changeset 2539:f7d55ea6ee56

7045514: SPARC assembly code for JSR 292 ricochet frames Reviewed-by: kvn, jrose
author never
date Fri, 03 Jun 2011 22:31:43 -0700
parents f56542cb325a
children 293f68bda347
files src/cpu/sparc/vm/assembler_sparc.cpp src/cpu/sparc/vm/assembler_sparc.hpp src/cpu/sparc/vm/assembler_sparc.inline.hpp src/cpu/sparc/vm/frame_sparc.cpp src/cpu/sparc/vm/methodHandles_sparc.cpp src/cpu/sparc/vm/methodHandles_sparc.hpp src/cpu/sparc/vm/registerMap_sparc.hpp src/cpu/sparc/vm/runtime_sparc.cpp src/cpu/sparc/vm/sharedRuntime_sparc.cpp src/cpu/sparc/vm/stubRoutines_sparc.hpp src/cpu/sparc/vm/templateInterpreter_sparc.cpp src/cpu/x86/vm/assembler_x86.cpp src/cpu/x86/vm/methodHandles_x86.cpp src/cpu/x86/vm/methodHandles_x86.hpp src/cpu/x86/vm/runtime_x86_32.cpp src/cpu/x86/vm/sharedRuntime_x86_32.cpp src/cpu/x86/vm/sharedRuntime_x86_64.cpp src/cpu/x86/vm/stubRoutines_x86_32.hpp src/cpu/x86/vm/stubRoutines_x86_64.hpp src/share/vm/compiler/oopMap.cpp src/share/vm/opto/runtime.cpp src/share/vm/prims/methodHandleWalk.cpp src/share/vm/prims/methodHandleWalk.hpp src/share/vm/prims/methodHandles.cpp src/share/vm/prims/methodHandles.hpp src/share/vm/runtime/sharedRuntime.cpp src/share/vm/runtime/sharedRuntime.hpp
diffstat 27 files changed, 2037 insertions(+), 561 deletions(-) [+]
line wrap: on
line diff
--- a/src/cpu/sparc/vm/assembler_sparc.cpp	Thu Jun 02 13:36:11 2011 -0700
+++ b/src/cpu/sparc/vm/assembler_sparc.cpp	Fri Jun 03 22:31:43 2011 -0700
@@ -42,6 +42,12 @@
 #include "gc_implementation/g1/heapRegion.hpp"
 #endif
 
+#ifdef PRODUCT
+#define BLOCK_COMMENT(str) /* nothing */
+#else
+#define BLOCK_COMMENT(str) block_comment(str)
+#endif
+
 // Convert the raw encoding form into the form expected by the
 // constructor for Address.
 Address Address::make_raw(int base, int index, int scale, int disp, bool disp_is_oop) {
@@ -1072,6 +1078,12 @@
     check_and_forward_exception(Gtemp);
   }
 
+#ifdef ASSERT
+  set(badHeapWordVal, G3);
+  set(badHeapWordVal, G4);
+  set(badHeapWordVal, G5);
+#endif
+
   // get oop result if there is one and reset the value in the thread
   if (oop_result->is_valid()) {
     get_vm_result(oop_result);
@@ -1177,6 +1189,11 @@
   call(entry_point, relocInfo::runtime_call_type);
   delayed()->nop();
   restore_thread(thread_cache);
+#ifdef ASSERT
+  set(badHeapWordVal, G3);
+  set(badHeapWordVal, G4);
+  set(badHeapWordVal, G5);
+#endif
 }
 
 
@@ -1518,7 +1535,7 @@
 // save_frame: given number of "extra" words in frame,
 // issue approp. save instruction (p 200, v8 manual)
 
-void MacroAssembler::save_frame(int extraWords = 0) {
+void MacroAssembler::save_frame(int extraWords) {
   int delta = -total_frame_size_in_bytes(extraWords);
   if (is_simm13(delta)) {
     save(SP, delta, SP);
@@ -1730,6 +1747,7 @@
 
   if (reg == G0)  return;       // always NULL, which is always an oop
 
+  BLOCK_COMMENT("verify_oop {");
   char buffer[64];
 #ifdef COMPILER1
   if (CommentedAssembly) {
@@ -1768,6 +1786,7 @@
   delayed()->nop();
   // recover frame size
   add(SP, 8*8,SP);
+  BLOCK_COMMENT("} verify_oop");
 }
 
 void MacroAssembler::_verify_oop_addr(Address addr, const char* msg, const char * file, int line) {
@@ -2040,7 +2059,7 @@
   }
   else
      ::tty->print_cr("=============== DEBUG MESSAGE: %s ================\n", msg);
-  assert(false, "error");
+  assert(false, err_msg("DEBUG MESSAGE: %s", msg));
 }
 
 
@@ -3230,6 +3249,7 @@
 
 
 RegisterOrConstant MacroAssembler::argument_offset(RegisterOrConstant arg_slot,
+                                                   Register temp_reg,
                                                    int extra_slot_offset) {
   // cf. TemplateTable::prepare_invoke(), if (load_receiver).
   int stackElementSize = Interpreter::stackElementSize;
@@ -3238,18 +3258,19 @@
     offset += arg_slot.as_constant() * stackElementSize;
     return offset;
   } else {
-    Register temp = arg_slot.as_register();
-    sll_ptr(temp, exact_log2(stackElementSize), temp);
+    assert(temp_reg != noreg, "must specify");
+    sll_ptr(arg_slot.as_register(), exact_log2(stackElementSize), temp_reg);
     if (offset != 0)
-      add(temp, offset, temp);
-    return temp;
+      add(temp_reg, offset, temp_reg);
+    return temp_reg;
   }
 }
 
 
 Address MacroAssembler::argument_address(RegisterOrConstant arg_slot,
+                                         Register temp_reg,
                                          int extra_slot_offset) {
-  return Address(Gargs, argument_offset(arg_slot, extra_slot_offset));
+  return Address(Gargs, argument_offset(arg_slot, temp_reg, extra_slot_offset));
 }
 
 
@@ -4906,4 +4927,3 @@
   // Caller should set it:
   // add(G0, 1, result); // equals
 }
-
--- a/src/cpu/sparc/vm/assembler_sparc.hpp	Thu Jun 02 13:36:11 2011 -0700
+++ b/src/cpu/sparc/vm/assembler_sparc.hpp	Fri Jun 03 22:31:43 2011 -0700
@@ -309,12 +309,14 @@
 #endif
 
   // accessors
-  Register base()      const { return _base; }
-  Register index()     const { return _index_or_disp.as_register(); }
-  int      disp()      const { return _index_or_disp.as_constant(); }
-
-  bool     has_index() const { return _index_or_disp.is_register(); }
-  bool     has_disp()  const { return _index_or_disp.is_constant(); }
+  Register base()             const { return _base; }
+  Register index()            const { return _index_or_disp.as_register(); }
+  int      disp()             const { return _index_or_disp.as_constant(); }
+
+  bool     has_index()        const { return _index_or_disp.is_register(); }
+  bool     has_disp()         const { return _index_or_disp.is_constant(); }
+
+  bool     uses(Register reg) const { return base() == reg || (has_index() && index() == reg); }
 
   const relocInfo::relocType rtype() { return _rspec.type(); }
   const RelocationHolder&    rspec() { return _rspec; }
@@ -330,6 +332,10 @@
     Address a(base(), disp() + plusdisp);
     return a;
   }
+  bool is_same_address(Address a) const {
+    // disregard _rspec
+    return base() == a.base() && (has_index() ? index() == a.index() : disp() == a.disp());
+  }
 
   Address after_save() const {
     Address a = (*this);
@@ -436,6 +442,10 @@
     : _address((address) addr),
       _rspec(rspec_from_rtype(rtype, (address) addr)) {}
 
+  AddressLiteral(oop* addr, relocInfo::relocType rtype = relocInfo::none)
+    : _address((address) addr),
+      _rspec(rspec_from_rtype(rtype, (address) addr)) {}
+
   AddressLiteral(float* addr, relocInfo::relocType rtype = relocInfo::none)
     : _address((address) addr),
       _rspec(rspec_from_rtype(rtype, (address) addr)) {}
@@ -455,6 +465,21 @@
   }
 };
 
+// Convenience classes
+class ExternalAddress: public AddressLiteral {
+ private:
+  static relocInfo::relocType reloc_for_target(address target) {
+    // Sometimes ExternalAddress is used for values which aren't
+    // exactly addresses, like the card table base.
+    // external_word_type can't be used for values in the first page
+    // so just skip the reloc in that case.
+    return external_word_Relocation::can_be_relocated(target) ? relocInfo::external_word_type : relocInfo::none;
+  }
+
+ public:
+  ExternalAddress(address target) : AddressLiteral(target, reloc_for_target(          target)) {}
+  ExternalAddress(oop*    target) : AddressLiteral(target, reloc_for_target((address) target)) {}
+};
 
 inline Address RegisterImpl::address_in_saved_window() const {
    return (Address(SP, (sp_offset_in_saved_window() * wordSize) + STACK_BIAS));
@@ -855,9 +880,8 @@
   // and be sign-extended. Check the range.
 
   static void assert_signed_range(intptr_t x, int nbits) {
-    assert( nbits == 32
-        ||  -(1 << nbits-1) <= x  &&  x < ( 1 << nbits-1),
-      "value out of range");
+    assert(nbits == 32 || (-(1 << nbits-1) <= x  &&  x < ( 1 << nbits-1)),
+           err_msg("value out of range: x=" INTPTR_FORMAT ", nbits=%d", x, nbits));
   }
 
   static void assert_signed_word_disp_range(intptr_t x, int nbits) {
@@ -2287,7 +2311,7 @@
   int total_frame_size_in_bytes(int extraWords);
 
   // used when extraWords known statically
-  void save_frame(int extraWords);
+  void save_frame(int extraWords = 0);
   void save_frame_c1(int size_in_bytes);
   // make a frame, and simultaneously pass up one or two register value
   // into the new register window
@@ -2456,9 +2480,11 @@
   // 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,
+                                     Register temp_reg,
                                      int extra_slot_offset = 0);
   // Address of Gargs and argument_offset.
   Address            argument_address(RegisterOrConstant arg_slot,
+                                      Register temp_reg,
                                       int extra_slot_offset = 0);
 
   // Stack overflow checking
--- a/src/cpu/sparc/vm/assembler_sparc.inline.hpp	Thu Jun 02 13:36:11 2011 -0700
+++ b/src/cpu/sparc/vm/assembler_sparc.inline.hpp	Fri Jun 03 22:31:43 2011 -0700
@@ -255,7 +255,11 @@
 inline void Assembler::stf(    FloatRegisterImpl::Width w, FloatRegister d, Register s1, Register s2) { emit_long( op(ldst_op) | fd(d, w) | alt_op3(stf_op3, w) | rs1(s1) | rs2(s2) ); }
 inline void Assembler::stf(    FloatRegisterImpl::Width w, FloatRegister d, Register s1, int simm13a) { emit_data( op(ldst_op) | fd(d, w) | alt_op3(stf_op3, w) | rs1(s1) | immed(true) | simm(simm13a, 13)); }
 
-inline void Assembler::stf(    FloatRegisterImpl::Width w, FloatRegister d, const Address& a, int offset) { relocate(a.rspec(offset)); stf(w, d, a.base(), a.disp() + offset); }
+inline void Assembler::stf(    FloatRegisterImpl::Width w, FloatRegister d, const Address& a, int offset) {
+  relocate(a.rspec(offset));
+  if (a.has_index()) { assert(offset == 0, ""); stf(w, d, a.base(), a.index()        ); }
+  else               {                          stf(w, d, a.base(), a.disp() + offset); }
+}
 
 inline void Assembler::stfsr(  Register s1, Register s2) { v9_dep();   emit_long( op(ldst_op) |             op3(stfsr_op3) | rs1(s1) | rs2(s2) ); }
 inline void Assembler::stfsr(  Register s1, int simm13a) { v9_dep();   emit_data( op(ldst_op) |             op3(stfsr_op3) | rs1(s1) | immed(true) | simm(simm13a, 13)); }
--- a/src/cpu/sparc/vm/frame_sparc.cpp	Thu Jun 02 13:36:11 2011 -0700
+++ b/src/cpu/sparc/vm/frame_sparc.cpp	Fri Jun 03 22:31:43 2011 -0700
@@ -513,6 +513,8 @@
   // interpreted but its pc is in the code cache (for c1 -> osr_frame_return_id stub), so it must be
   // explicitly recognized.
 
+  if (is_ricochet_frame())    return sender_for_ricochet_frame(map);
+
   bool frame_is_interpreted = is_interpreted_frame();
   if (frame_is_interpreted) {
     map->make_integer_regs_unsaved();
--- a/src/cpu/sparc/vm/methodHandles_sparc.cpp	Thu Jun 02 13:36:11 2011 -0700
+++ b/src/cpu/sparc/vm/methodHandles_sparc.cpp	Fri Jun 03 22:31:43 2011 -0700
@@ -69,6 +69,459 @@
   return me;
 }
 
+// stack walking support
+
+frame MethodHandles::ricochet_frame_sender(const frame& fr, RegisterMap *map) {
+  //RicochetFrame* f = RicochetFrame::from_frame(fr);
+  // Cf. is_interpreted_frame path of frame::sender
+  intptr_t* younger_sp = fr.sp();
+  intptr_t* sp         = fr.sender_sp();
+  map->make_integer_regs_unsaved();
+  map->shift_window(sp, younger_sp);
+  bool this_frame_adjusted_stack = true;  // I5_savedSP is live in this RF
+  return frame(sp, younger_sp, this_frame_adjusted_stack);
+}
+
+void MethodHandles::ricochet_frame_oops_do(const frame& fr, OopClosure* blk, const RegisterMap* reg_map) {
+  ResourceMark rm;
+  RicochetFrame* f = RicochetFrame::from_frame(fr);
+
+  // pick up the argument type descriptor:
+  Thread* thread = Thread::current();
+  Handle cookie(thread, f->compute_saved_args_layout(true, true));
+
+  // process fixed part
+  blk->do_oop((oop*)f->saved_target_addr());
+  blk->do_oop((oop*)f->saved_args_layout_addr());
+
+  // process variable arguments:
+  if (cookie.is_null())  return;  // no arguments to describe
+
+  // the cookie is actually the invokeExact method for my target
+  // his argument signature is what I'm interested in
+  assert(cookie->is_method(), "");
+  methodHandle invoker(thread, methodOop(cookie()));
+  assert(invoker->name() == vmSymbols::invokeExact_name(), "must be this kind of method");
+  assert(!invoker->is_static(), "must have MH argument");
+  int slot_count = invoker->size_of_parameters();
+  assert(slot_count >= 1, "must include 'this'");
+  intptr_t* base = f->saved_args_base();
+  intptr_t* retval = NULL;
+  if (f->has_return_value_slot())
+    retval = f->return_value_slot_addr();
+  int slot_num = slot_count - 1;
+  intptr_t* loc = &base[slot_num];
+  //blk->do_oop((oop*) loc);   // original target, which is irrelevant
+  int arg_num = 0;
+  for (SignatureStream ss(invoker->signature()); !ss.is_done(); ss.next()) {
+    if (ss.at_return_type())  continue;
+    BasicType ptype = ss.type();
+    if (ptype == T_ARRAY)  ptype = T_OBJECT; // fold all refs to T_OBJECT
+    assert(ptype >= T_BOOLEAN && ptype <= T_OBJECT, "not array or void");
+    slot_num -= type2size[ptype];
+    loc = &base[slot_num];
+    bool is_oop = (ptype == T_OBJECT && loc != retval);
+    if (is_oop)  blk->do_oop((oop*)loc);
+    arg_num += 1;
+  }
+  assert(slot_num == 0, "must have processed all the arguments");
+}
+
+// Ricochet Frames
+const Register MethodHandles::RicochetFrame::L1_continuation      = L1;
+const Register MethodHandles::RicochetFrame::L2_saved_target      = L2;
+const Register MethodHandles::RicochetFrame::L3_saved_args_layout = L3;
+const Register MethodHandles::RicochetFrame::L4_saved_args_base   = L4; // cf. Gargs = G4
+const Register MethodHandles::RicochetFrame::L5_conversion        = L5;
+#ifdef ASSERT
+const Register MethodHandles::RicochetFrame::L0_magic_number_1    = L0;
+#endif //ASSERT
+
+oop MethodHandles::RicochetFrame::compute_saved_args_layout(bool read_cache, bool write_cache) {
+  if (read_cache) {
+    oop cookie = saved_args_layout();
+    if (cookie != NULL)  return cookie;
+  }
+  oop target = saved_target();
+  oop mtype  = java_lang_invoke_MethodHandle::type(target);
+  oop mtform = java_lang_invoke_MethodType::form(mtype);
+  oop cookie = java_lang_invoke_MethodTypeForm::vmlayout(mtform);
+  if (write_cache)  {
+    (*saved_args_layout_addr()) = cookie;
+  }
+  return cookie;
+}
+
+void MethodHandles::RicochetFrame::generate_ricochet_blob(MacroAssembler* _masm,
+                                                          // output params:
+                                                          int* bounce_offset,
+                                                          int* exception_offset,
+                                                          int* frame_size_in_words) {
+  (*frame_size_in_words) = RicochetFrame::frame_size_in_bytes() / wordSize;
+
+  address start = __ pc();
+
+#ifdef ASSERT
+  __ illtrap(0); __ illtrap(0); __ illtrap(0);
+  // here's a hint of something special:
+  __ set(MAGIC_NUMBER_1, G0);
+  __ set(MAGIC_NUMBER_2, G0);
+#endif //ASSERT
+  __ illtrap(0);  // not reached
+
+  // Return values are in registers.
+  // L1_continuation contains a cleanup continuation we must return
+  // to.
+
+  (*bounce_offset) = __ pc() - start;
+  BLOCK_COMMENT("ricochet_blob.bounce");
+
+  if (VerifyMethodHandles)  RicochetFrame::verify_clean(_masm);
+  trace_method_handle(_masm, "ricochet_blob.bounce");
+
+  __ JMP(L1_continuation, 0);
+  __ delayed()->nop();
+  __ illtrap(0);
+
+  DEBUG_ONLY(__ set(MAGIC_NUMBER_2, G0));
+
+  (*exception_offset) = __ pc() - start;
+  BLOCK_COMMENT("ricochet_blob.exception");
+
+  // compare this to Interpreter::rethrow_exception_entry, which is parallel code
+  // for example, see TemplateInterpreterGenerator::generate_throw_exception
+  // Live registers in:
+  //   Oexception  (O0): exception
+  //   Oissuing_pc (O1): return address/pc that threw exception (ignored, always equal to bounce addr)
+  __ verify_oop(Oexception);
+
+  // Take down the frame.
+
+  // Cf. InterpreterMacroAssembler::remove_activation.
+  leave_ricochet_frame(_masm, /*recv_reg=*/ noreg, I5_savedSP, I7);
+
+  // We are done with this activation frame; find out where to go next.
+  // The continuation point will be an exception handler, which expects
+  // the following registers set up:
+  //
+  // Oexception: exception
+  // Oissuing_pc: the local call that threw exception
+  // Other On: garbage
+  // In/Ln:  the contents of the caller's register window
+  //
+  // We do the required restore at the last possible moment, because we
+  // need to preserve some state across a runtime call.
+  // (Remember that the caller activation is unknown--it might not be
+  // interpreted, so things like Lscratch are useless in the caller.)
+  __ mov(Oexception,  Oexception ->after_save());  // get exception in I0 so it will be on O0 after restore
+  __ add(I7, frame::pc_return_offset, Oissuing_pc->after_save());  // likewise set I1 to a value local to the caller
+  __ call_VM_leaf(L7_thread_cache,
+                  CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address),
+                  G2_thread, Oissuing_pc->after_save());
+
+  // The caller's SP was adjusted upon method entry to accomodate
+  // the callee's non-argument locals. Undo that adjustment.
+  __ JMP(O0, 0);                         // return exception handler in caller
+  __ delayed()->restore(I5_savedSP, G0, SP);
+
+  // (same old exception object is already in Oexception; see above)
+  // Note that an "issuing PC" is actually the next PC after the call
+}
+
+void MethodHandles::RicochetFrame::enter_ricochet_frame(MacroAssembler* _masm,
+                                                        Register recv_reg,
+                                                        Register argv_reg,
+                                                        address return_handler) {
+  // does not include the __ save()
+  assert(argv_reg == Gargs, "");
+  Address G3_mh_vmtarget(   recv_reg, java_lang_invoke_MethodHandle::vmtarget_offset_in_bytes());
+  Address G3_amh_conversion(recv_reg, java_lang_invoke_AdapterMethodHandle::conversion_offset_in_bytes());
+
+  // Create the RicochetFrame.
+  // Unlike on x86 we can store all required information in local
+  // registers.
+  BLOCK_COMMENT("push RicochetFrame {");
+  __ set(ExternalAddress(return_handler),          L1_continuation);
+  __ load_heap_oop(G3_mh_vmtarget,                 L2_saved_target);
+  __ mov(G0,                                       L3_saved_args_layout);
+  __ mov(Gargs,                                    L4_saved_args_base);
+  __ lduw(G3_amh_conversion,                       L5_conversion);  // 32-bit field
+  // I5, I6, I7 are already set up
+  DEBUG_ONLY(__ set((int32_t) MAGIC_NUMBER_1,      L0_magic_number_1));
+  BLOCK_COMMENT("} RicochetFrame");
+}
+
+void MethodHandles::RicochetFrame::leave_ricochet_frame(MacroAssembler* _masm,
+                                                        Register recv_reg,
+                                                        Register new_sp_reg,
+                                                        Register sender_pc_reg) {
+  assert(new_sp_reg == I5_savedSP, "exact_sender_sp already in place");
+  assert(sender_pc_reg == I7, "in a fixed place");
+  // does not include the __ ret() & __ restore()
+  assert_different_registers(recv_reg, new_sp_reg, sender_pc_reg);
+  // Take down the frame.
+  // Cf. InterpreterMacroAssembler::remove_activation.
+  BLOCK_COMMENT("end_ricochet_frame {");
+  if (recv_reg->is_valid())
+    __ mov(L2_saved_target, recv_reg);
+  BLOCK_COMMENT("} end_ricochet_frame");
+}
+
+// Emit code to verify that FP is pointing at a valid ricochet frame.
+#ifdef ASSERT
+enum {
+  ARG_LIMIT = 255, SLOP = 35,
+  // use this parameter for checking for garbage stack movements:
+  UNREASONABLE_STACK_MOVE = (ARG_LIMIT + SLOP)
+  // the slop defends against false alarms due to fencepost errors
+};
+
+void MethodHandles::RicochetFrame::verify_clean(MacroAssembler* _masm) {
+  // The stack should look like this:
+  //    ... keep1 | dest=42 | keep2 | magic | handler | magic | recursive args | [RF]
+  // Check various invariants.
+
+  Register O7_temp = O7, O5_temp = O5;
+
+  Label L_ok_1, L_ok_2, L_ok_3, L_ok_4;
+  BLOCK_COMMENT("verify_clean {");
+  // Magic numbers must check out:
+  __ set((int32_t) MAGIC_NUMBER_1, O7_temp);
+  __ cmp(O7_temp, L0_magic_number_1);
+  __ br(Assembler::equal, false, Assembler::pt, L_ok_1);
+  __ delayed()->nop();
+  __ stop("damaged ricochet frame: MAGIC_NUMBER_1 not found");
+
+  __ BIND(L_ok_1);
+
+  // Arguments pointer must look reasonable:
+#ifdef _LP64
+  Register FP_temp = O5_temp;
+  __ add(FP, STACK_BIAS, FP_temp);
+#else
+  Register FP_temp = FP;
+#endif
+  __ cmp(L4_saved_args_base, FP_temp);
+  __ br(Assembler::greaterEqualUnsigned, false, Assembler::pt, L_ok_2);
+  __ delayed()->nop();
+  __ stop("damaged ricochet frame: L4 < FP");
+
+  __ BIND(L_ok_2);
+  __ sub(L4_saved_args_base, UNREASONABLE_STACK_MOVE * Interpreter::stackElementSize, O7_temp);
+  __ cmp(O7_temp, FP_temp);
+  __ br(Assembler::lessEqualUnsigned, false, Assembler::pt, L_ok_3);
+  __ delayed()->nop();
+  __ stop("damaged ricochet frame: (L4 - UNREASONABLE_STACK_MOVE) > FP");
+
+  __ BIND(L_ok_3);
+  extract_conversion_dest_type(_masm, L5_conversion, O7_temp);
+  __ cmp(O7_temp, T_VOID);
+  __ br(Assembler::equal, false, Assembler::pt, L_ok_4);
+  __ delayed()->nop();
+  extract_conversion_vminfo(_masm, L5_conversion, O5_temp);
+  __ ld_ptr(L4_saved_args_base, __ argument_offset(O5_temp, O5_temp), O7_temp);
+  assert(__ is_simm13(RETURN_VALUE_PLACEHOLDER), "must be simm13");
+  __ cmp(O7_temp, (int32_t) RETURN_VALUE_PLACEHOLDER);
+  __ brx(Assembler::equal, false, Assembler::pt, L_ok_4);
+  __ delayed()->nop();
+  __ stop("damaged ricochet frame: RETURN_VALUE_PLACEHOLDER not found");
+  __ BIND(L_ok_4);
+  BLOCK_COMMENT("} verify_clean");
+}
+#endif //ASSERT
+
+void MethodHandles::load_klass_from_Class(MacroAssembler* _masm, Register klass_reg, Register temp_reg, Register temp2_reg) {
+  if (VerifyMethodHandles)
+    verify_klass(_masm, klass_reg, SystemDictionaryHandles::Class_klass(), temp_reg, temp2_reg,
+                 "AMH argument is a Class");
+  __ load_heap_oop(Address(klass_reg, java_lang_Class::klass_offset_in_bytes()), klass_reg);
+}
+
+void MethodHandles::load_conversion_vminfo(MacroAssembler* _masm, Address conversion_field_addr, Register reg) {
+  assert(CONV_VMINFO_SHIFT == 0, "preshifted");
+  assert(CONV_VMINFO_MASK == right_n_bits(BitsPerByte), "else change type of following load");
+  __ ldub(conversion_field_addr.plus_disp(BytesPerInt - 1), reg);
+}
+
+void MethodHandles::extract_conversion_vminfo(MacroAssembler* _masm, Register conversion_field_reg, Register reg) {
+  assert(CONV_VMINFO_SHIFT == 0, "preshifted");
+  __ and3(conversion_field_reg, CONV_VMINFO_MASK, reg);
+}
+
+void MethodHandles::extract_conversion_dest_type(MacroAssembler* _masm, Register conversion_field_reg, Register reg) {
+  __ srl(conversion_field_reg, CONV_DEST_TYPE_SHIFT, reg);
+  __ and3(reg, 0x0F, reg);
+}
+
+void MethodHandles::load_stack_move(MacroAssembler* _masm,
+                                    Address G3_amh_conversion,
+                                    Register stack_move_reg) {
+  BLOCK_COMMENT("load_stack_move {");
+  __ ldsw(G3_amh_conversion, stack_move_reg);
+  __ sra(stack_move_reg, CONV_STACK_MOVE_SHIFT, stack_move_reg);
+  if (VerifyMethodHandles) {
+    Label L_ok, L_bad;
+    int32_t stack_move_limit = 0x0800;  // extra-large
+    __ cmp(stack_move_reg, stack_move_limit);
+    __ br(Assembler::greaterEqual, false, Assembler::pn, L_bad);
+    __ delayed()->nop();
+    __ cmp(stack_move_reg, -stack_move_limit);
+    __ br(Assembler::greater, false, Assembler::pt, L_ok);
+    __ delayed()->nop();
+    __ BIND(L_bad);
+    __ stop("load_stack_move of garbage value");
+    __ BIND(L_ok);
+  }
+  BLOCK_COMMENT("} load_stack_move");
+}
+
+#ifdef ASSERT
+void MethodHandles::RicochetFrame::verify() const {
+  assert(magic_number_1() == MAGIC_NUMBER_1, "");
+  if (!Universe::heap()->is_gc_active()) {
+    if (saved_args_layout() != NULL) {
+      assert(saved_args_layout()->is_method(), "must be valid oop");
+    }
+    if (saved_target() != NULL) {
+      assert(java_lang_invoke_MethodHandle::is_instance(saved_target()), "checking frame value");
+    }
+  }
+  int conv_op = adapter_conversion_op(conversion());
+  assert(conv_op == java_lang_invoke_AdapterMethodHandle::OP_COLLECT_ARGS ||
+         conv_op == java_lang_invoke_AdapterMethodHandle::OP_FOLD_ARGS ||
+         conv_op == java_lang_invoke_AdapterMethodHandle::OP_PRIM_TO_REF,
+         "must be a sane conversion");
+  if (has_return_value_slot()) {
+    assert(*return_value_slot_addr() == RETURN_VALUE_PLACEHOLDER, "");
+  }
+}
+
+void MethodHandles::verify_argslot(MacroAssembler* _masm, Register argslot_reg, Register temp_reg, const char* error_message) {
+  // Verify that argslot lies within (Gargs, FP].
+  Label L_ok, L_bad;
+  BLOCK_COMMENT("verify_argslot {");
+  __ add(FP, STACK_BIAS, temp_reg);  // STACK_BIAS is zero on !_LP64
+  __ cmp(argslot_reg, temp_reg);
+  __ brx(Assembler::greaterUnsigned, false, Assembler::pn, L_bad);
+  __ delayed()->nop();
+  __ cmp(Gargs, argslot_reg);
+  __ brx(Assembler::lessEqualUnsigned, false, Assembler::pt, L_ok);
+  __ delayed()->nop();
+  __ BIND(L_bad);
+  __ stop(error_message);
+  __ BIND(L_ok);
+  BLOCK_COMMENT("} verify_argslot");
+}
+
+void MethodHandles::verify_argslots(MacroAssembler* _masm,
+                                    RegisterOrConstant arg_slots,
+                                    Register arg_slot_base_reg,
+                                    Register temp_reg,
+                                    Register temp2_reg,
+                                    bool negate_argslots,
+                                    const char* error_message) {
+  // Verify that [argslot..argslot+size) lies within (Gargs, FP).
+  Label L_ok, L_bad;
+  BLOCK_COMMENT("verify_argslots {");
+  if (negate_argslots) {
+    if (arg_slots.is_constant()) {
+      arg_slots = -1 * arg_slots.as_constant();
+    } else {
+      __ neg(arg_slots.as_register(), temp_reg);
+      arg_slots = temp_reg;
+    }
+  }
+  __ add(arg_slot_base_reg, __ argument_offset(arg_slots, temp_reg), temp_reg);
+  __ add(FP, STACK_BIAS, temp2_reg);  // STACK_BIAS is zero on !_LP64
+  __ cmp(temp_reg, temp2_reg);
+  __ brx(Assembler::greaterUnsigned, false, Assembler::pn, L_bad);
+  __ delayed()->nop();
+  // Gargs points to the first word so adjust by BytesPerWord
+  __ add(arg_slot_base_reg, BytesPerWord, temp_reg);
+  __ cmp(Gargs, temp_reg);
+  __ brx(Assembler::lessEqualUnsigned, false, Assembler::pt, L_ok);
+  __ delayed()->nop();
+  __ BIND(L_bad);
+  __ stop(error_message);
+  __ BIND(L_ok);
+  BLOCK_COMMENT("} verify_argslots");
+}
+
+// Make sure that arg_slots has the same sign as the given direction.
+// If (and only if) arg_slots is a assembly-time constant, also allow it to be zero.
+void MethodHandles::verify_stack_move(MacroAssembler* _masm,
+                                      RegisterOrConstant arg_slots, int direction) {
+  enum { UNREASONABLE_STACK_MOVE = 256 * 4 };  // limit of 255 arguments
+  bool allow_zero = arg_slots.is_constant();
+  if (direction == 0) { direction = +1; allow_zero = true; }
+  assert(stack_move_unit() == -1, "else add extra checks here");
+  if (arg_slots.is_register()) {
+    Label L_ok, L_bad;
+    BLOCK_COMMENT("verify_stack_move {");
+    // __ btst(-stack_move_unit() - 1, arg_slots.as_register());  // no need
+    // __ br(Assembler::notZero, false, Assembler::pn, L_bad);
+    // __ delayed()->nop();
+    __ cmp(arg_slots.as_register(), (int32_t) NULL_WORD);
+    if (direction > 0) {
+      __ br(allow_zero ? Assembler::less : Assembler::lessEqual, false, Assembler::pn, L_bad);
+      __ delayed()->nop();
+      __ cmp(arg_slots.as_register(), (int32_t) UNREASONABLE_STACK_MOVE);
+      __ br(Assembler::less, false, Assembler::pn, L_ok);
+      __ delayed()->nop();
+    } else {
+      __ br(allow_zero ? Assembler::greater : Assembler::greaterEqual, false, Assembler::pn, L_bad);
+      __ delayed()->nop();
+      __ cmp(arg_slots.as_register(), (int32_t) -UNREASONABLE_STACK_MOVE);
+      __ br(Assembler::greater, false, Assembler::pn, L_ok);
+      __ delayed()->nop();
+    }
+    __ BIND(L_bad);
+    if (direction > 0)
+      __ stop("assert arg_slots > 0");
+    else
+      __ stop("assert arg_slots < 0");
+    __ BIND(L_ok);
+    BLOCK_COMMENT("} verify_stack_move");
+  } else {
+    intptr_t size = arg_slots.as_constant();
+    if (direction < 0)  size = -size;
+    assert(size >= 0, "correct direction of constant move");
+    assert(size < UNREASONABLE_STACK_MOVE, "reasonable size of constant move");
+  }
+}
+
+void MethodHandles::verify_klass(MacroAssembler* _masm,
+                                 Register obj_reg, KlassHandle klass,
+                                 Register temp_reg, Register temp2_reg,
+                                 const char* error_message) {
+  oop* klass_addr = klass.raw_value();
+  assert(klass_addr >= SystemDictionaryHandles::Object_klass().raw_value() &&
+         klass_addr <= SystemDictionaryHandles::Long_klass().raw_value(),
+         "must be one of the SystemDictionaryHandles");
+  Label L_ok, L_bad;
+  BLOCK_COMMENT("verify_klass {");
+  __ verify_oop(obj_reg);
+  __ br_null(obj_reg, false, Assembler::pn, L_bad);
+  __ delayed()->nop();
+  __ load_klass(obj_reg, temp_reg);
+  __ set(ExternalAddress(klass_addr), temp2_reg);
+  __ ld_ptr(Address(temp2_reg, 0), temp2_reg);
+  __ cmp(temp_reg, temp2_reg);
+  __ brx(Assembler::equal, false, Assembler::pt, L_ok);
+  __ delayed()->nop();
+  intptr_t super_check_offset = klass->super_check_offset();
+  __ ld_ptr(Address(temp_reg, super_check_offset), temp_reg);
+  __ set(ExternalAddress(klass_addr), temp2_reg);
+  __ ld_ptr(Address(temp2_reg, 0), temp2_reg);
+  __ cmp(temp_reg, temp2_reg);
+  __ brx(Assembler::equal, false, Assembler::pt, L_ok);
+  __ delayed()->nop();
+  __ BIND(L_bad);
+  __ stop(error_message);
+  __ BIND(L_ok);
+  BLOCK_COMMENT("} verify_klass");
+}
+#endif // ASSERT
 
 // Code generation
 address MethodHandles::generate_method_handle_interpreter_entry(MacroAssembler* _masm) {
@@ -103,6 +556,9 @@
   address entry_point = __ pc();
 
   // fetch the MethodType from the method handle
+  // FIXME: Interpreter should transmit pre-popped stack pointer, to locate base of arg list.
+  // This would simplify several touchy bits of code.
+  // See 6984712: JSR 292 method handle calls need a clean argument base pointer
   {
     Register tem = G5_method;
     for (jint* pchase = methodOopDesc::method_type_offsets_chain(); (*pchase) != -1; pchase++) {
@@ -114,19 +570,25 @@
   // given the MethodType, find out where the MH argument is buried
   __ load_heap_oop(Address(O0_mtype,   __ delayed_value(java_lang_invoke_MethodType::form_offset_in_bytes,        O1_scratch)), O4_argslot);
   __ ldsw(         Address(O4_argslot, __ delayed_value(java_lang_invoke_MethodTypeForm::vmslots_offset_in_bytes, O1_scratch)), O4_argslot);
-  __ add(Gargs, __ argument_offset(O4_argslot, 1), O4_argbase);
+  __ add(__ argument_address(O4_argslot, O4_argslot, 1), O4_argbase);
   // Note: argument_address uses its input as a scratch register!
-  __ ld_ptr(Address(O4_argbase, -Interpreter::stackElementSize), G3_method_handle);
+  Address mh_receiver_slot_addr(O4_argbase, -Interpreter::stackElementSize);
+  __ ld_ptr(mh_receiver_slot_addr, G3_method_handle);
 
   trace_method_handle(_masm, "invokeExact");
 
   __ check_method_handle_type(O0_mtype, G3_method_handle, O1_scratch, wrong_method_type);
+
+  // Nobody uses the MH receiver slot after this.  Make sure.
+  DEBUG_ONLY(__ set((int32_t) 0x999999, O1_scratch); __ st_ptr(O1_scratch, mh_receiver_slot_addr));
+
   __ jump_to_method_handle_entry(G3_method_handle, O1_scratch);
 
   // for invokeGeneric (only), apply argument and result conversions on the fly
   __ bind(invoke_generic_slow_path);
 #ifdef ASSERT
-  { Label L;
+  if (VerifyMethodHandles) {
+    Label L;
     __ ldub(Address(G5_method, methodOopDesc::intrinsic_id_offset_in_bytes()), O1_scratch);
     __ cmp(O1_scratch, (int) vmIntrinsics::_invokeGeneric);
     __ brx(Assembler::equal, false, Assembler::pt, L);
@@ -137,7 +599,7 @@
 #endif //ASSERT
 
   // make room on the stack for another pointer:
-  insert_arg_slots(_masm, 2 * stack_move_unit(), _INSERT_REF_MASK, O4_argbase, O1_scratch, O2_scratch, O3_scratch);
+  insert_arg_slots(_masm, 2 * stack_move_unit(), O4_argbase, O1_scratch, O2_scratch, O3_scratch);
   // load up an adapter from the calling type (Java weaves this)
   Register O2_form    = O2_scratch;
   Register O3_adapter = O3_scratch;
@@ -157,74 +619,88 @@
   return entry_point;
 }
 
+// Workaround for C++ overloading nastiness on '0' for RegisterOrConstant.
+static RegisterOrConstant constant(int value) {
+  return RegisterOrConstant(value);
+}
 
+static void load_vmargslot(MacroAssembler* _masm, Address vmargslot_addr, Register result) {
+  __ ldsw(vmargslot_addr, result);
+}
+
+static RegisterOrConstant adjust_SP_and_Gargs_down_by_slots(MacroAssembler* _masm,
+                                                            RegisterOrConstant arg_slots,
+                                                            Register temp_reg, Register temp2_reg) {
+  // Keep the stack pointer 2*wordSize aligned.
+  const int TwoWordAlignmentMask = right_n_bits(LogBytesPerWord + 1);
+  if (arg_slots.is_constant()) {
+    const int        offset = arg_slots.as_constant() << LogBytesPerWord;
+    const int masked_offset = round_to(offset, 2 * BytesPerWord);
+    const int masked_offset2 = (offset + 1*BytesPerWord) & ~TwoWordAlignmentMask;
+    assert(masked_offset == masked_offset2, "must agree");
+    __ sub(Gargs,        offset, Gargs);
+    __ sub(SP,    masked_offset, SP   );
+    return offset;
+  } else {
 #ifdef ASSERT
-static void verify_argslot(MacroAssembler* _masm, Register argslot_reg, Register temp_reg, const char* error_message) {
-  // Verify that argslot lies within (Gargs, FP].
-  Label L_ok, L_bad;
-  BLOCK_COMMENT("{ verify_argslot");
-#ifdef _LP64
-  __ add(FP, STACK_BIAS, temp_reg);
-  __ cmp(argslot_reg, temp_reg);
-#else
-  __ cmp(argslot_reg, FP);
+    {
+      Label L_ok;
+      __ cmp(arg_slots.as_register(), 0);
+      __ br(Assembler::greaterEqual, false, Assembler::pt, L_ok);
+      __ delayed()->nop();
+      __ stop("negative arg_slots");
+      __ bind(L_ok);
+    }
 #endif
-  __ brx(Assembler::greaterUnsigned, false, Assembler::pn, L_bad);
-  __ delayed()->nop();
-  __ cmp(Gargs, argslot_reg);
-  __ brx(Assembler::lessEqualUnsigned, false, Assembler::pt, L_ok);
-  __ delayed()->nop();
-  __ bind(L_bad);
-  __ stop(error_message);
-  __ bind(L_ok);
-  BLOCK_COMMENT("} verify_argslot");
+    __ sll_ptr(arg_slots.as_register(), LogBytesPerWord, temp_reg);
+    __ add( temp_reg,  1*BytesPerWord,       temp2_reg);
+    __ andn(temp2_reg, TwoWordAlignmentMask, temp2_reg);
+    __ sub(Gargs, temp_reg,  Gargs);
+    __ sub(SP,    temp2_reg, SP   );
+    return temp_reg;
+  }
 }
-#endif
 
+static RegisterOrConstant adjust_SP_and_Gargs_up_by_slots(MacroAssembler* _masm,
+                                                          RegisterOrConstant arg_slots,
+                                                          Register temp_reg, Register temp2_reg) {
+  // Keep the stack pointer 2*wordSize aligned.
+  const int TwoWordAlignmentMask = right_n_bits(LogBytesPerWord + 1);
+  if (arg_slots.is_constant()) {
+    const int        offset = arg_slots.as_constant() << LogBytesPerWord;
+    const int masked_offset = offset & ~TwoWordAlignmentMask;
+    __ add(Gargs,        offset, Gargs);
+    __ add(SP,    masked_offset, SP   );
+    return offset;
+  } else {
+    __ sll_ptr(arg_slots.as_register(), LogBytesPerWord, temp_reg);
+    __ andn(temp_reg, TwoWordAlignmentMask, temp2_reg);
+    __ add(Gargs, temp_reg,  Gargs);
+    __ add(SP,    temp2_reg, SP   );
+    return temp_reg;
+  }
+}
 
 // Helper to insert argument slots into the stack.
-// arg_slots must be a multiple of stack_move_unit() and <= 0
+// arg_slots must be a multiple of stack_move_unit() and < 0
+// argslot_reg is decremented to point to the new (shifted) location of the argslot
+// But, temp_reg ends up holding the original value of argslot_reg.
 void MethodHandles::insert_arg_slots(MacroAssembler* _masm,
                                      RegisterOrConstant arg_slots,
-                                     int arg_mask,
                                      Register argslot_reg,
                                      Register temp_reg, Register temp2_reg, Register temp3_reg) {
-  assert(temp3_reg != noreg, "temp3 required");
+  // allow constant zero
+  if (arg_slots.is_constant() && arg_slots.as_constant() == 0)
+    return;
+
   assert_different_registers(argslot_reg, temp_reg, temp2_reg, temp3_reg,
                              (!arg_slots.is_register() ? Gargs : arg_slots.as_register()));
 
-#ifdef ASSERT
-  verify_argslot(_masm, argslot_reg, temp_reg, "insertion point must fall within current frame");
-  if (arg_slots.is_register()) {
-    Label L_ok, L_bad;
-    __ cmp(arg_slots.as_register(), (int32_t) NULL_WORD);
-    __ br(Assembler::greater, false, Assembler::pn, L_bad);
-    __ delayed()->nop();
-    __ btst(-stack_move_unit() - 1, arg_slots.as_register());
-    __ br(Assembler::zero, false, Assembler::pt, L_ok);
-    __ delayed()->nop();
-    __ 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()) {
-    // Was arg_slots register loaded as signed int?
-    Label L_ok;
-    __ sll(arg_slots.as_register(), BitsPerInt, temp_reg);
-    __ sra(temp_reg, BitsPerInt, temp_reg);
-    __ cmp(arg_slots.as_register(), temp_reg);
-    __ br(Assembler::equal, false, Assembler::pt, L_ok);
-    __ delayed()->nop();
-    __ stop("arg_slots register not loaded as signed int");
-    __ bind(L_ok);
-  }
-#endif
+  BLOCK_COMMENT("insert_arg_slots {");
+  if (VerifyMethodHandles)
+    verify_argslot(_masm, argslot_reg, temp_reg, "insertion point must fall within current frame");
+  if (VerifyMethodHandles)
+    verify_stack_move(_masm, arg_slots, -1);
 
   // Make space on the stack for the inserted argument(s).
   // Then pull down everything shallower than argslot_reg.
@@ -234,26 +710,20 @@
   //   for (temp = sp + size; temp < argslot; temp++)
   //     temp[-size] = temp[0]
   //   argslot -= size;
-  BLOCK_COMMENT("insert_arg_slots {");
-  RegisterOrConstant offset = __ regcon_sll_ptr(arg_slots, LogBytesPerWord, temp3_reg);
 
-  // Keep the stack pointer 2*wordSize aligned.
-  const int TwoWordAlignmentMask = right_n_bits(LogBytesPerWord + 1);
-  RegisterOrConstant masked_offset = __ regcon_andn_ptr(offset, TwoWordAlignmentMask, temp_reg);
-  __ add(SP, masked_offset, SP);
-
-  __ mov(Gargs, temp_reg);  // source pointer for copy
-  __ add(Gargs, offset, Gargs);
+  // offset is temp3_reg in case of arg_slots being a register.
+  RegisterOrConstant offset = adjust_SP_and_Gargs_up_by_slots(_masm, arg_slots, temp3_reg, temp_reg);
+  __ sub(Gargs, offset, temp_reg);  // source pointer for copy
 
   {
     Label loop;
     __ BIND(loop);
     // pull one word down each time through the loop
-    __ ld_ptr(Address(temp_reg, 0), temp2_reg);
-    __ st_ptr(temp2_reg, Address(temp_reg, offset));
+    __ ld_ptr(           Address(temp_reg, 0     ), temp2_reg);
+    __ st_ptr(temp2_reg, Address(temp_reg, offset)           );
     __ add(temp_reg, wordSize, temp_reg);
     __ cmp(temp_reg, argslot_reg);
-    __ brx(Assembler::less, false, Assembler::pt, loop);
+    __ brx(Assembler::lessUnsigned, false, Assembler::pt, loop);
     __ delayed()->nop();  // FILLME
   }
 
@@ -264,39 +734,24 @@
 
 
 // Helper to remove argument slots from the stack.
-// arg_slots must be a multiple of stack_move_unit() and >= 0
+// arg_slots must be a multiple of stack_move_unit() and > 0
 void MethodHandles::remove_arg_slots(MacroAssembler* _masm,
                                      RegisterOrConstant arg_slots,
                                      Register argslot_reg,
                                      Register temp_reg, Register temp2_reg, Register temp3_reg) {
-  assert(temp3_reg != noreg, "temp3 required");
+  // allow constant zero
+  if (arg_slots.is_constant() && arg_slots.as_constant() == 0)
+    return;
   assert_different_registers(argslot_reg, temp_reg, temp2_reg, temp3_reg,
                              (!arg_slots.is_register() ? Gargs : arg_slots.as_register()));
 
-  RegisterOrConstant offset = __ regcon_sll_ptr(arg_slots, LogBytesPerWord, temp3_reg);
+  BLOCK_COMMENT("remove_arg_slots {");
+  if (VerifyMethodHandles)
+    verify_argslots(_masm, arg_slots, argslot_reg, temp_reg, temp2_reg, false,
+                    "deleted argument(s) must fall within current frame");
+  if (VerifyMethodHandles)
+    verify_stack_move(_masm, arg_slots, +1);
 
-#ifdef ASSERT
-  // Verify that [argslot..argslot+size) lies within (Gargs, FP).
-  __ add(argslot_reg, offset, temp2_reg);
-  verify_argslot(_masm, temp2_reg, temp_reg, "deleted argument(s) must fall within current frame");
-  if (arg_slots.is_register()) {
-    Label L_ok, L_bad;
-    __ cmp(arg_slots.as_register(), (int32_t) NULL_WORD);
-    __ br(Assembler::less, false, Assembler::pn, L_bad);
-    __ delayed()->nop();
-    __ btst(-stack_move_unit() - 1, arg_slots.as_register());
-    __ br(Assembler::zero, false, Assembler::pt, L_ok);
-    __ delayed()->nop();
-    __ 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
-
-  BLOCK_COMMENT("remove_arg_slots {");
   // Pull up everything shallower than argslot.
   // Then remove the excess space on the stack.
   // The stacked return address gets pulled up with everything else.
@@ -305,39 +760,271 @@
   //     temp[size] = temp[0]
   //   argslot += size;
   //   sp += size;
+
+  RegisterOrConstant offset = __ regcon_sll_ptr(arg_slots, LogBytesPerWord, temp3_reg);
   __ sub(argslot_reg, wordSize, temp_reg);  // source pointer for copy
+
   {
-    Label loop;
-    __ BIND(loop);
+    Label L_loop;
+    __ BIND(L_loop);
     // pull one word up each time through the loop
-    __ ld_ptr(Address(temp_reg, 0), temp2_reg);
-    __ st_ptr(temp2_reg, Address(temp_reg, offset));
+    __ ld_ptr(           Address(temp_reg, 0     ), temp2_reg);
+    __ st_ptr(temp2_reg, Address(temp_reg, offset)           );
     __ sub(temp_reg, wordSize, temp_reg);
     __ cmp(temp_reg, Gargs);
-    __ brx(Assembler::greaterEqual, false, Assembler::pt, loop);
+    __ brx(Assembler::greaterEqualUnsigned, false, Assembler::pt, L_loop);
     __ delayed()->nop();  // FILLME
   }
 
-  // Now move the argslot up, to point to the just-copied block.
-  __ add(Gargs, offset, Gargs);
   // And adjust the argslot address to point at the deletion point.
   __ add(argslot_reg, offset, argslot_reg);
 
-  // Keep the stack pointer 2*wordSize aligned.
-  const int TwoWordAlignmentMask = right_n_bits(LogBytesPerWord + 1);
-  RegisterOrConstant masked_offset = __ regcon_andn_ptr(offset, TwoWordAlignmentMask, temp_reg);
-  __ add(SP, masked_offset, SP);
+  // We don't need the offset at this point anymore, just adjust SP and Gargs.
+  (void) adjust_SP_and_Gargs_up_by_slots(_masm, arg_slots, temp3_reg, temp_reg);
+
   BLOCK_COMMENT("} remove_arg_slots");
 }
 
+// Helper to copy argument slots to the top of the stack.
+// The sequence starts with argslot_reg and is counted by slot_count
+// slot_count must be a multiple of stack_move_unit() and >= 0
+// This function blows the temps but does not change argslot_reg.
+void MethodHandles::push_arg_slots(MacroAssembler* _masm,
+                                   Register argslot_reg,
+                                   RegisterOrConstant slot_count,
+                                   Register temp_reg, Register temp2_reg) {
+  // allow constant zero
+  if (slot_count.is_constant() && slot_count.as_constant() == 0)
+    return;
+  assert_different_registers(argslot_reg, temp_reg, temp2_reg,
+                             (!slot_count.is_register() ? Gargs : slot_count.as_register()),
+                             SP);
+  assert(Interpreter::stackElementSize == wordSize, "else change this code");
+
+  BLOCK_COMMENT("push_arg_slots {");
+  if (VerifyMethodHandles)
+    verify_stack_move(_masm, slot_count, 0);
+
+  RegisterOrConstant offset = adjust_SP_and_Gargs_down_by_slots(_masm, slot_count, temp2_reg, temp_reg);
+
+  if (slot_count.is_constant()) {
+    for (int i = slot_count.as_constant() - 1; i >= 0; i--) {
+      __ ld_ptr(          Address(argslot_reg, i * wordSize), temp_reg);
+      __ st_ptr(temp_reg, Address(Gargs,       i * wordSize));
+    }
+  } else {
+    Label L_plural, L_loop, L_break;
+    // Emit code to dynamically check for the common cases, zero and one slot.
+    __ cmp(slot_count.as_register(), (int32_t) 1);
+    __ br(Assembler::greater, false, Assembler::pn, L_plural);
+    __ delayed()->nop();
+    __ br(Assembler::less, false, Assembler::pn, L_break);
+    __ delayed()->nop();
+    __ ld_ptr(          Address(argslot_reg, 0), temp_reg);
+    __ st_ptr(temp_reg, Address(Gargs,       0));
+    __ ba(false, L_break);
+    __ delayed()->nop();  // FILLME
+    __ BIND(L_plural);
+
+    // Loop for 2 or more:
+    //   top = &argslot[slot_count]
+    //   while (top > argslot)  *(--Gargs) = *(--top)
+    Register top_reg = temp_reg;
+    __ add(argslot_reg, offset, top_reg);
+    __ add(Gargs,       offset, Gargs  );  // move back up again so we can go down
+    __ BIND(L_loop);
+    __ sub(top_reg, wordSize, top_reg);
+    __ sub(Gargs,   wordSize, Gargs  );
+    __ ld_ptr(           Address(top_reg, 0), temp2_reg);
+    __ st_ptr(temp2_reg, Address(Gargs,   0));
+    __ cmp(top_reg, argslot_reg);
+    __ brx(Assembler::greaterUnsigned, false, Assembler::pt, L_loop);
+    __ delayed()->nop();  // FILLME
+    __ BIND(L_break);
+  }
+  BLOCK_COMMENT("} push_arg_slots");
+}
+
+// in-place movement; no change to Gargs
+// blows temp_reg, temp2_reg
+void MethodHandles::move_arg_slots_up(MacroAssembler* _masm,
+                                      Register bottom_reg,  // invariant
+                                      Address  top_addr,    // can use temp_reg
+                                      RegisterOrConstant positive_distance_in_slots,  // destroyed if register
+                                      Register temp_reg, Register temp2_reg) {
+  assert_different_registers(bottom_reg,
+                             temp_reg, temp2_reg,
+                             positive_distance_in_slots.register_or_noreg());
+  BLOCK_COMMENT("move_arg_slots_up {");
+  Label L_loop, L_break;
+  Register top_reg = temp_reg;
+  if (!top_addr.is_same_address(Address(top_reg, 0))) {
+    __ add(top_addr, top_reg);
+  }
+  // Detect empty (or broken) loop:
+#ifdef ASSERT
+  if (VerifyMethodHandles) {
+    // Verify that &bottom < &top (non-empty interval)
+    Label L_ok, L_bad;
+    if (positive_distance_in_slots.is_register()) {
+      __ cmp(positive_distance_in_slots.as_register(), (int32_t) 0);
+      __ br(Assembler::lessEqual, false, Assembler::pn, L_bad);
+      __ delayed()->nop();
+    }
+    __ cmp(bottom_reg, top_reg);
+    __ brx(Assembler::lessUnsigned, false, Assembler::pt, L_ok);
+    __ delayed()->nop();
+    __ BIND(L_bad);
+    __ stop("valid bounds (copy up)");
+    __ BIND(L_ok);
+  }
+#endif
+  __ cmp(bottom_reg, top_reg);
+  __ brx(Assembler::greaterEqualUnsigned, false, Assembler::pn, L_break);
+  __ delayed()->nop();
+  // work top down to bottom, copying contiguous data upwards
+  // In pseudo-code:
+  //   while (--top >= bottom) *(top + distance) = *(top + 0);
+  RegisterOrConstant offset = __ argument_offset(positive_distance_in_slots, positive_distance_in_slots.register_or_noreg());
+  __ BIND(L_loop);
+  __ sub(top_reg, wordSize, top_reg);
+  __ ld_ptr(           Address(top_reg, 0     ), temp2_reg);
+  __ st_ptr(temp2_reg, Address(top_reg, offset)           );
+  __ cmp(top_reg, bottom_reg);
+  __ brx(Assembler::greaterUnsigned, false, Assembler::pt, L_loop);
+  __ delayed()->nop();  // FILLME
+  assert(Interpreter::stackElementSize == wordSize, "else change loop");
+  __ BIND(L_break);
+  BLOCK_COMMENT("} move_arg_slots_up");
+}
+
+// in-place movement; no change to rsp
+// blows temp_reg, temp2_reg
+void MethodHandles::move_arg_slots_down(MacroAssembler* _masm,
+                                        Address  bottom_addr,  // can use temp_reg
+                                        Register top_reg,      // invariant
+                                        RegisterOrConstant negative_distance_in_slots,  // destroyed if register
+                                        Register temp_reg, Register temp2_reg) {
+  assert_different_registers(top_reg,
+                             negative_distance_in_slots.register_or_noreg(),
+                             temp_reg, temp2_reg);
+  BLOCK_COMMENT("move_arg_slots_down {");
+  Label L_loop, L_break;
+  Register bottom_reg = temp_reg;
+  if (!bottom_addr.is_same_address(Address(bottom_reg, 0))) {
+    __ add(bottom_addr, bottom_reg);
+  }
+  // Detect empty (or broken) loop:
+#ifdef ASSERT
+  assert(!negative_distance_in_slots.is_constant() || negative_distance_in_slots.as_constant() < 0, "");
+  if (VerifyMethodHandles) {
+    // Verify that &bottom < &top (non-empty interval)
+    Label L_ok, L_bad;
+    if (negative_distance_in_slots.is_register()) {
+      __ cmp(negative_distance_in_slots.as_register(), (int32_t) 0);
+      __ br(Assembler::greaterEqual, false, Assembler::pn, L_bad);
+      __ delayed()->nop();
+    }
+    __ cmp(bottom_reg, top_reg);
+    __ brx(Assembler::lessUnsigned, false, Assembler::pt, L_ok);
+    __ delayed()->nop();
+    __ BIND(L_bad);
+    __ stop("valid bounds (copy down)");
+    __ BIND(L_ok);
+  }
+#endif
+  __ cmp(bottom_reg, top_reg);
+  __ brx(Assembler::greaterEqualUnsigned, false, Assembler::pn, L_break);
+  __ delayed()->nop();
+  // work bottom up to top, copying contiguous data downwards
+  // In pseudo-code:
+  //   while (bottom < top) *(bottom - distance) = *(bottom + 0), bottom++;
+  RegisterOrConstant offset = __ argument_offset(negative_distance_in_slots, negative_distance_in_slots.register_or_noreg());
+  __ BIND(L_loop);
+  __ ld_ptr(           Address(bottom_reg, 0     ), temp2_reg);
+  __ st_ptr(temp2_reg, Address(bottom_reg, offset)           );
+  __ add(bottom_reg, wordSize, bottom_reg);
+  __ cmp(bottom_reg, top_reg);
+  __ brx(Assembler::lessUnsigned, false, Assembler::pt, L_loop);
+  __ delayed()->nop();  // FILLME
+  assert(Interpreter::stackElementSize == wordSize, "else change loop");
+  __ BIND(L_break);
+  BLOCK_COMMENT("} move_arg_slots_down");
+}
+
+// Copy from a field or array element to a stacked argument slot.
+// is_element (ignored) says whether caller is loading an array element instead of an instance field.
+void MethodHandles::move_typed_arg(MacroAssembler* _masm,
+                                   BasicType type, bool is_element,
+                                   Address value_src, Address slot_dest,
+                                   Register temp_reg) {
+  assert(!slot_dest.uses(temp_reg), "must be different register");
+  BLOCK_COMMENT(!is_element ? "move_typed_arg {" : "move_typed_arg { (array element)");
+  if (type == T_OBJECT || type == T_ARRAY) {
+    __ load_heap_oop(value_src, temp_reg);
+    __ verify_oop(temp_reg);
+    __ st_ptr(temp_reg, slot_dest);
+  } else if (type != T_VOID) {
+    int  arg_size      = type2aelembytes(type);
+    bool arg_is_signed = is_signed_subword_type(type);
+    int  slot_size     = is_subword_type(type) ? type2aelembytes(T_INT) : arg_size;  // store int sub-words as int
+    __ load_sized_value( value_src, temp_reg, arg_size, arg_is_signed);
+    __ store_sized_value(temp_reg, slot_dest, slot_size              );
+  }
+  BLOCK_COMMENT("} move_typed_arg");
+}
+
+// Cf. TemplateInterpreterGenerator::generate_return_entry_for and
+// InterpreterMacroAssembler::save_return_value
+void MethodHandles::move_return_value(MacroAssembler* _masm, BasicType type,
+                                      Address return_slot) {
+  BLOCK_COMMENT("move_return_value {");
+  // Look at the type and pull the value out of the corresponding register.
+  if (type == T_VOID) {
+    // nothing to do
+  } else if (type == T_OBJECT) {
+    __ verify_oop(O0);
+    __ st_ptr(O0, return_slot);
+  } else if (type == T_INT || is_subword_type(type)) {
+    int type_size = type2aelembytes(T_INT);
+    __ store_sized_value(O0, return_slot, type_size);
+  } else if (type == T_LONG) {
+    // store the value by parts
+    // Note: We assume longs are continguous (if misaligned) on the interpreter stack.
+#if !defined(_LP64) && defined(COMPILER2)
+    __ stx(G1, return_slot);
+#else
+  #ifdef _LP64
+    __ stx(O0, return_slot);
+  #else
+    if (return_slot.has_disp()) {
+      // The displacement is a constant
+      __ st(O0, return_slot);
+      __ st(O1, return_slot.plus_disp(Interpreter::stackElementSize));
+    } else {
+      __ std(O0, return_slot);
+    }
+  #endif
+#endif
+  } else if (type == T_FLOAT) {
+    __ stf(FloatRegisterImpl::S, Ftos_f, return_slot);
+  } else if (type == T_DOUBLE) {
+    __ stf(FloatRegisterImpl::D, Ftos_f, return_slot);
+  } else {
+    ShouldNotReachHere();
+  }
+  BLOCK_COMMENT("} move_return_value");
+}
 
 #ifndef PRODUCT
 extern "C" void print_method_handle(oop mh);
 void trace_method_handle_stub(const char* adaptername,
                               oopDesc* mh,
                               intptr_t* saved_sp) {
+  bool has_mh = (strstr(adaptername, "return/") == NULL);  // return adapters don't have mh
   tty->print_cr("MH %s mh="INTPTR_FORMAT " saved_sp=" INTPTR_FORMAT, adaptername, (intptr_t) mh, saved_sp);
-  print_method_handle(mh);
+  if (has_mh)
+    print_method_handle(mh);
 }
 void MethodHandles::trace_method_handle(MacroAssembler* _masm, const char* adaptername) {
   if (!TraceMethodHandles)  return;
@@ -367,13 +1054,21 @@
          |(1<<java_lang_invoke_AdapterMethodHandle::OP_CHECK_CAST)
          |(1<<java_lang_invoke_AdapterMethodHandle::OP_PRIM_TO_PRIM)
          |(1<<java_lang_invoke_AdapterMethodHandle::OP_REF_TO_PRIM)
+          // OP_PRIM_TO_REF is below...
          |(1<<java_lang_invoke_AdapterMethodHandle::OP_SWAP_ARGS)
          |(1<<java_lang_invoke_AdapterMethodHandle::OP_ROT_ARGS)
          |(1<<java_lang_invoke_AdapterMethodHandle::OP_DUP_ARGS)
          |(1<<java_lang_invoke_AdapterMethodHandle::OP_DROP_ARGS)
-         //|(1<<java_lang_invoke_AdapterMethodHandle::OP_SPREAD_ARGS) //BUG!
+          // OP_COLLECT_ARGS is below...
+         |(1<<java_lang_invoke_AdapterMethodHandle::OP_SPREAD_ARGS)
+         |(!UseRicochetFrames ? 0 :
+           java_lang_invoke_MethodTypeForm::vmlayout_offset_in_bytes() <= 0 ? 0 :
+           ((1<<java_lang_invoke_AdapterMethodHandle::OP_PRIM_TO_REF)
+           |(1<<java_lang_invoke_AdapterMethodHandle::OP_COLLECT_ARGS)
+           |(1<<java_lang_invoke_AdapterMethodHandle::OP_FOLD_ARGS)
+           )
+          )
          );
-  // FIXME: MethodHandlesTest gets a crash if we enable OP_SPREAD_ARGS.
 }
 
 //------------------------------------------------------------------------------
@@ -382,19 +1077,25 @@
 // 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) {
+  MethodHandles::EntryKind ek_orig = ek_original_kind(ek);
+
   // Here is the register state during an interpreted call,
   // as set up by generate_method_handle_interpreter_entry():
   // - G5: garbage temp (was MethodHandle.invoke methodOop, unused)
   // - G3: receiver method handle
   // - O5_savedSP: sender SP (must preserve)
 
-  const Register O0_argslot = O0;
+  const Register O0_scratch = O0;
   const Register O1_scratch = O1;
   const Register O2_scratch = O2;
   const Register O3_scratch = O3;
-  const Register G5_index   = G5;
+  const Register O4_scratch = O4;
+  const Register G5_scratch = G5;
 
-  // Argument registers for _raise_exception.
+  // Often used names:
+  const Register O0_argslot = O0;
+
+  // Argument registers for _raise_exception:
   const Register O0_code     = O0;
   const Register O1_actual   = O1;
   const Register O2_required = O2;
@@ -427,6 +1128,8 @@
 
   trace_method_handle(_masm, entry_name(ek));
 
+  BLOCK_COMMENT(err_msg("Entry %s {", entry_name(ek)));
+
   switch ((int) ek) {
   case _raise_exception:
     {
@@ -472,7 +1175,7 @@
       if (ek == _invokespecial_mh) {
         // Must load & check the first argument before entering the target method.
         __ load_method_handle_vmslots(O0_argslot, G3_method_handle, O1_scratch);
-        __ ld_ptr(__ argument_address(O0_argslot, -1), G3_method_handle);
+        __ ld_ptr(__ argument_address(O0_argslot, O0_argslot, -1), G3_method_handle);
         __ null_check(G3_method_handle);
         __ verify_oop(G3_method_handle);
       }
@@ -488,10 +1191,11 @@
 
       // Pick out the vtable index and receiver offset from the MH,
       // and then we can discard it:
+      Register O2_index = O2_scratch;
       __ load_method_handle_vmslots(O0_argslot, G3_method_handle, O1_scratch);
-      __ ldsw(G3_dmh_vmindex, G5_index);
+      __ ldsw(G3_dmh_vmindex, O2_index);
       // Note:  The verifier allows us to ignore G3_mh_vmtarget.
-      __ ld_ptr(__ argument_address(O0_argslot, -1), G3_method_handle);
+      __ ld_ptr(__ argument_address(O0_argslot, O0_argslot, -1), G3_method_handle);
       __ null_check(G3_method_handle, oopDesc::klass_offset_in_bytes());
 
       // Get receiver klass:
@@ -503,8 +1207,8 @@
       const int base = instanceKlass::vtable_start_offset() * wordSize;
       assert(vtableEntry::size() * wordSize == wordSize, "adjust the scaling in the code below");
 
-      __ sll_ptr(G5_index, LogBytesPerWord, G5_index);
-      __ add(O0_klass, G5_index, O0_klass);
+      __ sll_ptr(O2_index, LogBytesPerWord, O2_index);
+      __ add(O0_klass, O2_index, O0_klass);
       Address vtable_entry_addr(O0_klass, base + vtableEntry::method_offset_in_bytes());
       __ ld_ptr(vtable_entry_addr, G5_method);
 
@@ -520,9 +1224,10 @@
       // minus the CP setup and profiling:
       __ load_method_handle_vmslots(O0_argslot, G3_method_handle, O1_scratch);
       Register O1_intf  = O1_scratch;
+      Register G5_index = G5_scratch;
       __ load_heap_oop(G3_mh_vmtarget, O1_intf);
       __ ldsw(G3_dmh_vmindex, G5_index);
-      __ ld_ptr(__ argument_address(O0_argslot, -1), G3_method_handle);
+      __ ld_ptr(__ argument_address(O0_argslot, O0_argslot, -1), G3_method_handle);
       __ null_check(G3_method_handle, oopDesc::klass_offset_in_bytes());
 
       // Get receiver klass:
@@ -563,16 +1268,14 @@
   case _bound_long_direct_mh:
     {
       const bool direct_to_method = (ek >= _bound_ref_direct_mh);
-      BasicType arg_type  = T_ILLEGAL;
-      int       arg_mask  = _INSERT_NO_MASK;
-      int       arg_slots = -1;
-      get_ek_bound_mh_info(ek, arg_type, arg_mask, arg_slots);
+      BasicType arg_type  = ek_bound_mh_arg_type(ek);
+      int       arg_slots = type2size[arg_type];
 
       // Make room for the new argument:
-      __ ldsw(G3_bmh_vmargslot, O0_argslot);
-      __ add(Gargs, __ argument_offset(O0_argslot), O0_argslot);
+      load_vmargslot(_masm, G3_bmh_vmargslot, O0_argslot);
+      __ add(__ argument_address(O0_argslot, O0_argslot), O0_argslot);
 
-      insert_arg_slots(_masm, arg_slots * stack_move_unit(), arg_mask, O0_argslot, O1_scratch, O2_scratch, G5_index);
+      insert_arg_slots(_masm, arg_slots * stack_move_unit(), O0_argslot, O1_scratch, O2_scratch, O3_scratch);
 
       // Store bound argument into the new stack slot:
       __ load_heap_oop(G3_bmh_argument, O1_scratch);
@@ -580,9 +1283,10 @@
         __ st_ptr(O1_scratch, Address(O0_argslot, 0));
       } else {
         Address prim_value_addr(O1_scratch, java_lang_boxing_object::value_offset_in_bytes(arg_type));
-        const int arg_size = type2aelembytes(arg_type);
-        __ load_sized_value(prim_value_addr, O2_scratch, arg_size, is_signed_subword_type(arg_type));
-        __ store_sized_value(O2_scratch, Address(O0_argslot, 0), arg_size);  // long store uses O2/O3 on !_LP64
+        move_typed_arg(_masm, arg_type, false,
+                       prim_value_addr,
+                       Address(O0_argslot, 0),
+                       O2_scratch);  // must be an even register for !_LP64 long moves (uses O2/O3)
       }
 
       if (direct_to_method) {
@@ -602,6 +1306,7 @@
   case _adapter_retype_raw:
     // Immediately jump to the next MH layer:
     __ load_heap_oop(G3_mh_vmtarget, G3_method_handle);
+    __ verify_oop(G3_method_handle);
     __ jump_to_method_handle_entry(G3_method_handle, O1_scratch);
     // This is OK when all parameter types widen.
     // It is also OK when a return type narrows.
@@ -609,30 +1314,28 @@
 
   case _adapter_check_cast:
     {
-      // Temps:
-      Register G5_klass = G5_index;  // Interesting AMH data.
-
       // Check a reference argument before jumping to the next layer of MH:
-      __ ldsw(G3_amh_vmargslot, O0_argslot);
-      Address vmarg = __ argument_address(O0_argslot);
+      load_vmargslot(_masm, G3_amh_vmargslot, O0_argslot);
+      Address vmarg = __ argument_address(O0_argslot, O0_argslot);
 
       // What class are we casting to?
-      __ load_heap_oop(G3_amh_argument, G5_klass);  // This is a Class object!
-      __ load_heap_oop(Address(G5_klass, java_lang_Class::klass_offset_in_bytes()), G5_klass);
+      Register O1_klass = O1_scratch;  // Interesting AMH data.
+      __ load_heap_oop(G3_amh_argument, O1_klass);  // This is a Class object!
+      load_klass_from_Class(_masm, O1_klass, O2_scratch, O3_scratch);
 
-      Label done;
-      __ ld_ptr(vmarg, O1_scratch);
-      __ tst(O1_scratch);
-      __ brx(Assembler::zero, false, Assembler::pn, done);  // No cast if null.
+      Label L_done;
+      __ ld_ptr(vmarg, O2_scratch);
+      __ tst(O2_scratch);
+      __ brx(Assembler::zero, false, Assembler::pn, L_done);  // No cast if null.
       __ delayed()->nop();
-      __ load_klass(O1_scratch, O1_scratch);
+      __ load_klass(O2_scratch, O2_scratch);
 
       // Live at this point:
-      // - G5_klass        :  klass required by the target method
       // - O0_argslot      :  argslot index in vmarg; may be required in the failing path
-      // - O1_scratch      :  argument klass to test
+      // - O1_klass        :  klass required by the target method
+      // - O2_scratch      :  argument klass to test
       // - G3_method_handle:  adapter method handle
-      __ check_klass_subtype(O1_scratch, G5_klass, O2_scratch, O3_scratch, done);
+      __ check_klass_subtype(O2_scratch, O1_klass, O3_scratch, O4_scratch, L_done);
 
       // If we get here, the type check failed!
       __ load_heap_oop(G3_amh_argument,        O2_required);  // required class
@@ -640,7 +1343,7 @@
       __ jump_to(AddressLiteral(from_interpreted_entry(_raise_exception)), O3_scratch);
       __ delayed()->mov(Bytecodes::_checkcast, O0_code);      // who is complaining?
 
-      __ bind(done);
+      __ BIND(L_done);
       // Get the new MH:
       __ load_heap_oop(G3_mh_vmtarget, G3_method_handle);
       __ jump_to_method_handle_entry(G3_method_handle, O1_scratch);
@@ -659,14 +1362,14 @@
   case _adapter_opt_unboxi:     // optimized subcase of adapt_ref_to_prim
     {
       // Perform an in-place conversion to int or an int subword.
-      __ ldsw(G3_amh_vmargslot, O0_argslot);
+      load_vmargslot(_masm, G3_amh_vmargslot, O0_argslot);
       Address value;
-      Address vmarg = __ argument_address(O0_argslot);
+      Address vmarg;
       bool value_left_justified = false;
 
       switch (ek) {
       case _adapter_opt_i2i:
-        value = vmarg;
+        value = vmarg = __ argument_address(O0_argslot, O0_argslot);
         break;
       case _adapter_opt_l2i:
         {
@@ -675,13 +1378,13 @@
           // In V9, longs are given 2 64-bit slots in the interpreter, but the
           // data is passed in only 1 slot.
           // Keep the second slot.
-          __ add(Gargs, __ argument_offset(O0_argslot, -1), O0_argslot);
+          __ add(__ argument_address(O0_argslot, O0_argslot, -1), O0_argslot);
           remove_arg_slots(_masm, -stack_move_unit(), O0_argslot, O1_scratch, O2_scratch, O3_scratch);
           value = Address(O0_argslot, 4);  // Get least-significant 32-bit of 64-bit value.
           vmarg = Address(O0_argslot, Interpreter::stackElementSize);
 #else
           // Keep the first slot.
-          __ add(Gargs, __ argument_offset(O0_argslot), O0_argslot);
+          __ add(__ argument_address(O0_argslot, O0_argslot), O0_argslot);
           remove_arg_slots(_masm, -stack_move_unit(), O0_argslot, O1_scratch, O2_scratch, O3_scratch);
           value = Address(O0_argslot, 0);
           vmarg = value;
@@ -690,6 +1393,7 @@
         break;
       case _adapter_opt_unboxi:
         {
+          vmarg = __ argument_address(O0_argslot, O0_argslot);
           // Load the value up from the heap.
           __ ld_ptr(vmarg, O1_scratch);
           int value_offset = java_lang_boxing_object::value_offset_in_bytes(T_INT);
@@ -712,7 +1416,7 @@
       }
 
       // This check is required on _BIG_ENDIAN
-      Register G5_vminfo = G5_index;
+      Register G5_vminfo = G5_scratch;
       __ ldsw(G3_amh_conversion, G5_vminfo);
       assert(CONV_VMINFO_SHIFT == 0, "preshifted");
 
@@ -748,13 +1452,13 @@
   case _adapter_opt_unboxl:     // optimized subcase of adapt_ref_to_prim
     {
       // Perform an in-place int-to-long or ref-to-long conversion.
-      __ ldsw(G3_amh_vmargslot, O0_argslot);
+      load_vmargslot(_masm, G3_amh_vmargslot, O0_argslot);
 
       // On big-endian machine we duplicate the slot and store the MSW
       // in the first slot.
-      __ add(Gargs, __ argument_offset(O0_argslot, 1), O0_argslot);
+      __ add(__ argument_address(O0_argslot, O0_argslot, 1), O0_argslot);
 
-      insert_arg_slots(_masm, stack_move_unit(), _INSERT_INT_MASK, O0_argslot, O1_scratch, O2_scratch, G5_index);
+      insert_arg_slots(_masm, stack_move_unit(), O0_argslot, O1_scratch, O2_scratch, O3_scratch);
 
       Address arg_lsw(O0_argslot, 0);
       Address arg_msw(O0_argslot, -Interpreter::stackElementSize);
@@ -816,103 +1520,81 @@
   case _adapter_opt_rot_2_up:
   case _adapter_opt_rot_2_down:
     {
-      int swap_bytes = 0, rotate = 0;
-      get_ek_adapter_opt_swap_rot_info(ek, swap_bytes, rotate);
+      int swap_slots = ek_adapter_opt_swap_slots(ek);
+      int rotate     = ek_adapter_opt_swap_mode(ek);
 
       // 'argslot' is the position of the first argument to swap.
-      __ ldsw(G3_amh_vmargslot, O0_argslot);
-      __ add(Gargs, __ argument_offset(O0_argslot), O0_argslot);
+      load_vmargslot(_masm, G3_amh_vmargslot, O0_argslot);
+      __ add(__ argument_address(O0_argslot, O0_argslot), O0_argslot);
+      if (VerifyMethodHandles)
+        verify_argslot(_masm, O0_argslot, O2_scratch, "swap point must fall within current frame");
 
       // 'vminfo' is the second.
       Register O1_destslot = O1_scratch;
-      __ ldsw(G3_amh_conversion, O1_destslot);
-      assert(CONV_VMINFO_SHIFT == 0, "preshifted");
-      __ and3(O1_destslot, CONV_VMINFO_MASK, O1_destslot);
-      __ add(Gargs, __ argument_offset(O1_destslot), O1_destslot);
+      load_conversion_vminfo(_masm, G3_amh_conversion, O1_destslot);
+      __ add(__ argument_address(O1_destslot, O1_destslot), O1_destslot);
+      if (VerifyMethodHandles)
+        verify_argslot(_masm, O1_destslot, O2_scratch, "swap point must fall within current frame");
 
+      assert(Interpreter::stackElementSize == wordSize, "else rethink use of wordSize here");
       if (!rotate) {
-        for (int i = 0; i < swap_bytes; i += wordSize) {
-          __ ld_ptr(Address(O0_argslot,  i), O2_scratch);
-          __ ld_ptr(Address(O1_destslot, i), O3_scratch);
-          __ st_ptr(O3_scratch, Address(O0_argslot,  i));
-          __ st_ptr(O2_scratch, Address(O1_destslot, i));
+        // simple swap
+        for (int i = 0; i < swap_slots; i++) {
+          __ ld_ptr(            Address(O0_argslot,  i * wordSize), O2_scratch);
+          __ ld_ptr(            Address(O1_destslot, i * wordSize), O3_scratch);
+          __ st_ptr(O3_scratch, Address(O0_argslot,  i * wordSize));
+          __ st_ptr(O2_scratch, Address(O1_destslot, i * wordSize));
         }
       } else {
-        // Save the first chunk, which is going to get overwritten.
-        switch (swap_bytes) {
-        case 4 : __ lduw(Address(O0_argslot, 0), O2_scratch); break;
-        case 16: __ ldx( Address(O0_argslot, 8), O3_scratch); //fall-thru
-        case 8 : __ ldx( Address(O0_argslot, 0), O2_scratch); break;
-        default: ShouldNotReachHere();
+        // A rotate is actually pair of moves, with an "odd slot" (or pair)
+        // changing place with a series of other slots.
+        // First, push the "odd slot", which is going to get overwritten
+        switch (swap_slots) {
+        case 2 :  __ ld_ptr(Address(O0_argslot, 1 * wordSize), O4_scratch); // fall-thru
+        case 1 :  __ ld_ptr(Address(O0_argslot, 0 * wordSize), O3_scratch); break;
+        default:  ShouldNotReachHere();
         }
-
         if (rotate > 0) {
-          // Rorate upward.
-          __ sub(O0_argslot, swap_bytes, O0_argslot);
-#if ASSERT
-          {
-            // Verify that argslot > destslot, by at least swap_bytes.
-            Label L_ok;
-            __ cmp(O0_argslot, O1_destslot);
-            __ brx(Assembler::greaterEqualUnsigned, false, Assembler::pt, L_ok);
-            __ delayed()->nop();
-            __ stop("source must be above destination (upward rotation)");
-            __ bind(L_ok);
-          }
-#endif
-          // Work argslot down to destslot, copying contiguous data upwards.
-          // Pseudo-code:
+          // Here is rotate > 0:
+          // (low mem)                                          (high mem)
+          //     | dest:     more_slots...     | arg: odd_slot :arg+1 |
+          // =>
+          //     | dest: odd_slot | dest+1: more_slots...      :arg+1 |
+          // work argslot down to destslot, copying contiguous data upwards
+          // pseudo-code:
           //   argslot  = src_addr - swap_bytes
           //   destslot = dest_addr
-          //   while (argslot >= destslot) {
-          //     *(argslot + swap_bytes) = *(argslot + 0);
-          //     argslot--;
-          //   }
-          Label loop;
-          __ bind(loop);
-          __ ld_ptr(Address(O0_argslot, 0), G5_index);
-          __ st_ptr(G5_index, Address(O0_argslot, swap_bytes));
-          __ sub(O0_argslot, wordSize, O0_argslot);
-          __ cmp(O0_argslot, O1_destslot);
-          __ brx(Assembler::greaterEqualUnsigned, false, Assembler::pt, loop);
-          __ delayed()->nop();  // FILLME
+          //   while (argslot >= destslot) *(argslot + swap_bytes) = *(argslot + 0), argslot--;
+          move_arg_slots_up(_masm,
+                            O1_destslot,
+                            Address(O0_argslot, 0),
+                            swap_slots,
+                            O0_argslot, O2_scratch);
         } else {
-          __ add(O0_argslot, swap_bytes, O0_argslot);
-#if ASSERT
-          {
-            // Verify that argslot < destslot, by at least swap_bytes.
-            Label L_ok;
-            __ cmp(O0_argslot, O1_destslot);
-            __ brx(Assembler::lessEqualUnsigned, false, Assembler::pt, L_ok);
-            __ delayed()->nop();
-            __ stop("source must be above destination (upward rotation)");
-            __ bind(L_ok);
-          }
-#endif
-          // Work argslot up to destslot, copying contiguous data downwards.
-          // Pseudo-code:
+          // Here is the other direction, rotate < 0:
+          // (low mem)                                          (high mem)
+          //     | arg: odd_slot | arg+1: more_slots...       :dest+1 |
+          // =>
+          //     | arg:    more_slots...     | dest: odd_slot :dest+1 |
+          // work argslot up to destslot, copying contiguous data downwards
+          // pseudo-code:
           //   argslot  = src_addr + swap_bytes
           //   destslot = dest_addr
-          //   while (argslot >= destslot) {
-          //     *(argslot - swap_bytes) = *(argslot + 0);
-          //     argslot++;
-          //   }
-          Label loop;
-          __ bind(loop);
-          __ ld_ptr(Address(O0_argslot, 0), G5_index);
-          __ st_ptr(G5_index, Address(O0_argslot, -swap_bytes));
-          __ add(O0_argslot, wordSize, O0_argslot);
-          __ cmp(O0_argslot, O1_destslot);
-          __ brx(Assembler::lessEqualUnsigned, false, Assembler::pt, loop);
-          __ delayed()->nop();  // FILLME
+          //   while (argslot <= destslot) *(argslot - swap_bytes) = *(argslot + 0), argslot++;
+          __ add(O1_destslot, wordSize, O1_destslot);
+          move_arg_slots_down(_masm,
+                              Address(O0_argslot, swap_slots * wordSize),
+                              O1_destslot,
+                              -swap_slots,
+                              O0_argslot, O2_scratch);
+
+          __ sub(O1_destslot, wordSize, O1_destslot);
         }
-
-        // Store the original first chunk into the destination slot, now free.
-        switch (swap_bytes) {
-        case 4 : __ stw(O2_scratch, Address(O1_destslot, 0)); break;
-        case 16: __ stx(O3_scratch, Address(O1_destslot, 8)); // fall-thru
-        case 8 : __ stx(O2_scratch, Address(O1_destslot, 0)); break;
-        default: ShouldNotReachHere();
+        // pop the original first chunk into the destination slot, now free
+        switch (swap_slots) {
+        case 2 :  __ st_ptr(O4_scratch, Address(O1_destslot, 1 * wordSize)); // fall-thru
+        case 1 :  __ st_ptr(O3_scratch, Address(O1_destslot, 0 * wordSize)); break;
+        default:  ShouldNotReachHere();
         }
       }
 
@@ -924,41 +1606,21 @@
   case _adapter_dup_args:
     {
       // 'argslot' is the position of the first argument to duplicate.
-      __ ldsw(G3_amh_vmargslot, O0_argslot);
-      __ add(Gargs, __ argument_offset(O0_argslot), O0_argslot);
+      load_vmargslot(_masm, G3_amh_vmargslot, O0_argslot);
+      __ add(__ argument_address(O0_argslot, O0_argslot), O0_argslot);
 
       // 'stack_move' is negative number of words to duplicate.
-      Register G5_stack_move = G5_index;
-      __ ldsw(G3_amh_conversion, G5_stack_move);
-      __ sra(G5_stack_move, CONV_STACK_MOVE_SHIFT, G5_stack_move);
+      Register O1_stack_move = O1_scratch;
+      load_stack_move(_masm, G3_amh_conversion, O1_stack_move);
 
-      // Remember the old Gargs (argslot[0]).
-      Register O1_oldarg = O1_scratch;
-      __ mov(Gargs, O1_oldarg);
+      if (VerifyMethodHandles) {
+        verify_argslots(_masm, O1_stack_move, O0_argslot, O2_scratch, O3_scratch, true,
+                        "copied argument(s) must fall within current frame");
+      }
 
-      // Move Gargs down to make room for dups.
-      __ sll_ptr(G5_stack_move, LogBytesPerWord, G5_stack_move);
-      __ add(Gargs, G5_stack_move, Gargs);
-
-      // Compute the new Gargs (argslot[0]).
-      Register O2_newarg = O2_scratch;
-      __ mov(Gargs, O2_newarg);
-
-      // Copy from oldarg[0...] down to newarg[0...]
-      // Pseude-code:
-      //   O1_oldarg  = old-Gargs
-      //   O2_newarg  = new-Gargs
-      //   O0_argslot = argslot
-      //   while (O2_newarg < O1_oldarg) *O2_newarg = *O0_argslot++
-      Label loop;
-      __ bind(loop);
-      __ ld_ptr(Address(O0_argslot, 0), O3_scratch);
-      __ st_ptr(O3_scratch, Address(O2_newarg, 0));
-      __ add(O0_argslot, wordSize, O0_argslot);
-      __ add(O2_newarg,  wordSize, O2_newarg);
-      __ cmp(O2_newarg, O1_oldarg);
-      __ brx(Assembler::less, false, Assembler::pt, loop);
-      __ delayed()->nop();  // FILLME
+      // insert location is always the bottom of the argument list:
+      __ neg(O1_stack_move);
+      push_arg_slots(_masm, O0_argslot, O1_stack_move, O2_scratch, O3_scratch);
 
       __ load_heap_oop(G3_mh_vmtarget, G3_method_handle);
       __ jump_to_method_handle_entry(G3_method_handle, O1_scratch);
@@ -968,15 +1630,14 @@
   case _adapter_drop_args:
     {
       // 'argslot' is the position of the first argument to nuke.
-      __ ldsw(G3_amh_vmargslot, O0_argslot);
-      __ add(Gargs, __ argument_offset(O0_argslot), O0_argslot);
+      load_vmargslot(_masm, G3_amh_vmargslot, O0_argslot);
+      __ add(__ argument_address(O0_argslot, O0_argslot), O0_argslot);
 
       // 'stack_move' is number of words to drop.
-      Register G5_stack_move = G5_index;
-      __ ldsw(G3_amh_conversion, G5_stack_move);
-      __ sra(G5_stack_move, CONV_STACK_MOVE_SHIFT, G5_stack_move);
+      Register O1_stack_move = O1_scratch;
+      load_stack_move(_masm, G3_amh_conversion, O1_stack_move);
 
-      remove_arg_slots(_masm, G5_stack_move, O0_argslot, O1_scratch, O2_scratch, O3_scratch);
+      remove_arg_slots(_masm, O1_stack_move, O0_argslot, O2_scratch, O3_scratch, O4_scratch);
 
       __ load_heap_oop(G3_mh_vmtarget, G3_method_handle);
       __ jump_to_method_handle_entry(G3_method_handle, O1_scratch);
@@ -984,31 +1645,686 @@
     break;
 
   case _adapter_collect_args:
-    __ unimplemented(entry_name(ek)); // %%% FIXME: NYI
-    break;
-
+  case _adapter_fold_args:
   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:
+  case _adapter_opt_collect_ref:
+  case _adapter_opt_collect_int:
+  case _adapter_opt_collect_long:
+  case _adapter_opt_collect_float:
+  case _adapter_opt_collect_double:
+  case _adapter_opt_collect_void:
+  case _adapter_opt_collect_0_ref:
+  case _adapter_opt_collect_1_ref:
+  case _adapter_opt_collect_2_ref:
+  case _adapter_opt_collect_3_ref:
+  case _adapter_opt_collect_4_ref:
+  case _adapter_opt_collect_5_ref:
+  case _adapter_opt_filter_S0_ref:
+  case _adapter_opt_filter_S1_ref:
+  case _adapter_opt_filter_S2_ref:
+  case _adapter_opt_filter_S3_ref:
+  case _adapter_opt_filter_S4_ref:
+  case _adapter_opt_filter_S5_ref:
+  case _adapter_opt_collect_2_S0_ref:
+  case _adapter_opt_collect_2_S1_ref:
+  case _adapter_opt_collect_2_S2_ref:
+  case _adapter_opt_collect_2_S3_ref:
+  case _adapter_opt_collect_2_S4_ref:
+  case _adapter_opt_collect_2_S5_ref:
+  case _adapter_opt_fold_ref:
+  case _adapter_opt_fold_int:
+  case _adapter_opt_fold_long:
+  case _adapter_opt_fold_float:
+  case _adapter_opt_fold_double:
+  case _adapter_opt_fold_void:
+  case _adapter_opt_fold_1_ref:
+  case _adapter_opt_fold_2_ref:
+  case _adapter_opt_fold_3_ref:
+  case _adapter_opt_fold_4_ref:
+  case _adapter_opt_fold_5_ref:
     {
-      // spread an array out into a group of arguments
-      __ unimplemented(entry_name(ek));
+      // Given a fresh incoming stack frame, build a new ricochet frame.
+      // On entry, TOS points at a return PC, and FP is the callers frame ptr.
+      // RSI/R13 has the caller's exact stack pointer, which we must also preserve.
+      // RCX contains an AdapterMethodHandle of the indicated kind.
+
+      // Relevant AMH fields:
+      // amh.vmargslot:
+      //   points to the trailing edge of the arguments
+      //   to filter, collect, or fold.  For a boxing operation,
+      //   it points just after the single primitive value.
+      // amh.argument:
+      //   recursively called MH, on |collect| arguments
+      // amh.vmtarget:
+      //   final destination MH, on return value, etc.
+      // amh.conversion.dest:
+      //   tells what is the type of the return value
+      //   (not needed here, since dest is also derived from ek)
+      // amh.conversion.vminfo:
+      //   points to the trailing edge of the return value
+      //   when the vmtarget is to be called; this is
+      //   equal to vmargslot + (retained ? |collect| : 0)
+
+      // Pass 0 or more argument slots to the recursive target.
+      int collect_count_constant = ek_adapter_opt_collect_count(ek);
+
+      // The collected arguments are copied from the saved argument list:
+      int collect_slot_constant = ek_adapter_opt_collect_slot(ek);
+
+      assert(ek_orig == _adapter_collect_args ||
+             ek_orig == _adapter_fold_args, "");
+      bool retain_original_args = (ek_orig == _adapter_fold_args);
+
+      // The return value is replaced (or inserted) at the 'vminfo' argslot.
+      // Sometimes we can compute this statically.
+      int dest_slot_constant = -1;
+      if (!retain_original_args)
+        dest_slot_constant = collect_slot_constant;
+      else if (collect_slot_constant >= 0 && collect_count_constant >= 0)
+        // We are preserving all the arguments, and the return value is prepended,
+        // so the return slot is to the left (above) the |collect| sequence.
+        dest_slot_constant = collect_slot_constant + collect_count_constant;
+
+      // Replace all those slots by the result of the recursive call.
+      // The result type can be one of ref, int, long, float, double, void.
+      // In the case of void, nothing is pushed on the stack after return.
+      BasicType dest = ek_adapter_opt_collect_type(ek);
+      assert(dest == type2wfield[dest], "dest is a stack slot type");
+      int dest_count = type2size[dest];
+      assert(dest_count == 1 || dest_count == 2 || (dest_count == 0 && dest == T_VOID), "dest has a size");
+
+      // Choose a return continuation.
+      EntryKind ek_ret = _adapter_opt_return_any;
+      if (dest != T_CONFLICT && OptimizeMethodHandles) {
+        switch (dest) {
+        case T_INT    : ek_ret = _adapter_opt_return_int;     break;
+        case T_LONG   : ek_ret = _adapter_opt_return_long;    break;
+        case T_FLOAT  : ek_ret = _adapter_opt_return_float;   break;
+        case T_DOUBLE : ek_ret = _adapter_opt_return_double;  break;
+        case T_OBJECT : ek_ret = _adapter_opt_return_ref;     break;
+        case T_VOID   : ek_ret = _adapter_opt_return_void;    break;
+        default       : ShouldNotReachHere();
+        }
+        if (dest == T_OBJECT && dest_slot_constant >= 0) {
+          EntryKind ek_try = EntryKind(_adapter_opt_return_S0_ref + dest_slot_constant);
+          if (ek_try <= _adapter_opt_return_LAST &&
+              ek_adapter_opt_return_slot(ek_try) == dest_slot_constant) {
+            ek_ret = ek_try;
+          }
+        }
+        assert(ek_adapter_opt_return_type(ek_ret) == dest, "");
+      }
+
+      // Already pushed:  ... keep1 | collect | keep2 |
+
+      // Push a few extra argument words, if we need them to store the return value.
+      {
+        int extra_slots = 0;
+        if (retain_original_args) {
+          extra_slots = dest_count;
+        } else if (collect_count_constant == -1) {
+          extra_slots = dest_count;  // collect_count might be zero; be generous
+        } else if (dest_count > collect_count_constant) {
+          extra_slots = (dest_count - collect_count_constant);
+        } else {
+          // else we know we have enough dead space in |collect| to repurpose for return values
+        }
+        if (extra_slots != 0) {
+          __ sub(SP, round_to(extra_slots, 2) * Interpreter::stackElementSize, SP);
+        }
+      }
+
+      // Set up Ricochet Frame.
+      __ mov(SP, O5_savedSP);  // record SP for the callee
+
+      // One extra (empty) slot for outgoing target MH (see Gargs computation below).
+      __ save_frame(2);  // Note: we need to add 2 slots since frame::memory_parameter_word_sp_offset is 23.
+
+      // Note: Gargs is live throughout the following, until we make our recursive call.
+      // And the RF saves a copy in L4_saved_args_base.
+
+      RicochetFrame::enter_ricochet_frame(_masm, G3_method_handle, Gargs,
+                                          entry(ek_ret)->from_interpreted_entry());
+
+      // Compute argument base:
+      // Set up Gargs for current frame, extra (empty) slot is for outgoing target MH (space reserved by save_frame above).
+      __ add(FP, STACK_BIAS - (1 * Interpreter::stackElementSize), Gargs);
+
+      // Now pushed:  ... keep1 | collect | keep2 | extra | [RF]
+
+#ifdef ASSERT
+      if (VerifyMethodHandles && dest != T_CONFLICT) {
+        BLOCK_COMMENT("verify AMH.conv.dest {");
+        extract_conversion_dest_type(_masm, RicochetFrame::L5_conversion, O1_scratch);
+        Label L_dest_ok;
+        __ cmp(O1_scratch, (int) dest);
+        __ br(Assembler::equal, false, Assembler::pt, L_dest_ok);
+        __ delayed()->nop();
+        if (dest == T_INT) {
+          for (int bt = T_BOOLEAN; bt < T_INT; bt++) {
+            if (is_subword_type(BasicType(bt))) {
+              __ cmp(O1_scratch, (int) bt);
+              __ br(Assembler::equal, false, Assembler::pt, L_dest_ok);
+              __ delayed()->nop();
+            }
+          }
+        }
+        __ stop("bad dest in AMH.conv");
+        __ BIND(L_dest_ok);
+        BLOCK_COMMENT("} verify AMH.conv.dest");
+      }
+#endif //ASSERT
+
+      // Find out where the original copy of the recursive argument sequence begins.
+      Register O0_coll = O0_scratch;
+      {
+        RegisterOrConstant collect_slot = collect_slot_constant;
+        if (collect_slot_constant == -1) {
+          load_vmargslot(_masm, G3_amh_vmargslot, O1_scratch);
+          collect_slot = O1_scratch;
+        }
+        // collect_slot might be 0, but we need the move anyway.
+        __ add(RicochetFrame::L4_saved_args_base, __ argument_offset(collect_slot, collect_slot.register_or_noreg()), O0_coll);
+        // O0_coll now points at the trailing edge of |collect| and leading edge of |keep2|
+      }
+
+      // Replace the old AMH with the recursive MH.  (No going back now.)
+      // In the case of a boxing call, the recursive call is to a 'boxer' method,
+      // such as Integer.valueOf or Long.valueOf.  In the case of a filter
+      // or collect call, it will take one or more arguments, transform them,
+      // and return some result, to store back into argument_base[vminfo].
+      __ load_heap_oop(G3_amh_argument, G3_method_handle);
+      if (VerifyMethodHandles)  verify_method_handle(_masm, G3_method_handle, O1_scratch, O2_scratch);
+
+      // Calculate |collect|, the number of arguments we are collecting.
+      Register O1_collect_count = O1_scratch;
+      RegisterOrConstant collect_count;
+      if (collect_count_constant < 0) {
+        __ load_method_handle_vmslots(O1_collect_count, G3_method_handle, O2_scratch);
+        collect_count = O1_collect_count;
+      } else {
+        collect_count = collect_count_constant;
+#ifdef ASSERT
+        if (VerifyMethodHandles) {
+          BLOCK_COMMENT("verify collect_count_constant {");
+          __ load_method_handle_vmslots(O3_scratch, G3_method_handle, O2_scratch);
+          Label L_count_ok;
+          __ cmp(O3_scratch, collect_count_constant);
+          __ br(Assembler::equal, false, Assembler::pt, L_count_ok);
+          __ delayed()->nop();
+          __ stop("bad vminfo in AMH.conv");
+          __ BIND(L_count_ok);
+          BLOCK_COMMENT("} verify collect_count_constant");
+        }
+#endif //ASSERT
+      }
+
+      // copy |collect| slots directly to TOS:
+      push_arg_slots(_masm, O0_coll, collect_count, O2_scratch, O3_scratch);
+      // Now pushed:  ... keep1 | collect | keep2 | RF... | collect |
+      // O0_coll still points at the trailing edge of |collect| and leading edge of |keep2|
+
+      // If necessary, adjust the saved arguments to make room for the eventual return value.
+      // Normal adjustment:  ... keep1 | +dest+ | -collect- | keep2 | RF... | collect |
+      // If retaining args:  ... keep1 | +dest+ |  collect  | keep2 | RF... | collect |
+      // In the non-retaining case, this might move keep2 either up or down.
+      // We don't have to copy the whole | RF... collect | complex,
+      // but we must adjust RF.saved_args_base.
+      // Also, from now on, we will forget about the original copy of |collect|.
+      // If we are retaining it, we will treat it as part of |keep2|.
+      // For clarity we will define |keep3| = |collect|keep2| or |keep2|.
+
+      BLOCK_COMMENT("adjust trailing arguments {");
+      // Compare the sizes of |+dest+| and |-collect-|, which are opposed opening and closing movements.
+      int                open_count  = dest_count;
+      RegisterOrConstant close_count = collect_count_constant;
+      Register O1_close_count = O1_collect_count;
+      if (retain_original_args) {
+        close_count = constant(0);
+      } else if (collect_count_constant == -1) {
+        close_count = O1_collect_count;
+      }
+
+      // How many slots need moving?  This is simply dest_slot (0 => no |keep3|).
+      RegisterOrConstant keep3_count;
+      Register O2_keep3_count = O2_scratch;
+      if (dest_slot_constant < 0) {
+        extract_conversion_vminfo(_masm, RicochetFrame::L5_conversion, O2_keep3_count);
+        keep3_count = O2_keep3_count;
+      } else  {
+        keep3_count = dest_slot_constant;
+#ifdef ASSERT
+        if (VerifyMethodHandles && dest_slot_constant < 0) {
+          BLOCK_COMMENT("verify dest_slot_constant {");
+          extract_conversion_vminfo(_masm, RicochetFrame::L5_conversion, O3_scratch);
+          Label L_vminfo_ok;
+          __ cmp(O3_scratch, dest_slot_constant);
+          __ br(Assembler::equal, false, Assembler::pt, L_vminfo_ok);
+          __ delayed()->nop();
+          __ stop("bad vminfo in AMH.conv");
+          __ BIND(L_vminfo_ok);
+          BLOCK_COMMENT("} verify dest_slot_constant");
+        }
+#endif //ASSERT
+      }
+
+      // tasks remaining:
+      bool move_keep3 = (!keep3_count.is_constant() || keep3_count.as_constant() != 0);
+      bool stomp_dest = (NOT_DEBUG(dest == T_OBJECT) DEBUG_ONLY(dest_count != 0));
+      bool fix_arg_base = (!close_count.is_constant() || open_count != close_count.as_constant());
+
+      // Old and new argument locations (based at slot 0).
+      // Net shift (&new_argv - &old_argv) is (close_count - open_count).
+      bool zero_open_count = (open_count == 0);  // remember this bit of info
+      if (move_keep3 && fix_arg_base) {
+        // It will be easier to have everything in one register:
+        if (close_count.is_register()) {
+          // Deduct open_count from close_count register to get a clean +/- value.
+          __ sub(close_count.as_register(), open_count, close_count.as_register());
+        } else {
+          close_count = close_count.as_constant() - open_count;
+        }
+        open_count = 0;
+      }
+      Register L4_old_argv = RicochetFrame::L4_saved_args_base;
+      Register O3_new_argv = O3_scratch;
+      if (fix_arg_base) {
+        __ add(L4_old_argv, __ argument_offset(close_count, O4_scratch), O3_new_argv,
+               -(open_count * Interpreter::stackElementSize));
+      }
+
+      // First decide if any actual data are to be moved.
+      // We can skip if (a) |keep3| is empty, or (b) the argument list size didn't change.
+      // (As it happens, all movements involve an argument list size change.)
+
+      // If there are variable parameters, use dynamic checks to skip around the whole mess.
+      Label L_done;
+      if (keep3_count.is_register()) {
+        __ tst(keep3_count.as_register());
+        __ br(Assembler::zero, false, Assembler::pn, L_done);
+        __ delayed()->nop();
+      }
+      if (close_count.is_register()) {
+        __ cmp(close_count.as_register(), open_count);
+        __ br(Assembler::equal, false, Assembler::pn, L_done);
+        __ delayed()->nop();
+      }
+
+      if (move_keep3 && fix_arg_base) {
+        bool emit_move_down = false, emit_move_up = false, emit_guard = false;
+        if (!close_count.is_constant()) {
+          emit_move_down = emit_guard = !zero_open_count;
+          emit_move_up   = true;
+        } else if (open_count != close_count.as_constant()) {
+          emit_move_down = (open_count > close_count.as_constant());
+          emit_move_up   = !emit_move_down;
+        }
+        Label L_move_up;
+        if (emit_guard) {
+          __ cmp(close_count.as_register(), open_count);
+          __ br(Assembler::greater, false, Assembler::pn, L_move_up);
+          __ delayed()->nop();
+        }
+
+        if (emit_move_down) {
+          // Move arguments down if |+dest+| > |-collect-|
+          // (This is rare, except when arguments are retained.)
+          // This opens space for the return value.
+          if (keep3_count.is_constant()) {
+            for (int i = 0; i < keep3_count.as_constant(); i++) {
+              __ ld_ptr(            Address(L4_old_argv, i * Interpreter::stackElementSize), O4_scratch);
+              __ st_ptr(O4_scratch, Address(O3_new_argv, i * Interpreter::stackElementSize)            );
+            }
+          } else {
+            // Live: O1_close_count, O2_keep3_count, O3_new_argv
+            Register argv_top = O0_scratch;
+            __ add(L4_old_argv, __ argument_offset(keep3_count, O4_scratch), argv_top);
+            move_arg_slots_down(_masm,
+                                Address(L4_old_argv, 0),  // beginning of old argv
+                                argv_top,                 // end of old argv
+                                close_count,              // distance to move down (must be negative)
+                                O4_scratch, G5_scratch);
+          }
+        }
+
+        if (emit_guard) {
+          __ ba(false, L_done);  // assumes emit_move_up is true also
+          __ delayed()->nop();
+          __ BIND(L_move_up);
+        }
+
+        if (emit_move_up) {
+          // Move arguments up if |+dest+| < |-collect-|
+          // (This is usual, except when |keep3| is empty.)
+          // This closes up the space occupied by the now-deleted collect values.
+          if (keep3_count.is_constant()) {
+            for (int i = keep3_count.as_constant() - 1; i >= 0; i--) {
+              __ ld_ptr(            Address(L4_old_argv, i * Interpreter::stackElementSize), O4_scratch);
+              __ st_ptr(O4_scratch, Address(O3_new_argv, i * Interpreter::stackElementSize)            );
+            }
+          } else {
+            Address argv_top(L4_old_argv, __ argument_offset(keep3_count, O4_scratch));
+            // Live: O1_close_count, O2_keep3_count, O3_new_argv
+            move_arg_slots_up(_masm,
+                              L4_old_argv,  // beginning of old argv
+                              argv_top,     // end of old argv
+                              close_count,  // distance to move up (must be positive)
+                              O4_scratch, G5_scratch);
+          }
+        }
+      }
+      __ BIND(L_done);
+
+      if (fix_arg_base) {
+        // adjust RF.saved_args_base
+        __ mov(O3_new_argv, RicochetFrame::L4_saved_args_base);
+      }
+
+      if (stomp_dest) {
+        // Stomp the return slot, so it doesn't hold garbage.
+        // This isn't strictly necessary, but it may help detect bugs.
+        __ set(RicochetFrame::RETURN_VALUE_PLACEHOLDER, O4_scratch);
+        __ st_ptr(O4_scratch, Address(RicochetFrame::L4_saved_args_base,
+                                      __ argument_offset(keep3_count, keep3_count.register_or_noreg())));  // uses O2_keep3_count
+      }
+      BLOCK_COMMENT("} adjust trailing arguments");
+
+      BLOCK_COMMENT("do_recursive_call");
+      __ mov(SP, O5_savedSP);  // record SP for the callee
+      __ set(ExternalAddress(SharedRuntime::ricochet_blob()->bounce_addr() - frame::pc_return_offset), O7);
+      // The globally unique bounce address has two purposes:
+      // 1. It helps the JVM recognize this frame (frame::is_ricochet_frame).
+      // 2. When returned to, it cuts back the stack and redirects control flow
+      //    to the return handler.
+      // The return handler will further cut back the stack when it takes
+      // down the RF.  Perhaps there is a way to streamline this further.
+
+      // State during recursive call:
+      // ... keep1 | dest | dest=42 | keep3 | RF... | collect | bounce_pc |
+      __ jump_to_method_handle_entry(G3_method_handle, O1_scratch);
     }
     break;
 
-  case _adapter_flyby:
-  case _adapter_ricochet:
-    __ unimplemented(entry_name(ek)); // %%% FIXME: NYI
+  case _adapter_opt_return_ref:
+  case _adapter_opt_return_int:
+  case _adapter_opt_return_long:
+  case _adapter_opt_return_float:
+  case _adapter_opt_return_double:
+  case _adapter_opt_return_void:
+  case _adapter_opt_return_S0_ref:
+  case _adapter_opt_return_S1_ref:
+  case _adapter_opt_return_S2_ref:
+  case _adapter_opt_return_S3_ref:
+  case _adapter_opt_return_S4_ref:
+  case _adapter_opt_return_S5_ref:
+    {
+      BasicType dest_type_constant = ek_adapter_opt_return_type(ek);
+      int       dest_slot_constant = ek_adapter_opt_return_slot(ek);
+
+      if (VerifyMethodHandles)  RicochetFrame::verify_clean(_masm);
+
+      if (dest_slot_constant == -1) {
+        // The current stub is a general handler for this dest_type.
+        // It can be called from _adapter_opt_return_any below.
+        // Stash the address in a little table.
+        assert((dest_type_constant & CONV_TYPE_MASK) == dest_type_constant, "oob");
+        address return_handler = __ pc();
+        _adapter_return_handlers[dest_type_constant] = return_handler;
+        if (dest_type_constant == T_INT) {
+          // do the subword types too
+          for (int bt = T_BOOLEAN; bt < T_INT; bt++) {
+            if (is_subword_type(BasicType(bt)) &&
+                _adapter_return_handlers[bt] == NULL) {
+              _adapter_return_handlers[bt] = return_handler;
+            }
+          }
+        }
+      }
+
+      // On entry to this continuation handler, make Gargs live again.
+      __ mov(RicochetFrame::L4_saved_args_base, Gargs);
+
+      Register O7_temp   = O7;
+      Register O5_vminfo = O5;
+
+      RegisterOrConstant dest_slot = dest_slot_constant;
+      if (dest_slot_constant == -1) {
+        extract_conversion_vminfo(_masm, RicochetFrame::L5_conversion, O5_vminfo);
+        dest_slot = O5_vminfo;
+      }
+      // Store the result back into the argslot.
+      // This code uses the interpreter calling sequence, in which the return value
+      // is usually left in the TOS register, as defined by InterpreterMacroAssembler::pop.
+      // There are certain irregularities with floating point values, which can be seen
+      // in TemplateInterpreterGenerator::generate_return_entry_for.
+      move_return_value(_masm, dest_type_constant, __ argument_address(dest_slot, O7_temp));
+
+      RicochetFrame::leave_ricochet_frame(_masm, G3_method_handle, I5_savedSP, I7);
+
+      // Load the final target and go.
+      if (VerifyMethodHandles)  verify_method_handle(_masm, G3_method_handle, O0_scratch, O1_scratch);
+      __ restore(I5_savedSP, G0, SP);
+      __ jump_to_method_handle_entry(G3_method_handle, O0_scratch);
+      __ illtrap(0);
+    }
+    break;
+
+  case _adapter_opt_return_any:
+    {
+      Register O7_temp      = O7;
+      Register O5_dest_type = O5;
+
+      if (VerifyMethodHandles)  RicochetFrame::verify_clean(_masm);
+      extract_conversion_dest_type(_masm, RicochetFrame::L5_conversion, O5_dest_type);
+      __ set(ExternalAddress((address) &_adapter_return_handlers[0]), O7_temp);
+      __ sll_ptr(O5_dest_type, LogBytesPerWord, O5_dest_type);
+      __ ld_ptr(O7_temp, O5_dest_type, O7_temp);
+
+#ifdef ASSERT
+      { Label L_ok;
+        __ br_notnull(O7_temp, false, Assembler::pt, L_ok);
+        __ delayed()->nop();
+        __ stop("bad method handle return");
+        __ BIND(L_ok);
+      }
+#endif //ASSERT
+      __ JMP(O7_temp, 0);
+      __ delayed()->nop();
+    }
+    break;
+
+  case _adapter_opt_spread_0:
+  case _adapter_opt_spread_1_ref:
+  case _adapter_opt_spread_2_ref:
+  case _adapter_opt_spread_3_ref:
+  case _adapter_opt_spread_4_ref:
+  case _adapter_opt_spread_5_ref:
+  case _adapter_opt_spread_ref:
+  case _adapter_opt_spread_byte:
+  case _adapter_opt_spread_char:
+  case _adapter_opt_spread_short:
+  case _adapter_opt_spread_int:
+  case _adapter_opt_spread_long:
+  case _adapter_opt_spread_float:
+  case _adapter_opt_spread_double:
+    {
+      // spread an array out into a group of arguments
+      int  length_constant    = ek_adapter_opt_spread_count(ek);
+      bool length_can_be_zero = (length_constant == 0);
+      if (length_constant < 0) {
+        // some adapters with variable length must handle the zero case
+        if (!OptimizeMethodHandles ||
+            ek_adapter_opt_spread_type(ek) != T_OBJECT)
+          length_can_be_zero = true;
+      }
+
+      // find the address of the array argument
+      load_vmargslot(_masm, G3_amh_vmargslot, O0_argslot);
+      __ add(__ argument_address(O0_argslot, O0_argslot), O0_argslot);
+
+      // O0_argslot points both to the array and to the first output arg
+      Address vmarg = Address(O0_argslot, 0);
+
+      // Get the array value.
+      Register  O1_array       = O1_scratch;
+      Register  O2_array_klass = O2_scratch;
+      BasicType elem_type      = ek_adapter_opt_spread_type(ek);
+      int       elem_slots     = type2size[elem_type];  // 1 or 2
+      int       array_slots    = 1;  // array is always a T_OBJECT
+      int       length_offset  = arrayOopDesc::length_offset_in_bytes();
+      int       elem0_offset   = arrayOopDesc::base_offset_in_bytes(elem_type);
+      __ ld_ptr(vmarg, O1_array);
+
+      Label L_array_is_empty, L_insert_arg_space, L_copy_args, L_args_done;
+      if (length_can_be_zero) {
+        // handle the null pointer case, if zero is allowed
+        Label L_skip;
+        if (length_constant < 0) {
+          load_conversion_vminfo(_masm, G3_amh_conversion, O3_scratch);
+          __ br_zero(Assembler::notZero, false, Assembler::pn, O3_scratch, L_skip);
+          __ delayed()->nop();
+        }
+        __ br_null(O1_array, false, Assembler::pn, L_array_is_empty);
+        __ delayed()->nop();
+        __ BIND(L_skip);
+      }
+      __ null_check(O1_array, oopDesc::klass_offset_in_bytes());
+      __ load_klass(O1_array, O2_array_klass);
+
+      // Check the array type.
+      Register O3_klass = O3_scratch;
+      __ load_heap_oop(G3_amh_argument, O3_klass);  // this is a Class object!
+      load_klass_from_Class(_masm, O3_klass, O4_scratch, G5_scratch);
+
+      Label L_ok_array_klass, L_bad_array_klass, L_bad_array_length;
+      __ check_klass_subtype(O2_array_klass, O3_klass, O4_scratch, G5_scratch, L_ok_array_klass);
+      // If we get here, the type check failed!
+      __ ba(false, L_bad_array_klass);
+      __ delayed()->nop();
+      __ BIND(L_ok_array_klass);
+
+      // Check length.
+      if (length_constant >= 0) {
+        __ ldsw(Address(O1_array, length_offset), O4_scratch);
+        __ cmp(O4_scratch, length_constant);
+      } else {
+        Register O3_vminfo = O3_scratch;
+        load_conversion_vminfo(_masm, G3_amh_conversion, O3_vminfo);
+        __ ldsw(Address(O1_array, length_offset), O4_scratch);
+        __ cmp(O3_vminfo, O4_scratch);
+      }
+      __ br(Assembler::notEqual, false, Assembler::pn, L_bad_array_length);
+      __ delayed()->nop();
+
+      Register O2_argslot_limit = O2_scratch;
+
+      // Array length checks out.  Now insert any required stack slots.
+      if (length_constant == -1) {
+        // Form a pointer to the end of the affected region.
+        __ add(O0_argslot, Interpreter::stackElementSize, O2_argslot_limit);
+        // 'stack_move' is negative number of words to insert
+        // This number already accounts for elem_slots.
+        Register O3_stack_move = O3_scratch;
+        load_stack_move(_masm, G3_amh_conversion, O3_stack_move);
+        __ cmp(O3_stack_move, 0);
+        assert(stack_move_unit() < 0, "else change this comparison");
+        __ br(Assembler::less, false, Assembler::pn, L_insert_arg_space);
+        __ delayed()->nop();
+        __ br(Assembler::equal, false, Assembler::pn, L_copy_args);
+        __ delayed()->nop();
+        // single argument case, with no array movement
+        __ BIND(L_array_is_empty);
+        remove_arg_slots(_masm, -stack_move_unit() * array_slots,
+                         O0_argslot, O1_scratch, O2_scratch, O3_scratch);
+        __ ba(false, L_args_done);  // no spreading to do
+        __ delayed()->nop();
+        __ BIND(L_insert_arg_space);
+        // come here in the usual case, stack_move < 0 (2 or more spread arguments)
+        // Live: O1_array, O2_argslot_limit, O3_stack_move
+        insert_arg_slots(_masm, O3_stack_move,
+                         O0_argslot, O4_scratch, G5_scratch, O1_scratch);
+        // reload from rdx_argslot_limit since rax_argslot is now decremented
+        __ ld_ptr(Address(O2_argslot_limit, -Interpreter::stackElementSize), O1_array);
+      } else if (length_constant >= 1) {
+        int new_slots = (length_constant * elem_slots) - array_slots;
+        insert_arg_slots(_masm, new_slots * stack_move_unit(),
+                         O0_argslot, O2_scratch, O3_scratch, O4_scratch);
+      } else if (length_constant == 0) {
+        __ BIND(L_array_is_empty);
+        remove_arg_slots(_masm, -stack_move_unit() * array_slots,
+                         O0_argslot, O1_scratch, O2_scratch, O3_scratch);
+      } else {
+        ShouldNotReachHere();
+      }
+
+      // Copy from the array to the new slots.
+      // Note: Stack change code preserves integrity of O0_argslot pointer.
+      // So even after slot insertions, O0_argslot still points to first argument.
+      // Beware:  Arguments that are shallow on the stack are deep in the array,
+      // and vice versa.  So a downward-growing stack (the usual) has to be copied
+      // elementwise in reverse order from the source array.
+      __ BIND(L_copy_args);
+      if (length_constant == -1) {
+        // [O0_argslot, O2_argslot_limit) is the area we are inserting into.
+        // Array element [0] goes at O0_argslot_limit[-wordSize].
+        Register O1_source = O1_array;
+        __ add(Address(O1_array, elem0_offset), O1_source);
+        Register O4_fill_ptr = O4_scratch;
+        __ mov(O2_argslot_limit, O4_fill_ptr);
+        Label L_loop;
+        __ BIND(L_loop);
+        __ add(O4_fill_ptr, -Interpreter::stackElementSize * elem_slots, O4_fill_ptr);
+        move_typed_arg(_masm, elem_type, true,
+                       Address(O1_source, 0), Address(O4_fill_ptr, 0),
+                       O2_scratch);  // must be an even register for !_LP64 long moves (uses O2/O3)
+        __ add(O1_source, type2aelembytes(elem_type), O1_source);
+        __ cmp(O4_fill_ptr, O0_argslot);
+        __ brx(Assembler::greaterUnsigned, false, Assembler::pt, L_loop);
+        __ delayed()->nop();  // FILLME
+      } else if (length_constant == 0) {
+        // nothing to copy
+      } else {
+        int elem_offset = elem0_offset;
+        int slot_offset = length_constant * Interpreter::stackElementSize;
+        for (int index = 0; index < length_constant; index++) {
+          slot_offset -= Interpreter::stackElementSize * elem_slots;  // fill backward
+          move_typed_arg(_masm, elem_type, true,
+                         Address(O1_array, elem_offset), Address(O0_argslot, slot_offset),
+                         O2_scratch);  // must be an even register for !_LP64 long moves (uses O2/O3)
+          elem_offset += type2aelembytes(elem_type);
+        }
+      }
+      __ BIND(L_args_done);
+
+      // Arguments are spread.  Move to next method handle.
+      __ load_heap_oop(G3_mh_vmtarget, G3_method_handle);
+      __ jump_to_method_handle_entry(G3_method_handle, O1_scratch);
+
+      __ BIND(L_bad_array_klass);
+      assert(!vmarg.uses(O2_required), "must be different registers");
+      __ load_heap_oop(Address(O2_array_klass, java_mirror_offset), O2_required);  // required class
+      __ ld_ptr(       vmarg,                                       O1_actual);    // bad object
+      __ jump_to(AddressLiteral(from_interpreted_entry(_raise_exception)), O3_scratch);
+      __ delayed()->mov(Bytecodes::_aaload,                         O0_code);      // who is complaining?
+
+      __ bind(L_bad_array_length);
+      assert(!vmarg.uses(O2_required), "must be different registers");
+      __ mov(   G3_method_handle,                O2_required);  // required class
+      __ ld_ptr(vmarg,                           O1_actual);    // bad object
+      __ jump_to(AddressLiteral(from_interpreted_entry(_raise_exception)), O3_scratch);
+      __ delayed()->mov(Bytecodes::_arraylength, O0_code);      // who is complaining?
+    }
     break;
 
   default:
+    DEBUG_ONLY(tty->print_cr("bad ek=%d (%s)", (int)ek, entry_name(ek)));
     ShouldNotReachHere();
   }
+  BLOCK_COMMENT(err_msg("} Entry %s", entry_name(ek)));
 
   address me_cookie = MethodHandleEntry::start_compiled_entry(_masm, interp_entry);
   __ unimplemented(entry_name(ek)); // %%% FIXME: NYI
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/src/cpu/sparc/vm/methodHandles_sparc.hpp	Fri Jun 03 22:31:43 2011 -0700
@@ -0,0 +1,224 @@
+/*
+ * Copyright (c) 2011, Oracle and/or its affiliates. All rights reserved.
+ * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
+ *
+ * This code is free software; you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License version 2 only, as
+ * published by the Free Software Foundation.
+ *
+ * This code is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
+ * version 2 for more details (a copy is included in the LICENSE file that
+ * accompanied this code).
+ *
+ * You should have received a copy of the GNU General Public License version
+ * 2 along with this work; if not, write to the Free Software Foundation,
+ * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
+ *
+ * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
+ * or visit www.oracle.com if you need additional information or have any
+ * questions.
+ *
+ */
+
+// Platform-specific definitions for method handles.
+// These definitions are inlined into class MethodHandles.
+
+// Adapters
+enum /* platform_dependent_constants */ {
+  adapter_code_size = NOT_LP64(22000 DEBUG_ONLY(+ 40000)) LP64_ONLY(32000 DEBUG_ONLY(+ 80000))
+};
+
+public:
+
+class RicochetFrame : public ResourceObj {
+  friend class MethodHandles;
+
+ private:
+  /*
+    RF field            x86                 SPARC
+    sender_pc           *(rsp+0)            I7-0x8
+    sender_link         rbp                 I6+BIAS
+    exact_sender_sp     rsi/r13             I5_savedSP
+    conversion          *(rcx+&amh_conv)    L5_conv
+    saved_args_base     rax                 L4_sab (cf. Gargs = G4)
+    saved_args_layout   #NULL               L3_sal
+    saved_target        *(rcx+&mh_vmtgt)    L2_stgt
+    continuation        #STUB_CON           L1_cont
+   */
+  static const Register L1_continuation     ;  // what to do when control gets back here
+  static const Register L2_saved_target     ;  // target method handle to invoke on saved_args
+  static const Register L3_saved_args_layout;  // caching point for MethodTypeForm.vmlayout cookie
+  static const Register L4_saved_args_base  ;  // base of pushed arguments (slot 0, arg N) (-3)
+  static const Register L5_conversion       ;  // misc. information from original AdapterMethodHandle (-2)
+
+  frame _fr;
+
+  RicochetFrame(const frame& fr) : _fr(fr) { }
+
+  intptr_t* register_addr(Register reg) const  {
+    assert((_fr.sp() + reg->sp_offset_in_saved_window()) == _fr.register_addr(reg), "must agree");
+    return _fr.register_addr(reg);
+  }
+  intptr_t  register_value(Register reg) const { return *register_addr(reg); }
+
+ public:
+  intptr_t* continuation() const        { return (intptr_t*) register_value(L1_continuation); }
+  oop       saved_target() const        { return (oop)       register_value(L2_saved_target); }
+  oop       saved_args_layout() const   { return (oop)       register_value(L3_saved_args_layout); }
+  intptr_t* saved_args_base() const     { return (intptr_t*) register_value(L4_saved_args_base); }
+  intptr_t  conversion() const          { return             register_value(L5_conversion); }
+  intptr_t* exact_sender_sp() const     { return (intptr_t*) register_value(I5_savedSP); }
+  intptr_t* sender_link() const         { return _fr.sender_sp(); }  // XXX
+  address   sender_pc() const           { return _fr.sender_pc(); }
+
+  // This value is not used for much, but it apparently must be nonzero.
+  static int frame_size_in_bytes()              { return wordSize * 4; }
+
+  intptr_t* extended_sender_sp() const  { return saved_args_base(); }
+
+  intptr_t  return_value_slot_number() const {
+    return adapter_conversion_vminfo(conversion());
+  }
+  BasicType return_value_type() const {
+    return adapter_conversion_dest_type(conversion());
+  }
+  bool has_return_value_slot() const {
+    return return_value_type() != T_VOID;
+  }
+  intptr_t* return_value_slot_addr() const {
+    assert(has_return_value_slot(), "");
+    return saved_arg_slot_addr(return_value_slot_number());
+  }
+  intptr_t* saved_target_slot_addr() const {
+    return saved_arg_slot_addr(saved_args_length());
+  }
+  intptr_t* saved_arg_slot_addr(int slot) const {
+    assert(slot >= 0, "");
+    return (intptr_t*)( (address)saved_args_base() + (slot * Interpreter::stackElementSize) );
+  }
+
+  jint      saved_args_length() const;
+  jint      saved_arg_offset(int arg) const;
+
+  // GC interface
+  oop*  saved_target_addr()                     { return (oop*)register_addr(L2_saved_target); }
+  oop*  saved_args_layout_addr()                { return (oop*)register_addr(L3_saved_args_layout); }
+
+  oop  compute_saved_args_layout(bool read_cache, bool write_cache);
+
+#ifdef ASSERT
+  // The magic number is supposed to help find ricochet frames within the bytes of stack dumps.
+  enum { MAGIC_NUMBER_1 = 0xFEED03E, MAGIC_NUMBER_2 = 0xBEEF03E };
+  static const Register L0_magic_number_1   ;  // cookie for debugging, at start of RSA
+  static Address magic_number_2_addr()  { return Address(L4_saved_args_base, -wordSize); }
+  intptr_t magic_number_1() const       { return register_value(L0_magic_number_1); }
+  intptr_t magic_number_2() const       { return saved_args_base()[-1]; }
+#endif //ASSERT
+
+ public:
+  enum { RETURN_VALUE_PLACEHOLDER = (NOT_DEBUG(0) DEBUG_ONLY(42)) };
+
+  void verify() const NOT_DEBUG_RETURN; // check for MAGIC_NUMBER, etc.
+
+  static void generate_ricochet_blob(MacroAssembler* _masm,
+                                     // output params:
+                                     int* bounce_offset,
+                                     int* exception_offset,
+                                     int* frame_size_in_words);
+
+  static void enter_ricochet_frame(MacroAssembler* _masm,
+                                   Register recv_reg,
+                                   Register argv_reg,
+                                   address return_handler);
+
+  static void leave_ricochet_frame(MacroAssembler* _masm,
+                                   Register recv_reg,
+                                   Register new_sp_reg,
+                                   Register sender_pc_reg);
+
+  static RicochetFrame* from_frame(const frame& fr) {
+    RicochetFrame* rf = new RicochetFrame(fr);
+    rf->verify();
+    return rf;
+  }
+
+  static void verify_clean(MacroAssembler* _masm) NOT_DEBUG_RETURN;
+};
+
+// Additional helper methods for MethodHandles code generation:
+public:
+  static void load_klass_from_Class(MacroAssembler* _masm, Register klass_reg, Register temp_reg, Register temp2_reg);
+  static void load_conversion_vminfo(MacroAssembler* _masm, Address conversion_field_addr, Register reg);
+  static void extract_conversion_vminfo(MacroAssembler* _masm, Register conversion_field_reg, Register reg);
+  static void extract_conversion_dest_type(MacroAssembler* _masm, Register conversion_field_reg, Register reg);
+
+  static void load_stack_move(MacroAssembler* _masm,
+                              Address G3_amh_conversion,
+                              Register G5_stack_move);
+
+  static void insert_arg_slots(MacroAssembler* _masm,
+                               RegisterOrConstant arg_slots,
+                               Register argslot_reg,
+                               Register temp_reg, Register temp2_reg, Register temp3_reg);
+
+  static void remove_arg_slots(MacroAssembler* _masm,
+                               RegisterOrConstant arg_slots,
+                               Register argslot_reg,
+                               Register temp_reg, Register temp2_reg, Register temp3_reg);
+
+  static void push_arg_slots(MacroAssembler* _masm,
+                             Register argslot_reg,
+                             RegisterOrConstant slot_count,
+                             Register temp_reg, Register temp2_reg);
+
+  static void move_arg_slots_up(MacroAssembler* _masm,
+                                Register bottom_reg,  // invariant
+                                Address  top_addr,    // can use temp_reg
+                                RegisterOrConstant positive_distance_in_slots,
+                                Register temp_reg, Register temp2_reg);
+
+  static void move_arg_slots_down(MacroAssembler* _masm,
+                                  Address  bottom_addr,  // can use temp_reg
+                                  Register top_reg,      // invariant
+                                  RegisterOrConstant negative_distance_in_slots,
+                                  Register temp_reg, Register temp2_reg);
+
+  static void move_typed_arg(MacroAssembler* _masm,
+                             BasicType type, bool is_element,
+                             Address value_src, Address slot_dest,
+                             Register temp_reg);
+
+  static void move_return_value(MacroAssembler* _masm, BasicType type,
+                                Address return_slot);
+
+  static void verify_argslot(MacroAssembler* _masm, Register argslot_reg,
+                             Register temp_reg,
+                             const char* error_message) NOT_DEBUG_RETURN;
+
+  static void verify_argslots(MacroAssembler* _masm,
+                              RegisterOrConstant argslot_count,
+                              Register argslot_reg,
+                              Register temp_reg,
+                              Register temp2_reg,
+                              bool negate_argslot,
+                              const char* error_message) NOT_DEBUG_RETURN;
+
+  static void verify_stack_move(MacroAssembler* _masm,
+                                RegisterOrConstant arg_slots,
+                                int direction) NOT_DEBUG_RETURN;
+
+  static void verify_klass(MacroAssembler* _masm,
+                           Register obj_reg, KlassHandle klass,
+                           Register temp_reg, Register temp2_reg,
+                           const char* error_message = "wrong klass") NOT_DEBUG_RETURN;
+
+  static void verify_method_handle(MacroAssembler* _masm, Register mh_reg,
+                                   Register temp_reg, Register temp2_reg) {
+    verify_klass(_masm, mh_reg, SystemDictionaryHandles::MethodHandle_klass(),
+                 temp_reg, temp2_reg,
+                 "reference is a MH");
+  }
+
+  static void trace_method_handle(MacroAssembler* _masm, const char* adaptername) PRODUCT_RETURN;
--- a/src/cpu/sparc/vm/registerMap_sparc.hpp	Thu Jun 02 13:36:11 2011 -0700
+++ b/src/cpu/sparc/vm/registerMap_sparc.hpp	Fri Jun 03 22:31:43 2011 -0700
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 1998, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1998, 2011, Oracle and/or its affiliates. All rights reserved.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
  * This code is free software; you can redistribute it and/or modify it
@@ -27,6 +27,7 @@
 
 // machine-dependent implemention for register maps
   friend class frame;
+  friend class MethodHandles;
 
  private:
   intptr_t* _window;         // register window save area (for L and I regs)
--- a/src/cpu/sparc/vm/runtime_sparc.cpp	Thu Jun 02 13:36:11 2011 -0700
+++ b/src/cpu/sparc/vm/runtime_sparc.cpp	Fri Jun 03 22:31:43 2011 -0700
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 1998, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1998, 2011, Oracle and/or its affiliates. All rights reserved.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
  * This code is free software; you can redistribute it and/or modify it
@@ -42,8 +42,6 @@
 
 #define __ masm->
 
-ExceptionBlob      *OptoRuntime::_exception_blob;
-
 //------------------------------ generate_exception_blob ---------------------------
 // creates exception blob at the end
 // Using exception blob, this code is jumped from a compiled method.
--- a/src/cpu/sparc/vm/sharedRuntime_sparc.cpp	Thu Jun 02 13:36:11 2011 -0700
+++ b/src/cpu/sparc/vm/sharedRuntime_sparc.cpp	Fri Jun 03 22:31:43 2011 -0700
@@ -47,18 +47,6 @@
 
 #define __ masm->
 
-#ifdef COMPILER2
-UncommonTrapBlob*   SharedRuntime::_uncommon_trap_blob;
-#endif // COMPILER2
-
-DeoptimizationBlob* SharedRuntime::_deopt_blob;
-SafepointBlob*      SharedRuntime::_polling_page_safepoint_handler_blob;
-SafepointBlob*      SharedRuntime::_polling_page_return_handler_blob;
-RuntimeStub*        SharedRuntime::_wrong_method_blob;
-RuntimeStub*        SharedRuntime::_ic_miss_blob;
-RuntimeStub*        SharedRuntime::_resolve_opt_virtual_call_blob;
-RuntimeStub*        SharedRuntime::_resolve_virtual_call_blob;
-RuntimeStub*        SharedRuntime::_resolve_static_call_blob;
 
 class RegisterSaver {
 
@@ -3492,7 +3480,7 @@
 // the 64-bit %o's, then do a save, then fixup the caller's SP (our FP).
 // Tricky, tricky, tricky...
 
-static SafepointBlob* generate_handler_blob(address call_ptr, bool cause_return) {
+SafepointBlob* SharedRuntime::generate_handler_blob(address call_ptr, bool cause_return) {
   assert (StubRoutines::forward_exception_entry() != NULL, "must be generated before");
 
   // allocate space for the code
@@ -3587,7 +3575,7 @@
 // but since this is generic code we don't know what they are and the caller
 // must do any gc of the args.
 //
-static RuntimeStub* generate_resolve_blob(address destination, const char* name) {
+RuntimeStub* SharedRuntime::generate_resolve_blob(address destination, const char* name) {
   assert (StubRoutines::forward_exception_entry() != NULL, "must be generated before");
 
   // allocate space for the code
@@ -3677,35 +3665,3 @@
   // frame_size_words or bytes??
   return RuntimeStub::new_runtime_stub(name, &buffer, frame_complete, frame_size_words, oop_maps, true);
 }
-
-void SharedRuntime::generate_stubs() {
-
-  _wrong_method_blob = generate_resolve_blob(CAST_FROM_FN_PTR(address, SharedRuntime::handle_wrong_method),
-                                             "wrong_method_stub");
-
-  _ic_miss_blob = generate_resolve_blob(CAST_FROM_FN_PTR(address, SharedRuntime::handle_wrong_method_ic_miss),
-                                        "ic_miss_stub");
-
-  _resolve_opt_virtual_call_blob = generate_resolve_blob(CAST_FROM_FN_PTR(address, SharedRuntime::resolve_opt_virtual_call_C),
-                                        "resolve_opt_virtual_call");
-
-  _resolve_virtual_call_blob = generate_resolve_blob(CAST_FROM_FN_PTR(address, SharedRuntime::resolve_virtual_call_C),
-                                        "resolve_virtual_call");
-
-  _resolve_static_call_blob = generate_resolve_blob(CAST_FROM_FN_PTR(address, SharedRuntime::resolve_static_call_C),
-                                        "resolve_static_call");
-
-  _polling_page_safepoint_handler_blob =
-    generate_handler_blob(CAST_FROM_FN_PTR(address,
-                   SafepointSynchronize::handle_polling_page_exception), false);
-
-  _polling_page_return_handler_blob =
-    generate_handler_blob(CAST_FROM_FN_PTR(address,
-                   SafepointSynchronize::handle_polling_page_exception), true);
-
-  generate_deopt_blob();
-
-#ifdef COMPILER2
-  generate_uncommon_trap_blob();
-#endif // COMPILER2
-}
--- a/src/cpu/sparc/vm/stubRoutines_sparc.hpp	Thu Jun 02 13:36:11 2011 -0700
+++ b/src/cpu/sparc/vm/stubRoutines_sparc.hpp	Fri Jun 03 22:31:43 2011 -0700
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 1997, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1997, 2011, Oracle and/or its affiliates. All rights reserved.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
  * This code is free software; you can redistribute it and/or modify it
@@ -44,11 +44,6 @@
   code_size2 = 20000            // simply increase if too small (assembler will crash if too small)
 };
 
-// MethodHandles adapters
-enum method_handles_platform_dependent_constants {
-  method_handles_adapters_code_size = 15000
-};
-
 class Sparc {
  friend class StubGenerator;
 
--- a/src/cpu/sparc/vm/templateInterpreter_sparc.cpp	Thu Jun 02 13:36:11 2011 -0700
+++ b/src/cpu/sparc/vm/templateInterpreter_sparc.cpp	Fri Jun 03 22:31:43 2011 -0700
@@ -1712,7 +1712,7 @@
       int computed_sp_adjustment = (delta > 0) ? round_to(delta, WordsPerLong) : 0;
       *interpreter_frame->register_addr(I5_savedSP)    = (intptr_t) (fp + computed_sp_adjustment) - STACK_BIAS;
     } else {
-      assert(caller->is_compiled_frame() || caller->is_entry_frame(), "only possible cases");
+      assert(caller->is_compiled_frame() || caller->is_entry_frame() || caller->is_ricochet_frame(), "only possible cases");
       // Don't have Lesp available; lay out locals block in the caller
       // adjacent to the register window save area.
       //
--- a/src/cpu/x86/vm/assembler_x86.cpp	Thu Jun 02 13:36:11 2011 -0700
+++ b/src/cpu/x86/vm/assembler_x86.cpp	Fri Jun 03 22:31:43 2011 -0700
@@ -5090,7 +5090,7 @@
   } else {
     ttyLocker ttyl;
     ::tty->print_cr("=============== DEBUG MESSAGE: %s ================\n", msg);
-    assert(false, "DEBUG MESSAGE");
+    assert(false, err_msg("DEBUG MESSAGE: %s", msg));
   }
   ThreadStateTransition::transition(thread, _thread_in_vm, saved_state);
 }
@@ -5653,6 +5653,7 @@
     ttyLocker ttyl;
     ::tty->print_cr("=============== DEBUG MESSAGE: %s ================\n",
                     msg);
+    assert(false, err_msg("DEBUG MESSAGE: %s", msg));
   }
 }
 
--- a/src/cpu/x86/vm/methodHandles_x86.cpp	Thu Jun 02 13:36:11 2011 -0700
+++ b/src/cpu/x86/vm/methodHandles_x86.cpp	Fri Jun 03 22:31:43 2011 -0700
@@ -139,9 +139,9 @@
 
 void MethodHandles::RicochetFrame::generate_ricochet_blob(MacroAssembler* _masm,
                                                           // output params:
-                                                          int* frame_size_in_words,
                                                           int* bounce_offset,
-                                                          int* exception_offset) {
+                                                          int* exception_offset,
+                                                          int* frame_size_in_words) {
   (*frame_size_in_words) = RicochetFrame::frame_size_in_bytes() / wordSize;
 
   address start = __ pc();
@@ -366,7 +366,7 @@
                                     Register rdi_stack_move,
                                     Register rcx_amh,
                                     bool might_be_negative) {
-  BLOCK_COMMENT("load_stack_move");
+  BLOCK_COMMENT("load_stack_move {");
   Address rcx_amh_conversion(rcx_amh, java_lang_invoke_AdapterMethodHandle::conversion_offset_in_bytes());
   __ movl(rdi_stack_move, rcx_amh_conversion);
   __ sarl(rdi_stack_move, CONV_STACK_MOVE_SHIFT);
@@ -387,6 +387,7 @@
     __ stop("load_stack_move of garbage value");
     __ BIND(L_ok);
   }
+  BLOCK_COMMENT("} load_stack_move");
 }
 
 #ifdef ASSERT
@@ -1147,7 +1148,7 @@
 
   trace_method_handle(_masm, entry_name(ek));
 
-  BLOCK_COMMENT(entry_name(ek));
+  BLOCK_COMMENT(err_msg("Entry %s {", entry_name(ek)));
 
   switch ((int) ek) {
   case _raise_exception:
@@ -1292,7 +1293,7 @@
   case _bound_int_direct_mh:
   case _bound_long_direct_mh:
     {
-      bool direct_to_method = (ek >= _bound_ref_direct_mh);
+      const bool direct_to_method = (ek >= _bound_ref_direct_mh);
       BasicType arg_type  = ek_bound_mh_arg_type(ek);
       int       arg_slots = type2size[arg_type];
 
@@ -1929,7 +1930,7 @@
       // In the non-retaining case, this might move keep2 either up or down.
       // We don't have to copy the whole | RF... collect | complex,
       // but we must adjust RF.saved_args_base.
-      // Also, from now on, we will forget about the origial copy of |collect|.
+      // Also, from now on, we will forget about the original copy of |collect|.
       // If we are retaining it, we will treat it as part of |keep2|.
       // For clarity we will define |keep3| = |collect|keep2| or |keep2|.
 
@@ -1986,7 +1987,7 @@
       // Net shift (&new_argv - &old_argv) is (close_count - open_count).
       bool zero_open_count = (open_count == 0);  // remember this bit of info
       if (move_keep3 && fix_arg_base) {
-        // It will be easier t have everything in one register:
+        // It will be easier to have everything in one register:
         if (close_count.is_register()) {
           // Deduct open_count from close_count register to get a clean +/- value.
           __ subptr(close_count.as_register(), open_count);
@@ -2396,6 +2397,7 @@
     __ nop();
     return;
   }
+  BLOCK_COMMENT(err_msg("} Entry %s", entry_name(ek)));
   __ hlt();
 
   address me_cookie = MethodHandleEntry::start_compiled_entry(_masm, interp_entry);
--- a/src/cpu/x86/vm/methodHandles_x86.hpp	Thu Jun 02 13:36:11 2011 -0700
+++ b/src/cpu/x86/vm/methodHandles_x86.hpp	Fri Jun 03 22:31:43 2011 -0700
@@ -25,6 +25,11 @@
 // Platform-specific definitions for method handles.
 // These definitions are inlined into class MethodHandles.
 
+// Adapters
+enum /* platform_dependent_constants */ {
+  adapter_code_size = NOT_LP64(30000 DEBUG_ONLY(+ 10000)) LP64_ONLY(80000 DEBUG_ONLY(+ 120000))
+};
+
 public:
 
 // The stack just after the recursive call from a ricochet frame
@@ -188,7 +193,9 @@
 
   static void generate_ricochet_blob(MacroAssembler* _masm,
                                      // output params:
-                                     int* frame_size_in_words, int* bounce_offset, int* exception_offset);
+                                     int* bounce_offset,
+                                     int* exception_offset,
+                                     int* frame_size_in_words);
 
   static void enter_ricochet_frame(MacroAssembler* _masm,
                                    Register rcx_recv,
--- a/src/cpu/x86/vm/runtime_x86_32.cpp	Thu Jun 02 13:36:11 2011 -0700
+++ b/src/cpu/x86/vm/runtime_x86_32.cpp	Fri Jun 03 22:31:43 2011 -0700
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 1998, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1998, 2011, Oracle and/or its affiliates. All rights reserved.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
  * This code is free software; you can redistribute it and/or modify it
@@ -42,8 +42,6 @@
 
 #define __ masm->
 
-ExceptionBlob*     OptoRuntime::_exception_blob;
-
 //------------------------------generate_exception_blob---------------------------
 // creates exception blob at the end
 // Using exception blob, this code is jumped from a compiled method.
--- a/src/cpu/x86/vm/sharedRuntime_x86_32.cpp	Thu Jun 02 13:36:11 2011 -0700
+++ b/src/cpu/x86/vm/sharedRuntime_x86_32.cpp	Fri Jun 03 22:31:43 2011 -0700
@@ -42,18 +42,6 @@
 #endif
 
 #define __ masm->
-#ifdef COMPILER2
-UncommonTrapBlob   *SharedRuntime::_uncommon_trap_blob;
-#endif // COMPILER2
-
-DeoptimizationBlob *SharedRuntime::_deopt_blob;
-SafepointBlob      *SharedRuntime::_polling_page_safepoint_handler_blob;
-SafepointBlob      *SharedRuntime::_polling_page_return_handler_blob;
-RuntimeStub*       SharedRuntime::_wrong_method_blob;
-RuntimeStub*       SharedRuntime::_ic_miss_blob;
-RuntimeStub*       SharedRuntime::_resolve_opt_virtual_call_blob;
-RuntimeStub*       SharedRuntime::_resolve_virtual_call_blob;
-RuntimeStub*       SharedRuntime::_resolve_static_call_blob;
 
 const int StackAlignmentInSlots = StackAlignmentInBytes / VMRegImpl::stack_slot_size;
 
@@ -2253,31 +2241,6 @@
   return 0;
 }
 
-//----------------------------generate_ricochet_blob---------------------------
-void SharedRuntime::generate_ricochet_blob() {
-  if (!EnableInvokeDynamic)  return;  // leave it as a null
-
-  // allocate space for the code
-  ResourceMark rm;
-  // setup code generation tools
-  CodeBuffer   buffer("ricochet_blob", 256, 256);
-  MacroAssembler* masm = new MacroAssembler(&buffer);
-
-  int frame_size_in_words = -1, bounce_offset = -1, exception_offset = -1;
-  MethodHandles::RicochetFrame::generate_ricochet_blob(masm, &frame_size_in_words, &bounce_offset, &exception_offset);
-
-  // -------------
-  // make sure all code is generated
-  masm->flush();
-
-  // failed to generate?
-  if (frame_size_in_words < 0 || bounce_offset < 0 || exception_offset < 0) {
-    assert(false, "bad ricochet blob");
-    return;
-  }
-
-  _ricochet_blob = RicochetBlob::create(&buffer, bounce_offset, exception_offset, frame_size_in_words);
-}
 
 //------------------------------generate_deopt_blob----------------------------
 void SharedRuntime::generate_deopt_blob() {
@@ -2816,7 +2779,7 @@
 // setup oopmap, and calls safepoint code to stop the compiled code for
 // a safepoint.
 //
-static SafepointBlob* generate_handler_blob(address call_ptr, bool cause_return) {
+SafepointBlob* SharedRuntime::generate_handler_blob(address call_ptr, bool cause_return) {
 
   // Account for thread arg in our frame
   const int additional_words = 1;
@@ -2913,7 +2876,7 @@
 // but since this is generic code we don't know what they are and the caller
 // must do any gc of the args.
 //
-static RuntimeStub* generate_resolve_blob(address destination, const char* name) {
+RuntimeStub* SharedRuntime::generate_resolve_blob(address destination, const char* name) {
   assert (StubRoutines::forward_exception_entry() != NULL, "must be generated before");
 
   // allocate space for the code
@@ -2995,36 +2958,3 @@
   // frame_size_words or bytes??
   return RuntimeStub::new_runtime_stub(name, &buffer, frame_complete, frame_size_words, oop_maps, true);
 }
-
-void SharedRuntime::generate_stubs() {
-
-  _wrong_method_blob = generate_resolve_blob(CAST_FROM_FN_PTR(address, SharedRuntime::handle_wrong_method),
-                                        "wrong_method_stub");
-
-  _ic_miss_blob      = generate_resolve_blob(CAST_FROM_FN_PTR(address, SharedRuntime::handle_wrong_method_ic_miss),
-                                        "ic_miss_stub");
-
-  _resolve_opt_virtual_call_blob = generate_resolve_blob(CAST_FROM_FN_PTR(address, SharedRuntime::resolve_opt_virtual_call_C),
-                                        "resolve_opt_virtual_call");
-
-  _resolve_virtual_call_blob = generate_resolve_blob(CAST_FROM_FN_PTR(address, SharedRuntime::resolve_virtual_call_C),
-                                        "resolve_virtual_call");
-
-  _resolve_static_call_blob = generate_resolve_blob(CAST_FROM_FN_PTR(address, SharedRuntime::resolve_static_call_C),
-                                        "resolve_static_call");
-
-  _polling_page_safepoint_handler_blob =
-    generate_handler_blob(CAST_FROM_FN_PTR(address,
-                   SafepointSynchronize::handle_polling_page_exception), false);
-
-  _polling_page_return_handler_blob =
-    generate_handler_blob(CAST_FROM_FN_PTR(address,
-                   SafepointSynchronize::handle_polling_page_exception), true);
-
-  generate_ricochet_blob();
-
-  generate_deopt_blob();
-#ifdef COMPILER2
-  generate_uncommon_trap_blob();
-#endif // COMPILER2
-}
--- a/src/cpu/x86/vm/sharedRuntime_x86_64.cpp	Thu Jun 02 13:36:11 2011 -0700
+++ b/src/cpu/x86/vm/sharedRuntime_x86_64.cpp	Fri Jun 03 22:31:43 2011 -0700
@@ -41,24 +41,10 @@
 #include "opto/runtime.hpp"
 #endif
 
-DeoptimizationBlob *SharedRuntime::_deopt_blob;
-#ifdef COMPILER2
-UncommonTrapBlob   *SharedRuntime::_uncommon_trap_blob;
-ExceptionBlob      *OptoRuntime::_exception_blob;
-#endif // COMPILER2
-
-SafepointBlob      *SharedRuntime::_polling_page_safepoint_handler_blob;
-SafepointBlob      *SharedRuntime::_polling_page_return_handler_blob;
-RuntimeStub*       SharedRuntime::_wrong_method_blob;
-RuntimeStub*       SharedRuntime::_ic_miss_blob;
-RuntimeStub*       SharedRuntime::_resolve_opt_virtual_call_blob;
-RuntimeStub*       SharedRuntime::_resolve_virtual_call_blob;
-RuntimeStub*       SharedRuntime::_resolve_static_call_blob;
+#define __ masm->
 
 const int StackAlignmentInSlots = StackAlignmentInBytes / VMRegImpl::stack_slot_size;
 
-#define __ masm->
-
 class SimpleRuntimeFrame {
 
   public:
@@ -2530,32 +2516,6 @@
 }
 
 
-//----------------------------generate_ricochet_blob---------------------------
-void SharedRuntime::generate_ricochet_blob() {
-  if (!EnableInvokeDynamic)  return;  // leave it as a null
-
-  // allocate space for the code
-  ResourceMark rm;
-  // setup code generation tools
-  CodeBuffer   buffer("ricochet_blob", 512, 512);
-  MacroAssembler* masm = new MacroAssembler(&buffer);
-
-  int frame_size_in_words = -1, bounce_offset = -1, exception_offset = -1;
-  MethodHandles::RicochetFrame::generate_ricochet_blob(masm, &frame_size_in_words, &bounce_offset, &exception_offset);
-
-  // -------------
-  // make sure all code is generated
-  masm->flush();
-
-  // failed to generate?
-  if (frame_size_in_words < 0 || bounce_offset < 0 || exception_offset < 0) {
-    assert(false, "bad ricochet blob");
-    return;
-  }
-
-  _ricochet_blob = RicochetBlob::create(&buffer, bounce_offset, exception_offset, frame_size_in_words);
-}
-
 //------------------------------generate_deopt_blob----------------------------
 void SharedRuntime::generate_deopt_blob() {
   // Allocate space for the code
@@ -3046,7 +3006,7 @@
 // Generate a special Compile2Runtime blob that saves all registers,
 // and setup oopmap.
 //
-static SafepointBlob* generate_handler_blob(address call_ptr, bool cause_return) {
+SafepointBlob* SharedRuntime::generate_handler_blob(address call_ptr, bool cause_return) {
   assert(StubRoutines::forward_exception_entry() != NULL,
          "must be generated before");
 
@@ -3132,7 +3092,7 @@
 // but since this is generic code we don't know what they are and the caller
 // must do any gc of the args.
 //
-static RuntimeStub* generate_resolve_blob(address destination, const char* name) {
+RuntimeStub* SharedRuntime::generate_resolve_blob(address destination, const char* name) {
   assert (StubRoutines::forward_exception_entry() != NULL, "must be generated before");
 
   // allocate space for the code
@@ -3209,38 +3169,6 @@
 }
 
 
-void SharedRuntime::generate_stubs() {
-
-  _wrong_method_blob = generate_resolve_blob(CAST_FROM_FN_PTR(address, SharedRuntime::handle_wrong_method),
-                                        "wrong_method_stub");
-  _ic_miss_blob =      generate_resolve_blob(CAST_FROM_FN_PTR(address, SharedRuntime::handle_wrong_method_ic_miss),
-                                        "ic_miss_stub");
-  _resolve_opt_virtual_call_blob = generate_resolve_blob(CAST_FROM_FN_PTR(address, SharedRuntime::resolve_opt_virtual_call_C),
-                                        "resolve_opt_virtual_call");
-
-  _resolve_virtual_call_blob = generate_resolve_blob(CAST_FROM_FN_PTR(address, SharedRuntime::resolve_virtual_call_C),
-                                        "resolve_virtual_call");
-
-  _resolve_static_call_blob = generate_resolve_blob(CAST_FROM_FN_PTR(address, SharedRuntime::resolve_static_call_C),
-                                        "resolve_static_call");
-  _polling_page_safepoint_handler_blob =
-    generate_handler_blob(CAST_FROM_FN_PTR(address,
-                   SafepointSynchronize::handle_polling_page_exception), false);
-
-  _polling_page_return_handler_blob =
-    generate_handler_blob(CAST_FROM_FN_PTR(address,
-                   SafepointSynchronize::handle_polling_page_exception), true);
-
-  generate_ricochet_blob();
-
-  generate_deopt_blob();
-
-#ifdef COMPILER2
-  generate_uncommon_trap_blob();
-#endif // COMPILER2
-}
-
-
 #ifdef COMPILER2
 // This is here instead of runtime_x86_64.cpp because it uses SimpleRuntimeFrame
 //
--- a/src/cpu/x86/vm/stubRoutines_x86_32.hpp	Thu Jun 02 13:36:11 2011 -0700
+++ b/src/cpu/x86/vm/stubRoutines_x86_32.hpp	Fri Jun 03 22:31:43 2011 -0700
@@ -34,11 +34,6 @@
   code_size2 = 22000            // simply increase if too small (assembler will crash if too small)
 };
 
-// MethodHandles adapters
-enum method_handles_platform_dependent_constants {
-  method_handles_adapters_code_size = 30000 DEBUG_ONLY(+ 10000)
-};
-
 class x86 {
  friend class StubGenerator;
  friend class VMStructs;
--- a/src/cpu/x86/vm/stubRoutines_x86_64.hpp	Thu Jun 02 13:36:11 2011 -0700
+++ b/src/cpu/x86/vm/stubRoutines_x86_64.hpp	Fri Jun 03 22:31:43 2011 -0700
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2003, 2011, Oracle and/or its affiliates. All rights reserved.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
  * This code is free software; you can redistribute it and/or modify it
@@ -36,11 +36,6 @@
   code_size2 = 22000           // simply increase if too small (assembler will crash if too small)
 };
 
-// MethodHandles adapters
-enum method_handles_platform_dependent_constants {
-  method_handles_adapters_code_size = 80000 DEBUG_ONLY(+ 120000)
-};
-
 class x86 {
  friend class StubGenerator;
 
--- a/src/share/vm/compiler/oopMap.cpp	Thu Jun 02 13:36:11 2011 -0700
+++ b/src/share/vm/compiler/oopMap.cpp	Fri Jun 03 22:31:43 2011 -0700
@@ -638,7 +638,9 @@
     assert(*derived_loc != (oop)base_loc, "location already added");
     assert(_list != NULL, "list must exist");
     intptr_t offset = value_of_loc(derived_loc) - value_of_loc(base_loc);
-    assert(offset >= -1000000, "wrong derived pointer info");
+    // This assert is invalid because derived pointers can be
+    // arbitrarily far away from their base.
+    // assert(offset >= -1000000, "wrong derived pointer info");
 
     if (TraceDerivedPointers) {
       tty->print_cr(
--- a/src/share/vm/opto/runtime.cpp	Thu Jun 02 13:36:11 2011 -0700
+++ b/src/share/vm/opto/runtime.cpp	Fri Jun 03 22:31:43 2011 -0700
@@ -120,6 +120,7 @@
 address OptoRuntime::_zap_dead_native_locals_Java                 = NULL;
 # endif
 
+ExceptionBlob* OptoRuntime::_exception_blob;
 
 // This should be called in an assertion at the start of OptoRuntime routines
 // which are entered from compiled code (all of them)
--- a/src/share/vm/prims/methodHandleWalk.cpp	Thu Jun 02 13:36:11 2011 -0700
+++ b/src/share/vm/prims/methodHandleWalk.cpp	Fri Jun 03 22:31:43 2011 -0700
@@ -178,6 +178,12 @@
   return "unknown_op";
 }
 
+void MethodHandleChain::print(oopDesc* m) {
+  HandleMark hm;
+  ResourceMark rm;
+  Handle mh(m);
+  print(mh);
+}
 
 void MethodHandleChain::print(Handle mh) {
   EXCEPTION_MARK;
@@ -414,8 +420,7 @@
         assert(dest == T_OBJECT, "");
         ArgToken arg = _outgoing.at(arg_slot);
         assert(dest == arg.basic_type(), "");
-        ArgToken new_arg = make_conversion(T_OBJECT, dest_klass, Bytecodes::_checkcast, arg, CHECK_(empty));
-        assert(!arg.has_index() || arg.index() == new_arg.index(), "should be the same index");
+        arg = make_conversion(T_OBJECT, dest_klass, Bytecodes::_checkcast, arg, CHECK_(empty));
         debug_only(dest_klass = (klassOop)badOop);
         break;
       }
@@ -1248,8 +1253,9 @@
       index = src.index();
     }
     emit_bc(op, cpool_klass_put(tk));
-    if (index == -1)
-      index = new_local_index(type);
+    // Allocate a new local for the type so that we don't hide the
+    // previous type from the verifier.
+    index = new_local_index(type);
     emit_store(srctype, index);
     break;
 
--- a/src/share/vm/prims/methodHandleWalk.hpp	Thu Jun 02 13:36:11 2011 -0700
+++ b/src/share/vm/prims/methodHandleWalk.hpp	Fri Jun 03 22:31:43 2011 -0700
@@ -110,6 +110,7 @@
   // slot order to make it easier to understand.
   void print();
   static void print(Handle mh);
+  static void print(oopDesc* mh);
 #endif
 };
 
@@ -273,7 +274,7 @@
   static jvalue one_jvalue;
 
   // Fake constant pool entry.
-  class ConstantValue {
+  class ConstantValue : public ResourceObj {
   private:
     int       _tag;   // Constant pool tag type.
     JavaValue _value;
--- a/src/share/vm/prims/methodHandles.cpp	Thu Jun 02 13:36:11 2011 -0700
+++ b/src/share/vm/prims/methodHandles.cpp	Fri Jun 03 22:31:43 2011 -0700
@@ -160,8 +160,7 @@
 };
 
 // Adapters.
-MethodHandlesAdapterBlob* MethodHandles::_adapter_code      = NULL;
-int                       MethodHandles::_adapter_code_size = StubRoutines::method_handles_adapters_code_size;
+MethodHandlesAdapterBlob* MethodHandles::_adapter_code = NULL;
 
 jobject MethodHandles::_raise_exception_method;
 
@@ -200,12 +199,15 @@
 
   ResourceMark rm;
   TraceTime timer("MethodHandles adapters generation", TraceStartupTime);
-  _adapter_code = MethodHandlesAdapterBlob::create(_adapter_code_size);
+  _adapter_code = MethodHandlesAdapterBlob::create(adapter_code_size);
   if (_adapter_code == NULL)
-    vm_exit_out_of_memory(_adapter_code_size, "CodeCache: no room for MethodHandles adapters");
+    vm_exit_out_of_memory(adapter_code_size, "CodeCache: no room for MethodHandles adapters");
   CodeBuffer code(_adapter_code);
   MethodHandlesAdapterGenerator g(&code);
   g.generate();
+
+  // Transfer code comments
+  _adapter_code->set_comments(code.comments());
 }
 
 //------------------------------------------------------------------------------
@@ -1721,6 +1723,7 @@
 
 void MethodHandles::verify_BoundMethodHandle(Handle mh, Handle target, int argnum,
                                              bool direct_to_method, TRAPS) {
+  ResourceMark rm;
   Handle ptype_handle(THREAD,
                            java_lang_invoke_MethodType::ptype(java_lang_invoke_MethodHandle::type(target()), argnum));
   KlassHandle ptype_klass;
@@ -1884,6 +1887,7 @@
 }
 
 void MethodHandles::verify_AdapterMethodHandle(Handle mh, int argnum, TRAPS) {
+  ResourceMark rm;
   jint conversion = java_lang_invoke_AdapterMethodHandle::conversion(mh());
   int  argslot    = java_lang_invoke_AdapterMethodHandle::vmargslot(mh());
 
@@ -1987,7 +1991,7 @@
         bool rotate_up = (src_slot > dest_slot); // upward rotation
         int src_arg    = argnum;
         int dest_arg   = argument_slot_to_argnum(dst_mtype(), dest_slot);
-        verify_vmargslot(mh, dest_arg, dest_slot, CHECK);
+        verify_vmargslot(target, dest_arg, dest_slot, CHECK);
         if (!(dest_slot >= src_slot + swap_size) &&
             !(src_slot >= dest_slot + swap_size)) {
           err = "source, destination slots must be distinct";
@@ -2563,7 +2567,7 @@
   }
 
   if (err != NULL) {
-    throw_InternalError_for_bad_conversion(conversion, err, THREAD);
+    throw_InternalError_for_bad_conversion(conversion, err_msg("%s: conv_op %d ek_opt %d", err, conv_op, ek_opt), THREAD);
     return;
   }
 
--- a/src/share/vm/prims/methodHandles.hpp	Thu Jun 02 13:36:11 2011 -0700
+++ b/src/share/vm/prims/methodHandles.hpp	Fri Jun 03 22:31:43 2011 -0700
@@ -217,7 +217,6 @@
 
   // Adapters.
   static MethodHandlesAdapterBlob* _adapter_code;
-  static int                       _adapter_code_size;
 
   static bool ek_valid(EntryKind ek)            { return (uint)ek < (uint)_EK_LIMIT; }
   static bool conv_op_valid(int op)             { return (uint)op < (uint)CONV_OP_LIMIT; }
@@ -716,12 +715,10 @@
 # include "methodHandles_x86.hpp"
 #endif
 #ifdef TARGET_ARCH_sparc
-#define TARGET_ARCH_NYI_6939861 1 //FIXME
-//# include "methodHandles_sparc.hpp"
+# include "methodHandles_sparc.hpp"
 #endif
 #ifdef TARGET_ARCH_zero
-#define TARGET_ARCH_NYI_6939861 1 //FIXME
-//# include "methodHandles_zero.hpp"
+# include "methodHandles_zero.hpp"
 #endif
 #ifdef TARGET_ARCH_arm
 # include "methodHandles_arm.hpp"
--- a/src/share/vm/runtime/sharedRuntime.cpp	Thu Jun 02 13:36:11 2011 -0700
+++ b/src/share/vm/runtime/sharedRuntime.cpp	Fri Jun 03 22:31:43 2011 -0700
@@ -80,6 +80,72 @@
 #include "c1/c1_Runtime1.hpp"
 #endif
 
+// Shared stub locations
+RuntimeStub*        SharedRuntime::_wrong_method_blob;
+RuntimeStub*        SharedRuntime::_ic_miss_blob;
+RuntimeStub*        SharedRuntime::_resolve_opt_virtual_call_blob;
+RuntimeStub*        SharedRuntime::_resolve_virtual_call_blob;
+RuntimeStub*        SharedRuntime::_resolve_static_call_blob;
+
+DeoptimizationBlob* SharedRuntime::_deopt_blob;
+RicochetBlob*       SharedRuntime::_ricochet_blob;
+
+SafepointBlob*      SharedRuntime::_polling_page_safepoint_handler_blob;
+SafepointBlob*      SharedRuntime::_polling_page_return_handler_blob;
+
+#ifdef COMPILER2
+UncommonTrapBlob*   SharedRuntime::_uncommon_trap_blob;
+#endif // COMPILER2
+
+
+//----------------------------generate_stubs-----------------------------------
+void SharedRuntime::generate_stubs() {
+  _wrong_method_blob                   = generate_resolve_blob(CAST_FROM_FN_PTR(address, SharedRuntime::handle_wrong_method),         "wrong_method_stub");
+  _ic_miss_blob                        = generate_resolve_blob(CAST_FROM_FN_PTR(address, SharedRuntime::handle_wrong_method_ic_miss), "ic_miss_stub");
+  _resolve_opt_virtual_call_blob       = generate_resolve_blob(CAST_FROM_FN_PTR(address, SharedRuntime::resolve_opt_virtual_call_C),  "resolve_opt_virtual_call");
+  _resolve_virtual_call_blob           = generate_resolve_blob(CAST_FROM_FN_PTR(address, SharedRuntime::resolve_virtual_call_C),      "resolve_virtual_call");
+  _resolve_static_call_blob            = generate_resolve_blob(CAST_FROM_FN_PTR(address, SharedRuntime::resolve_static_call_C),       "resolve_static_call");
+
+  _polling_page_safepoint_handler_blob = generate_handler_blob(CAST_FROM_FN_PTR(address, SafepointSynchronize::handle_polling_page_exception), false);
+  _polling_page_return_handler_blob    = generate_handler_blob(CAST_FROM_FN_PTR(address, SafepointSynchronize::handle_polling_page_exception), true);
+
+  generate_ricochet_blob();
+  generate_deopt_blob();
+
+#ifdef COMPILER2
+  generate_uncommon_trap_blob();
+#endif // COMPILER2
+}
+
+//----------------------------generate_ricochet_blob---------------------------
+void SharedRuntime::generate_ricochet_blob() {
+  if (!EnableInvokeDynamic)  return;  // leave it as a null
+
+#ifndef TARGET_ARCH_NYI_6939861
+  // allocate space for the code
+  ResourceMark rm;
+  // setup code generation tools
+  CodeBuffer buffer("ricochet_blob", 256 LP64_ONLY(+ 256), 256);  // XXX x86 LP64L: 512, 512
+  MacroAssembler* masm = new MacroAssembler(&buffer);
+
+  int bounce_offset = -1, exception_offset = -1, frame_size_in_words = -1;
+  MethodHandles::RicochetFrame::generate_ricochet_blob(masm, &bounce_offset, &exception_offset, &frame_size_in_words);
+
+  // -------------
+  // make sure all code is generated
+  masm->flush();
+
+  // failed to generate?
+  if (bounce_offset < 0 || exception_offset < 0 || frame_size_in_words < 0) {
+    assert(false, "bad ricochet blob");
+    return;
+  }
+
+  _ricochet_blob = RicochetBlob::create(&buffer, bounce_offset, exception_offset, frame_size_in_words);
+#endif
+}
+
+
 #include <math.h>
 
 HS_DTRACE_PROBE_DECL4(hotspot, object__alloc, Thread*, char*, int, size_t);
@@ -88,8 +154,6 @@
 HS_DTRACE_PROBE_DECL7(hotspot, method__return, int,
                       char*, int, char*, int, char*, int);
 
-RicochetBlob*      SharedRuntime::_ricochet_blob = NULL;
-
 // Implementation of SharedRuntime
 
 #ifndef PRODUCT
@@ -143,6 +207,7 @@
 int     SharedRuntime::_ICmiss_count[SharedRuntime::maxICmiss_count];
 address SharedRuntime::_ICmiss_at[SharedRuntime::maxICmiss_count];
 
+
 void SharedRuntime::trace_ic_miss(address at) {
   for (int i = 0; i < _ICmiss_index; i++) {
     if (_ICmiss_at[i] == at) {
--- a/src/share/vm/runtime/sharedRuntime.hpp	Thu Jun 02 13:36:11 2011 -0700
+++ b/src/share/vm/runtime/sharedRuntime.hpp	Fri Jun 03 22:31:43 2011 -0700
@@ -52,28 +52,33 @@
 
   // Shared stub locations
 
-  static RuntimeStub* _wrong_method_blob;
-  static RuntimeStub* _ic_miss_blob;
-  static RuntimeStub* _resolve_opt_virtual_call_blob;
-  static RuntimeStub* _resolve_virtual_call_blob;
-  static RuntimeStub* _resolve_static_call_blob;
+  static RuntimeStub*        _wrong_method_blob;
+  static RuntimeStub*        _ic_miss_blob;
+  static RuntimeStub*        _resolve_opt_virtual_call_blob;
+  static RuntimeStub*        _resolve_virtual_call_blob;
+  static RuntimeStub*        _resolve_static_call_blob;
 
-  static RicochetBlob* _ricochet_blob;
+  static DeoptimizationBlob* _deopt_blob;
+  static RicochetBlob*       _ricochet_blob;
 
-  static SafepointBlob* _polling_page_safepoint_handler_blob;
-  static SafepointBlob* _polling_page_return_handler_blob;
+  static SafepointBlob*      _polling_page_safepoint_handler_blob;
+  static SafepointBlob*      _polling_page_return_handler_blob;
+
 #ifdef COMPILER2
-  static ExceptionBlob*       _exception_blob;
-  static UncommonTrapBlob*    _uncommon_trap_blob;
+  static UncommonTrapBlob*   _uncommon_trap_blob;
 #endif // COMPILER2
 
 #ifndef PRODUCT
-
   // Counters
   static int     _nof_megamorphic_calls;         // total # of megamorphic calls (through vtable)
+#endif // !PRODUCT
 
-#endif // !PRODUCT
+ private:
+  static SafepointBlob* generate_handler_blob(address call_ptr, bool cause_return);
+  static RuntimeStub*   generate_resolve_blob(address destination, const char* name);
+
  public:
+  static void generate_stubs(void);
 
   // max bytes for each dtrace string parameter
   enum { max_dtrace_string_size = 256 };
@@ -326,12 +331,9 @@
                                      bool is_virtual,
                                      bool is_optimized, TRAPS);
 
-  static void generate_stubs(void);
-
   private:
   // deopt blob
   static void generate_deopt_blob(void);
-  static DeoptimizationBlob* _deopt_blob;
 
   public:
   static DeoptimizationBlob* deopt_blob(void)      { return _deopt_blob; }