, 37 deletions

diff --git a/src/cpu/sparc/vm/assembler_sparc.cpp b/src/cpu/sparc/vm/assembler_sparc.cpp
index 597a91ae3..d36caf55f 100644
--- a/src/cpu/sparc/vm/assembler_sparc.cpp
+++ b/src/cpu/sparc/vm/assembler_sparc.cpp
@@ -3029,6 +3029,58 @@ void MacroAssembler::check_klass_subtype_slow_path(Register sub_klass,
 
 
 
+void MacroAssembler::check_method_handle_type(Register mtype_reg, Register mh_reg,
+                                              Register temp_reg,
+                                              Label& wrong_method_type) {
+  assert_different_registers(mtype_reg, mh_reg, temp_reg);
+  // compare method type against that of the receiver
+  RegisterOrConstant mhtype_offset = delayed_value(java_dyn_MethodHandle::type_offset_in_bytes, temp_reg);
+  ld_ptr(mh_reg, mhtype_offset, temp_reg);
+  cmp(temp_reg, mtype_reg);
+  br(Assembler::notEqual, false, Assembler::pn, wrong_method_type);
+  delayed()->nop();
+}
+
+
+void MacroAssembler::jump_to_method_handle_entry(Register mh_reg, Register temp_reg) {
+  assert(mh_reg == G3_method_handle, "caller must put MH object in G3");
+  assert_different_registers(mh_reg, temp_reg);
+
+  // pick out the interpreted side of the handler
+  ld_ptr(mh_reg, delayed_value(java_dyn_MethodHandle::vmentry_offset_in_bytes, temp_reg), temp_reg);
+
+  // off we go...
+  ld_ptr(temp_reg, MethodHandleEntry::from_interpreted_entry_offset_in_bytes(), temp_reg);
+  jmp(temp_reg, 0);
+
+  // for the various stubs which take control at this point,
+  // see MethodHandles::generate_method_handle_stub
+
+  // (Can any caller use this delay slot?  If so, add an option for supression.)
+  delayed()->nop();
+}
+
+RegisterOrConstant MacroAssembler::argument_offset(RegisterOrConstant arg_slot,
+                                                   int extra_slot_offset) {
+  // cf. TemplateTable::prepare_invoke(), if (load_receiver).
+  int stackElementSize = Interpreter::stackElementWords() * wordSize;
+  int offset = Interpreter::expr_offset_in_bytes(extra_slot_offset+0);
+  int offset1 = Interpreter::expr_offset_in_bytes(extra_slot_offset+1);
+  assert(offset1 - offset == stackElementSize, "correct arithmetic");
+  if (arg_slot.is_constant()) {
+    offset += arg_slot.as_constant() * stackElementSize;
+    return offset;
+  } else {
+    Register temp = arg_slot.as_register();
+    sll_ptr(temp, exact_log2(stackElementSize), temp);
+    if (offset != 0)
+      add(temp, offset, temp);
+    return temp;
+  }
+}
+
+
+
 void MacroAssembler::biased_locking_enter(Register obj_reg, Register mark_reg,
                                           Register temp_reg,
                                           Label& done, Label* slow_case,
diff --git a/src/cpu/sparc/vm/assembler_sparc.hpp b/src/cpu/sparc/vm/assembler_sparc.hpp
index 5c756c4b6..cfd935987 100644
--- a/src/cpu/sparc/vm/assembler_sparc.hpp
+++ b/src/cpu/sparc/vm/assembler_sparc.hpp
@@ -84,6 +84,10 @@ REGISTER_DECLARATION(Register, G4_scratch    , G4);
 
 REGISTER_DECLARATION(Register, Gtemp  , G5);
 
+// JSR 292 fixed register usages:
+REGISTER_DECLARATION(Register, G5_method_type        , G5);
+REGISTER_DECLARATION(Register, G3_method_handle      , G3);
+
 // The compiler requires that G5_megamorphic_method is G5_inline_cache_klass,
 // because a single patchable "set" instruction (NativeMovConstReg,
 // or NativeMovConstPatching for compiler1) instruction
@@ -91,9 +95,13 @@ REGISTER_DECLARATION(Register, Gtemp  , G5);
 // call site is an inline cache or is megamorphic.  See the function
 // CompiledIC::set_to_megamorphic.
 //
-// On the other hand, G5_inline_cache_klass must differ from G5_method,
-// because both registers are needed for an inline cache that calls
-// an interpreted method.
+// If a inline cache targets an interpreted method, then the
+// G5 register will be used twice during the call.  First,
+// the call site will be patched to load a compiledICHolder
+// into G5. (This is an ordered pair of ic_klass, method.)
+// The c2i adapter will first check the ic_klass, then load
+// G5_method with the method part of the pair just before
+// jumping into the interpreter.
 //
 // Note that G5_method is only the method-self for the interpreter,
 // and is logically unrelated to G5_megamorphic_method.
@@ -1931,6 +1939,7 @@ class MacroAssembler: public Assembler {
   inline void store_ptr_contents( Register s, Address& a, int offset = 0 );
   inline void jumpl_to( Address& a, Register d, int offset = 0 );
   inline void jump_to(  Address& a,             int offset = 0 );
+  inline void jump_indirect_to(  Address& a, Register temp, int ld_offset = 0, int jmp_offset = 0 );
 
   // ring buffer traceable jumps
 
@@ -2366,6 +2375,16 @@ class MacroAssembler: public Assembler {
                            Register temp2_reg,
                            Label& L_success);
 
+  // method handles (JSR 292)
+  void check_method_handle_type(Register mtype_reg, Register mh_reg,
+                                Register temp_reg,
+                                Label& wrong_method_type);
+  void jump_to_method_handle_entry(Register mh_reg, Register temp_reg);
+  // offset relative to Gargs of argument at tos[arg_slot].
+  // (arg_slot == 0 means the last argument, not the first).
+  RegisterOrConstant argument_offset(RegisterOrConstant arg_slot,
+                                     int extra_slot_offset = 0);
+
 
   // Stack overflow checking
 
diff --git a/src/cpu/sparc/vm/assembler_sparc.inline.hpp b/src/cpu/sparc/vm/assembler_sparc.inline.hpp
index d9053f7f6..692c8650b 100644
--- a/src/cpu/sparc/vm/assembler_sparc.inline.hpp
+++ b/src/cpu/sparc/vm/assembler_sparc.inline.hpp
@@ -671,6 +671,15 @@ inline void MacroAssembler::jump_to( Address& a, int offset ) {
 }
 
 
+inline void MacroAssembler::jump_indirect_to(  Address& a, Register temp,
+                                               int ld_offset, int jmp_offset ) {
+  assert_not_delayed();
+  //sethi(a);                   // sethi is caller responsibility for this one
+  ld_ptr(a, temp, ld_offset);
+  jmp(temp, jmp_offset);
+}
+
+
 inline void MacroAssembler::set_oop( jobject obj, Register d ) {
   set_oop(allocate_oop_address(obj, d));
 }
diff --git a/src/cpu/sparc/vm/cppInterpreter_sparc.cpp b/src/cpu/sparc/vm/cppInterpreter_sparc.cpp
index f2d6486f8..c89d6e169 100644
--- a/src/cpu/sparc/vm/cppInterpreter_sparc.cpp
+++ b/src/cpu/sparc/vm/cppInterpreter_sparc.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright 2007-2008 Sun Microsystems, Inc.  All Rights Reserved.
+ * Copyright 2007-2009 Sun Microsystems, Inc.  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
@@ -1017,6 +1017,7 @@ void CppInterpreterGenerator::generate_compute_interpreter_state(const Register
   const int slop_factor = 2*wordSize;
 
   const int fixed_size = ((sizeof(BytecodeInterpreter) + slop_factor) >> LogBytesPerWord) + // what is the slop factor?
+                         //6815692//methodOopDesc::extra_stack_words() +  // extra push slots for MH adapters
                          frame::memory_parameter_word_sp_offset +  // register save area + param window
                          (native ?  frame::interpreter_frame_extra_outgoing_argument_words : 0); // JNI, class
 
@@ -1163,6 +1164,9 @@ void CppInterpreterGenerator::generate_compute_interpreter_state(const Register
   __ st_ptr(O2, XXX_STATE(_stack));                // PREPUSH
 
   __ lduh(max_stack, O3);                      // Full size expression stack
+  guarantee(!EnableMethodHandles, "no support yet for java.dyn.MethodHandle"); //6815692
+  //6815692//if (EnableMethodHandles)
+  //6815692//  __ inc(O3, methodOopDesc::extra_stack_entries());
   __ sll(O3, LogBytesPerWord, O3);
   __ sub(O2, O3, O3);
 //  __ sub(O3, wordSize, O3);                    // so prepush doesn't look out of bounds
@@ -2017,7 +2021,9 @@ static int size_activation_helper(int callee_extra_locals, int max_stack, int mo
 
   const int fixed_size = sizeof(BytecodeInterpreter)/wordSize +           // interpreter state object
                          frame::memory_parameter_word_sp_offset;   // register save area + param window
+  const int extra_stack = 0; //6815692//methodOopDesc::extra_stack_entries();
   return (round_to(max_stack +
+                   extra_stack +
                    slop_factor +
                    fixed_size +
                    monitor_size +
@@ -2104,7 +2110,8 @@ void BytecodeInterpreter::layout_interpreterState(interpreterState to_fill,
   // Need +1 here because stack_base points to the word just above the first expr stack entry
   // and stack_limit is supposed to point to the word just below the last expr stack entry.
   // See generate_compute_interpreter_state.
-  to_fill->_stack_limit = stack_base - (method->max_stack() + 1);
+  int extra_stack = 0; //6815692//methodOopDesc::extra_stack_entries();
+  to_fill->_stack_limit = stack_base - (method->max_stack() + 1 + extra_stack);
   to_fill->_monitor_base = (BasicObjectLock*) monitor_base;
 
   // sparc specific
diff --git a/src/cpu/sparc/vm/interpreterGenerator_sparc.hpp b/src/cpu/sparc/vm/interpreterGenerator_sparc.hpp
index 3352d2abd..3ed9bc471 100644
--- a/src/cpu/sparc/vm/interpreterGenerator_sparc.hpp
+++ b/src/cpu/sparc/vm/interpreterGenerator_sparc.hpp
@@ -1,5 +1,5 @@
 /*
- * Copyright 1997-2007 Sun Microsystems, Inc.  All Rights Reserved.
+ * Copyright 1997-2009 Sun Microsystems, Inc.  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
@@ -29,6 +29,7 @@
   address generate_normal_entry(bool synchronized);
   address generate_native_entry(bool synchronized);
   address generate_abstract_entry(void);
+  address generate_method_handle_entry(void);
   address generate_math_entry(AbstractInterpreter::MethodKind kind);
   address generate_empty_entry(void);
   address generate_accessor_entry(void);
diff --git a/src/cpu/sparc/vm/interpreter_sparc.cpp b/src/cpu/sparc/vm/interpreter_sparc.cpp
index 898d37060..9f0dd7166 100644
--- a/src/cpu/sparc/vm/interpreter_sparc.cpp
+++ b/src/cpu/sparc/vm/interpreter_sparc.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright 1997-2007 Sun Microsystems, Inc.  All Rights Reserved.
+ * Copyright 1997-2009 Sun Microsystems, Inc.  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
@@ -235,6 +235,19 @@ address InterpreterGenerator::generate_abstract_entry(void) {
 }
 
 
+
+// Method handle invoker
+// Dispatch a method of the form java.dyn.MethodHandles::invoke(...)
+address InterpreterGenerator::generate_method_handle_entry(void) {
+  if (!EnableMethodHandles) {
+    return generate_abstract_entry();
+  }
+  return generate_abstract_entry(); //6815692//
+}
+
+
+
+
 //----------------------------------------------------------------------------------------------------
 // Entry points & stack frame layout
 //
@@ -364,6 +377,7 @@ address AbstractInterpreterGenerator::generate_method_entry(AbstractInterpreter:
     case Interpreter::empty                  : entry_point = ((InterpreterGenerator*)this)->generate_empty_entry();        break;
     case Interpreter::accessor               : entry_point = ((InterpreterGenerator*)this)->generate_accessor_entry();     break;
     case Interpreter::abstract               : entry_point = ((InterpreterGenerator*)this)->generate_abstract_entry();     break;
+    case Interpreter::method_handle          : entry_point = ((InterpreterGenerator*)this)->generate_method_handle_entry(); break;
     case Interpreter::java_lang_math_sin     :                                                                             break;
     case Interpreter::java_lang_math_cos     :                                                                             break;
     case Interpreter::java_lang_math_tan     :                                                                             break;
diff --git a/src/cpu/sparc/vm/methodHandles_sparc.cpp b/src/cpu/sparc/vm/methodHandles_sparc.cpp
new file mode 100644
index 000000000..dd40006e1
--- /dev/null
+++ b/src/cpu/sparc/vm/methodHandles_sparc.cpp
@@ -0,0 +1,70 @@
+/*
+ * Copyright 1997-2009 Sun Microsystems, Inc.  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 Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
+ * CA 95054 USA or visit www.sun.com if you need additional information or
+ * have any questions.
+ *
+ */
+
+#include "incls/_precompiled.incl"
+#include "incls/_methodHandles_sparc.cpp.incl"
+
+#define __ _masm->
+
+address MethodHandleEntry::start_compiled_entry(MacroAssembler* _masm,
+                                                address interpreted_entry) {
+  __ align(wordSize);
+  address target = __ pc() + sizeof(Data);
+  while (__ pc() < target) {
+    __ nop();
+    __ align(wordSize);
+  }
+
+  MethodHandleEntry* me = (MethodHandleEntry*) __ pc();
+  me->set_end_address(__ pc());         // set a temporary end_address
+  me->set_from_interpreted_entry(interpreted_entry);
+  me->set_type_checking_entry(NULL);
+
+  return (address) me;
+}
+
+MethodHandleEntry* MethodHandleEntry::finish_compiled_entry(MacroAssembler* _masm,
+                                                address start_addr) {
+  MethodHandleEntry* me = (MethodHandleEntry*) start_addr;
+  assert(me->end_address() == start_addr, "valid ME");
+
+  // Fill in the real end_address:
+  __ align(wordSize);
+  me->set_end_address(__ pc());
+
+  return me;
+}
+
+
+// Code generation
+address MethodHandles::generate_method_handle_interpreter_entry(MacroAssembler* _masm) {
+  ShouldNotReachHere(); //NYI, 6815692
+  return NULL;
+}
+
+// Generate an "entry" field for a method handle.
+// This determines how the method handle will respond to calls.
+void MethodHandles::generate_method_handle_stub(MacroAssembler* _masm, MethodHandles::EntryKind ek) {
+  ShouldNotReachHere(); //NYI, 6815692
+}
diff --git a/src/cpu/sparc/vm/register_definitions_sparc.cpp b/src/cpu/sparc/vm/register_definitions_sparc.cpp
index d3e805570..410f62aeb 100644
--- a/src/cpu/sparc/vm/register_definitions_sparc.cpp
+++ b/src/cpu/sparc/vm/register_definitions_sparc.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright 2002-2008 Sun Microsystems, Inc.  All Rights Reserved.
+ * Copyright 2002-2009 Sun Microsystems, Inc.  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
@@ -142,6 +142,8 @@ REGISTER_DEFINITION(Register, G1_scratch);
 REGISTER_DEFINITION(Register, G3_scratch);
 REGISTER_DEFINITION(Register, G4_scratch);
 REGISTER_DEFINITION(Register, Gtemp);
+REGISTER_DEFINITION(Register, G5_method_type);
+REGISTER_DEFINITION(Register, G3_method_handle);
 REGISTER_DEFINITION(Register, Lentry_args);
 
 #ifdef CC_INTERP
diff --git a/src/cpu/sparc/vm/sharedRuntime_sparc.cpp b/src/cpu/sparc/vm/sharedRuntime_sparc.cpp
index 5804fbbfe..e2bd04436 100644
--- a/src/cpu/sparc/vm/sharedRuntime_sparc.cpp
+++ b/src/cpu/sparc/vm/sharedRuntime_sparc.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright 2003-2008 Sun Microsystems, Inc.  All Rights Reserved.
+ * Copyright 2003-2009 Sun Microsystems, Inc.  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
@@ -937,12 +937,12 @@ void AdapterGenerator::gen_i2c_adapter(
   // Inputs:
   // G2_thread      - TLS
   // G5_method      - Method oop
-  // O0             - Flag telling us to restore SP from O5
-  // O4_args        - Pointer to interpreter's args
-  // O5             - Caller's saved SP, to be restored if needed
+  // G4 (Gargs)     - Pointer to interpreter's args
+  // O0..O4         - free for scratch
+  // O5_savedSP     - Caller's saved SP, to be restored if needed
   // O6             - Current SP!
   // O7             - Valid return address
-  // L0-L7, I0-I7    - Caller's temps (no frame pushed yet)
+  // L0-L7, I0-I7   - Caller's temps (no frame pushed yet)
 
   // Outputs:
   // G2_thread      - TLS
@@ -954,7 +954,7 @@ void AdapterGenerator::gen_i2c_adapter(
   // F0-F7          - more outgoing args
 
 
-  // O4 is about to get loaded up with compiled callee's args
+  // Gargs is the incoming argument base, and also an outgoing argument.
   __ sub(Gargs, BytesPerWord, Gargs);
 
 #ifdef ASSERT
diff --git a/src/cpu/sparc/vm/templateInterpreter_sparc.cpp b/src/cpu/sparc/vm/templateInterpreter_sparc.cpp
index 090242c17..6cbcf9272 100644
--- a/src/cpu/sparc/vm/templateInterpreter_sparc.cpp
+++ b/src/cpu/sparc/vm/templateInterpreter_sparc.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright 1997-2008 Sun Microsystems, Inc.  All Rights Reserved.
+ * Copyright 1997-2009 Sun Microsystems, Inc.  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
@@ -108,6 +108,24 @@ address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
 }
 
 
+// Arguments are: required type in G5_method_type, and
+// failing object (or NULL) in G3_method_handle.
+address TemplateInterpreterGenerator::generate_WrongMethodType_handler() {
+  address entry = __ pc();
+  // expression stack must be empty before entering the VM if an exception
+  // happened
+  __ empty_expression_stack();
+  // load exception object
+  __ call_VM(Oexception,
+             CAST_FROM_FN_PTR(address,
+                              InterpreterRuntime::throw_WrongMethodTypeException),
+             G5_method_type,    // required
+             G3_method_handle); // actual
+  __ should_not_reach_here();
+  return entry;
+}
+
+
 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler(const char* name) {
   address entry = __ pc();
   // expression stack must be empty before entering the VM if an exception happened
@@ -448,6 +466,7 @@ void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
 
   const int extra_space =
     rounded_vm_local_words +                   // frame local scratch space
+    //6815692//methodOopDesc::extra_stack_words() +       // extra push slots for MH adapters
     frame::memory_parameter_word_sp_offset +   // register save area
     (native_call ? frame::interpreter_frame_extra_outgoing_argument_words : 0);
 
@@ -1447,6 +1466,7 @@ static int size_activation_helper(int callee_extra_locals, int max_stack, int mo
        round_to(callee_extra_locals * Interpreter::stackElementWords(), WordsPerLong);
   const int max_stack_words = max_stack * Interpreter::stackElementWords();
   return (round_to((max_stack_words
+                   //6815692//+ methodOopDesc::extra_stack_words()
                    + rounded_vm_local_words
                    + frame::memory_parameter_word_sp_offset), WordsPerLong)
                    // already rounded
diff --git a/src/cpu/x86/vm/assembler_x86.cpp b/src/cpu/x86/vm/assembler_x86.cpp
index 35ee7e540..be4c842bb 100644
--- a/src/cpu/x86/vm/assembler_x86.cpp
+++ b/src/cpu/x86/vm/assembler_x86.cpp
@@ -7609,6 +7609,83 @@ RegisterOrConstant MacroAssembler::delayed_value_impl(intptr_t* delayed_value_ad
 }
 
 
+// registers on entry:
+//  - rax ('check' register): required MethodType
+//  - rcx: method handle
+//  - rdx, rsi, or ?: killable temp
+void MacroAssembler::check_method_handle_type(Register mtype_reg, Register mh_reg,
+                                              Register temp_reg,
+                                              Label& wrong_method_type) {
+  if (UseCompressedOops)  unimplemented();  // field accesses must decode
+  // compare method type against that of the receiver
+  cmpptr(mtype_reg, Address(mh_reg, delayed_value(java_dyn_MethodHandle::type_offset_in_bytes, temp_reg)));
+  jcc(Assembler::notEqual, wrong_method_type);
+}
+
+
+// A method handle has a "vmslots" field which gives the size of its
+// argument list in JVM stack slots.  This field is either located directly
+// in every method handle, or else is indirectly accessed through the
+// method handle's MethodType.  This macro hides the distinction.
+void MacroAssembler::load_method_handle_vmslots(Register vmslots_reg, Register mh_reg,
+                                                Register temp_reg) {
+  if (UseCompressedOops)  unimplemented();  // field accesses must decode
+  // load mh.type.form.vmslots
+  if (java_dyn_MethodHandle::vmslots_offset_in_bytes() != 0) {
+    // hoist vmslots into every mh to avoid dependent load chain
+    movl(vmslots_reg, Address(mh_reg, delayed_value(java_dyn_MethodHandle::vmslots_offset_in_bytes, temp_reg)));
+  } else {
+    Register temp2_reg = vmslots_reg;
+    movptr(temp2_reg, Address(mh_reg,    delayed_value(java_dyn_MethodHandle::type_offset_in_bytes, temp_reg)));
+    movptr(temp2_reg, Address(temp2_reg, delayed_value(java_dyn_MethodType::form_offset_in_bytes, temp_reg)));
+    movl(vmslots_reg, Address(temp2_reg, delayed_value(java_dyn_MethodTypeForm::vmslots_offset_in_bytes, temp_reg)));
+  }
+}
+
+
+// registers on entry:
+//  - rcx: method handle
+//  - rdx: killable temp (interpreted only)
+//  - rax: killable temp (compiled only)
+void MacroAssembler::jump_to_method_handle_entry(Register mh_reg, Register temp_reg) {
+  assert(mh_reg == rcx, "caller must put MH object in rcx");
+  assert_different_registers(mh_reg, temp_reg);
+
+  if (UseCompressedOops)  unimplemented();  // field accesses must decode
+
+  // pick out the interpreted side of the handler
+  movptr(temp_reg, Address(mh_reg, delayed_value(java_dyn_MethodHandle::vmentry_offset_in_bytes, temp_reg)));
+
+  // off we go...
+  jmp(Address(temp_reg, MethodHandleEntry::from_interpreted_entry_offset_in_bytes()));
+
+  // for the various stubs which take control at this point,
+  // see MethodHandles::generate_method_handle_stub
+}
+
+
+Address MacroAssembler::argument_address(RegisterOrConstant arg_slot,
+                                         int extra_slot_offset) {
+  // cf. TemplateTable::prepare_invoke(), if (load_receiver).
+  int stackElementSize = Interpreter::stackElementSize();
+  int offset = Interpreter::expr_offset_in_bytes(extra_slot_offset+0);
+#ifdef ASSERT
+  int offset1 = Interpreter::expr_offset_in_bytes(extra_slot_offset+1);
+  assert(offset1 - offset == stackElementSize, "correct arithmetic");
+#endif
+  Register             scale_reg    = noreg;
+  Address::ScaleFactor scale_factor = Address::no_scale;
+  if (arg_slot.is_constant()) {
+    offset += arg_slot.as_constant() * stackElementSize;
+  } else {
+    scale_reg    = arg_slot.as_register();
+    scale_factor = Address::times(stackElementSize);
+  }
+  offset += wordSize;           // return PC is on stack
+  return Address(rsp, scale_reg, scale_factor, offset);
+}
+
+
 void MacroAssembler::verify_oop_addr(Address addr, const char* s) {
   if (!VerifyOops) return;
 
diff --git a/src/cpu/x86/vm/assembler_x86.hpp b/src/cpu/x86/vm/assembler_x86.hpp
index 89ac6dd53..92b55b9de 100644
--- a/src/cpu/x86/vm/assembler_x86.hpp
+++ b/src/cpu/x86/vm/assembler_x86.hpp
@@ -1857,6 +1857,16 @@ class MacroAssembler: public Assembler {
                            Register temp_reg,
                            Label& L_success);
 
+  // method handles (JSR 292)
+  void check_method_handle_type(Register mtype_reg, Register mh_reg,
+                                Register temp_reg,
+                                Label& wrong_method_type);
+  void load_method_handle_vmslots(Register vmslots_reg, Register mh_reg,
+                                  Register temp_reg);
+  void jump_to_method_handle_entry(Register mh_reg, Register temp_reg);
+  Address argument_address(RegisterOrConstant arg_slot, int extra_slot_offset = 0);
+
+
   //----
   void set_word_if_not_zero(Register reg); // sets reg to 1 if not zero, otherwise 0
 
diff --git a/src/cpu/x86/vm/cppInterpreter_x86.cpp b/src/cpu/x86/vm/cppInterpreter_x86.cpp
index 31afca773..98af0c244 100644
--- a/src/cpu/x86/vm/cppInterpreter_x86.cpp
+++ b/src/cpu/x86/vm/cppInterpreter_x86.cpp
@@ -513,10 +513,11 @@ void CppInterpreterGenerator::generate_compute_interpreter_state(const Register
     // compute full expression stack limit
 
     const Address size_of_stack    (rbx, methodOopDesc::max_stack_offset());
+    const int extra_stack = 0; //6815692//methodOopDesc::extra_stack_words();
     __ load_unsigned_short(rdx, size_of_stack);                           // get size of expression stack in words
     __ negptr(rdx);                                                       // so we can subtract in next step
     // Allocate expression stack
-    __ lea(rsp, Address(rsp, rdx, Address::times_ptr));
+    __ lea(rsp, Address(rsp, rdx, Address::times_ptr, -extra_stack));
     __ movptr(STATE(_stack_limit), rsp);
   }
 
@@ -659,8 +660,9 @@ void InterpreterGenerator::generate_stack_overflow_check(void) {
     // Always give one monitor to allow us to start interp if sync method.
     // Any additional monitors need a check when moving the expression stack
     const int one_monitor = frame::interpreter_frame_monitor_size() * wordSize;
+    const int extra_stack = 0; //6815692//methodOopDesc::extra_stack_entries();
   __ load_unsigned_short(rax, size_of_stack);                           // get size of expression stack in words
-  __ lea(rax, Address(noreg, rax, Interpreter::stackElementScale(), one_monitor));
+  __ lea(rax, Address(noreg, rax, Interpreter::stackElementScale(), extra_stack + one_monitor));
   __ lea(rax, Address(rax, rdx, Interpreter::stackElementScale(), overhead_size));
 
 #ifdef ASSERT
@@ -2185,6 +2187,7 @@ address AbstractInterpreterGenerator::generate_method_entry(AbstractInterpreter:
     case Interpreter::empty                  : entry_point = ((InterpreterGenerator*)this)->generate_empty_entry();        break;
     case Interpreter::accessor               : entry_point = ((InterpreterGenerator*)this)->generate_accessor_entry();     break;
     case Interpreter::abstract               : entry_point = ((InterpreterGenerator*)this)->generate_abstract_entry();     break;
+    case Interpreter::method_handle          : entry_point = ((InterpreterGenerator*)this)->generate_method_handle_entry(); break;
 
     case Interpreter::java_lang_math_sin     : // fall thru
     case Interpreter::java_lang_math_cos     : // fall thru
@@ -2224,7 +2227,8 @@ int AbstractInterpreter::size_top_interpreter_activation(methodOop method) {
   const int overhead_size = sizeof(BytecodeInterpreter)/wordSize +
     ( frame::sender_sp_offset - frame::link_offset) + 2;
 
-  const int method_stack = (method->max_locals() + method->max_stack()) *
+  const int extra_stack = 0; //6815692//methodOopDesc::extra_stack_entries();
+  const int method_stack = (method->max_locals() + method->max_stack() + extra_stack) *
                            Interpreter::stackElementWords();
   return overhead_size + method_stack + stub_code;
 }
@@ -2289,7 +2293,8 @@ void BytecodeInterpreter::layout_interpreterState(interpreterState to_fill,
   // Need +1 here because stack_base points to the word just above the first expr stack entry
   // and stack_limit is supposed to point to the word just below the last expr stack entry.
   // See generate_compute_interpreter_state.
-  to_fill->_stack_limit = stack_base - (method->max_stack() + 1);
+  int extra_stack = 0; //6815692//methodOopDesc::extra_stack_entries();
+  to_fill->_stack_limit = stack_base - (method->max_stack() + extra_stack + 1);
   to_fill->_monitor_base = (BasicObjectLock*) monitor_base;
 
   to_fill->_self_link = to_fill;
@@ -2335,7 +2340,8 @@ int AbstractInterpreter::layout_activation(methodOop method,
                                                 monitor_size);
 
   // Now with full size expression stack
-  int full_frame_size = short_frame_size + method->max_stack() * BytesPerWord;
+  int extra_stack = 0; //6815692//methodOopDesc::extra_stack_entries();
+  int full_frame_size = short_frame_size + (method->max_stack() + extra_stack) * BytesPerWord;
 
   // and now with only live portion of the expression stack
   short_frame_size = short_frame_size + tempcount * BytesPerWord;
diff --git a/src/cpu/x86/vm/interp_masm_x86_32.cpp b/src/cpu/x86/vm/interp_masm_x86_32.cpp
index bb14db905..413e59735 100644
--- a/src/cpu/x86/vm/interp_masm_x86_32.cpp
+++ b/src/cpu/x86/vm/interp_masm_x86_32.cpp
@@ -555,13 +555,18 @@ void InterpreterMacroAssembler::super_call_VM_leaf(address entry_point, Register
 }
 
 
-// Jump to from_interpreted entry of a call unless single stepping is possible
-// in this thread in which case we must call the i2i entry
-void InterpreterMacroAssembler::jump_from_interpreted(Register method, Register temp) {
+void InterpreterMacroAssembler::prepare_to_jump_from_interpreted() {
   // set sender sp
   lea(rsi, Address(rsp, wordSize));
   // record last_sp
   movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), rsi);
+}
+
+
+// Jump to from_interpreted entry of a call unless single stepping is possible
+// in this thread in which case we must call the i2i entry
+void InterpreterMacroAssembler::jump_from_interpreted(Register method, Register temp) {
+  prepare_to_jump_from_interpreted();
 
   if (JvmtiExport::can_post_interpreter_events()) {
     Label run_compiled_code;
diff --git a/src/cpu/x86/vm/interp_masm_x86_32.hpp b/src/cpu/x86/vm/interp_masm_x86_32.hpp
index dc6c7bce2..3500b5fbd 100644
--- a/src/cpu/x86/vm/interp_masm_x86_32.hpp
+++ b/src/cpu/x86/vm/interp_masm_x86_32.hpp
@@ -161,6 +161,7 @@ class InterpreterMacroAssembler: public MacroAssembler {
 
 
   // jump to an invoked target
+  void prepare_to_jump_from_interpreted();
   void jump_from_interpreted(Register method, Register temp);
 
   // Returning from interpreted functions
diff --git a/src/cpu/x86/vm/interp_masm_x86_64.cpp b/src/cpu/x86/vm/interp_masm_x86_64.cpp
index 56bd0c8c9..dbd858da2 100644
--- a/src/cpu/x86/vm/interp_masm_x86_64.cpp
+++ b/src/cpu/x86/vm/interp_masm_x86_64.cpp
@@ -551,13 +551,18 @@ void InterpreterMacroAssembler::super_call_VM_leaf(address entry_point,
   MacroAssembler::call_VM_leaf_base(entry_point, 3);
 }
 
-// Jump to from_interpreted entry of a call unless single stepping is possible
-// in this thread in which case we must call the i2i entry
-void InterpreterMacroAssembler::jump_from_interpreted(Register method, Register temp) {
+void InterpreterMacroAssembler::prepare_to_jump_from_interpreted() {
   // set sender sp
   lea(r13, Address(rsp, wordSize));
   // record last_sp
   movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), r13);
+}
+
+
+// Jump to from_interpreted entry of a call unless single stepping is possible
+// in this thread in which case we must call the i2i entry
+void InterpreterMacroAssembler::jump_from_interpreted(Register method, Register temp) {
+  prepare_to_jump_from_interpreted();
 
   if (JvmtiExport::can_post_interpreter_events()) {
     Label run_compiled_code;
diff --git a/src/cpu/x86/vm/interp_masm_x86_64.hpp b/src/cpu/x86/vm/interp_masm_x86_64.hpp
index 62396c94b..c35cb3a19 100644
--- a/src/cpu/x86/vm/interp_masm_x86_64.hpp
+++ b/src/cpu/x86/vm/interp_masm_x86_64.hpp
@@ -1,5 +1,5 @@
 /*
- * Copyright 2003-2008 Sun Microsystems, Inc.  All Rights Reserved.
+ * Copyright 2003-2009 Sun Microsystems, Inc.  All Rights Reserved.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
  * This code is free software; you can redistribute it and/or modify it
@@ -176,6 +176,7 @@ class InterpreterMacroAssembler: public MacroAssembler {
   void dispatch_via (TosState state, address* table);
 
   // jump to an invoked target
+  void prepare_to_jump_from_interpreted();
   void jump_from_interpreted(Register method, Register temp);
 
 
diff --git a/src/cpu/x86/vm/interpreterGenerator_x86.hpp b/src/cpu/x86/vm/interpreterGenerator_x86.hpp
index 066d89d5f..168579511 100644
--- a/src/cpu/x86/vm/interpreterGenerator_x86.hpp
+++ b/src/cpu/x86/vm/interpreterGenerator_x86.hpp
@@ -1,5 +1,5 @@
 /*
- * Copyright 1997-2007 Sun Microsystems, Inc.  All Rights Reserved.
+ * Copyright 1997-2009 Sun Microsystems, Inc.  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
@@ -32,6 +32,7 @@
   address generate_normal_entry(bool synchronized);
   address generate_native_entry(bool synchronized);
   address generate_abstract_entry(void);
+  address generate_method_handle_entry(void);
   address generate_math_entry(AbstractInterpreter::MethodKind kind);
   address generate_empty_entry(void);
   address generate_accessor_entry(void);
diff --git a/src/cpu/x86/vm/interpreter_x86_32.cpp b/src/cpu/x86/vm/interpreter_x86_32.cpp
index 221cd8561..2db290c43 100644
--- a/src/cpu/x86/vm/interpreter_x86_32.cpp
+++ b/src/cpu/x86/vm/interpreter_x86_32.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright 1997-2008 Sun Microsystems, Inc.  All Rights Reserved.
+ * Copyright 1997-2009 Sun Microsystems, Inc.  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
@@ -201,11 +201,12 @@ address InterpreterGenerator::generate_abstract_entry(void) {
   address entry_point = __ pc();
 
   // abstract method entry
-  // remove return address. Not really needed, since exception handling throws away expression stack
-  __ pop(rbx);
 
-  // adjust stack to what a normal return would do
-  __ mov(rsp, rsi);
+  //  pop return address, reset last_sp to NULL
+  __ empty_expression_stack();
+  __ restore_bcp();      // rsi must be correct for exception handler   (was destroyed)
+  __ restore_locals();   // make sure locals pointer is correct as well (was destroyed)
+
   // throw exception
   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError));
   // the call_VM checks for exception, so we should never return here.
@@ -214,6 +215,20 @@ address InterpreterGenerator::generate_abstract_entry(void) {
   return entry_point;
 }
 
+
+// Method handle invoker
+// Dispatch a method of the form java.dyn.MethodHandles::invoke(...)
+address InterpreterGenerator::generate_method_handle_entry(void) {
+  if (!EnableMethodHandles) {
+    return generate_abstract_entry();
+  }
+
+  address entry_point = MethodHandles::generate_method_handle_interpreter_entry(_masm);
+
+  return entry_point;
+}
+
+
 // This method tells the deoptimizer how big an interpreted frame must be:
 int AbstractInterpreter::size_activation(methodOop method,
                                          int tempcount,
diff --git a/src/cpu/x86/vm/interpreter_x86_64.cpp b/src/cpu/x86/vm/interpreter_x86_64.cpp
index 34e5b56b8..c3cbf56ca 100644
--- a/src/cpu/x86/vm/interpreter_x86_64.cpp
+++ b/src/cpu/x86/vm/interpreter_x86_64.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright 2003-2008 Sun Microsystems, Inc.  All Rights Reserved.
+ * Copyright 2003-2009 Sun Microsystems, Inc.  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
@@ -294,6 +294,16 @@ address InterpreterGenerator::generate_abstract_entry(void) {
 }
 
 
+// Method handle invoker
+// Dispatch a method of the form java.dyn.MethodHandles::invoke(...)
+address InterpreterGenerator::generate_method_handle_entry(void) {
+  if (!EnableMethodHandles) {
+    return generate_abstract_entry();
+  }
+  return generate_abstract_entry(); //6815692//
+}
+
+
 // Empty method, generate a very fast return.
 
 address InterpreterGenerator::generate_empty_entry(void) {
diff --git a/src/cpu/x86/vm/methodHandles_x86.cpp b/src/cpu/x86/vm/methodHandles_x86.cpp
new file mode 100644
index 000000000..24210bf5d
--- /dev/null
+++ b/src/cpu/x86/vm/methodHandles_x86.cpp
@@ -0,0 +1,1133 @@
+/*
+ * Copyright 1997-2009 Sun Microsystems, Inc.  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 Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
+ * CA 95054 USA or visit www.sun.com if you need additional information or
+ * have any questions.
+ *
+ */
+
+#include "incls/_precompiled.incl"
+#include "incls/_methodHandles_x86.cpp.incl"
+
+#define __ _masm->
+
+address MethodHandleEntry::start_compiled_entry(MacroAssembler* _masm,
+                                                address interpreted_entry) {
+  // Just before the actual machine code entry point, allocate space
+  // for a MethodHandleEntry::Data record, so that we can manage everything
+  // from one base pointer.
+  __ align(wordSize);
+  address target = __ pc() + sizeof(Data);
+  while (__ pc() < target) {
+    __ nop();
+    __ align(wordSize);
+  }
+
+  MethodHandleEntry* me = (MethodHandleEntry*) __ pc();
+  me->set_end_address(__ pc());         // set a temporary end_address
+  me->set_from_interpreted_entry(interpreted_entry);
+  me->set_type_checking_entry(NULL);
+
+  return (address) me;
+}
+
+MethodHandleEntry* MethodHandleEntry::finish_compiled_entry(MacroAssembler* _masm,
+                                                address start_addr) {
+  MethodHandleEntry* me = (MethodHandleEntry*) start_addr;
+  assert(me->end_address() == start_addr, "valid ME");
+
+  // Fill in the real end_address:
+  __ align(wordSize);
+  me->set_end_address(__ pc());
+
+  return me;
+}
+
+#ifdef ASSERT
+static void verify_argslot(MacroAssembler* _masm, Register rax_argslot,
+                           const char* error_message) {
+  // Verify that argslot lies within (rsp, rbp].
+  Label L_ok, L_bad;
+  __ cmpptr(rax_argslot, rbp);
+  __ jcc(Assembler::above, L_bad);
+  __ cmpptr(rsp, rax_argslot);
+  __ jcc(Assembler::below, L_ok);
+  __ bind(L_bad);
+  __ stop(error_message);
+  __ bind(L_ok);
+}
+#endif
+
+
+// Code generation
+address MethodHandles::generate_method_handle_interpreter_entry(MacroAssembler* _masm) {
+  // rbx: methodOop
+  // rcx: receiver method handle (must load from sp[MethodTypeForm.vmslots])
+  // rsi/r13: sender SP (must preserve; see prepare_to_jump_from_interpreted)
+  // rdx: garbage temp, blown away
+
+  Register rbx_method = rbx;
+  Register rcx_recv   = rcx;
+  Register rax_mtype  = rax;
+  Register rdx_temp   = rdx;
+
+  // emit WrongMethodType path first, to enable jccb back-branch from main path
+  Label wrong_method_type;
+  __ bind(wrong_method_type);
+  __ push(rax_mtype);       // required mtype
+  __ push(rcx_recv);        // bad mh (1st stacked argument)
+  __ jump(ExternalAddress(Interpreter::throw_WrongMethodType_entry()));
+
+  // here's where control starts out:
+  __ align(CodeEntryAlignment);
+  address entry_point = __ pc();
+
+  // fetch the MethodType from the method handle into rax (the 'check' register)
+  {
+    Register tem = rbx_method;
+    for (jint* pchase = methodOopDesc::method_type_offsets_chain(); (*pchase) != -1; pchase++) {
+      __ movptr(rax_mtype, Address(tem, *pchase));
+      tem = rax_mtype;          // in case there is another indirection
+    }
+  }
+  Register rbx_temp = rbx_method; // done with incoming methodOop
+
+  // given the MethodType, find out where the MH argument is buried
+  __ movptr(rdx_temp, Address(rax_mtype,
+                              __ delayed_value(java_dyn_MethodType::form_offset_in_bytes, rbx_temp)));
+  __ movl(rdx_temp, Address(rdx_temp,
+                            __ delayed_value(java_dyn_MethodTypeForm::vmslots_offset_in_bytes, rbx_temp)));
+  __ movptr(rcx_recv, __ argument_address(rdx_temp));
+
+  __ check_method_handle_type(rax_mtype, rcx_recv, rdx_temp, wrong_method_type);
+  __ jump_to_method_handle_entry(rcx_recv, rdx_temp);
+
+  return entry_point;
+}
+
+// Helper to insert argument slots into the stack.
+// arg_slots must be a multiple of stack_move_unit() and <= 0
+void MethodHandles::insert_arg_slots(MacroAssembler* _masm,
+                                     RegisterOrConstant arg_slots,
+                                     int arg_mask,
+                                     Register rax_argslot,
+                                     Register rbx_temp, Register rdx_temp) {
+  assert_different_registers(rax_argslot, rbx_temp, rdx_temp,
+                             (!arg_slots.is_register() ? rsp : arg_slots.as_register()));
+
+#ifdef ASSERT
+  verify_argslot(_masm, rax_argslot, "insertion point must fall within current frame");
+  if (arg_slots.is_register()) {
+    Label L_ok, L_bad;
+    __ cmpptr(arg_slots.as_register(), (int32_t) NULL_WORD);
+    __ jcc(Assembler::greater, L_bad);
+    __ testl(arg_slots.as_register(), -stack_move_unit() - 1);
+    __ jcc(Assembler::zero, L_ok);
+    __ bind(L_bad);
+    __ stop("assert arg_slots <= 0 and clear low bits");
+    __ bind(L_ok);
+  } else {
+    assert(arg_slots.as_constant() <= 0, "");
+    assert(arg_slots.as_constant() % -stack_move_unit() == 0, "");
+  }
+#endif //ASSERT
+
+#ifdef _LP64
+  if (arg_slots.is_register()) {
+    // clean high bits of stack motion register (was loaded as an int)
+    __ movslq(arg_slots.as_register(), arg_slots.as_register());
+  }
+#endif
+
+  // Make space on the stack for the inserted argument(s).
+  // Then pull down everything shallower than rax_argslot.
+  // The stacked return address gets pulled down with everything else.
+  // That is, copy [rsp, argslot) downward by -size words.  In pseudo-code:
+  //   rsp -= size;
+  //   for (rdx = rsp + size; rdx < argslot; rdx++)
+  //     rdx[-size] = rdx[0]
+  //   argslot -= size;
+  __ mov(rdx_temp, rsp);                        // source pointer for copy
+  __ lea(rsp, Address(rsp, arg_slots, Address::times_ptr));
+  {
+    Label loop;
+    __ bind(loop);
+    // pull one word down each time through the loop
+    __ movptr(rbx_temp, Address(rdx_temp, 0));
+    __ movptr(Address(rdx_temp, arg_slots, Address::times_ptr), rbx_temp);
+    __ addptr(rdx_temp, wordSize);
+    __ cmpptr(rdx_temp, rax_argslot);
+    __ jcc(Assembler::less, loop);
+  }
+
+  // Now move the argslot down, to point to the opened-up space.
+  __ lea(rax_argslot, Address(rax_argslot, arg_slots, Address::times_ptr));
+
+  if (TaggedStackInterpreter && arg_mask != _INSERT_NO_MASK) {
+    // The caller has specified a bitmask of tags to put into the opened space.
+    // This only works when the arg_slots value is an assembly-time constant.
+    int constant_arg_slots = arg_slots.as_constant() / stack_move_unit();
+    int tag_offset = Interpreter::tag_offset_in_bytes() - Interpreter::value_offset_in_bytes();
+    for (int slot = 0; slot < constant_arg_slots; slot++) {
+      BasicType slot_type   = ((arg_mask & (1 << slot)) == 0 ? T_OBJECT : T_INT);
+      int       slot_offset = Interpreter::stackElementSize() * slot;
+      Address   tag_addr(rax_argslot, slot_offset + tag_offset);
+      __ movptr(tag_addr, frame::tag_for_basic_type(slot_type));
+    }
+    // Note that the new argument slots are tagged properly but contain
+    // garbage at this point.  The value portions must be initialized
+    // by the caller.  (Especially references!)
+  }
+}
+
+// Helper to remove argument slots from the stack.
+// arg_slots must be a multiple of stack_move_unit() and >= 0
+void MethodHandles::remove_arg_slots(MacroAssembler* _masm,
+                                    RegisterOrConstant arg_slots,
+                                    Register rax_argslot,
+                                    Register rbx_temp, Register rdx_temp) {
+  assert_different_registers(rax_argslot, rbx_temp, rdx_temp,
+                             (!arg_slots.is_register() ? rsp : arg_slots.as_register()));
+
+#ifdef ASSERT
+  {
+    // Verify that [argslot..argslot+size) lies within (rsp, rbp).
+    Label L_ok, L_bad;
+    __ lea(rbx_temp, Address(rax_argslot, arg_slots, Address::times_ptr));
+    __ cmpptr(rbx_temp, rbp);
+    __ jcc(Assembler::above, L_bad);
+    __ cmpptr(rsp, rax_argslot);
+    __ jcc(Assembler::below, L_ok);
+    __ bind(L_bad);
+    __ stop("deleted argument(s) must fall within current frame");
+    __ bind(L_ok);
+  }
+  if (arg_slots.is_register()) {
+    Label L_ok, L_bad;
+    __ cmpptr(arg_slots.as_register(), (int32_t) NULL_WORD);
+    __ jcc(Assembler::less, L_bad);
+    __ testl(arg_slots.as_register(), -stack_move_unit() - 1);
+    __ jcc(Assembler::zero, L_ok);
+    __ bind(L_bad);
+    __ stop("assert arg_slots >= 0 and clear low bits");
+    __ bind(L_ok);
+  } else {
+    assert(arg_slots.as_constant() >= 0, "");
+    assert(arg_slots.as_constant() % -stack_move_unit() == 0, "");
+  }
+#endif //ASSERT
+
+#ifdef _LP64
+  if (false) {                  // not needed, since register is positive
+    // clean high bits of stack motion register (was loaded as an int)
+    if (arg_slots.is_register())
+      __ movslq(arg_slots.as_register(), arg_slots.as_register());
+  }
+#endif
+
+  // Pull up everything shallower than rax_argslot.
+  // Then remove the excess space on the stack.
+  // The stacked return address gets pulled up with everything else.
+  // That is, copy [rsp, argslot) upward by size words.  In pseudo-code:
+  //   for (rdx = argslot-1; rdx >= rsp; --rdx)
+  //     rdx[size] = rdx[0]
+  //   argslot += size;
+  //   rsp += size;
+  __ lea(rdx_temp, Address(rax_argslot, -wordSize)); // source pointer for copy
+  {
+    Label loop;
+    __ bind(loop);
+    // pull one word up each time through the loop
+    __ movptr(rbx_temp, Address(rdx_temp, 0));
+    __ movptr(Address(rdx_temp, arg_slots, Address::times_ptr), rbx_temp);
+    __ addptr(rdx_temp, -wordSize);
+    __ cmpptr(rdx_temp, rsp);
+    __ jcc(Assembler::greaterEqual, loop);
+  }
+
+  // Now move the argslot up, to point to the just-copied block.
+  __ lea(rsp, Address(rsp, arg_slots, Address::times_ptr));
+  // And adjust the argslot address to point at the deletion point.
+  __ lea(rax_argslot, Address(rax_argslot, arg_slots, Address::times_ptr));
+}
+
+#ifndef PRODUCT
+void trace_method_handle_stub(const char* adaptername,
+                              oop mh,
+                              intptr_t* entry_sp,
+                              intptr_t* saved_sp) {
+  // called as a leaf from native code: do not block the JVM!
+  printf("MH %s "PTR_FORMAT" "PTR_FORMAT" "INTX_FORMAT"\n", adaptername, mh, entry_sp, entry_sp - saved_sp);
+}
+#endif //PRODUCT
+
+// Generate an "entry" field for a method handle.
+// This determines how the method handle will respond to calls.
+void MethodHandles::generate_method_handle_stub(MacroAssembler* _masm, MethodHandles::EntryKind ek) {
+  // Here is the register state during an interpreted call,
+  // as set up by generate_method_handle_interpreter_entry():
+  // - rbx: garbage temp (was MethodHandle.invoke methodOop, unused)
+  // - rcx: receiver method handle
+  // - rax: method handle type (only used by the check_mtype entry point)
+  // - rsi/r13: sender SP (must preserve; see prepare_to_jump_from_interpreted)
+  // - rdx: garbage temp, can blow away
+
+  Register rcx_recv    = rcx;
+  Register rax_argslot = rax;
+  Register rbx_temp    = rbx;
+  Register rdx_temp    = rdx;
+
+  guarantee(java_dyn_MethodHandle::vmentry_offset_in_bytes() != 0, "must have offsets");
+
+  // some handy addresses
+  Address rbx_method_fie(     rbx,      methodOopDesc::from_interpreted_offset() );
+
+  Address rcx_mh_vmtarget(    rcx_recv, java_dyn_MethodHandle::vmtarget_offset_in_bytes() );
+  Address rcx_dmh_vmindex(    rcx_recv, sun_dyn_DirectMethodHandle::vmindex_offset_in_bytes() );
+
+  Address rcx_bmh_vmargslot(  rcx_recv, sun_dyn_BoundMethodHandle::vmargslot_offset_in_bytes() );
+  Address rcx_bmh_argument(   rcx_recv, sun_dyn_BoundMethodHandle::argument_offset_in_bytes() );
+
+  Address rcx_amh_vmargslot(  rcx_recv, sun_dyn_AdapterMethodHandle::vmargslot_offset_in_bytes() );
+  Address rcx_amh_argument(   rcx_recv, sun_dyn_AdapterMethodHandle::argument_offset_in_bytes() );
+  Address rcx_amh_conversion( rcx_recv, sun_dyn_AdapterMethodHandle::conversion_offset_in_bytes() );
+  Address vmarg;                // __ argument_address(vmargslot)
+
+  int tag_offset = -1;
+  if (TaggedStackInterpreter) {
+    tag_offset = Interpreter::tag_offset_in_bytes() - Interpreter::value_offset_in_bytes();
+    assert(tag_offset = wordSize, "stack grows as expected");
+  }
+
+  if (have_entry(ek)) {
+    __ nop();                   // empty stubs make SG sick
+    return;
+  }
+
+  address interp_entry = __ pc();
+  if (UseCompressedOops)  __ unimplemented("UseCompressedOops");
+
+#ifndef PRODUCT
+  if (TraceMethodHandles) {
+    __ push(rax); __ push(rbx); __ push(rcx); __ push(rdx); __ push(rsi); __ push(rdi);
+    __ lea(rax, Address(rsp, wordSize*6)); // entry_sp
+    // arguments:
+    __ push(rsi);               // saved_sp
+    __ push(rax);               // entry_sp
+    __ push(rcx);               // mh
+    __ push(rcx);
+    __ movptr(Address(rsp, 0), (intptr_t)entry_name(ek));
+    __ call_VM_leaf(CAST_FROM_FN_PTR(address, trace_method_handle_stub), 4);
+    __ pop(rdi); __ pop(rsi); __ pop(rdx); __ pop(rcx); __ pop(rbx); __ pop(rax);
+  }
+#endif //PRODUCT
+
+  switch ((int) ek) {
+  case _check_mtype:
+    {
+      // this stub is special, because it requires a live mtype argument
+      Register rax_mtype = rax;
+
+      // emit WrongMethodType path first, to enable jccb back-branch
+      Label wrong_method_type;
+      __ bind(wrong_method_type);
+      __ movptr(rdx_temp, ExternalAddress((address) &_entries[_wrong_method_type]));
+      __ jmp(Address(rdx_temp, MethodHandleEntry::from_interpreted_entry_offset_in_bytes()));
+      __ hlt();
+
+      interp_entry = __ pc();
+      __ check_method_handle_type(rax_mtype, rcx_recv, rdx_temp, wrong_method_type);
+      // now rax_mtype is dead; subsequent stubs will use it as a temp
+
+      __ jump_to_method_handle_entry(rcx_recv, rdx_temp);
+    }
+    break;
+
+  case _wrong_method_type:
+    {
+      // this stub is special, because it requires a live mtype argument
+      Register rax_mtype = rax;
+
+      interp_entry = __ pc();
+      __ push(rax_mtype);       // required mtype
+      __ push(rcx_recv);        // random mh (1st stacked argument)
+      __ jump(ExternalAddress(Interpreter::throw_WrongMethodType_entry()));
+    }
+    break;
+
+  case _invokestatic_mh:
+  case _invokespecial_mh:
+    {
+      Register rbx_method = rbx_temp;
+      __ movptr(rbx_method, rcx_mh_vmtarget); // target is a methodOop
+      __ verify_oop(rbx_method);
+      // same as TemplateTable::invokestatic or invokespecial,
+      // minus the CP setup and profiling:
+      if (ek == _invokespecial_mh) {
+        // Must load & check the first argument before entering the target method.
+        __ load_method_handle_vmslots(rax_argslot, rcx_recv, rdx_temp);
+        __ movptr(rcx_recv, __ argument_address(rax_argslot, -1));
+        __ null_check(rcx_recv);
+        __ verify_oop(rcx_recv);
+      }
+      __ jmp(rbx_method_fie);
+    }
+    break;
+
+  case _invokevirtual_mh:
+    {
+      // same as TemplateTable::invokevirtual,
+      // minus the CP setup and profiling:
+
+      // pick out the vtable index and receiver offset from the MH,
+      // and then we can discard it:
+      __ load_method_handle_vmslots(rax_argslot, rcx_recv, rdx_temp);
+      Register rbx_index = rbx_temp;
+      __ movl(rbx_index, rcx_dmh_vmindex);
+      // Note:  The verifier allows us to ignore rcx_mh_vmtarget.
+      __ movptr(rcx_recv, __ argument_address(rax_argslot, -1));
+      __ null_check(rcx_recv, oopDesc::klass_offset_in_bytes());
+
+      // get receiver klass
+      Register rax_klass = rax_argslot;
+      __ load_klass(rax_klass, rcx_recv);
+      __ verify_oop(rax_klass);
+
+      // get target methodOop & entry point
+      const int base = instanceKlass::vtable_start_offset() * wordSize;
+      assert(vtableEntry::size() * wordSize == wordSize, "adjust the scaling in the code below");
+      Address vtable_entry_addr(rax_klass,
+                                rbx_index, Address::times_ptr,
+                                base + vtableEntry::method_offset_in_bytes());
+      Register rbx_method = rbx_temp;
+      __ movl(rbx_method, vtable_entry_addr);
+
+      __ verify_oop(rbx_method);
+      __ jmp(rbx_method_fie);
+    }
+    break;
+
+  case _invokeinterface_mh:
+    {
+      // same as TemplateTable::invokeinterface,
+      // minus the CP setup and profiling:
+
+      // pick out the interface and itable index from the MH.
+      __ load_method_handle_vmslots(rax_argslot, rcx_recv, rdx_temp);
+      Register rdx_intf  = rdx_temp;
+      Register rbx_index = rbx_temp;
+      __ movptr(rdx_intf,  rcx_mh_vmtarget);
+      __ movl(rbx_index,   rcx_dmh_vmindex);
+      __ movptr(rcx_recv, __ argument_address(rax_argslot, -1));
+      __ null_check(rcx_recv, oopDesc::klass_offset_in_bytes());
+
+      // get receiver klass
+      Register rax_klass = rax_argslot;
+      __ load_klass(rax_klass, rcx_recv);
+      __ verify_oop(rax_klass);
+
+      Register rcx_temp   = rcx_recv;
+      Register rbx_method = rbx_index;
+
+      // get interface klass
+      Label no_such_interface;
+      __ verify_oop(rdx_intf);
+      __ lookup_interface_method(rax_klass, rdx_intf,
+                                 // note: next two args must be the same:
+                                 rbx_index, rbx_method,
+                                 rcx_temp,
+                                 no_such_interface);
+
+      __ verify_oop(rbx_method);
+      __ jmp(rbx_method_fie);
+      __ hlt();
+
+      __ bind(no_such_interface);
+      // Throw an exception.
+      // For historical reasons, it will be IncompatibleClassChangeError.
+      __ should_not_reach_here(); // %%% FIXME NYI
+    }
+    break;
+
+  case _bound_ref_mh:
+  case _bound_int_mh:
+  case _bound_long_mh:
+  case _bound_ref_direct_mh:
+  case _bound_int_direct_mh:
+  case _bound_long_direct_mh:
+    {
+      bool direct_to_method = (ek >= _bound_ref_direct_mh);
+      BasicType arg_type = T_ILLEGAL;
+      if (ek == _bound_long_mh || ek == _bound_long_direct_mh) {
+        arg_type = T_LONG;
+      } else if (ek == _bound_int_mh || ek == _bound_int_direct_mh) {
+        arg_type = T_INT;
+      } else {
+        assert(ek == _bound_ref_mh || ek == _bound_ref_direct_mh, "must be ref");
+        arg_type = T_OBJECT;
+      }
+      int arg_slots = type2size[arg_type];
+      int arg_mask  = (arg_type == T_OBJECT ? _INSERT_REF_MASK :
+                       arg_slots == 1       ? _INSERT_INT_MASK :  _INSERT_LONG_MASK);
+
+      // make room for the new argument:
+      __ movl(rax_argslot, rcx_bmh_vmargslot);
+      __ lea(rax_argslot, __ argument_address(rax_argslot));
+      insert_arg_slots(_masm, arg_slots * stack_move_unit(), arg_mask,
+                       rax_argslot, rbx_temp, rdx_temp);
+
+      // store bound argument into the new stack slot:
+      __ movptr(rbx_temp, rcx_bmh_argument);
+      Address prim_value_addr(rbx_temp, java_lang_boxing_object::value_offset_in_bytes(arg_type));
+      if (arg_type == T_OBJECT) {
+        __ movptr(Address(rax_argslot, 0), rbx_temp);
+      } else {
+        __ load_sized_value(rbx_temp, prim_value_addr,
+                            type2aelembytes(arg_type), is_signed_subword_type(arg_type));
+        __ movptr(Address(rax_argslot, 0), rbx_temp);
+#ifndef _LP64
+        if (arg_slots == 2) {
+          __ movl(rbx_temp, prim_value_addr.plus_disp(wordSize));
+          __ movl(Address(rax_argslot, Interpreter::stackElementSize()), rbx_temp);
+        }
+#endif //_LP64
+        break;
+      }
+
+      if (direct_to_method) {
+        Register rbx_method = rbx_temp;
+        __ movptr(rbx_method, rcx_mh_vmtarget);
+        __ verify_oop(rbx_method);
+        __ jmp(rbx_method_fie);
+      } else {
+        __ movptr(rcx_recv, rcx_mh_vmtarget);
+        __ verify_oop(rcx_recv);
+        __ jump_to_method_handle_entry(rcx_recv, rdx_temp);
+      }
+    }
+    break;
+
+  case _adapter_retype_only:
+    // immediately jump to the next MH layer:
+    __ movptr(rcx_recv, rcx_mh_vmtarget);
+    __ verify_oop(rcx_recv);
+    __ jump_to_method_handle_entry(rcx_recv, rdx_temp);
+    // This is OK when all parameter types widen.
+    // It is also OK when a return type narrows.
+    break;
+
+  case _adapter_check_cast:
+    {
+      // temps:
+      Register rbx_klass = rbx_temp; // interesting AMH data
+
+      // check a reference argument before jumping to the next layer of MH:
+      __ movl(rax_argslot, rcx_amh_vmargslot);
+      vmarg = __ argument_address(rax_argslot);
+
+      // What class are we casting to?
+      __ movptr(rbx_klass, rcx_amh_argument); // this is a Class object!
+      __ movptr(rbx_klass, Address(rbx_klass, java_lang_Class::klass_offset_in_bytes()));
+
+      // get the new MH:
+      __ movptr(rcx_recv, rcx_mh_vmtarget);
+      // (now we are done with the old MH)
+
+      Label done;
+      __ movptr(rdx_temp, vmarg);
+      __ testl(rdx_temp, rdx_temp);
+      __ jcc(Assembler::zero, done);          // no cast if null
+      __ load_klass(rdx_temp, rdx_temp);
+
+      // live at this point:
+      // - rbx_klass:  klass required by the target method
+      // - rdx_temp:   argument klass to test
+      // - rcx_recv:   method handle to invoke (after cast succeeds)
+      __ check_klass_subtype(rdx_temp, rbx_klass, rax_argslot, done);
+
+      // If we get here, the type check failed!
+      // Call the wrong_method_type stub, passing the failing argument type in rax.
+      Register rax_mtype = rax_argslot;
+      __ push(rbx_klass);       // missed klass (required type)
+      __ push(rdx_temp);        // bad actual type (1st stacked argument)
+      __ jump(ExternalAddress(Interpreter::throw_WrongMethodType_entry()));
+
+      __ bind(done);
+      __ jump_to_method_handle_entry(rcx_recv, rdx_temp);
+    }
+    break;
+
+  case _adapter_prim_to_prim:
+  case _adapter_ref_to_prim:
+    // handled completely by optimized cases
+    __ stop("init_AdapterMethodHandle should not issue this");
+    break;
+
+  case _adapter_opt_i2i:        // optimized subcase of adapt_prim_to_prim
+//case _adapter_opt_f2i:        // optimized subcase of adapt_prim_to_prim
+  case _adapter_opt_l2i:        // optimized subcase of adapt_prim_to_prim
+  case _adapter_opt_unboxi:     // optimized subcase of adapt_ref_to_prim
+    {
+      // perform an in-place conversion to int or an int subword
+      __ movl(rax_argslot, rcx_amh_vmargslot);
+      vmarg = __ argument_address(rax_argslot);
+
+      switch (ek) {
+      case _adapter_opt_i2i:
+        __ movl(rdx_temp, vmarg);
+        break;
+      case _adapter_opt_l2i:
+        {
+          // just delete the extra slot; on a little-endian machine we keep the first
+          __ lea(rax_argslot, __ argument_address(rax_argslot, 1));
+          remove_arg_slots(_masm, -stack_move_unit(),
+                           rax_argslot, rbx_temp, rdx_temp);
+          vmarg = Address(rax_argslot, -Interpreter::stackElementSize());
+          __ movl(rdx_temp, vmarg);
+        }
+        break;
+      case _adapter_opt_unboxi:
+        {
+          // Load the value up from the heap.
+          __ movptr(rdx_temp, vmarg);
+          int value_offset = java_lang_boxing_object::value_offset_in_bytes(T_INT);
+#ifdef ASSERT
+          for (int bt = T_BOOLEAN; bt < T_INT; bt++) {
+            if (is_subword_type(BasicType(bt)))
+              assert(value_offset == java_lang_boxing_object::value_offset_in_bytes(BasicType(bt)), "");
+          }
+#endif
+          __ null_check(rdx_temp, value_offset);
+          __ movl(rdx_temp, Address(rdx_temp, value_offset));
+          // We load this as a word.  Because we are little-endian,
+          // the low bits will be correct, but the high bits may need cleaning.
+          // The vminfo will guide us to clean those bits.
+        }
+        break;
+      default:
+        assert(false, "");
+      }
+      goto finish_int_conversion;
+    }
+
+  finish_int_conversion:
+    {
+      Register rbx_vminfo = rbx_temp;
+      __ movl(rbx_vminfo, rcx_amh_conversion);
+      assert(CONV_VMINFO_SHIFT == 0, "preshifted");
+
+      // get the new MH:
+      __ movptr(rcx_recv, rcx_mh_vmtarget);
+      // (now we are done with the old MH)
+
+      // original 32-bit vmdata word must be of this form:
+      //    | MBZ:16 | signBitCount:8 | srcDstTypes:8 | conversionOp:8 |
+      __ xchgl(rcx, rbx_vminfo);                // free rcx for shifts
+      __ shll(rdx_temp /*, rcx*/);
+      Label zero_extend, done;
+      __ testl(rcx, CONV_VMINFO_SIGN_FLAG);
+      __ jcc(Assembler::zero, zero_extend);
+
+      // this path is taken for int->byte, int->short
+      __ sarl(rdx_temp /*, rcx*/);
+      __ jmp(done);
+
+      __ bind(zero_extend);
+      // this is taken for int->char
+      __ shrl(rdx_temp /*, rcx*/);
+
+      __ bind(done);
+      __ movptr(vmarg, rdx_temp);
+      __ xchgl(rcx, rbx_vminfo);                // restore rcx_recv
+
+      __ jump_to_method_handle_entry(rcx_recv, rdx_temp);
+    }
+    break;
+
+  case _adapter_opt_i2l:        // optimized subcase of adapt_prim_to_prim
+  case _adapter_opt_unboxl:     // optimized subcase of adapt_ref_to_prim
+    {
+      // perform an in-place int-to-long or ref-to-long conversion
+      __ movl(rax_argslot, rcx_amh_vmargslot);
+
+      // on a little-endian machine we keep the first slot and add another after
+      __ lea(rax_argslot, __ argument_address(rax_argslot, 1));
+      insert_arg_slots(_masm, stack_move_unit(), _INSERT_INT_MASK,
+                       rax_argslot, rbx_temp, rdx_temp);
+      Address vmarg1(rax_argslot, -Interpreter::stackElementSize());
+      Address vmarg2 = vmarg1.plus_disp(Interpreter::stackElementSize());
+
+      switch (ek) {
+      case _adapter_opt_i2l:
+        {
+          __ movl(rdx_temp, vmarg1);
+          __ sarl(rdx_temp, 31);  // __ extend_sign()
+          __ movl(vmarg2, rdx_temp); // store second word
+        }
+        break;
+      case _adapter_opt_unboxl:
+        {
+          // Load the value up from the heap.
+          __ movptr(rdx_temp, vmarg1);
+          int value_offset = java_lang_boxing_object::value_offset_in_bytes(T_LONG);
+          assert(value_offset == java_lang_boxing_object::value_offset_in_bytes(T_DOUBLE), "");
+          __ null_check(rdx_temp, value_offset);
+          __ movl(rbx_temp, Address(rdx_temp, value_offset + 0*BytesPerInt));
+          __ movl(rdx_temp, Address(rdx_temp, value_offset + 1*BytesPerInt));
+          __ movl(vmarg1, rbx_temp);
+          __ movl(vmarg2, rdx_temp);
+        }
+        break;
+      default:
+        assert(false, "");
+      }
+
+      __ movptr(rcx_recv, rcx_mh_vmtarget);
+      __ jump_to_method_handle_entry(rcx_recv, rdx_temp);
+    }
+    break;
+
+  case _adapter_opt_f2d:        // optimized subcase of adapt_prim_to_prim
+  case _adapter_opt_d2f:        // optimized subcase of adapt_prim_to_prim
+    {
+      // perform an in-place floating primitive conversion
+      __ movl(rax_argslot, rcx_amh_vmargslot);
+      __ lea(rax_argslot, __ argument_address(rax_argslot, 1));
+      if (ek == _adapter_opt_f2d) {
+        insert_arg_slots(_masm, stack_move_unit(), _INSERT_INT_MASK,
+                         rax_argslot, rbx_temp, rdx_temp);
+      }
+      Address vmarg(rax_argslot, -Interpreter::stackElementSize());
+
+#ifdef _LP64
+      if (ek == _adapter_opt_f2d) {
+        __ movflt(xmm0, vmarg);
+        __ cvtss2sd(xmm0, xmm0);
+        __ movdbl(vmarg, xmm0);
+      } else {
+        __ movdbl(xmm0, vmarg);
+        __ cvtsd2ss(xmm0, xmm0);
+        __ movflt(vmarg, xmm0);
+      }
+#else //_LP64
+      if (ek == _adapter_opt_f2d) {
+        __ fld_s(vmarg);        // load float to ST0
+        __ fstp_s(vmarg);       // store single
+      } else if (!TaggedStackInterpreter) {
+        __ fld_d(vmarg);        // load double to ST0
+        __ fstp_s(vmarg);       // store single
+      } else {
+        Address vmarg_tag = vmarg.plus_disp(tag_offset);
+        Address vmarg2    = vmarg.plus_disp(Interpreter::stackElementSize());
+        // vmarg2_tag does not participate in this code
+        Register rbx_tag = rbx_temp;
+        __ movl(rbx_tag, vmarg_tag); // preserve tag
+        __ movl(rdx_temp, vmarg2); // get second word of double
+        __ movl(vmarg_tag, rdx_temp); // align with first word
+        __ fld_d(vmarg);        // load double to ST0
+        __ movl(vmarg_tag, rbx_tag); // restore tag
+        __ fstp_s(vmarg);       // store single
+      }
+#endif //_LP64
+
+      if (ek == _adapter_opt_d2f) {
+        remove_arg_slots(_masm, -stack_move_unit(),
+                         rax_argslot, rbx_temp, rdx_temp);
+      }
+
+      __ movptr(rcx_recv, rcx_mh_vmtarget);
+      __ jump_to_method_handle_entry(rcx_recv, rdx_temp);
+    }
+    break;
+
+  case _adapter_prim_to_ref:
+    __ unimplemented(entry_name(ek)); // %%% FIXME: NYI
+    break;
+
+  case _adapter_swap_args:
+  case _adapter_rot_args:
+    // handled completely by optimized cases
+    __ stop("init_AdapterMethodHandle should not issue this");
+    break;
+
+  case _adapter_opt_swap_1:
+  case _adapter_opt_swap_2:
+  case _adapter_opt_rot_1_up:
+  case _adapter_opt_rot_1_down:
+  case _adapter_opt_rot_2_up:
+  case _adapter_opt_rot_2_down:
+    {
+      int rotate = 0, swap_slots = 0;
+      switch ((int)ek) {
+      case _adapter_opt_swap_1:     swap_slots = 1; break;
+      case _adapter_opt_swap_2:     swap_slots = 2; break;
+      case _adapter_opt_rot_1_up:   swap_slots = 1; rotate++; break;
+      case _adapter_opt_rot_1_down: swap_slots = 1; rotate--; break;
+      case _adapter_opt_rot_2_up:   swap_slots = 2; rotate++; break;
+      case _adapter_opt_rot_2_down: swap_slots = 2; rotate--; break;
+      default: assert(false, "");
+      }
+
+      // the real size of the move must be doubled if TaggedStackInterpreter:
+      int swap_bytes = (int)( swap_slots * Interpreter::stackElementWords() * wordSize );
+
+      // 'argslot' is the position of the first argument to swap
+      __ movl(rax_argslot, rcx_amh_vmargslot);
+      __ lea(rax_argslot, __ argument_address(rax_argslot));
+
+      // 'vminfo' is the second
+      Register rbx_destslot = rbx_temp;
+      __ movl(rbx_destslot, rcx_amh_conversion);
+      assert(CONV_VMINFO_SHIFT == 0, "preshifted");
+      __ andl(rbx_destslot, CONV_VMINFO_MASK);
+      __ lea(rbx_destslot, __ argument_address(rbx_destslot));
+      DEBUG_ONLY(verify_argslot(_masm, rbx_destslot, "swap point must fall within current frame"));
+
+      if (!rotate) {
+        for (int i = 0; i < swap_bytes; i += wordSize) {
+          __ movptr(rdx_temp, Address(rax_argslot , i));
+          __ push(rdx_temp);
+          __ movptr(rdx_temp, Address(rbx_destslot, i));
+          __ movptr(Address(rax_argslot, i), rdx_temp);
+          __ pop(rdx_temp);
+          __ movptr(Address(rbx_destslot, i), rdx_temp);
+        }
+      } else {
+        // push the first chunk, which is going to get overwritten
+        for (int i = swap_bytes; (i -= wordSize) >= 0; ) {
+          __ movptr(rdx_temp, Address(rax_argslot, i));
+          __ push(rdx_temp);
+        }
+
+        if (rotate > 0) {
+          // rotate upward
+          __ subptr(rax_argslot, swap_bytes);
+#ifdef ASSERT
+          {
+            // Verify that argslot > destslot, by at least swap_bytes.
+            Label L_ok;
+            __ cmpptr(rax_argslot, rbx_destslot);
+            __ jcc(Assembler::aboveEqual, L_ok);
+            __ stop("source must be above destination (upward rotation)");
+            __ bind(L_ok);
+          }
+#endif
+          // work argslot down to destslot, copying contiguous data upwards
+          // pseudo-code:
+          //   rax = src_addr - swap_bytes
+          //   rbx = dest_addr
+          //   while (rax >= rbx) *(rax + swap_bytes) = *(rax + 0), rax--;
+          Label loop;
+          __ bind(loop);
+          __ movptr(rdx_temp, Address(rax_argslot, 0));
+          __ movptr(Address(rax_argslot, swap_bytes), rdx_temp);
+          __ addptr(rax_argslot, -wordSize);
+          __ cmpptr(rax_argslot, rbx_destslot);
+          __ jcc(Assembler::aboveEqual, loop);
+        } else {
+          __ addptr(rax_argslot, swap_bytes);
+#ifdef ASSERT
+          {
+            // Verify that argslot < destslot, by at least swap_bytes.
+            Label L_ok;
+            __ cmpptr(rax_argslot, rbx_destslot);
+            __ jcc(Assembler::belowEqual, L_ok);
+            __ stop("source must be below destination (downward rotation)");
+            __ bind(L_ok);
+          }
+#endif
+          // work argslot up to destslot, copying contiguous data downwards
+          // pseudo-code:
+          //   rax = src_addr + swap_bytes
+          //   rbx = dest_addr
+          //   while (rax <= rbx) *(rax - swap_bytes) = *(rax + 0), rax++;
+          Label loop;
+          __ bind(loop);
+          __ movptr(rdx_temp, Address(rax_argslot, 0));
+          __ movptr(Address(rax_argslot, -swap_bytes), rdx_temp);
+          __ addptr(rax_argslot, wordSize);
+          __ cmpptr(rax_argslot, rbx_destslot);
+          __ jcc(Assembler::belowEqual, loop);
+        }
+
+        // pop the original first chunk into the destination slot, now free
+        for (int i = 0; i < swap_bytes; i += wordSize) {
+          __ pop(rdx_temp);
+          __ movptr(Address(rbx_destslot, i), rdx_temp);
+        }
+      }
+
+      __ movptr(rcx_recv, rcx_mh_vmtarget);
+      __ jump_to_method_handle_entry(rcx_recv, rdx_temp);
+    }
+    break;
+
+  case _adapter_dup_args:
+    {
+      // 'argslot' is the position of the first argument to duplicate
+      __ movl(rax_argslot, rcx_amh_vmargslot);
+      __ lea(rax_argslot, __ argument_address(rax_argslot));
+
+      // 'stack_move' is negative number of words to duplicate
+      Register rdx_stack_move = rdx_temp;
+      __ movl(rdx_stack_move, rcx_amh_conversion);
+      __ sarl(rdx_stack_move, CONV_STACK_MOVE_SHIFT);
+
+      int argslot0_num = 0;
+      Address argslot0 = __ argument_address(RegisterOrConstant(argslot0_num));
+      assert(argslot0.base() == rsp, "");
+      int pre_arg_size = argslot0.disp();
+      assert(pre_arg_size % wordSize == 0, "");
+      assert(pre_arg_size > 0, "must include PC");
+
+      // remember the old rsp+1 (argslot[0])
+      Register rbx_oldarg = rbx_temp;
+      __ lea(rbx_oldarg, argslot0);
+
+      // move rsp down to make room for dups
+      __ lea(rsp, Address(rsp, rdx_stack_move, Address::times_ptr));
+
+      // compute the new rsp+1 (argslot[0])
+      Register rdx_newarg = rdx_temp;
+      __ lea(rdx_newarg, argslot0);
+
+      __ push(rdi);             // need a temp
+      // (preceding push must be done after arg addresses are taken!)
+
+      // pull down the pre_arg_size data (PC)
+      for (int i = -pre_arg_size; i < 0; i += wordSize) {
+        __ movptr(rdi, Address(rbx_oldarg, i));
+        __ movptr(Address(rdx_newarg, i), rdi);
+      }
+
+      // copy from rax_argslot[0...] down to new_rsp[1...]
+      // pseudo-code:
+      //   rbx = old_rsp+1
+      //   rdx = new_rsp+1
+      //   rax = argslot
+      //   while (rdx < rbx) *rdx++ = *rax++
+      Label loop;
+      __ bind(loop);
+      __ movptr(rdi, Address(rax_argslot, 0));
+      __ movptr(Address(rdx_newarg, 0), rdi);
+      __ addptr(rax_argslot, wordSize);
+      __ addptr(rdx_newarg, wordSize);
+      __ cmpptr(rdx_newarg, rbx_oldarg);
+      __ jcc(Assembler::less, loop);
+
+      __ pop(rdi);              // restore temp
+
+      __ movptr(rcx_recv, rcx_mh_vmtarget);
+      __ jump_to_method_handle_entry(rcx_recv, rdx_temp);
+    }
+    break;
+
+  case _adapter_drop_args:
+    {
+      // 'argslot' is the position of the first argument to nuke
+      __ movl(rax_argslot, rcx_amh_vmargslot);
+      __ lea(rax_argslot, __ argument_address(rax_argslot));
+
+      __ push(rdi);             // need a temp
+      // (must do previous push after argslot address is taken)
+
+      // 'stack_move' is number of words to drop
+      Register rdi_stack_move = rdi;
+      __ movl(rdi_stack_move, rcx_amh_conversion);
+      __ sarl(rdi_stack_move, CONV_STACK_MOVE_SHIFT);
+      remove_arg_slots(_masm, rdi_stack_move,
+                       rax_argslot, rbx_temp, rdx_temp);
+
+      __ pop(rdi);              // restore temp
+
+      __ movptr(rcx_recv, rcx_mh_vmtarget);
+      __ jump_to_method_handle_entry(rcx_recv, rdx_temp);
+    }
+    break;
+
+  case _adapter_collect_args:
+    __ unimplemented(entry_name(ek)); // %%% FIXME: NYI
+    break;
+
+  case _adapter_spread_args:
+    // handled completely by optimized cases
+    __ stop("init_AdapterMethodHandle should not issue this");
+    break;
+
+  case _adapter_opt_spread_0:
+  case _adapter_opt_spread_1:
+  case _adapter_opt_spread_more:
+    {
+      // spread an array out into a group of arguments
+      int length_constant = -1;
+      switch (ek) {
+      case _adapter_opt_spread_0: length_constant = 0; break;
+      case _adapter_opt_spread_1: length_constant = 1; break;
+      }
+
+      // find the address of the array argument
+      __ movl(rax_argslot, rcx_amh_vmargslot);
+      __ lea(rax_argslot, __ argument_address(rax_argslot));
+
+      // grab some temps
+      { __ push(rsi); __ push(rdi); }
+      // (preceding pushes must be done after argslot address is taken!)
+#define UNPUSH_RSI_RDI \
+      { __ pop(rdi); __ pop(rsi); }
+
+      // arx_argslot points both to the array and to the first output arg
+      vmarg = Address(rax_argslot, 0);
+
+      // Get the array value.
+      Register  rsi_array       = rsi;
+      Register  rdx_array_klass = rdx_temp;
+      BasicType elem_type       = T_OBJECT;
+      int       length_offset   = arrayOopDesc::length_offset_in_bytes();
+      int       elem0_offset    = arrayOopDesc::base_offset_in_bytes(elem_type);
+      __ movptr(rsi_array, vmarg);
+      Label skip_array_check;
+      if (length_constant == 0) {
+        __ testptr(rsi_array, rsi_array);
+        __ jcc(Assembler::zero, skip_array_check);
+      }
+      __ null_check(rsi_array, oopDesc::klass_offset_in_bytes());
+      __ load_klass(rdx_array_klass, rsi_array);
+
+      // Check the array type.
+      Register rbx_klass = rbx_temp;
+      __ movptr(rbx_klass, rcx_amh_argument); // this is a Class object!
+      __ movptr(rbx_klass, Address(rbx_klass, java_lang_Class::klass_offset_in_bytes()));
+
+      Label ok_array_klass, bad_array_klass, bad_array_length;
+      __ check_klass_subtype(rdx_array_klass, rbx_klass, rdi, ok_array_klass);
+      // If we get here, the type check failed!
+      __ jmp(bad_array_klass);
+      __ bind(ok_array_klass);
+
+      // Check length.
+      if (length_constant >= 0) {
+        __ cmpl(Address(rsi_array, length_offset), length_constant);
+      } else {
+        Register rbx_vminfo = rbx_temp;
+        __ movl(rbx_vminfo, rcx_amh_conversion);
+        assert(CONV_VMINFO_SHIFT == 0, "preshifted");
+        __ andl(rbx_vminfo, CONV_VMINFO_MASK);
+        __ cmpl(rbx_vminfo, Address(rsi_array, length_offset));
+      }
+      __ jcc(Assembler::notEqual, bad_array_length);
+
+      Register rdx_argslot_limit = rdx_temp;
+
+      // Array length checks out.  Now insert any required stack slots.
+      if (length_constant == -1) {
+        // Form a pointer to the end of the affected region.
+        __ lea(rdx_argslot_limit, Address(rax_argslot, Interpreter::stackElementSize()));
+        // 'stack_move' is negative number of words to insert
+        Register rdi_stack_move = rdi;
+        __ movl(rdi_stack_move, rcx_amh_conversion);
+        __ sarl(rdi_stack_move, CONV_STACK_MOVE_SHIFT);
+        Register rsi_temp = rsi_array;  // spill this
+        insert_arg_slots(_masm, rdi_stack_move, -1,
+                         rax_argslot, rbx_temp, rsi_temp);
+        // reload the array (since rsi was killed)
+        __ movptr(rsi_array, vmarg);
+      } else if (length_constant > 1) {
+        int arg_mask = 0;
+        int new_slots = (length_constant - 1);
+        for (int i = 0; i < new_slots; i++) {
+          arg_mask <<= 1;
+          arg_mask |= _INSERT_REF_MASK;
+        }
+        insert_arg_slots(_masm, new_slots * stack_move_unit(), arg_mask,
+                         rax_argslot, rbx_temp, rdx_temp);
+      } else if (length_constant == 1) {
+        // no stack resizing required
+      } else if (length_constant == 0) {
+        remove_arg_slots(_masm, -stack_move_unit(),
+                         rax_argslot, rbx_temp, rdx_temp);
+      }
+
+      // Copy from the array to the new slots.
+      // Note: Stack change code preserves integrity of rax_argslot pointer.
+      // So even after slot insertions, rax_argslot still points to first argument.
+      if (length_constant == -1) {
+        // [rax_argslot, rdx_argslot_limit) is the area we are inserting into.
+        Register rsi_source = rsi_array;
+        __ lea(rsi_source, Address(rsi_array, elem0_offset));
+        Label loop;
+        __ bind(loop);
+        __ movptr(rbx_temp, Address(rsi_source, 0));
+        __ movptr(Address(rax_argslot, 0), rbx_temp);
+        __ addptr(rsi_source, type2aelembytes(elem_type));
+        if (TaggedStackInterpreter) {
+          __ movptr(Address(rax_argslot, tag_offset),
+                    frame::tag_for_basic_type(elem_type));
+        }
+        __ addptr(rax_argslot, Interpreter::stackElementSize());
+        __ cmpptr(rax_argslot, rdx_argslot_limit);
+        __ jcc(Assembler::less, loop);
+      } else if (length_constant == 0) {
+        __ bind(skip_array_check);
+        // nothing to copy
+      } else {
+        int elem_offset = elem0_offset;
+        int slot_offset = 0;
+        for (int index = 0; index < length_constant; index++) {
+          __ movptr(rbx_temp, Address(rsi_array, elem_offset));
+          __ movptr(Address(rax_argslot, slot_offset), rbx_temp);
+          elem_offset += type2aelembytes(elem_type);
+          if (TaggedStackInterpreter) {
+            __ movptr(Address(rax_argslot, slot_offset + tag_offset),
+                      frame::tag_for_basic_type(elem_type));
+          }
+          slot_offset += Interpreter::stackElementSize();
+        }
+      }
+
+      // Arguments are spread.  Move to next method handle.
+      UNPUSH_RSI_RDI;
+      __ movptr(rcx_recv, rcx_mh_vmtarget);
+      __ jump_to_method_handle_entry(rcx_recv, rdx_temp);
+
+      __ bind(bad_array_klass);
+      UNPUSH_RSI_RDI;
+      __ stop("bad array klass NYI");
+
+      __ bind(bad_array_length);
+      UNPUSH_RSI_RDI;
+      __ stop("bad array length NYI");
+
+#undef UNPUSH_RSI_RDI
+    }
+    break;
+
+  case _adapter_flyby:
+  case _adapter_ricochet:
+    __ unimplemented(entry_name(ek)); // %%% FIXME: NYI
+    break;
+
+  default:  ShouldNotReachHere();
+  }
+  __ hlt();
+
+  address me_cookie = MethodHandleEntry::start_compiled_entry(_masm, interp_entry);
+  __ unimplemented(entry_name(ek)); // %%% FIXME: NYI
+
+  init_entry(ek, MethodHandleEntry::finish_compiled_entry(_masm, me_cookie));
+}
diff --git a/src/cpu/x86/vm/stubGenerator_x86_32.cpp b/src/cpu/x86/vm/stubGenerator_x86_32.cpp
index aca937780..e38fc1d0f 100644
--- a/src/cpu/x86/vm/stubGenerator_x86_32.cpp
+++ b/src/cpu/x86/vm/stubGenerator_x86_32.cpp
@@ -2219,6 +2219,16 @@ class StubGenerator: public StubCodeGenerator {
 
     // arraycopy stubs used by compilers
     generate_arraycopy_stubs();
+
+    // generic method handle stubs
+    if (EnableMethodHandles && SystemDictionary::MethodHandle_klass() != NULL) {
+      for (MethodHandles::EntryKind ek = MethodHandles::_EK_FIRST;
+           ek < MethodHandles::_EK_LIMIT;
+           ek = MethodHandles::EntryKind(1 + (int)ek)) {
+        StubCodeMark mark(this, "MethodHandle", MethodHandles::entry_name(ek));
+        MethodHandles::generate_method_handle_stub(_masm, ek);
+      }
+    }
   }
 
 
diff --git a/src/cpu/x86/vm/templateInterpreter_x86_32.cpp b/src/cpu/x86/vm/templateInterpreter_x86_32.cpp
index 8d94671e6..5769873dc 100644
--- a/src/cpu/x86/vm/templateInterpreter_x86_32.cpp
+++ b/src/cpu/x86/vm/templateInterpreter_x86_32.cpp
@@ -92,6 +92,33 @@ address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
   return entry;
 }
 
+// Arguments are: required type at TOS+8, failing object (or NULL) at TOS+4.
+// pc at TOS (just for debugging)
+address TemplateInterpreterGenerator::generate_WrongMethodType_handler() {
+  address entry = __ pc();
+
+  __ pop(rbx);                  // actual failing object is at TOS
+  __ pop(rax);                  // required type is at TOS+4
+
+  __ verify_oop(rbx);
+  __ verify_oop(rax);
+
+  // Various method handle types use interpreter registers as temps.
+  __ restore_bcp();
+  __ restore_locals();
+
+  // Expression stack must be empty before entering the VM for an exception.
+  __ empty_expression_stack();
+  __ empty_FPU_stack();
+  __ call_VM(noreg,
+             CAST_FROM_FN_PTR(address,
+                              InterpreterRuntime::throw_WrongMethodTypeException),
+             // pass required type, failing object (or NULL)
+             rax, rbx);
+  return entry;
+}
+
+
 address TemplateInterpreterGenerator::generate_exception_handler_common(const char* name, const char* message, bool pass_oop) {
   assert(!pass_oop || message == NULL, "either oop or message but not both");
   address entry = __ pc();
@@ -1370,6 +1397,7 @@ address AbstractInterpreterGenerator::generate_method_entry(AbstractInterpreter:
     case Interpreter::empty                  : entry_point = ((InterpreterGenerator*)this)->generate_empty_entry();        break;
     case Interpreter::accessor               : entry_point = ((InterpreterGenerator*)this)->generate_accessor_entry();     break;
     case Interpreter::abstract               : entry_point = ((InterpreterGenerator*)this)->generate_abstract_entry();     break;
+    case Interpreter::method_handle          : entry_point = ((InterpreterGenerator*)this)->generate_method_handle_entry(); break;
 
     case Interpreter::java_lang_math_sin     : // fall thru
     case Interpreter::java_lang_math_cos     : // fall thru
@@ -1400,7 +1428,8 @@ int AbstractInterpreter::size_top_interpreter_activation(methodOop method) {
   // be sure to change this if you add/subtract anything to/from the overhead area
   const int overhead_size = -frame::interpreter_frame_initial_sp_offset;
 
-  const int method_stack = (method->max_locals() + method->max_stack()) *
+  const int extra_stack = methodOopDesc::extra_stack_entries();
+  const int method_stack = (method->max_locals() + method->max_stack() + extra_stack) *
                            Interpreter::stackElementWords();
   return overhead_size + method_stack + stub_code;
 }
diff --git a/src/cpu/x86/vm/templateInterpreter_x86_64.cpp b/src/cpu/x86/vm/templateInterpreter_x86_64.cpp
index ec7712760..cd9d08915 100644
--- a/src/cpu/x86/vm/templateInterpreter_x86_64.cpp
+++ b/src/cpu/x86/vm/templateInterpreter_x86_64.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright 2003-2008 Sun Microsystems, Inc.  All Rights Reserved.
+ * Copyright 2003-2009 Sun Microsystems, Inc.  All Rights Reserved.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
  * This code is free software; you can redistribute it and/or modify it
@@ -100,6 +100,26 @@ address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
   return entry;
 }
 
+// Arguments are: required type in rarg1, failing object (or NULL) in rarg2
+address TemplateInterpreterGenerator::generate_WrongMethodType_handler() {
+  address entry = __ pc();
+
+  __ pop(c_rarg2);              // failing object is at TOS
+  __ pop(c_rarg1);              // required type is at TOS+8
+
+  // expression stack must be empty before entering the VM if an
+  // exception happened
+  __ empty_expression_stack();
+
+  __ call_VM(noreg,
+             CAST_FROM_FN_PTR(address,
+                              InterpreterRuntime::
+                              throw_WrongMethodTypeException),
+             // pass required type, failing object (or NULL)
+             c_rarg1, c_rarg2);
+  return entry;
+}
+
 address TemplateInterpreterGenerator::generate_exception_handler_common(
         const char* name, const char* message, bool pass_oop) {
   assert(!pass_oop || message == NULL, "either oop or message but not both");
@@ -1393,6 +1413,7 @@ address AbstractInterpreterGenerator::generate_method_entry(
   case Interpreter::empty                  : entry_point = ((InterpreterGenerator*) this)->generate_empty_entry();       break;
   case Interpreter::accessor               : entry_point = ((InterpreterGenerator*) this)->generate_accessor_entry();    break;
   case Interpreter::abstract               : entry_point = ((InterpreterGenerator*) this)->generate_abstract_entry();    break;
+  case Interpreter::method_handle          : entry_point = ((InterpreterGenerator*) this)->generate_method_handle_entry();break;
 
   case Interpreter::java_lang_math_sin     : // fall thru
   case Interpreter::java_lang_math_cos     : // fall thru
@@ -1423,7 +1444,8 @@ int AbstractInterpreter::size_top_interpreter_activation(methodOop method) {
     -(frame::interpreter_frame_initial_sp_offset) + entry_size;
 
   const int stub_code = frame::entry_frame_after_call_words;
-  const int method_stack = (method->max_locals() + method->max_stack()) *
+  const int extra_stack = methodOopDesc::extra_stack_entries();
+  const int method_stack = (method->max_locals() + method->max_stack() + extra_stack) *
                            Interpreter::stackElementWords();
   return (overhead_size + method_stack + stub_code);
 }