changeset 2460:167b70ff3abc

6939861: JVM should handle more conversion operations Reviewed-by: twisti, jrose
author never
date Fri, 06 May 2011 16:33:13 -0700
parents d4c1fbc3de95
children 566ea7a12419
files src/cpu/x86/vm/assembler_x86.hpp src/cpu/x86/vm/frame_x86.cpp src/cpu/x86/vm/frame_x86.hpp src/cpu/x86/vm/frame_x86.inline.hpp src/cpu/x86/vm/interpreter_x86.hpp src/cpu/x86/vm/methodHandles_x86.cpp src/cpu/x86/vm/methodHandles_x86.hpp 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/classfile/javaClasses.cpp src/share/vm/classfile/javaClasses.hpp src/share/vm/classfile/systemDictionary.cpp src/share/vm/classfile/vmSymbols.hpp src/share/vm/code/codeBlob.cpp src/share/vm/code/codeBlob.hpp src/share/vm/code/codeCache.cpp src/share/vm/compiler/disassembler.cpp src/share/vm/prims/jvmtiTagMap.cpp src/share/vm/prims/methodHandleWalk.cpp src/share/vm/prims/methodHandles.cpp src/share/vm/prims/methodHandles.hpp src/share/vm/runtime/frame.cpp src/share/vm/runtime/frame.hpp src/share/vm/runtime/globals.hpp src/share/vm/runtime/sharedRuntime.cpp src/share/vm/runtime/sharedRuntime.hpp src/share/vm/services/heapDumper.cpp
diffstat 29 files changed, 3009 insertions(+), 643 deletions(-) [+]
line wrap: on
line diff
--- a/src/cpu/x86/vm/assembler_x86.hpp	Fri May 06 12:12:29 2011 -0700
+++ b/src/cpu/x86/vm/assembler_x86.hpp	Fri May 06 16:33:13 2011 -0700
@@ -234,6 +234,20 @@
     a._disp += disp;
     return a;
   }
+  Address plus_disp(RegisterOrConstant disp, ScaleFactor scale = times_1) const {
+    Address a = (*this);
+    a._disp += disp.constant_or_zero() * scale_size(scale);
+    if (disp.is_register()) {
+      assert(!a.index()->is_valid(), "competing indexes");
+      a._index = disp.as_register();
+      a._scale = scale;
+    }
+    return a;
+  }
+  bool is_same_address(Address a) const {
+    // disregard _rspec
+    return _base == a._base && _disp == a._disp && _index == a._index && _scale == a._scale;
+  }
 
   // The following two overloads are used in connection with the
   // ByteSize type (see sizes.hpp).  They simplify the use of
@@ -2029,6 +2043,10 @@
   void addptr(Register dst, Address src) { LP64_ONLY(addq(dst, src)) NOT_LP64(addl(dst, src)); }
   void addptr(Register dst, int32_t src);
   void addptr(Register dst, Register src);
+  void addptr(Register dst, RegisterOrConstant src) {
+    if (src.is_constant()) addptr(dst, (int) src.as_constant());
+    else                   addptr(dst,       src.as_register());
+  }
 
   void andptr(Register dst, int32_t src);
   void andptr(Register src1, Register src2) { LP64_ONLY(andq(src1, src2)) NOT_LP64(andl(src1, src2)) ; }
@@ -2090,7 +2108,10 @@
   void subptr(Register dst, Address src) { LP64_ONLY(subq(dst, src)) NOT_LP64(subl(dst, src)); }
   void subptr(Register dst, int32_t src);
   void subptr(Register dst, Register src);
-
+  void subptr(Register dst, RegisterOrConstant src) {
+    if (src.is_constant()) subptr(dst, (int) src.as_constant());
+    else                   subptr(dst,       src.as_register());
+  }
 
   void sbbptr(Address dst, int32_t src) { LP64_ONLY(sbbq(dst, src)) NOT_LP64(sbbl(dst, src)); }
   void sbbptr(Register dst, int32_t src) { LP64_ONLY(sbbq(dst, src)) NOT_LP64(sbbl(dst, src)); }
@@ -2288,6 +2309,11 @@
 
   void movptr(Address dst, Register src);
 
+  void movptr(Register dst, RegisterOrConstant src) {
+    if (src.is_constant()) movptr(dst, src.as_constant());
+    else                   movptr(dst, src.as_register());
+  }
+
 #ifdef _LP64
   // Generally the next two are only used for moving NULL
   // Although there are situations in initializing the mark word where
--- a/src/cpu/x86/vm/frame_x86.cpp	Fri May 06 12:12:29 2011 -0700
+++ b/src/cpu/x86/vm/frame_x86.cpp	Fri May 06 16:33:13 2011 -0700
@@ -339,7 +339,6 @@
   return fr;
 }
 
-
 //------------------------------------------------------------------------------
 // frame::verify_deopt_original_pc
 //
@@ -361,6 +360,55 @@
 }
 #endif
 
+//------------------------------------------------------------------------------
+// frame::adjust_unextended_sp
+void frame::adjust_unextended_sp() {
+  // If we are returning to a compiled MethodHandle call site, the
+  // saved_fp will in fact be a saved value of the unextended SP.  The
+  // simplest way to tell whether we are returning to such a call site
+  // is as follows:
+
+  nmethod* sender_nm = (_cb == NULL) ? NULL : _cb->as_nmethod_or_null();
+  if (sender_nm != NULL) {
+    // If the sender PC is a deoptimization point, get the original
+    // PC.  For MethodHandle call site the unextended_sp is stored in
+    // saved_fp.
+    if (sender_nm->is_deopt_mh_entry(_pc)) {
+      DEBUG_ONLY(verify_deopt_mh_original_pc(sender_nm, _fp));
+      _unextended_sp = _fp;
+    }
+    else if (sender_nm->is_deopt_entry(_pc)) {
+      DEBUG_ONLY(verify_deopt_original_pc(sender_nm, _unextended_sp));
+    }
+    else if (sender_nm->is_method_handle_return(_pc)) {
+      _unextended_sp = _fp;
+    }
+  }
+}
+
+//------------------------------------------------------------------------------
+// frame::update_map_with_saved_link
+void frame::update_map_with_saved_link(RegisterMap* map, intptr_t** link_addr) {
+  // The interpreter and compiler(s) always save EBP/RBP in a known
+  // location on entry. We must record where that location is
+  // so this if EBP/RBP was live on callout from c2 we can find
+  // the saved copy no matter what it called.
+
+  // Since the interpreter always saves EBP/RBP if we record where it is then
+  // we don't have to always save EBP/RBP on entry and exit to c2 compiled
+  // code, on entry will be enough.
+  map->set_location(rbp->as_VMReg(), (address) link_addr);
+#ifdef AMD64
+  // this is weird "H" ought to be at a higher address however the
+  // oopMaps seems to have the "H" regs at the same address and the
+  // vanilla register.
+  // XXXX make this go away
+  if (true) {
+    map->set_location(rbp->as_VMReg()->next(), (address) link_addr);
+  }
+#endif // AMD64
+}
+
 
 //------------------------------------------------------------------------------
 // frame::sender_for_interpreter_frame
@@ -372,54 +420,13 @@
   // This is the sp before any possible extension (adapter/locals).
   intptr_t* unextended_sp = interpreter_frame_sender_sp();
 
-  // Stored FP.
-  intptr_t* saved_fp = link();
-
-  address sender_pc = this->sender_pc();
-  CodeBlob* sender_cb = CodeCache::find_blob_unsafe(sender_pc);
-  assert(sender_cb, "sanity");
-  nmethod* sender_nm = sender_cb->as_nmethod_or_null();
-
-  if (sender_nm != NULL) {
-    // If the sender PC is a deoptimization point, get the original
-    // PC.  For MethodHandle call site the unextended_sp is stored in
-    // saved_fp.
-    if (sender_nm->is_deopt_mh_entry(sender_pc)) {
-      DEBUG_ONLY(verify_deopt_mh_original_pc(sender_nm, saved_fp));
-      unextended_sp = saved_fp;
-    }
-    else if (sender_nm->is_deopt_entry(sender_pc)) {
-      DEBUG_ONLY(verify_deopt_original_pc(sender_nm, unextended_sp));
-    }
-    else if (sender_nm->is_method_handle_return(sender_pc)) {
-      unextended_sp = saved_fp;
-    }
-  }
-
-  // The interpreter and compiler(s) always save EBP/RBP in a known
-  // location on entry. We must record where that location is
-  // so this if EBP/RBP was live on callout from c2 we can find
-  // the saved copy no matter what it called.
-
-  // Since the interpreter always saves EBP/RBP if we record where it is then
-  // we don't have to always save EBP/RBP on entry and exit to c2 compiled
-  // code, on entry will be enough.
 #ifdef COMPILER2
   if (map->update_map()) {
-    map->set_location(rbp->as_VMReg(), (address) addr_at(link_offset));
-#ifdef AMD64
-    // this is weird "H" ought to be at a higher address however the
-    // oopMaps seems to have the "H" regs at the same address and the
-    // vanilla register.
-    // XXXX make this go away
-    if (true) {
-      map->set_location(rbp->as_VMReg()->next(), (address)addr_at(link_offset));
-    }
-#endif // AMD64
+    update_map_with_saved_link(map, (intptr_t**) addr_at(link_offset));
   }
 #endif // COMPILER2
 
-  return frame(sender_sp, unextended_sp, saved_fp, sender_pc);
+  return frame(sender_sp, unextended_sp, link(), sender_pc());
 }
 
 
@@ -427,6 +434,7 @@
 // frame::sender_for_compiled_frame
 frame frame::sender_for_compiled_frame(RegisterMap* map) const {
   assert(map != NULL, "map must be set");
+  assert(!is_ricochet_frame(), "caller must handle this");
 
   // frame owned by optimizing compiler
   assert(_cb->frame_size() >= 0, "must have non-zero frame size");
@@ -438,31 +446,7 @@
 
   // This is the saved value of EBP which may or may not really be an FP.
   // It is only an FP if the sender is an interpreter frame (or C1?).
-  intptr_t* saved_fp = (intptr_t*) *(sender_sp - frame::sender_sp_offset);
-
-  // If we are returning to a compiled MethodHandle call site, the
-  // saved_fp will in fact be a saved value of the unextended SP.  The
-  // simplest way to tell whether we are returning to such a call site
-  // is as follows:
-  CodeBlob* sender_cb = CodeCache::find_blob_unsafe(sender_pc);
-  assert(sender_cb, "sanity");
-  nmethod* sender_nm = sender_cb->as_nmethod_or_null();
-
-  if (sender_nm != NULL) {
-    // If the sender PC is a deoptimization point, get the original
-    // PC.  For MethodHandle call site the unextended_sp is stored in
-    // saved_fp.
-    if (sender_nm->is_deopt_mh_entry(sender_pc)) {
-      DEBUG_ONLY(verify_deopt_mh_original_pc(sender_nm, saved_fp));
-      unextended_sp = saved_fp;
-    }
-    else if (sender_nm->is_deopt_entry(sender_pc)) {
-      DEBUG_ONLY(verify_deopt_original_pc(sender_nm, unextended_sp));
-    }
-    else if (sender_nm->is_method_handle_return(sender_pc)) {
-      unextended_sp = saved_fp;
-    }
-  }
+  intptr_t** saved_fp_addr = (intptr_t**) (sender_sp - frame::sender_sp_offset);
 
   if (map->update_map()) {
     // Tell GC to use argument oopmaps for some runtime stubs that need it.
@@ -472,23 +456,15 @@
     if (_cb->oop_maps() != NULL) {
       OopMapSet::update_register_map(this, map);
     }
+
     // Since the prolog does the save and restore of EBP there is no oopmap
     // for it so we must fill in its location as if there was an oopmap entry
     // since if our caller was compiled code there could be live jvm state in it.
-    map->set_location(rbp->as_VMReg(), (address) (sender_sp - frame::sender_sp_offset));
-#ifdef AMD64
-    // this is weird "H" ought to be at a higher address however the
-    // oopMaps seems to have the "H" regs at the same address and the
-    // vanilla register.
-    // XXXX make this go away
-    if (true) {
-      map->set_location(rbp->as_VMReg()->next(), (address) (sender_sp - frame::sender_sp_offset));
-    }
-#endif // AMD64
+    update_map_with_saved_link(map, saved_fp_addr);
   }
 
   assert(sender_sp != sp(), "must have changed");
-  return frame(sender_sp, unextended_sp, saved_fp, sender_pc);
+  return frame(sender_sp, unextended_sp, *saved_fp_addr, sender_pc);
 }
 
 
@@ -502,6 +478,7 @@
   if (is_entry_frame())       return sender_for_entry_frame(map);
   if (is_interpreted_frame()) return sender_for_interpreter_frame(map);
   assert(_cb == CodeCache::find_blob(pc()),"Must be the same");
+  if (is_ricochet_frame())    return sender_for_ricochet_frame(map);
 
   if (_cb != NULL) {
     return sender_for_compiled_frame(map);
--- a/src/cpu/x86/vm/frame_x86.hpp	Fri May 06 12:12:29 2011 -0700
+++ b/src/cpu/x86/vm/frame_x86.hpp	Fri May 06 16:33:13 2011 -0700
@@ -164,6 +164,7 @@
   // original sp we use that convention.
 
   intptr_t*     _unextended_sp;
+  void adjust_unextended_sp();
 
   intptr_t* ptr_at_addr(int offset) const {
     return (intptr_t*) addr_at(offset);
@@ -197,6 +198,9 @@
   // expression stack tos if we are nested in a java call
   intptr_t* interpreter_frame_last_sp() const;
 
+  // helper to update a map with callee-saved RBP
+  static void update_map_with_saved_link(RegisterMap* map, intptr_t** link_addr);
+
 #ifndef CC_INTERP
   // deoptimization support
   void interpreter_frame_set_last_sp(intptr_t* sp);
--- a/src/cpu/x86/vm/frame_x86.inline.hpp	Fri May 06 12:12:29 2011 -0700
+++ b/src/cpu/x86/vm/frame_x86.inline.hpp	Fri May 06 16:33:13 2011 -0700
@@ -62,6 +62,7 @@
   _pc = pc;
   assert(pc != NULL, "no pc?");
   _cb = CodeCache::find_blob(pc);
+  adjust_unextended_sp();
 
   address original_pc = nmethod::get_deopt_original_pc(this);
   if (original_pc != NULL) {
--- a/src/cpu/x86/vm/interpreter_x86.hpp	Fri May 06 12:12:29 2011 -0700
+++ b/src/cpu/x86/vm/interpreter_x86.hpp	Fri May 06 16:33:13 2011 -0700
@@ -26,7 +26,9 @@
 #define CPU_X86_VM_INTERPRETER_X86_HPP
 
  public:
-  static Address::ScaleFactor stackElementScale() { return Address::times_4; }
+  static Address::ScaleFactor stackElementScale() {
+    return NOT_LP64(Address::times_4) LP64_ONLY(Address::times_8);
+  }
 
   // Offset from rsp (which points to the last stack element)
   static int expr_offset_in_bytes(int i) { return stackElementSize * i; }
--- a/src/cpu/x86/vm/methodHandles_x86.cpp	Fri May 06 12:12:29 2011 -0700
+++ b/src/cpu/x86/vm/methodHandles_x86.cpp	Fri May 06 16:33:13 2011 -0700
@@ -69,23 +69,475 @@
   return me;
 }
 
+// stack walking support
+
+frame MethodHandles::ricochet_frame_sender(const frame& fr, RegisterMap *map) {
+  RicochetFrame* f = RicochetFrame::from_frame(fr);
+  if (map->update_map())
+    frame::update_map_with_saved_link(map, &f->_sender_link);
+  return frame(f->extended_sender_sp(), f->exact_sender_sp(), f->sender_link(), f->sender_pc());
+}
+
+void MethodHandles::ricochet_frame_oops_do(const frame& fr, OopClosure* blk, const RegisterMap* reg_map) {
+  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;
+  intptr_t* loc = &base[slot_num -= 1];
+  //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");
+    loc = &base[slot_num -= type2size[ptype]];
+    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");
+}
+
+oop MethodHandles::RicochetFrame::compute_saved_args_layout(bool read_cache, bool write_cache) {
+  oop cookie = NULL;
+  if (read_cache) {
+    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);
+  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* frame_size_in_words,
+                                                          int* bounce_offset,
+                                                          int* exception_offset) {
+  (*frame_size_in_words) = RicochetFrame::frame_size_in_bytes() / wordSize;
+
+  address start = __ pc();
+
 #ifdef ASSERT
-static void verify_argslot(MacroAssembler* _masm, Register argslot_reg,
-                           const char* error_message) {
+  __ hlt(); __ hlt(); __ hlt();
+  // here's a hint of something special:
+  __ push(MAGIC_NUMBER_1);
+  __ push(MAGIC_NUMBER_2);
+#endif //ASSERT
+  __ hlt();  // not reached
+
+  // A return PC has just been popped from the stack.
+  // Return values are in registers.
+  // The ebp points into the RicochetFrame, which 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(frame_address(continuation_offset_in_bytes()));
+  __ hlt();
+  DEBUG_ONLY(__ push(MAGIC_NUMBER_2));
+
+  (*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:
+  //   rax: exception
+  //   rdx: return address/pc that threw exception (ignored, always equal to bounce addr)
+  __ verify_oop(rax);
+
+  // no need to empty_FPU_stack or reinit_heapbase, since caller frame will do the same if needed
+
+  // Take down the frame.
+
+  // Cf. InterpreterMacroAssembler::remove_activation.
+  leave_ricochet_frame(_masm, /*rcx_recv=*/ noreg,
+                       saved_last_sp_register(),
+                       /*sender_pc_reg=*/ rdx);
+
+  // In between activations - previous activation type unknown yet
+  // compute continuation point - the continuation point expects the
+  // following registers set up:
+  //
+  // rax: exception
+  // rdx: return address/pc that threw exception
+  // rsp: expression stack of caller
+  // rbp: ebp of caller
+  __ push(rax);                                  // save exception
+  __ push(rdx);                                  // save return address
+  Register thread_reg = LP64_ONLY(r15_thread) NOT_LP64(rdi);
+  NOT_LP64(__ get_thread(thread_reg));
+  __ call_VM_leaf(CAST_FROM_FN_PTR(address,
+                                   SharedRuntime::exception_handler_for_return_address),
+                  thread_reg, rdx);
+  __ mov(rbx, rax);                              // save exception handler
+  __ pop(rdx);                                   // restore return address
+  __ pop(rax);                                   // restore exception
+  __ jmp(rbx);                                   // jump to exception
+                                                 // handler of caller
+}
+
+void MethodHandles::RicochetFrame::enter_ricochet_frame(MacroAssembler* _masm,
+                                                        Register rcx_recv,
+                                                        Register rax_argv,
+                                                        address return_handler,
+                                                        Register rbx_temp) {
+  const Register saved_last_sp = saved_last_sp_register();
+  Address rcx_mh_vmtarget(    rcx_recv, java_lang_invoke_MethodHandle::vmtarget_offset_in_bytes() );
+  Address rcx_amh_conversion( rcx_recv, java_lang_invoke_AdapterMethodHandle::conversion_offset_in_bytes() );
+
+  // Push the RicochetFrame a word at a time.
+  // This creates something similar to an interpreter frame.
+  // Cf. TemplateInterpreterGenerator::generate_fixed_frame.
+  BLOCK_COMMENT("push RicochetFrame {");
+  DEBUG_ONLY(int rfo = (int) sizeof(RicochetFrame));
+  assert((rfo -= wordSize) == RicochetFrame::sender_pc_offset_in_bytes(), "");
+#define RF_FIELD(push_value, name)                                      \
+  { push_value;                                                         \
+    assert((rfo -= wordSize) == RicochetFrame::name##_offset_in_bytes(), ""); }
+  RF_FIELD(__ push(rbp),                   sender_link);
+  RF_FIELD(__ push(saved_last_sp),         exact_sender_sp);  // rsi/r13
+  RF_FIELD(__ pushptr(rcx_amh_conversion), conversion);
+  RF_FIELD(__ push(rax_argv),              saved_args_base);   // can be updated if args are shifted
+  RF_FIELD(__ push((int32_t) NULL_WORD),   saved_args_layout); // cache for GC layout cookie
+  if (UseCompressedOops) {
+    __ load_heap_oop(rbx_temp, rcx_mh_vmtarget);
+    RF_FIELD(__ push(rbx_temp),            saved_target);
+  } else {
+    RF_FIELD(__ pushptr(rcx_mh_vmtarget),  saved_target);
+  }
+  __ lea(rbx_temp, ExternalAddress(return_handler));
+  RF_FIELD(__ push(rbx_temp),              continuation);
+#undef RF_FIELD
+  assert(rfo == 0, "fully initialized the RicochetFrame");
+  // compute new frame pointer:
+  __ lea(rbp, Address(rsp, RicochetFrame::sender_link_offset_in_bytes()));
+  // Push guard word #1 in debug mode.
+  DEBUG_ONLY(__ push((int32_t) RicochetFrame::MAGIC_NUMBER_1));
+  // For debugging, leave behind an indication of which stub built this frame.
+  DEBUG_ONLY({ Label L; __ call(L, relocInfo::none); __ bind(L); });
+  BLOCK_COMMENT("} RicochetFrame");
+}
+
+void MethodHandles::RicochetFrame::leave_ricochet_frame(MacroAssembler* _masm,
+                                                        Register rcx_recv,
+                                                        Register new_sp_reg,
+                                                        Register sender_pc_reg) {
+  assert_different_registers(rcx_recv, new_sp_reg, sender_pc_reg);
+  const Register saved_last_sp = saved_last_sp_register();
+  // Take down the frame.
+  // Cf. InterpreterMacroAssembler::remove_activation.
+  BLOCK_COMMENT("end_ricochet_frame {");
+  // TO DO: If (exact_sender_sp - extended_sender_sp) > THRESH, compact the frame down.
+  // This will keep stack in bounds even with unlimited tailcalls, each with an adapter.
+  if (rcx_recv->is_valid())
+    __ movptr(rcx_recv,    RicochetFrame::frame_address(RicochetFrame::saved_target_offset_in_bytes()));
+  __ movptr(sender_pc_reg, RicochetFrame::frame_address(RicochetFrame::sender_pc_offset_in_bytes()));
+  __ movptr(saved_last_sp, RicochetFrame::frame_address(RicochetFrame::exact_sender_sp_offset_in_bytes()));
+  __ movptr(rbp,           RicochetFrame::frame_address(RicochetFrame::sender_link_offset_in_bytes()));
+  __ mov(rsp, new_sp_reg);
+  BLOCK_COMMENT("} end_ricochet_frame");
+}
+
+// Emit code to verify that RBP is pointing at a valid ricochet frame.
+#ifdef ASSERT
+enum {
+  ARG_LIMIT = 255, SLOP = 4,
+  // 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 | RF | magic | handler | magic | recursive args |
+  // Check various invariants.
+  verify_offsets();
+
+  Register rdi_temp = rdi;
+  Register rcx_temp = rcx;
+  { __ push(rdi_temp); __ push(rcx_temp); }
+#define UNPUSH_TEMPS \
+  { __ pop(rcx_temp);  __ pop(rdi_temp); }
+
+  Address magic_number_1_addr  = RicochetFrame::frame_address(RicochetFrame::magic_number_1_offset_in_bytes());
+  Address magic_number_2_addr  = RicochetFrame::frame_address(RicochetFrame::magic_number_2_offset_in_bytes());
+  Address continuation_addr    = RicochetFrame::frame_address(RicochetFrame::continuation_offset_in_bytes());
+  Address conversion_addr      = RicochetFrame::frame_address(RicochetFrame::conversion_offset_in_bytes());
+  Address saved_args_base_addr = RicochetFrame::frame_address(RicochetFrame::saved_args_base_offset_in_bytes());
+
+  Label L_bad, L_ok;
+  BLOCK_COMMENT("verify_clean {");
+  // Magic numbers must check out:
+  __ cmpptr(magic_number_1_addr, (int32_t) MAGIC_NUMBER_1);
+  __ jcc(Assembler::notEqual, L_bad);
+  __ cmpptr(magic_number_2_addr, (int32_t) MAGIC_NUMBER_2);
+  __ jcc(Assembler::notEqual, L_bad);
+
+  // Arguments pointer must look reasonable:
+  __ movptr(rcx_temp, saved_args_base_addr);
+  __ cmpptr(rcx_temp, rbp);
+  __ jcc(Assembler::below, L_bad);
+  __ subptr(rcx_temp, UNREASONABLE_STACK_MOVE * Interpreter::stackElementSize);
+  __ cmpptr(rcx_temp, rbp);
+  __ jcc(Assembler::above, L_bad);
+
+  load_conversion_dest_type(_masm, rdi_temp, conversion_addr);
+  __ cmpl(rdi_temp, T_VOID);
+  __ jcc(Assembler::equal, L_ok);
+  __ movptr(rcx_temp, saved_args_base_addr);
+  load_conversion_vminfo(_masm, rdi_temp, conversion_addr);
+  __ cmpptr(Address(rcx_temp, rdi_temp, Interpreter::stackElementScale()),
+            (int32_t) RETURN_VALUE_PLACEHOLDER);
+  __ jcc(Assembler::equal, L_ok);
+  __ BIND(L_bad);
+  UNPUSH_TEMPS;
+  __ stop("damaged ricochet frame");
+  __ BIND(L_ok);
+  UNPUSH_TEMPS;
+  BLOCK_COMMENT("} verify_clean");
+
+#undef UNPUSH_TEMPS
+
+}
+#endif //ASSERT
+
+void MethodHandles::load_klass_from_Class(MacroAssembler* _masm, Register klass_reg) {
+  if (VerifyMethodHandles)
+    verify_klass(_masm, klass_reg, SystemDictionaryHandles::Class_klass(),
+                 "AMH argument is a Class");
+  __ load_heap_oop(klass_reg, Address(klass_reg, java_lang_Class::klass_offset_in_bytes()));
+}
+
+void MethodHandles::load_conversion_vminfo(MacroAssembler* _masm, Register reg, Address conversion_field_addr) {
+  int bits   = BitsPerByte;
+  int offset = (CONV_VMINFO_SHIFT / bits);
+  int shift  = (CONV_VMINFO_SHIFT % bits);
+  __ load_unsigned_byte(reg, conversion_field_addr.plus_disp(offset));
+  assert(CONV_VMINFO_MASK == right_n_bits(bits - shift), "else change type of previous load");
+  assert(shift == 0, "no shift needed");
+}
+
+void MethodHandles::load_conversion_dest_type(MacroAssembler* _masm, Register reg, Address conversion_field_addr) {
+  int bits   = BitsPerByte;
+  int offset = (CONV_DEST_TYPE_SHIFT / bits);
+  int shift  = (CONV_DEST_TYPE_SHIFT % bits);
+  __ load_unsigned_byte(reg, conversion_field_addr.plus_disp(offset));
+  assert(CONV_TYPE_MASK == right_n_bits(bits - shift), "else change type of previous load");
+  __ shrl(reg, shift);
+  DEBUG_ONLY(int conv_type_bits = (int) exact_log2(CONV_TYPE_MASK+1));
+  assert((shift + conv_type_bits) == bits, "left justified in byte");
+}
+
+void MethodHandles::load_stack_move(MacroAssembler* _masm,
+                                    Register rdi_stack_move,
+                                    Register rcx_amh,
+                                    bool might_be_negative) {
+  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);
+#ifdef _LP64
+  if (might_be_negative) {
+    // clean high bits of stack motion register (was loaded as an int)
+    __ movslq(rdi_stack_move, rdi_stack_move);
+  }
+#endif //_LP64
+  if (VerifyMethodHandles) {
+    Label L_ok, L_bad;
+    int32_t stack_move_limit = 0x4000;  // extra-large
+    __ cmpptr(rdi_stack_move, stack_move_limit);
+    __ jcc(Assembler::greaterEqual, L_bad);
+    __ cmpptr(rdi_stack_move, -stack_move_limit);
+    __ jcc(Assembler::greater, L_ok);
+    __ bind(L_bad);
+    __ stop("load_stack_move of garbage value");
+    __ BIND(L_ok);
+  }
+}
+
+#ifndef PRODUCT
+void MethodHandles::RicochetFrame::verify_offsets() {
+  // Check compatibility of this struct with the more generally used offsets of class frame:
+  int ebp_off = sender_link_offset_in_bytes();  // offset from struct base to local rbp value
+  assert(ebp_off + wordSize*frame::interpreter_frame_method_offset      == saved_args_base_offset_in_bytes(), "");
+  assert(ebp_off + wordSize*frame::interpreter_frame_last_sp_offset     == conversion_offset_in_bytes(), "");
+  assert(ebp_off + wordSize*frame::interpreter_frame_sender_sp_offset   == exact_sender_sp_offset_in_bytes(), "");
+  // These last two have to be exact:
+  assert(ebp_off + wordSize*frame::link_offset                          == sender_link_offset_in_bytes(), "");
+  assert(ebp_off + wordSize*frame::return_addr_offset                   == sender_pc_offset_in_bytes(), "");
+}
+
+void MethodHandles::RicochetFrame::verify() const {
+  verify_offsets();
+  assert(magic_number_1() == MAGIC_NUMBER_1, "");
+  assert(magic_number_2() == MAGIC_NUMBER_2, "");
+  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, "");
+  }
+}
+#endif //PRODUCT
+
+#ifdef ASSERT
+void MethodHandles::verify_argslot(MacroAssembler* _masm,
+                                   Register argslot_reg,
+                                   const char* error_message) {
   // Verify that argslot lies within (rsp, rbp].
   Label L_ok, L_bad;
-  BLOCK_COMMENT("{ verify_argslot");
+  BLOCK_COMMENT("verify_argslot {");
   __ cmpptr(argslot_reg, rbp);
   __ jccb(Assembler::above, L_bad);
   __ cmpptr(rsp, argslot_reg);
   __ jccb(Assembler::below, L_ok);
   __ bind(L_bad);
   __ stop(error_message);
-  __ bind(L_ok);
+  __ BIND(L_ok);
   BLOCK_COMMENT("} verify_argslot");
 }
-#endif
 
+void MethodHandles::verify_argslots(MacroAssembler* _masm,
+                                    RegisterOrConstant arg_slots,
+                                    Register arg_slot_base_reg,
+                                    bool negate_argslots,
+                                    const char* error_message) {
+  // Verify that [argslot..argslot+size) lies within (rsp, rbp).
+  Label L_ok, L_bad;
+  Register rdi_temp = rdi;
+  BLOCK_COMMENT("verify_argslots {");
+  __ push(rdi_temp);
+  if (negate_argslots) {
+    if (arg_slots.is_constant()) {
+      arg_slots = -1 * arg_slots.as_constant();
+    } else {
+      __ movptr(rdi_temp, arg_slots);
+      __ negptr(rdi_temp);
+      arg_slots = rdi_temp;
+    }
+  }
+  __ lea(rdi_temp, Address(arg_slot_base_reg, arg_slots, Interpreter::stackElementScale()));
+  __ cmpptr(rdi_temp, rbp);
+  __ pop(rdi_temp);
+  __ jcc(Assembler::above, L_bad);
+  __ cmpptr(rsp, arg_slot_base_reg);
+  __ jcc(Assembler::below, L_ok);
+  __ 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) {
+  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 {");
+    // testl(arg_slots.as_register(), -stack_move_unit() - 1);  // no need
+    // jcc(Assembler::notZero, L_bad);
+    __ cmpptr(arg_slots.as_register(), (int32_t) NULL_WORD);
+    if (direction > 0) {
+      __ jcc(allow_zero ? Assembler::less : Assembler::lessEqual, L_bad);
+      __ cmpptr(arg_slots.as_register(), (int32_t) UNREASONABLE_STACK_MOVE);
+      __ jcc(Assembler::less, L_ok);
+    } else {
+      __ jcc(allow_zero ? Assembler::greater : Assembler::greaterEqual, L_bad);
+      __ cmpptr(arg_slots.as_register(), (int32_t) -UNREASONABLE_STACK_MOVE);
+      __ jcc(Assembler::greater, L_ok);
+    }
+    __ 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, KlassHandle klass,
+                                 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");
+  Register temp = rdi;
+  Label L_ok, L_bad;
+  BLOCK_COMMENT("verify_klass {");
+  __ verify_oop(obj);
+  __ testptr(obj, obj);
+  __ jcc(Assembler::zero, L_bad);
+  __ push(temp);
+  __ load_klass(temp, obj);
+  __ cmpptr(temp, ExternalAddress((address) klass_addr));
+  __ jcc(Assembler::equal, L_ok);
+  intptr_t super_check_offset = klass->super_check_offset();
+  __ movptr(temp, Address(temp, super_check_offset));
+  __ cmpptr(temp, ExternalAddress((address) klass_addr));
+  __ jcc(Assembler::equal, L_ok);
+  __ pop(temp);
+  __ bind(L_bad);
+  __ stop(error_message);
+  __ BIND(L_ok);
+  __ pop(temp);
+  BLOCK_COMMENT("} verify_klass");
+}
+#endif //ASSERT
 
 // Code generation
 address MethodHandles::generate_method_handle_interpreter_entry(MacroAssembler* _masm) {
@@ -116,6 +568,9 @@
   address entry_point = __ pc();
 
   // fetch the MethodType from the method handle into rax (the 'check' register)
+  // 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 = rbx_method;
     for (jint* pchase = methodOopDesc::method_type_offsets_chain(); (*pchase) != -1; pchase++) {
@@ -128,17 +583,23 @@
   __ load_heap_oop(rdx_temp, Address(rax_mtype, __ delayed_value(java_lang_invoke_MethodType::form_offset_in_bytes, rdi_temp)));
   Register rdx_vmslots = rdx_temp;
   __ movl(rdx_vmslots, Address(rdx_temp, __ delayed_value(java_lang_invoke_MethodTypeForm::vmslots_offset_in_bytes, rdi_temp)));
-  __ movptr(rcx_recv, __ argument_address(rdx_vmslots));
+  Address mh_receiver_slot_addr = __ argument_address(rdx_vmslots);
+  __ movptr(rcx_recv, mh_receiver_slot_addr);
 
   trace_method_handle(_masm, "invokeExact");
 
   __ check_method_handle_type(rax_mtype, rcx_recv, rdi_temp, wrong_method_type);
+
+  // Nobody uses the MH receiver slot after this.  Make sure.
+  DEBUG_ONLY(__ movptr(mh_receiver_slot_addr, (int32_t)0x999999));
+
   __ jump_to_method_handle_entry(rcx_recv, rdi_temp);
 
   // for invokeGeneric (only), apply argument and result conversions on the fly
   __ bind(invoke_generic_slow_path);
 #ifdef ASSERT
-  { Label L;
+  if (VerifyMethodHandles) {
+    Label L;
     __ cmpb(Address(rbx_method, methodOopDesc::intrinsic_id_offset_in_bytes()), (int) vmIntrinsics::_invokeGeneric);
     __ jcc(Assembler::equal, L);
     __ stop("bad methodOop::intrinsic_id");
@@ -150,7 +611,7 @@
   // make room on the stack for another pointer:
   Register rcx_argslot = rcx_recv;
   __ lea(rcx_argslot, __ argument_address(rdx_vmslots, 1));
-  insert_arg_slots(_masm, 2 * stack_move_unit(), _INSERT_REF_MASK,
+  insert_arg_slots(_masm, 2 * stack_move_unit(),
                    rcx_argslot, rbx_temp, rdx_temp);
 
   // load up an adapter from the calling type (Java weaves this)
@@ -185,40 +646,28 @@
   return entry_point;
 }
 
+// Workaround for C++ overloading nastiness on '0' for RegisterOrConstant.
+static RegisterOrConstant constant(int value) {
+  return RegisterOrConstant(value);
+}
+
 // 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
+// rax_argslot is decremented to point to the new (shifted) location of the argslot
+// But, rdx_temp ends up holding the original value of rax_argslot.
 void MethodHandles::insert_arg_slots(MacroAssembler* _masm,
                                      RegisterOrConstant arg_slots,
-                                     int arg_mask,
                                      Register rax_argslot,
-                                     Register rbx_temp, Register rdx_temp, Register temp3_reg) {
-  assert(temp3_reg == noreg, "temp3 not required");
+                                     Register rbx_temp, Register rdx_temp) {
+  // allow constant zero
+  if (arg_slots.is_constant() && arg_slots.as_constant() == 0)
+    return;
   assert_different_registers(rax_argslot, rbx_temp, rdx_temp,
                              (!arg_slots.is_register() ? rsp : arg_slots.as_register()));
-
-#ifdef ASSERT
-  verify_argslot(_masm, rax_argslot, "insertion point must fall within current frame");
-  if (arg_slots.is_register()) {
-    Label L_ok, L_bad;
-    __ cmpptr(arg_slots.as_register(), (int32_t) NULL_WORD);
-    __ jccb(Assembler::greater, L_bad);
-    __ testl(arg_slots.as_register(), -stack_move_unit() - 1);
-    __ jccb(Assembler::zero, L_ok);
-    __ bind(L_bad);
-    __ stop("assert arg_slots <= 0 and clear low bits");
-    __ bind(L_ok);
-  } else {
-    assert(arg_slots.as_constant() <= 0, "");
-    assert(arg_slots.as_constant() % -stack_move_unit() == 0, "");
-  }
-#endif //ASSERT
-
-#ifdef _LP64
-  if (arg_slots.is_register()) {
-    // clean high bits of stack motion register (was loaded as an int)
-    __ movslq(arg_slots.as_register(), arg_slots.as_register());
-  }
-#endif
+  if (VerifyMethodHandles)
+    verify_argslot(_masm, rax_argslot, "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 rax_argslot.
@@ -230,59 +679,39 @@
   //   argslot -= size;
   BLOCK_COMMENT("insert_arg_slots {");
   __ mov(rdx_temp, rsp);                        // source pointer for copy
-  __ lea(rsp, Address(rsp, arg_slots, Address::times_ptr));
+  __ lea(rsp, Address(rsp, arg_slots, Interpreter::stackElementScale()));
   {
     Label loop;
     __ BIND(loop);
     // pull one word down each time through the loop
     __ movptr(rbx_temp, Address(rdx_temp, 0));
-    __ movptr(Address(rdx_temp, arg_slots, Address::times_ptr), rbx_temp);
+    __ movptr(Address(rdx_temp, arg_slots, Interpreter::stackElementScale()), rbx_temp);
     __ addptr(rdx_temp, wordSize);
     __ cmpptr(rdx_temp, rax_argslot);
-    __ jccb(Assembler::less, loop);
+    __ jcc(Assembler::less, loop);
   }
 
   // Now move the argslot down, to point to the opened-up space.
-  __ lea(rax_argslot, Address(rax_argslot, arg_slots, Address::times_ptr));
+  __ lea(rax_argslot, Address(rax_argslot, arg_slots, Interpreter::stackElementScale()));
   BLOCK_COMMENT("} insert_arg_slots");
 }
 
 // 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 rax_argslot,
-                                     Register rbx_temp, Register rdx_temp, Register temp3_reg) {
-  assert(temp3_reg == noreg, "temp3 not required");
+                                     RegisterOrConstant arg_slots,
+                                     Register rax_argslot,
+                                     Register rbx_temp, Register rdx_temp) {
+  // allow constant zero
+  if (arg_slots.is_constant() && arg_slots.as_constant() == 0)
+    return;
   assert_different_registers(rax_argslot, rbx_temp, rdx_temp,
                              (!arg_slots.is_register() ? rsp : arg_slots.as_register()));
-
-#ifdef ASSERT
-  // Verify that [argslot..argslot+size) lies within (rsp, rbp).
-  __ lea(rbx_temp, Address(rax_argslot, arg_slots, Address::times_ptr));
-  verify_argslot(_masm, rbx_temp, "deleted argument(s) must fall within current frame");
-  if (arg_slots.is_register()) {
-    Label L_ok, L_bad;
-    __ cmpptr(arg_slots.as_register(), (int32_t) NULL_WORD);
-    __ jccb(Assembler::less, L_bad);
-    __ testl(arg_slots.as_register(), -stack_move_unit() - 1);
-    __ jccb(Assembler::zero, L_ok);
-    __ bind(L_bad);
-    __ stop("assert arg_slots >= 0 and clear low bits");
-    __ bind(L_ok);
-  } else {
-    assert(arg_slots.as_constant() >= 0, "");
-    assert(arg_slots.as_constant() % -stack_move_unit() == 0, "");
-  }
-#endif //ASSERT
-
-#ifdef _LP64
-  if (false) {                  // not needed, since register is positive
-    // clean high bits of stack motion register (was loaded as an int)
-    if (arg_slots.is_register())
-      __ movslq(arg_slots.as_register(), arg_slots.as_register());
-  }
-#endif
+  if (VerifyMethodHandles)
+    verify_argslots(_masm, arg_slots, rax_argslot, false,
+                    "deleted argument(s) must fall within current frame");
+  if (VerifyMethodHandles)
+    verify_stack_move(_masm, arg_slots, +1);
 
   BLOCK_COMMENT("remove_arg_slots {");
   // Pull up everything shallower than rax_argslot.
@@ -299,54 +728,332 @@
     __ BIND(loop);
     // pull one word up each time through the loop
     __ movptr(rbx_temp, Address(rdx_temp, 0));
-    __ movptr(Address(rdx_temp, arg_slots, Address::times_ptr), rbx_temp);
+    __ movptr(Address(rdx_temp, arg_slots, Interpreter::stackElementScale()), rbx_temp);
     __ addptr(rdx_temp, -wordSize);
     __ cmpptr(rdx_temp, rsp);
-    __ jccb(Assembler::greaterEqual, loop);
+    __ jcc(Assembler::greaterEqual, loop);
   }
 
   // Now move the argslot up, to point to the just-copied block.
-  __ lea(rsp, Address(rsp, arg_slots, Address::times_ptr));
+  __ lea(rsp, Address(rsp, arg_slots, Interpreter::stackElementScale()));
   // And adjust the argslot address to point at the deletion point.
-  __ lea(rax_argslot, Address(rax_argslot, arg_slots, Address::times_ptr));
+  __ lea(rax_argslot, Address(rax_argslot, arg_slots, Interpreter::stackElementScale()));
   BLOCK_COMMENT("} remove_arg_slots");
 }
 
+// Helper to copy argument slots to the top of the stack.
+// The sequence starts with rax_argslot 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 rax_argslot.
+void MethodHandles::push_arg_slots(MacroAssembler* _masm,
+                                   Register rax_argslot,
+                                   RegisterOrConstant slot_count,
+                                   int skip_words_count,
+                                   Register rbx_temp, Register rdx_temp) {
+  assert_different_registers(rax_argslot, rbx_temp, rdx_temp,
+                             (!slot_count.is_register() ? rbp : slot_count.as_register()),
+                             rsp);
+  assert(Interpreter::stackElementSize == wordSize, "else change this code");
+
+  if (VerifyMethodHandles)
+    verify_stack_move(_masm, slot_count, 0);
+
+  // allow constant zero
+  if (slot_count.is_constant() && slot_count.as_constant() == 0)
+    return;
+
+  BLOCK_COMMENT("push_arg_slots {");
+
+  Register rbx_top = rbx_temp;
+
+  // There is at most 1 word to carry down with the TOS.
+  switch (skip_words_count) {
+  case 1: __ pop(rdx_temp); break;
+  case 0:                   break;
+  default: ShouldNotReachHere();
+  }
+
+  if (slot_count.is_constant()) {
+    for (int i = slot_count.as_constant() - 1; i >= 0; i--) {
+      __ pushptr(Address(rax_argslot, i * wordSize));
+    }
+  } else {
+    Label L_plural, L_loop, L_break;
+    // Emit code to dynamically check for the common cases, zero and one slot.
+    __ cmpl(slot_count.as_register(), (int32_t) 1);
+    __ jccb(Assembler::greater, L_plural);
+    __ jccb(Assembler::less, L_break);
+    __ pushptr(Address(rax_argslot, 0));
+    __ jmpb(L_break);
+    __ BIND(L_plural);
+
+    // Loop for 2 or more:
+    //   rbx = &rax[slot_count]
+    //   while (rbx > rax)  *(--rsp) = *(--rbx)
+    __ lea(rbx_top, Address(rax_argslot, slot_count, Address::times_ptr));
+    __ BIND(L_loop);
+    __ subptr(rbx_top, wordSize);
+    __ pushptr(Address(rbx_top, 0));
+    __ cmpptr(rbx_top, rax_argslot);
+    __ jcc(Assembler::above, L_loop);
+    __ bind(L_break);
+  }
+  switch (skip_words_count) {
+  case 1: __ push(rdx_temp); break;
+  case 0:                    break;
+  default: ShouldNotReachHere();
+  }
+  BLOCK_COMMENT("} push_arg_slots");
+}
+
+// in-place movement; no change to rsp
+// blows rax_temp, rdx_temp
+void MethodHandles::move_arg_slots_up(MacroAssembler* _masm,
+                                      Register rbx_bottom,  // invariant
+                                      Address  top_addr,     // can use rax_temp
+                                      RegisterOrConstant positive_distance_in_slots,
+                                      Register rax_temp, Register rdx_temp) {
+  BLOCK_COMMENT("move_arg_slots_up {");
+  assert_different_registers(rbx_bottom,
+                             rax_temp, rdx_temp,
+                             positive_distance_in_slots.register_or_noreg());
+  Label L_loop, L_break;
+  Register rax_top = rax_temp;
+  if (!top_addr.is_same_address(Address(rax_top, 0)))
+    __ lea(rax_top, top_addr);
+  // 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()) {
+      __ cmpptr(positive_distance_in_slots.as_register(), (int32_t) 0);
+      __ jcc(Assembler::lessEqual, L_bad);
+    }
+    __ cmpptr(rbx_bottom, rax_top);
+    __ jcc(Assembler::below, L_ok);
+    __ bind(L_bad);
+    __ stop("valid bounds (copy up)");
+    __ BIND(L_ok);
+  }
+#endif
+  __ cmpptr(rbx_bottom, rax_top);
+  __ jccb(Assembler::aboveEqual, L_break);
+  // work rax down to rbx, copying contiguous data upwards
+  // In pseudo-code:
+  //   [rbx, rax) = &[bottom, top)
+  //   while (--rax >= rbx) *(rax + distance) = *(rax + 0), rax--;
+  __ BIND(L_loop);
+  __ subptr(rax_top, wordSize);
+  __ movptr(rdx_temp, Address(rax_top, 0));
+  __ movptr(          Address(rax_top, positive_distance_in_slots, Address::times_ptr), rdx_temp);
+  __ cmpptr(rax_top, rbx_bottom);
+  __ jcc(Assembler::above, L_loop);
+  assert(Interpreter::stackElementSize == wordSize, "else change loop");
+  __ bind(L_break);
+  BLOCK_COMMENT("} move_arg_slots_up");
+}
+
+// in-place movement; no change to rsp
+// blows rax_temp, rdx_temp
+void MethodHandles::move_arg_slots_down(MacroAssembler* _masm,
+                                        Address  bottom_addr,  // can use rax_temp
+                                        Register rbx_top,      // invariant
+                                        RegisterOrConstant negative_distance_in_slots,
+                                        Register rax_temp, Register rdx_temp) {
+  BLOCK_COMMENT("move_arg_slots_down {");
+  assert_different_registers(rbx_top,
+                             negative_distance_in_slots.register_or_noreg(),
+                             rax_temp, rdx_temp);
+  Label L_loop, L_break;
+  Register rax_bottom = rax_temp;
+  if (!bottom_addr.is_same_address(Address(rax_bottom, 0)))
+    __ lea(rax_bottom, bottom_addr);
+  // 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()) {
+      __ cmpptr(negative_distance_in_slots.as_register(), (int32_t) 0);
+      __ jcc(Assembler::greaterEqual, L_bad);
+    }
+    __ cmpptr(rax_bottom, rbx_top);
+    __ jcc(Assembler::below, L_ok);
+    __ bind(L_bad);
+    __ stop("valid bounds (copy down)");
+    __ BIND(L_ok);
+  }
+#endif
+  __ cmpptr(rax_bottom, rbx_top);
+  __ jccb(Assembler::aboveEqual, L_break);
+  // work rax up to rbx, copying contiguous data downwards
+  // In pseudo-code:
+  //   [rax, rbx) = &[bottom, top)
+  //   while (rax < rbx) *(rax - distance) = *(rax + 0), rax++;
+  __ BIND(L_loop);
+  __ movptr(rdx_temp, Address(rax_bottom, 0));
+  __ movptr(          Address(rax_bottom, negative_distance_in_slots, Address::times_ptr), rdx_temp);
+  __ addptr(rax_bottom, wordSize);
+  __ cmpptr(rax_bottom, rbx_top);
+  __ jcc(Assembler::below, L_loop);
+  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 slot_dest, Address value_src,
+                                   Register rbx_temp, Register rdx_temp) {
+  BLOCK_COMMENT(!is_element ? "move_typed_arg {" : "move_typed_arg { (array element)");
+  if (type == T_OBJECT || type == T_ARRAY) {
+    __ load_heap_oop(rbx_temp, value_src);
+    __ movptr(slot_dest, rbx_temp);
+  } else if (type != T_VOID) {
+    int  arg_size      = type2aelembytes(type);
+    bool arg_is_signed = is_signed_subword_type(type);
+    int  slot_size     = (arg_size > wordSize) ? arg_size : wordSize;
+    __ load_sized_value(  rdx_temp,  value_src, arg_size, arg_is_signed, rbx_temp);
+    __ store_sized_value( slot_dest, rdx_temp,  slot_size,               rbx_temp);
+  }
+  BLOCK_COMMENT("} move_typed_arg");
+}
+
+void MethodHandles::move_return_value(MacroAssembler* _masm, BasicType type,
+                                      Address return_slot) {
+  BLOCK_COMMENT("move_return_value {");
+  // Old versions of the JVM must clean the FPU stack after every return.
+#ifndef _LP64
+#ifdef COMPILER2
+  // The FPU stack is clean if UseSSE >= 2 but must be cleaned in other cases
+  if ((type == T_FLOAT && UseSSE < 1) || (type == T_DOUBLE && UseSSE < 2)) {
+    for (int i = 1; i < 8; i++) {
+        __ ffree(i);
+    }
+  } else if (UseSSE < 2) {
+    __ empty_FPU_stack();
+  }
+#endif //COMPILER2
+#endif //!_LP64
+
+  // 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) {
+    __ movptr(return_slot, rax);
+  } else if (type == T_INT || is_subword_type(type)) {
+    // write the whole word, even if only 32 bits is significant
+    __ movptr(return_slot, rax);
+  } else if (type == T_LONG) {
+    // store the value by parts
+    // Note: We assume longs are continguous (if misaligned) on the interpreter stack.
+    __ store_sized_value(return_slot, rax, BytesPerLong, rdx);
+  } else if (NOT_LP64((type == T_FLOAT  && UseSSE < 1) ||
+                      (type == T_DOUBLE && UseSSE < 2) ||)
+             false) {
+    // Use old x86 FPU registers:
+    if (type == T_FLOAT)
+      __ fstp_s(return_slot);
+    else
+      __ fstp_d(return_slot);
+  } else if (type == T_FLOAT) {
+    __ movflt(return_slot, xmm0);
+  } else if (type == T_DOUBLE) {
+    __ movdbl(return_slot, xmm0);
+  } 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,
+                              oop mh,
+                              intptr_t* saved_regs,
+                              intptr_t* entry_sp,
                               intptr_t* saved_sp,
-                              oop mh,
-                              intptr_t* sp) {
+                              intptr_t* saved_bp) {
   // called as a leaf from native code: do not block the JVM!
-  intptr_t* entry_sp = sp + LP64_ONLY(16) NOT_LP64(8);
-  tty->print_cr("MH %s mh="INTPTR_FORMAT" sp="INTPTR_FORMAT" saved_sp="INTPTR_FORMAT")",
-                adaptername, (intptr_t)mh, (intptr_t)entry_sp, saved_sp);
+  intptr_t* last_sp = (intptr_t*) saved_bp[frame::interpreter_frame_last_sp_offset];
+  intptr_t* base_sp = (intptr_t*) saved_bp[frame::interpreter_frame_monitor_block_top_offset];
+  tty->print_cr("MH %s mh="INTPTR_FORMAT" sp=("INTPTR_FORMAT"+"INTX_FORMAT") stack_size="INTX_FORMAT" bp="INTPTR_FORMAT,
+                adaptername, (intptr_t)mh, (intptr_t)entry_sp, (intptr_t)(saved_sp - entry_sp), (intptr_t)(base_sp - last_sp), (intptr_t)saved_bp);
+  if (last_sp != saved_sp && last_sp != NULL)
+    tty->print_cr("*** last_sp="INTPTR_FORMAT, (intptr_t)last_sp);
   if (Verbose) {
+    tty->print(" reg dump: ");
+    int saved_regs_count = (entry_sp-1) - saved_regs;
+    // 32 bit: rdi rsi rbp rsp; rbx rdx rcx (*) rax
+    int i;
+    for (i = 0; i <= saved_regs_count; i++) {
+      if (i > 0 && i % 4 == 0 && i != saved_regs_count) {
+        tty->cr();
+        tty->print("   + dump: ");
+      }
+      tty->print(" %d: "INTPTR_FORMAT, i, saved_regs[i]);
+    }
+    tty->cr();
+    int stack_dump_count = 16;
+    if (stack_dump_count < (int)(saved_bp + 2 - saved_sp))
+      stack_dump_count = (int)(saved_bp + 2 - saved_sp);
+    if (stack_dump_count > 64)  stack_dump_count = 48;
+    for (i = 0; i < stack_dump_count; i += 4) {
+      tty->print_cr(" dump at SP[%d] "INTPTR_FORMAT": "INTPTR_FORMAT" "INTPTR_FORMAT" "INTPTR_FORMAT" "INTPTR_FORMAT,
+                    i, (intptr_t) &entry_sp[i+0], entry_sp[i+0], entry_sp[i+1], entry_sp[i+2], entry_sp[i+3]);
+    }
     print_method_handle(mh);
   }
 }
+
+// The stub wraps the arguments in a struct on the stack to avoid
+// dealing with the different calling conventions for passing 6
+// arguments.
+struct MethodHandleStubArguments {
+  const char* adaptername;
+  oopDesc* mh;
+  intptr_t* saved_regs;
+  intptr_t* entry_sp;
+  intptr_t* saved_sp;
+  intptr_t* saved_bp;
+};
+void trace_method_handle_stub_wrapper(MethodHandleStubArguments* args) {
+  trace_method_handle_stub(args->adaptername,
+                           args->mh,
+                           args->saved_regs,
+                           args->entry_sp,
+                           args->saved_sp,
+                           args->saved_bp);
+}
+
 void MethodHandles::trace_method_handle(MacroAssembler* _masm, const char* adaptername) {
   if (!TraceMethodHandles)  return;
   BLOCK_COMMENT("trace_method_handle {");
+  __ push(rax);
+  __ lea(rax, Address(rsp, wordSize * NOT_LP64(6) LP64_ONLY(14))); // entry_sp  __ pusha();
   __ pusha();
-#ifdef _LP64
-  // Pass arguments carefully since the registers overlap with the calling convention.
+  __ mov(rbx, rsp);
+  __ enter();
+  // incoming state:
   // rcx: method handle
-  // r13: saved sp
-  __ mov(c_rarg2, rcx); // mh
-  __ mov(c_rarg1, r13); // saved sp
-  __ mov(c_rarg3, rsp); // sp
-  __ movptr(c_rarg0, (intptr_t) adaptername);
-  __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, trace_method_handle_stub), c_rarg0, c_rarg1, c_rarg2, c_rarg3);
-#else
-  // arguments:
-  // rcx: method handle
-  // rsi: saved sp
-  __ movptr(rbx, (intptr_t) adaptername);
-  __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, trace_method_handle_stub), rbx, rsi, rcx, rsp);
-#endif
+  // r13 or rsi: saved sp
+  // To avoid calling convention issues, build a record on the stack and pass the pointer to that instead.
+  __ push(rbp);               // saved_bp
+  __ push(rsi);               // saved_sp
+  __ push(rax);               // entry_sp
+  __ push(rbx);               // pusha saved_regs
+  __ push(rcx);               // mh
+  __ push(rcx);               // adaptername
+  __ movptr(Address(rsp, 0), (intptr_t) adaptername);
+  __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, trace_method_handle_stub_wrapper), rsp);
+  __ leave();
   __ popa();
+  __ pop(rax);
   BLOCK_COMMENT("} trace_method_handle");
 }
 #endif //PRODUCT
@@ -358,13 +1065,21 @@
          |(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 :
+           LP64_ONLY(FLAG_IS_DEFAULT(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.
 }
 
 //------------------------------------------------------------------------------
@@ -373,6 +1088,8 @@
 // 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():
   // - rbx: garbage temp (was MethodHandle.invoke methodOop, unused)
@@ -385,10 +1102,11 @@
   const Register rax_argslot = rax;
   const Register rbx_temp    = rbx;
   const Register rdx_temp    = rdx;
+  const Register rdi_temp    = rdi;
 
   // This guy is set up by prepare_to_jump_from_interpreted (from interpreted calls)
   // and gen_c2i_adapter (from compiled calls):
-  const Register saved_last_sp = LP64_ONLY(r13) NOT_LP64(rsi);
+  const Register saved_last_sp = saved_last_sp_register();
 
   // Argument registers for _raise_exception.
   // 32-bit: Pass first two oop/int args in registers ECX and EDX.
@@ -421,6 +1139,13 @@
     return;
   }
 
+#ifdef ASSERT
+  __ push((int32_t) 0xEEEEEEEE);
+  __ push((int32_t) (intptr_t) entry_name(ek));
+  LP64_ONLY(__ push((int32_t) high((intptr_t) entry_name(ek))));
+  __ push((int32_t) 0x33333333);
+#endif //ASSERT
+
   address interp_entry = __ pc();
 
   trace_method_handle(_masm, entry_name(ek));
@@ -536,7 +1261,6 @@
       __ load_klass(rax_klass, rcx_recv);
       __ verify_oop(rax_klass);
 
-      Register rdi_temp   = rdi;
       Register rbx_method = rbx_index;
 
       // get interface klass
@@ -572,16 +1296,14 @@
   case _bound_long_direct_mh:
     {
       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:
       __ movl(rax_argslot, rcx_bmh_vmargslot);
       __ lea(rax_argslot, __ argument_address(rax_argslot));
 
-      insert_arg_slots(_masm, arg_slots * stack_move_unit(), arg_mask, rax_argslot, rbx_temp, rdx_temp);
+      insert_arg_slots(_masm, arg_slots * stack_move_unit(), rax_argslot, rbx_temp, rdx_temp);
 
       // store bound argument into the new stack slot:
       __ load_heap_oop(rbx_temp, rcx_bmh_argument);
@@ -589,9 +1311,10 @@
         __ movptr(Address(rax_argslot, 0), rbx_temp);
       } else {
         Address prim_value_addr(rbx_temp, java_lang_boxing_object::value_offset_in_bytes(arg_type));
-        const int arg_size = type2aelembytes(arg_type);
-        __ load_sized_value(rdx_temp, prim_value_addr, arg_size, is_signed_subword_type(arg_type), rbx_temp);
-        __ store_sized_value(Address(rax_argslot, 0), rdx_temp, arg_size, rbx_temp);
+        move_typed_arg(_masm, arg_type, false,
+                       Address(rax_argslot, 0),
+                       prim_value_addr,
+                       rbx_temp, rdx_temp);
       }
 
       if (direct_to_method) {
@@ -628,7 +1351,7 @@
 
       // What class are we casting to?
       __ load_heap_oop(rbx_klass, rcx_amh_argument); // this is a Class object!
-      __ load_heap_oop(rbx_klass, Address(rbx_klass, java_lang_Class::klass_offset_in_bytes()));
+      load_klass_from_Class(_masm, rbx_klass);
 
       Label done;
       __ movptr(rdx_temp, vmarg);
@@ -663,6 +1386,7 @@
 
   case _adapter_prim_to_prim:
   case _adapter_ref_to_prim:
+  case _adapter_prim_to_ref:
     // handled completely by optimized cases
     __ stop("init_AdapterMethodHandle should not issue this");
     break;
@@ -714,8 +1438,7 @@
 
       // Do the requested conversion and store the value.
       Register rbx_vminfo = rbx_temp;
-      __ movl(rbx_vminfo, rcx_amh_conversion);
-      assert(CONV_VMINFO_SHIFT == 0, "preshifted");
+      load_conversion_vminfo(_masm, rbx_vminfo, rcx_amh_conversion);
 
       // get the new MH:
       __ load_heap_oop(rcx_recv, rcx_mh_vmtarget);
@@ -753,7 +1476,7 @@
 
       // on a little-endian machine we keep the first slot and add another after
       __ lea(rax_argslot, __ argument_address(rax_argslot, 1));
-      insert_arg_slots(_masm, stack_move_unit(), _INSERT_INT_MASK,
+      insert_arg_slots(_masm, stack_move_unit(),
                        rax_argslot, rbx_temp, rdx_temp);
       Address vmarg1(rax_argslot, -Interpreter::stackElementSize);
       Address vmarg2 = vmarg1.plus_disp(Interpreter::stackElementSize);
@@ -805,7 +1528,7 @@
       __ movl(rax_argslot, rcx_amh_vmargslot);
       __ lea(rax_argslot, __ argument_address(rax_argslot, 1));
       if (ek == _adapter_opt_f2d) {
-        insert_arg_slots(_masm, stack_move_unit(), _INSERT_INT_MASK,
+        insert_arg_slots(_masm, stack_move_unit(),
                          rax_argslot, rbx_temp, rdx_temp);
       }
       Address vmarg(rax_argslot, -Interpreter::stackElementSize);
@@ -840,10 +1563,6 @@
     }
     break;
 
-  case _adapter_prim_to_ref:
-    __ unimplemented(entry_name(ek)); // %%% FIXME: NYI
-    break;
-
   case _adapter_swap_args:
   case _adapter_rot_args:
     // handled completely by optimized cases
@@ -857,8 +1576,8 @@
   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
       __ movl(rax_argslot, rcx_amh_vmargslot);
@@ -866,83 +1585,69 @@
 
       // 'vminfo' is the second
       Register rbx_destslot = rbx_temp;
-      __ movl(rbx_destslot, rcx_amh_conversion);
-      assert(CONV_VMINFO_SHIFT == 0, "preshifted");
-      __ andl(rbx_destslot, CONV_VMINFO_MASK);
+      load_conversion_vminfo(_masm, rbx_destslot, rcx_amh_conversion);
       __ lea(rbx_destslot, __ argument_address(rbx_destslot));
-      DEBUG_ONLY(verify_argslot(_masm, rbx_destslot, "swap point must fall within current frame"));
+      if (VerifyMethodHandles)
+        verify_argslot(_masm, rbx_destslot, "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) {
-          __ movptr(rdx_temp, Address(rax_argslot , i));
-          __ push(rdx_temp);
-          __ movptr(rdx_temp, Address(rbx_destslot, i));
-          __ movptr(Address(rax_argslot, i), rdx_temp);
-          __ pop(rdx_temp);
-          __ movptr(Address(rbx_destslot, i), rdx_temp);
+        // simple swap
+        for (int i = 0; i < swap_slots; i++) {
+          __ movptr(rdi_temp, Address(rax_argslot,  i * wordSize));
+          __ movptr(rdx_temp, Address(rbx_destslot, i * wordSize));
+          __ movptr(Address(rax_argslot,  i * wordSize), rdx_temp);
+          __ movptr(Address(rbx_destslot, i * wordSize), rdi_temp);
         }
       } else {
-        // push the first chunk, which is going to get overwritten
-        for (int i = swap_bytes; (i -= wordSize) >= 0; ) {
-          __ movptr(rdx_temp, Address(rax_argslot, i));
-          __ push(rdx_temp);
+        // 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
+        for (int i = swap_slots - 1; i >= 0; i--) {
+          // handle one with rdi_temp instead of a push:
+          if (i == 0)  __ movptr(rdi_temp, Address(rax_argslot, i * wordSize));
+          else         __ pushptr(         Address(rax_argslot, i * wordSize));
         }
-
         if (rotate > 0) {
-          // rotate upward
-          __ subptr(rax_argslot, swap_bytes);
-#ifdef ASSERT
-          {
-            // Verify that argslot > destslot, by at least swap_bytes.
-            Label L_ok;
-            __ cmpptr(rax_argslot, rbx_destslot);
-            __ jccb(Assembler::aboveEqual, L_ok);
-            __ stop("source must be above destination (upward rotation)");
-            __ bind(L_ok);
-          }
-#endif
+          // 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:
           //   rax = src_addr - swap_bytes
           //   rbx = dest_addr
           //   while (rax >= rbx) *(rax + swap_bytes) = *(rax + 0), rax--;
-          Label loop;
-          __ bind(loop);
-          __ movptr(rdx_temp, Address(rax_argslot, 0));
-          __ movptr(Address(rax_argslot, swap_bytes), rdx_temp);
-          __ addptr(rax_argslot, -wordSize);
-          __ cmpptr(rax_argslot, rbx_destslot);
-          __ jccb(Assembler::aboveEqual, loop);
+          move_arg_slots_up(_masm,
+                            rbx_destslot,
+                            Address(rax_argslot, 0),
+                            swap_slots,
+                            rax_argslot, rdx_temp);
         } else {
-          __ addptr(rax_argslot, swap_bytes);
-#ifdef ASSERT
-          {
-            // Verify that argslot < destslot, by at least swap_bytes.
-            Label L_ok;
-            __ cmpptr(rax_argslot, rbx_destslot);
-            __ jccb(Assembler::belowEqual, L_ok);
-            __ stop("source must be below destination (downward rotation)");
-            __ bind(L_ok);
-          }
-#endif
+          // 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:
           //   rax = src_addr + swap_bytes
           //   rbx = dest_addr
           //   while (rax <= rbx) *(rax - swap_bytes) = *(rax + 0), rax++;
-          Label loop;
-          __ bind(loop);
-          __ movptr(rdx_temp, Address(rax_argslot, 0));
-          __ movptr(Address(rax_argslot, -swap_bytes), rdx_temp);
-          __ addptr(rax_argslot, wordSize);
-          __ cmpptr(rax_argslot, rbx_destslot);
-          __ jccb(Assembler::belowEqual, loop);
+          __ addptr(rbx_destslot, wordSize);
+          move_arg_slots_down(_masm,
+                              Address(rax_argslot, swap_slots * wordSize),
+                              rbx_destslot,
+                              -swap_slots,
+                              rax_argslot, rdx_temp);
+
+          __ subptr(rbx_destslot, wordSize);
         }
-
         // pop the original first chunk into the destination slot, now free
-        for (int i = 0; i < swap_bytes; i += wordSize) {
-          __ pop(rdx_temp);
-          __ movptr(Address(rbx_destslot, i), rdx_temp);
+        for (int i = 0; i < swap_slots; i++) {
+          if (i == 0)  __ movptr(Address(rbx_destslot, i * wordSize), rdi_temp);
+          else         __ popptr(Address(rbx_destslot, i * wordSize));
         }
       }
 
@@ -958,53 +1663,22 @@
       __ lea(rax_argslot, __ argument_address(rax_argslot));
 
       // 'stack_move' is negative number of words to duplicate
-      Register rdx_stack_move = rdx_temp;
-      __ movl2ptr(rdx_stack_move, rcx_amh_conversion);
-      __ sarptr(rdx_stack_move, CONV_STACK_MOVE_SHIFT);
+      Register rdi_stack_move = rdi_temp;
+      load_stack_move(_masm, rdi_stack_move, rcx_recv, true);
 
-      int argslot0_num = 0;
-      Address argslot0 = __ argument_address(RegisterOrConstant(argslot0_num));
-      assert(argslot0.base() == rsp, "");
-      int pre_arg_size = argslot0.disp();
-      assert(pre_arg_size % wordSize == 0, "");
-      assert(pre_arg_size > 0, "must include PC");
-
-      // remember the old rsp+1 (argslot[0])
-      Register rbx_oldarg = rbx_temp;
-      __ lea(rbx_oldarg, argslot0);
-
-      // move rsp down to make room for dups
-      __ lea(rsp, Address(rsp, rdx_stack_move, Address::times_ptr));
-
-      // compute the new rsp+1 (argslot[0])
-      Register rdx_newarg = rdx_temp;
-      __ lea(rdx_newarg, argslot0);
-
-      __ push(rdi);             // need a temp
-      // (preceding push must be done after arg addresses are taken!)
-
-      // pull down the pre_arg_size data (PC)
-      for (int i = -pre_arg_size; i < 0; i += wordSize) {
-        __ movptr(rdi, Address(rbx_oldarg, i));
-        __ movptr(Address(rdx_newarg, i), rdi);
+      if (VerifyMethodHandles) {
+        verify_argslots(_masm, rdi_stack_move, rax_argslot, true,
+                        "copied argument(s) must fall within current frame");
       }
 
-      // copy from rax_argslot[0...] down to new_rsp[1...]
-      // pseudo-code:
-      //   rbx = old_rsp+1
-      //   rdx = new_rsp+1
-      //   rax = argslot
-      //   while (rdx < rbx) *rdx++ = *rax++
-      Label loop;
-      __ bind(loop);
-      __ movptr(rdi, Address(rax_argslot, 0));
-      __ movptr(Address(rdx_newarg, 0), rdi);
-      __ addptr(rax_argslot, wordSize);
-      __ addptr(rdx_newarg, wordSize);
-      __ cmpptr(rdx_newarg, rbx_oldarg);
-      __ jccb(Assembler::less, loop);
+      // insert location is always the bottom of the argument list:
+      Address insert_location = __ argument_address(constant(0));
+      int pre_arg_words = insert_location.disp() / wordSize;   // return PC is pushed
+      assert(insert_location.base() == rsp, "");
 
-      __ pop(rdi);              // restore temp
+      __ negl(rdi_stack_move);
+      push_arg_slots(_masm, rax_argslot, rdi_stack_move,
+                     pre_arg_words, rbx_temp, rdx_temp);
 
       __ load_heap_oop(rcx_recv, rcx_mh_vmtarget);
       __ jump_to_method_handle_entry(rcx_recv, rdx_temp);
@@ -1017,63 +1691,583 @@
       __ movl(rax_argslot, rcx_amh_vmargslot);
       __ lea(rax_argslot, __ argument_address(rax_argslot));
 
-      __ push(rdi);             // need a temp
       // (must do previous push after argslot address is taken)
 
       // 'stack_move' is number of words to drop
-      Register rdi_stack_move = rdi;
-      __ movl2ptr(rdi_stack_move, rcx_amh_conversion);
-      __ sarptr(rdi_stack_move, CONV_STACK_MOVE_SHIFT);
+      Register rdi_stack_move = rdi_temp;
+      load_stack_move(_masm, rdi_stack_move, rcx_recv, false);
       remove_arg_slots(_masm, rdi_stack_move,
                        rax_argslot, rbx_temp, rdx_temp);
 
-      __ pop(rdi);              // restore temp
-
       __ load_heap_oop(rcx_recv, rcx_mh_vmtarget);
       __ jump_to_method_handle_entry(rcx_recv, rdx_temp);
     }
     break;
 
   case _adapter_collect_args:
-    __ unimplemented(entry_name(ek)); // %%% FIXME: NYI
-    break;
-
+  case _adapter_fold_args:
   case _adapter_spread_args:
     // handled completely by optimized cases
     __ stop("init_AdapterMethodHandle should not issue this");
     break;
 
+  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:
+    {
+      // Given a fresh incoming stack frame, build a new ricochet frame.
+      // On entry, TOS points at a return PC, and RBP 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 | sender_pc |
+      // push(sender_pc);
+
+      // Compute argument base:
+      Register rax_argv = rax_argslot;
+      __ lea(rax_argv, __ argument_address(constant(0)));
+
+      // 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
+        }
+        DEBUG_ONLY(extra_slots += 1);
+        if (extra_slots > 0) {
+          __ pop(rbx_temp);   // return value
+          __ subptr(rsp, (extra_slots * Interpreter::stackElementSize));
+          // Push guard word #2 in debug mode.
+          DEBUG_ONLY(__ movptr(Address(rsp, 0), (int32_t) RicochetFrame::MAGIC_NUMBER_2));
+          __ push(rbx_temp);
+        }
+      }
+
+      RicochetFrame::enter_ricochet_frame(_masm, rcx_recv, rax_argv,
+                                          entry(ek_ret)->from_interpreted_entry(), rbx_temp);
+
+      // Now pushed:  ... keep1 | collect | keep2 | RF |
+      // some handy frame slots:
+      Address exact_sender_sp_addr = RicochetFrame::frame_address(RicochetFrame::exact_sender_sp_offset_in_bytes());
+      Address conversion_addr      = RicochetFrame::frame_address(RicochetFrame::conversion_offset_in_bytes());
+      Address saved_args_base_addr = RicochetFrame::frame_address(RicochetFrame::saved_args_base_offset_in_bytes());
+
+#ifdef ASSERT
+      if (VerifyMethodHandles && dest != T_CONFLICT) {
+        BLOCK_COMMENT("verify AMH.conv.dest");
+        load_conversion_dest_type(_masm, rbx_temp, conversion_addr);
+        Label L_dest_ok;
+        __ cmpl(rbx_temp, (int) dest);
+        __ jcc(Assembler::equal, L_dest_ok);
+        if (dest == T_INT) {
+          for (int bt = T_BOOLEAN; bt < T_INT; bt++) {
+            if (is_subword_type(BasicType(bt))) {
+              __ cmpl(rbx_temp, (int) bt);
+              __ jcc(Assembler::equal, L_dest_ok);
+            }
+          }
+        }
+        __ stop("bad dest in AMH.conv");
+        __ BIND(L_dest_ok);
+      }
+#endif //ASSERT
+
+      // Find out where the original copy of the recursive argument sequence begins.
+      Register rax_coll = rax_argv;
+      {
+        RegisterOrConstant collect_slot = collect_slot_constant;
+        if (collect_slot_constant == -1) {
+          __ movl(rdi_temp, rcx_amh_vmargslot);
+          collect_slot = rdi_temp;
+        }
+        if (collect_slot_constant != 0)
+          __ lea(rax_coll, Address(rax_argv, collect_slot, Interpreter::stackElementScale()));
+        // rax_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(rcx_recv, rcx_amh_argument);
+      if (VerifyMethodHandles)  verify_method_handle(_masm, rcx_recv);
+
+      // Push a space for the recursively called MH first:
+      __ push((int32_t)NULL_WORD);
+
+      // Calculate |collect|, the number of arguments we are collecting.
+      Register rdi_collect_count = rdi_temp;
+      RegisterOrConstant collect_count;
+      if (collect_count_constant >= 0) {
+        collect_count = collect_count_constant;
+      } else {
+        __ load_method_handle_vmslots(rdi_collect_count, rcx_recv, rdx_temp);
+        collect_count = rdi_collect_count;
+      }
+#ifdef ASSERT
+      if (VerifyMethodHandles && collect_count_constant >= 0) {
+        __ load_method_handle_vmslots(rbx_temp, rcx_recv, rdx_temp);
+        Label L_count_ok;
+        __ cmpl(rbx_temp, collect_count_constant);
+        __ jcc(Assembler::equal, L_count_ok);
+        __ stop("bad vminfo in AMH.conv");
+        __ BIND(L_count_ok);
+      }
+#endif //ASSERT
+
+      // copy |collect| slots directly to TOS:
+      push_arg_slots(_masm, rax_coll, collect_count, 0, rbx_temp, rdx_temp);
+      // Now pushed:  ... keep1 | collect | keep2 | RF... | collect |
+      // rax_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 origial 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 rdi_close_count = rdi_collect_count;
+      if (retain_original_args) {
+        close_count = constant(0);
+      } else if (collect_count_constant == -1) {
+        close_count = rdi_collect_count;
+      }
+
+      // How many slots need moving?  This is simply dest_slot (0 => no |keep3|).
+      RegisterOrConstant keep3_count;
+      Register rsi_keep3_count = rsi;  // can repair from RF.exact_sender_sp
+      if (dest_slot_constant >= 0) {
+        keep3_count = dest_slot_constant;
+      } else  {
+        load_conversion_vminfo(_masm, rsi_keep3_count, conversion_addr);
+        keep3_count = rsi_keep3_count;
+      }
+#ifdef ASSERT
+      if (VerifyMethodHandles && dest_slot_constant >= 0) {
+        load_conversion_vminfo(_masm, rbx_temp, conversion_addr);
+        Label L_vminfo_ok;
+        __ cmpl(rbx_temp, dest_slot_constant);
+        __ jcc(Assembler::equal, L_vminfo_ok);
+        __ stop("bad vminfo in AMH.conv");
+        __ BIND(L_vminfo_ok);
+      }
+#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());
+
+      if (stomp_dest | fix_arg_base) {
+        // we will probably need an updated rax_argv value
+        if (collect_slot_constant >= 0) {
+          // rax_coll already holds the leading edge of |keep2|, so tweak it
+          assert(rax_coll == rax_argv, "elided a move");
+          if (collect_slot_constant != 0)
+            __ subptr(rax_argv, collect_slot_constant * Interpreter::stackElementSize);
+        } else {
+          // Just reload from RF.saved_args_base.
+          __ movptr(rax_argv, saved_args_base_addr);
+        }
+      }
+
+      // 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 t 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);
+        } else {
+          close_count = close_count.as_constant() - open_count;
+        }
+        open_count = 0;
+      }
+      Address old_argv(rax_argv, 0);
+      Address new_argv(rax_argv, close_count,  Interpreter::stackElementScale(),
+                                - 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_constant()) {
+        __ testl(keep3_count.as_register(), keep3_count.as_register());
+        __ jcc(Assembler::zero, L_done);
+      }
+      if (!close_count.is_constant()) {
+        __ cmpl(close_count.as_register(), open_count);
+        __ jcc(Assembler::equal, L_done);
+      }
+
+      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) {
+          __ cmpl(close_count.as_register(), open_count);
+          __ jcc(Assembler::greater, L_move_up);
+        }
+
+        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++) {
+              __ movptr(rdx_temp, old_argv.plus_disp(i * Interpreter::stackElementSize));
+              __ movptr(          new_argv.plus_disp(i * Interpreter::stackElementSize), rdx_temp);
+            }
+          } else {
+            Register rbx_argv_top = rbx_temp;
+            __ lea(rbx_argv_top, old_argv.plus_disp(keep3_count, Interpreter::stackElementScale()));
+            move_arg_slots_down(_masm,
+                                old_argv,     // beginning of old argv
+                                rbx_argv_top, // end of old argv
+                                close_count,  // distance to move down (must be negative)
+                                rax_argv, rdx_temp);
+            // Used argv as an iteration variable; reload from RF.saved_args_base.
+            __ movptr(rax_argv, saved_args_base_addr);
+          }
+        }
+
+        if (emit_guard) {
+          __ jmp(L_done);  // assumes emit_move_up is true also
+          __ 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--) {
+              __ movptr(rdx_temp, old_argv.plus_disp(i * Interpreter::stackElementSize));
+              __ movptr(          new_argv.plus_disp(i * Interpreter::stackElementSize), rdx_temp);
+            }
+          } else {
+            Address argv_top = old_argv.plus_disp(keep3_count, Interpreter::stackElementScale());
+            move_arg_slots_up(_masm,
+                              rax_argv,     // beginning of old argv
+                              argv_top,     // end of old argv
+                              close_count,  // distance to move up (must be positive)
+                              rbx_temp, rdx_temp);
+          }
+        }
+      }
+      __ BIND(L_done);
+
+      if (fix_arg_base) {
+        // adjust RF.saved_args_base by adding (close_count - open_count)
+        if (!new_argv.is_same_address(Address(rax_argv, 0)))
+          __ lea(rax_argv, new_argv);
+        __ movptr(saved_args_base_addr, rax_argv);
+      }
+
+      if (stomp_dest) {
+        // Stomp the return slot, so it doesn't hold garbage.
+        // This isn't strictly necessary, but it may help detect bugs.
+        int forty_two = RicochetFrame::RETURN_VALUE_PLACEHOLDER;
+        __ movptr(Address(rax_argv, keep3_count, Address::times_ptr),
+                  (int32_t) forty_two);
+        // uses rsi_keep3_count
+      }
+      BLOCK_COMMENT("} adjust trailing arguments");
+
+      BLOCK_COMMENT("do_recursive_call");
+      __ mov(saved_last_sp, rsp);    // set rsi/r13 for callee
+      __ pushptr(ExternalAddress(SharedRuntime::ricochet_blob()->bounce_addr()).addr());
+      // 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(rcx_recv, rdx_temp);
+
+      break;
+    }
+
+  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;
+            }
+          }
+        }
+      }
+
+      Register rbx_arg_base = rbx_temp;
+      assert_different_registers(rax, rdx,  // possibly live return value registers
+                                 rdi_temp, rbx_arg_base);
+
+      Address conversion_addr      = RicochetFrame::frame_address(RicochetFrame::conversion_offset_in_bytes());
+      Address saved_args_base_addr = RicochetFrame::frame_address(RicochetFrame::saved_args_base_offset_in_bytes());
+
+      __ movptr(rbx_arg_base, saved_args_base_addr);
+      RegisterOrConstant dest_slot = dest_slot_constant;
+      if (dest_slot_constant == -1) {
+        load_conversion_vminfo(_masm, rdi_temp, conversion_addr);
+        dest_slot = rdi_temp;
+      }
+      // 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, Address(rbx_arg_base, dest_slot, Interpreter::stackElementScale()));
+
+      RicochetFrame::leave_ricochet_frame(_masm, rcx_recv, rbx_arg_base, rdx_temp);
+      __ push(rdx_temp);  // repush the return PC
+
+      // Load the final target and go.
+      if (VerifyMethodHandles)  verify_method_handle(_masm, rcx_recv);
+      __ jump_to_method_handle_entry(rcx_recv, rdx_temp);
+      __ hlt(); // --------------------
+      break;
+    }
+
+  case _adapter_opt_return_any:
+    {
+      if (VerifyMethodHandles)  RicochetFrame::verify_clean(_masm);
+      Register rdi_conv = rdi_temp;
+      assert_different_registers(rax, rdx,  // possibly live return value registers
+                                 rdi_conv, rbx_temp);
+
+      Address conversion_addr = RicochetFrame::frame_address(RicochetFrame::conversion_offset_in_bytes());
+      load_conversion_dest_type(_masm, rdi_conv, conversion_addr);
+      __ lea(rbx_temp, ExternalAddress((address) &_adapter_return_handlers[0]));
+      __ movptr(rbx_temp, Address(rbx_temp, rdi_conv, Address::times_ptr));
+
+#ifdef ASSERT
+      { Label L_badconv;
+        __ testptr(rbx_temp, rbx_temp);
+        __ jccb(Assembler::zero, L_badconv);
+        __ jmp(rbx_temp);
+        __ bind(L_badconv);
+        __ stop("bad method handle return");
+      }
+#else //ASSERT
+      __ jmp(rbx_temp);
+#endif //ASSERT
+      break;
+    }
+
   case _adapter_opt_spread_0:
-  case _adapter_opt_spread_1:
-  case _adapter_opt_spread_more:
+  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 = get_ek_adapter_opt_spread_info(ek);
+      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
       __ movl(rax_argslot, rcx_amh_vmargslot);
       __ lea(rax_argslot, __ argument_address(rax_argslot));
 
-      // grab some temps
-      { __ push(rsi); __ push(rdi); }
-      // (preceding pushes must be done after argslot address is taken!)
-#define UNPUSH_RSI_RDI \
-      { __ pop(rdi); __ pop(rsi); }
+      // grab another temp
+      Register rsi_temp = rsi;
+      { if (rsi_temp == saved_last_sp)  __ push(saved_last_sp); }
+      // (preceding push must be done after argslot address is taken!)
+#define UNPUSH_RSI \
+      { if (rsi_temp == saved_last_sp)  __ pop(saved_last_sp); }
 
       // arx_argslot points both to the array and to the first output arg
       vmarg = Address(rax_argslot, 0);
 
       // Get the array value.
-      Register  rsi_array       = rsi;
+      Register  rsi_array       = rsi_temp;
       Register  rdx_array_klass = rdx_temp;
-      BasicType elem_type       = T_OBJECT;
+      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);
       __ movptr(rsi_array, vmarg);
-      Label skip_array_check;
-      if (length_constant == 0) {
+
+      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, rbx_temp, rcx_amh_conversion);
+          __ testl(rbx_temp, rbx_temp);
+          __ jcc(Assembler::notZero, L_skip);
+        }
         __ testptr(rsi_array, rsi_array);
-        __ jcc(Assembler::zero, skip_array_check);
+        __ jcc(Assembler::zero, L_array_is_empty);
+        __ bind(L_skip);
       }
       __ null_check(rsi_array, oopDesc::klass_offset_in_bytes());
       __ load_klass(rdx_array_klass, rsi_array);
@@ -1081,22 +2275,20 @@
       // Check the array type.
       Register rbx_klass = rbx_temp;
       __ load_heap_oop(rbx_klass, rcx_amh_argument); // this is a Class object!
-      __ load_heap_oop(rbx_klass, Address(rbx_klass, java_lang_Class::klass_offset_in_bytes()));
+      load_klass_from_Class(_masm, rbx_klass);
 
       Label ok_array_klass, bad_array_klass, bad_array_length;
-      __ check_klass_subtype(rdx_array_klass, rbx_klass, rdi, ok_array_klass);
+      __ check_klass_subtype(rdx_array_klass, rbx_klass, rdi_temp, ok_array_klass);
       // If we get here, the type check failed!
       __ jmp(bad_array_klass);
-      __ bind(ok_array_klass);
+      __ BIND(ok_array_klass);
 
       // Check length.
       if (length_constant >= 0) {
         __ cmpl(Address(rsi_array, length_offset), length_constant);
       } else {
         Register rbx_vminfo = rbx_temp;
-        __ movl(rbx_vminfo, rcx_amh_conversion);
-        assert(CONV_VMINFO_SHIFT == 0, "preshifted");
-        __ andl(rbx_vminfo, CONV_VMINFO_MASK);
+        load_conversion_vminfo(_masm, rbx_vminfo, rcx_amh_conversion);
         __ cmpl(rbx_vminfo, Address(rsi_array, length_offset));
       }
       __ jcc(Assembler::notEqual, bad_array_length);
@@ -1108,90 +2300,104 @@
         // Form a pointer to the end of the affected region.
         __ lea(rdx_argslot_limit, Address(rax_argslot, Interpreter::stackElementSize));
         // 'stack_move' is negative number of words to insert
-        Register rdi_stack_move = rdi;
-        __ movl2ptr(rdi_stack_move, rcx_amh_conversion);
-        __ sarptr(rdi_stack_move, CONV_STACK_MOVE_SHIFT);
+        // This number already accounts for elem_slots.
+        Register rdi_stack_move = rdi_temp;
+        load_stack_move(_masm, rdi_stack_move, rcx_recv, true);
+        __ cmpptr(rdi_stack_move, 0);
+        assert(stack_move_unit() < 0, "else change this comparison");
+        __ jcc(Assembler::less, L_insert_arg_space);
+        __ jcc(Assembler::equal, L_copy_args);
+        // single argument case, with no array movement
+        __ BIND(L_array_is_empty);
+        remove_arg_slots(_masm, -stack_move_unit() * array_slots,
+                         rax_argslot, rbx_temp, rdx_temp);
+        __ jmp(L_args_done);  // no spreading to do
+        __ BIND(L_insert_arg_space);
+        // come here in the usual case, stack_move < 0 (2 or more spread arguments)
         Register rsi_temp = rsi_array;  // spill this
-        insert_arg_slots(_masm, rdi_stack_move, -1,
+        insert_arg_slots(_masm, rdi_stack_move,
                          rax_argslot, rbx_temp, rsi_temp);
-        // reload the array (since rsi was killed)
-        __ movptr(rsi_array, vmarg);
-      } else if (length_constant > 1) {
-        int arg_mask = 0;
-        int new_slots = (length_constant - 1);
-        for (int i = 0; i < new_slots; i++) {
-          arg_mask <<= 1;
-          arg_mask |= _INSERT_REF_MASK;
-        }
-        insert_arg_slots(_masm, new_slots * stack_move_unit(), arg_mask,
+        // reload the array since rsi was killed
+        // reload from rdx_argslot_limit since rax_argslot is now decremented
+        __ movptr(rsi_array, Address(rdx_argslot_limit, -Interpreter::stackElementSize));
+      } else if (length_constant >= 1) {
+        int new_slots = (length_constant * elem_slots) - array_slots;
+        insert_arg_slots(_masm, new_slots * stack_move_unit(),
                          rax_argslot, rbx_temp, rdx_temp);
-      } else if (length_constant == 1) {
-        // no stack resizing required
       } else if (length_constant == 0) {
-        remove_arg_slots(_masm, -stack_move_unit(),
+        __ BIND(L_array_is_empty);
+        remove_arg_slots(_masm, -stack_move_unit() * array_slots,
                          rax_argslot, rbx_temp, rdx_temp);
+      } else {
+        ShouldNotReachHere();
       }
 
       // Copy from the array to the new slots.
       // Note: Stack change code preserves integrity of rax_argslot pointer.
       // So even after slot insertions, rax_argslot still points to first argument.
+      // 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) {
         // [rax_argslot, rdx_argslot_limit) is the area we are inserting into.
+        // Array element [0] goes at rdx_argslot_limit[-wordSize].
         Register rsi_source = rsi_array;
         __ lea(rsi_source, Address(rsi_array, elem0_offset));
+        Register rdx_fill_ptr = rdx_argslot_limit;
         Label loop;
-        __ bind(loop);
-        __ movptr(rbx_temp, Address(rsi_source, 0));
-        __ movptr(Address(rax_argslot, 0), rbx_temp);
+        __ BIND(loop);
+        __ addptr(rdx_fill_ptr, -Interpreter::stackElementSize * elem_slots);
+        move_typed_arg(_masm, elem_type, true,
+                       Address(rdx_fill_ptr, 0), Address(rsi_source, 0),
+                       rbx_temp, rdi_temp);
         __ addptr(rsi_source, type2aelembytes(elem_type));
-        __ addptr(rax_argslot, Interpreter::stackElementSize);
-        __ cmpptr(rax_argslot, rdx_argslot_limit);
-        __ jccb(Assembler::less, loop);
+        __ cmpptr(rdx_fill_ptr, rax_argslot);
+        __ jcc(Assembler::greater, loop);
       } else if (length_constant == 0) {
-        __ bind(skip_array_check);
         // nothing to copy
       } else {
         int elem_offset = elem0_offset;
-        int slot_offset = 0;
+        int slot_offset = length_constant * Interpreter::stackElementSize;
         for (int index = 0; index < length_constant; index++) {
-          __ movptr(rbx_temp, Address(rsi_array, elem_offset));
-          __ movptr(Address(rax_argslot, slot_offset), rbx_temp);
+          slot_offset -= Interpreter::stackElementSize * elem_slots;  // fill backward
+          move_typed_arg(_masm, elem_type, true,
+                         Address(rax_argslot, slot_offset), Address(rsi_array, elem_offset),
+                         rbx_temp, rdi_temp);
           elem_offset += type2aelembytes(elem_type);
-           slot_offset += Interpreter::stackElementSize;
         }
       }
+      __ BIND(L_args_done);
 
       // Arguments are spread.  Move to next method handle.
-      UNPUSH_RSI_RDI;
+      UNPUSH_RSI;
       __ load_heap_oop(rcx_recv, rcx_mh_vmtarget);
       __ jump_to_method_handle_entry(rcx_recv, rdx_temp);
 
       __ bind(bad_array_klass);
-      UNPUSH_RSI_RDI;
+      UNPUSH_RSI;
       assert(!vmarg.uses(rarg2_required), "must be different registers");
-      __ movptr(rarg2_required, Address(rdx_array_klass, java_mirror_offset));  // required type
-      __ movptr(rarg1_actual,   vmarg);                                         // bad array
-      __ movl(  rarg0_code,     (int) Bytecodes::_aaload);                      // who is complaining?
+      __ load_heap_oop( rarg2_required, Address(rdx_array_klass, java_mirror_offset));  // required type
+      __ movptr(        rarg1_actual,   vmarg);                                         // bad array
+      __ movl(          rarg0_code,     (int) Bytecodes::_aaload);                      // who is complaining?
       __ jump(ExternalAddress(from_interpreted_entry(_raise_exception)));
 
       __ bind(bad_array_length);
-      UNPUSH_RSI_RDI;
+      UNPUSH_RSI;
       assert(!vmarg.uses(rarg2_required), "must be different registers");
-      __ mov   (rarg2_required, rcx_recv);                       // AMH requiring a certain length
-      __ movptr(rarg1_actual,   vmarg);                          // bad array
-      __ movl(  rarg0_code,     (int) Bytecodes::_arraylength);  // who is complaining?
+      __ mov(    rarg2_required, rcx_recv);                       // AMH requiring a certain length
+      __ movptr( rarg1_actual,   vmarg);                          // bad array
+      __ movl(   rarg0_code,     (int) Bytecodes::_arraylength);  // who is complaining?
       __ jump(ExternalAddress(from_interpreted_entry(_raise_exception)));
+#undef UNPUSH_RSI
 
-#undef UNPUSH_RSI_RDI
+      break;
     }
-    break;
 
-  case _adapter_flyby:
-  case _adapter_ricochet:
-    __ unimplemented(entry_name(ek)); // %%% FIXME: NYI
-    break;
-
-  default:  ShouldNotReachHere();
+  default:
+    // do not require all platforms to recognize all adapter types
+    __ nop();
+    return;
   }
   __ hlt();
 
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/src/cpu/x86/vm/methodHandles_x86.hpp	Fri May 06 16:33:13 2011 -0700
@@ -0,0 +1,292 @@
+/*
+ * Copyright (c) 2010, 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.
+
+public:
+
+// The stack just after the recursive call from a ricochet frame
+// looks something like this.  Offsets are marked in words, not bytes.
+// rsi (r13 on LP64) is part of the interpreter calling sequence
+// which tells the callee where is my real rsp (for frame walking).
+// (...lower memory addresses)
+// rsp:     [ return pc                 ]   always the global RicochetBlob::bounce_addr
+// rsp+1:   [ recursive arg N           ]
+// rsp+2:   [ recursive arg N-1         ]
+// ...
+// rsp+N:   [ recursive arg 1           ]
+// rsp+N+1: [ recursive method handle   ]
+// ...
+// rbp-6:   [ cleanup continuation pc   ]   <-- (struct RicochetFrame)
+// rbp-5:   [ saved target MH           ]   the MH we will call on the saved args
+// rbp-4:   [ saved args layout oop     ]   an int[] array which describes argument layout
+// rbp-3:   [ saved args pointer        ]   address of transformed adapter arg M (slot 0)
+// rbp-2:   [ conversion                ]   information about how the return value is used
+// rbp-1:   [ exact sender sp           ]   exact TOS (rsi/r13) of original sender frame
+// rbp+0:   [ saved sender fp           ]   (for original sender of AMH)
+// rbp+1:   [ saved sender pc           ]   (back to original sender of AMH)
+// rbp+2:   [ transformed adapter arg M ]   <-- (extended TOS of original sender)
+// rbp+3:   [ transformed adapter arg M-1]
+// ...
+// rbp+M+1: [ transformed adapter arg 1 ]
+// rbp+M+2: [ padding                   ] <-- (rbp + saved args base offset)
+// ...      [ optional padding]
+// (higher memory addresses...)
+//
+// The arguments originally passed by the original sender
+// are lost, and arbitrary amounts of stack motion might have
+// happened due to argument transformation.
+// (This is done by C2I/I2C adapters and non-direct method handles.)
+// This is why there is an unpredictable amount of memory between
+// the extended and exact TOS of the sender.
+// The ricochet adapter itself will also (in general) perform
+// transformations before the recursive call.
+//
+// The transformed and saved arguments, immediately above the saved
+// return PC, are a well-formed method handle invocation ready to execute.
+// When the GC needs to walk the stack, these arguments are described
+// via the saved arg types oop, an int[] array with a private format.
+// This array is derived from the type of the transformed adapter
+// method handle, which also sits at the base of the saved argument
+// bundle.  Since the GC may not be able to fish out the int[]
+// array, so it is pushed explicitly on the stack.  This may be
+// an unnecessary expense.
+//
+// The following register conventions are significant at this point:
+// rsp       the thread stack, as always; preserved by caller
+// rsi/r13   exact TOS of recursive frame (contents of [rbp-2])
+// rcx       recursive method handle (contents of [rsp+N+1])
+// rbp       preserved by caller (not used by caller)
+// Unless otherwise specified, all registers can be blown by the call.
+//
+// If this frame must be walked, the transformed adapter arguments
+// will be found with the help of the saved arguments descriptor.
+//
+// Therefore, the descriptor must match the referenced arguments.
+// The arguments must be followed by at least one word of padding,
+// which will be necessary to complete the final method handle call.
+// That word is not treated as holding an oop.  Neither is the word
+//
+// The word pointed to by the return argument pointer is not
+// treated as an oop, even if points to a saved argument.
+// This allows the saved argument list to have a "hole" in it
+// to receive an oop from the recursive call.
+// (The hole might temporarily contain RETURN_VALUE_PLACEHOLDER.)
+//
+// When the recursive callee returns, RicochetBlob::bounce_addr will
+// immediately jump to the continuation stored in the RF.
+// This continuation will merge the recursive return value
+// into the saved argument list.  At that point, the original
+// rsi, rbp, and rsp will be reloaded, the ricochet frame will
+// disappear, and the final target of the adapter method handle
+// will be invoked on the transformed argument list.
+
+class RicochetFrame {
+  friend class MethodHandles;
+
+ private:
+  intptr_t* _continuation;          // what to do when control gets back here
+  oopDesc*  _saved_target;          // target method handle to invoke on saved_args
+  oopDesc*  _saved_args_layout;     // caching point for MethodTypeForm.vmlayout cookie
+  intptr_t* _saved_args_base;       // base of pushed arguments (slot 0, arg N) (-3)
+  intptr_t  _conversion;            // misc. information from original AdapterMethodHandle (-2)
+  intptr_t* _exact_sender_sp;       // parallel to interpreter_frame_sender_sp (-1)
+  intptr_t* _sender_link;           // *must* coincide with frame::link_offset (0)
+  address   _sender_pc;             // *must* coincide with frame::return_addr_offset (1)
+
+ public:
+  intptr_t* continuation() const        { return _continuation; }
+  oop       saved_target() const        { return _saved_target; }
+  oop       saved_args_layout() const   { return _saved_args_layout; }
+  intptr_t* saved_args_base() const     { return _saved_args_base; }
+  intptr_t  conversion() const          { return _conversion; }
+  intptr_t* exact_sender_sp() const     { return _exact_sender_sp; }
+  intptr_t* sender_link() const         { return _sender_link; }
+  address   sender_pc() const           { return _sender_pc; }
+
+  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*)&_saved_target; }
+  oop*  saved_args_layout_addr()                { return (oop*)&_saved_args_layout; }
+
+  oop  compute_saved_args_layout(bool read_cache, bool write_cache);
+
+  // Compiler/assembler interface.
+  static int continuation_offset_in_bytes()     { return offset_of(RicochetFrame, _continuation); }
+  static int saved_target_offset_in_bytes()     { return offset_of(RicochetFrame, _saved_target); }
+  static int saved_args_layout_offset_in_bytes(){ return offset_of(RicochetFrame, _saved_args_layout); }
+  static int saved_args_base_offset_in_bytes()  { return offset_of(RicochetFrame, _saved_args_base); }
+  static int conversion_offset_in_bytes()       { return offset_of(RicochetFrame, _conversion); }
+  static int exact_sender_sp_offset_in_bytes()  { return offset_of(RicochetFrame, _exact_sender_sp); }
+  static int sender_link_offset_in_bytes()      { return offset_of(RicochetFrame, _sender_link); }
+  static int sender_pc_offset_in_bytes()        { return offset_of(RicochetFrame, _sender_pc); }
+
+  // This value is not used for much, but it apparently must be nonzero.
+  static int frame_size_in_bytes()              { return sender_link_offset_in_bytes(); }
+
+#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 int magic_number_1_offset_in_bytes()   { return -wordSize; }
+  static int magic_number_2_offset_in_bytes()   { return sizeof(RicochetFrame); }
+  intptr_t magic_number_1() const               { return *(intptr_t*)((address)this + magic_number_1_offset_in_bytes()); };
+  intptr_t magic_number_2() const               { return *(intptr_t*)((address)this + magic_number_2_offset_in_bytes()); };
+#endif //ASSERT
+
+  enum { RETURN_VALUE_PLACEHOLDER = (NOT_DEBUG(0) DEBUG_ONLY(42)) };
+
+  static void verify_offsets() NOT_DEBUG_RETURN;
+  void verify() const NOT_DEBUG_RETURN; // check for MAGIC_NUMBER, etc.
+  void zap_arguments() NOT_DEBUG_RETURN;
+
+  static void generate_ricochet_blob(MacroAssembler* _masm,
+                                     // output params:
+                                     int* frame_size_in_words, int* bounce_offset, int* exception_offset);
+
+  static void enter_ricochet_frame(MacroAssembler* _masm,
+                                   Register rcx_recv,
+                                   Register rax_argv,
+                                   address return_handler,
+                                   Register rbx_temp);
+  static void leave_ricochet_frame(MacroAssembler* _masm,
+                                   Register rcx_recv,
+                                   Register new_sp_reg,
+                                   Register sender_pc_reg);
+
+  static Address frame_address(int offset = 0) {
+    // The RicochetFrame is found by subtracting a constant offset from rbp.
+    return Address(rbp, - sender_link_offset_in_bytes() + offset);
+  }
+
+  static RicochetFrame* from_frame(const frame& fr) {
+    address bp = (address) fr.fp();
+    RicochetFrame* rf = (RicochetFrame*)(bp - sender_link_offset_in_bytes());
+    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);
+  static void load_conversion_vminfo(MacroAssembler* _masm, Register reg, Address conversion_field_addr);
+  static void load_conversion_dest_type(MacroAssembler* _masm, Register reg, Address conversion_field_addr);
+
+  static void load_stack_move(MacroAssembler* _masm,
+                              Register rdi_stack_move,
+                              Register rcx_amh,
+                              bool might_be_negative);
+
+  static void insert_arg_slots(MacroAssembler* _masm,
+                               RegisterOrConstant arg_slots,
+                               Register rax_argslot,
+                               Register rbx_temp, Register rdx_temp);
+
+  static void remove_arg_slots(MacroAssembler* _masm,
+                               RegisterOrConstant arg_slots,
+                               Register rax_argslot,
+                               Register rbx_temp, Register rdx_temp);
+
+  static void push_arg_slots(MacroAssembler* _masm,
+                                   Register rax_argslot,
+                                   RegisterOrConstant slot_count,
+                                   int skip_words_count,
+                                   Register rbx_temp, Register rdx_temp);
+
+  static void move_arg_slots_up(MacroAssembler* _masm,
+                                Register rbx_bottom,  // invariant
+                                Address  top_addr,    // can use rax_temp
+                                RegisterOrConstant positive_distance_in_slots,
+                                Register rax_temp, Register rdx_temp);
+
+  static void move_arg_slots_down(MacroAssembler* _masm,
+                                  Address  bottom_addr,  // can use rax_temp
+                                  Register rbx_top,      // invariant
+                                  RegisterOrConstant negative_distance_in_slots,
+                                  Register rax_temp, Register rdx_temp);
+
+  static void move_typed_arg(MacroAssembler* _masm,
+                             BasicType type, bool is_element,
+                             Address slot_dest, Address value_src,
+                             Register rbx_temp, Register rdx_temp);
+
+  static void move_return_value(MacroAssembler* _masm, BasicType type,
+                                Address return_slot);
+
+  static void verify_argslot(MacroAssembler* _masm, Register argslot_reg,
+                             const char* error_message) NOT_DEBUG_RETURN;
+
+  static void verify_argslots(MacroAssembler* _masm,
+                              RegisterOrConstant argslot_count,
+                              Register argslot_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, KlassHandle klass,
+                           const char* error_message = "wrong klass") NOT_DEBUG_RETURN;
+
+  static void verify_method_handle(MacroAssembler* _masm, Register mh_reg) {
+    verify_klass(_masm, mh_reg, SystemDictionaryHandles::MethodHandle_klass(),
+                 "reference is a MH");
+  }
+
+  static void trace_method_handle(MacroAssembler* _masm, const char* adaptername) PRODUCT_RETURN;
+
+  static Register saved_last_sp_register() {
+    // Should be in sharedRuntime, not here.
+    return LP64_ONLY(r13) NOT_LP64(rsi);
+  }
--- a/src/cpu/x86/vm/sharedRuntime_x86_32.cpp	Fri May 06 12:12:29 2011 -0700
+++ b/src/cpu/x86/vm/sharedRuntime_x86_32.cpp	Fri May 06 16:33:13 2011 -0700
@@ -2253,6 +2253,31 @@
   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() {
@@ -2996,6 +3021,8 @@
     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();
--- a/src/cpu/x86/vm/sharedRuntime_x86_64.cpp	Fri May 06 12:12:29 2011 -0700
+++ b/src/cpu/x86/vm/sharedRuntime_x86_64.cpp	Fri May 06 16:33:13 2011 -0700
@@ -2530,6 +2530,32 @@
 }
 
 
+//----------------------------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
@@ -3205,6 +3231,8 @@
     generate_handler_blob(CAST_FROM_FN_PTR(address,
                    SafepointSynchronize::handle_polling_page_exception), true);
 
+  generate_ricochet_blob();
+
   generate_deopt_blob();
 
 #ifdef COMPILER2
--- a/src/cpu/x86/vm/stubRoutines_x86_32.hpp	Fri May 06 12:12:29 2011 -0700
+++ b/src/cpu/x86/vm/stubRoutines_x86_32.hpp	Fri May 06 16:33:13 2011 -0700
@@ -36,7 +36,7 @@
 
 // MethodHandles adapters
 enum method_handles_platform_dependent_constants {
-  method_handles_adapters_code_size = 10000
+  method_handles_adapters_code_size = 30000 DEBUG_ONLY(+ 10000)
 };
 
 class x86 {
--- a/src/cpu/x86/vm/stubRoutines_x86_64.hpp	Fri May 06 12:12:29 2011 -0700
+++ b/src/cpu/x86/vm/stubRoutines_x86_64.hpp	Fri May 06 16:33:13 2011 -0700
@@ -38,7 +38,7 @@
 
 // MethodHandles adapters
 enum method_handles_platform_dependent_constants {
-  method_handles_adapters_code_size = 40000
+  method_handles_adapters_code_size = 80000 DEBUG_ONLY(+ 120000)
 };
 
 class x86 {
--- a/src/share/vm/classfile/javaClasses.cpp	Fri May 06 12:12:29 2011 -0700
+++ b/src/share/vm/classfile/javaClasses.cpp	Fri May 06 16:33:13 2011 -0700
@@ -2602,6 +2602,7 @@
 // Support for java_lang_invoke_MethodTypeForm
 
 int java_lang_invoke_MethodTypeForm::_vmslots_offset;
+int java_lang_invoke_MethodTypeForm::_vmlayout_offset;
 int java_lang_invoke_MethodTypeForm::_erasedType_offset;
 int java_lang_invoke_MethodTypeForm::_genericInvoker_offset;
 
@@ -2609,6 +2610,7 @@
   klassOop k = SystemDictionary::MethodTypeForm_klass();
   if (k != NULL) {
     compute_optional_offset(_vmslots_offset,    k, vmSymbols::vmslots_name(),    vmSymbols::int_signature(), true);
+    compute_optional_offset(_vmlayout_offset,   k, vmSymbols::vmlayout_name(),   vmSymbols::object_signature());
     compute_optional_offset(_erasedType_offset, k, vmSymbols::erasedType_name(), vmSymbols::java_lang_invoke_MethodType_signature(), true);
     compute_optional_offset(_genericInvoker_offset, k, vmSymbols::genericInvoker_name(), vmSymbols::java_lang_invoke_MethodHandle_signature(), true);
     if (_genericInvoker_offset == 0)  _genericInvoker_offset = -1;  // set to explicit "empty" value
@@ -2617,9 +2619,31 @@
 
 int java_lang_invoke_MethodTypeForm::vmslots(oop mtform) {
   assert(mtform->klass() == SystemDictionary::MethodTypeForm_klass(), "MTForm only");
+  assert(_vmslots_offset > 0, "");
   return mtform->int_field(_vmslots_offset);
 }
 
+oop java_lang_invoke_MethodTypeForm::vmlayout(oop mtform) {
+  assert(mtform->klass() == SystemDictionary::MethodTypeForm_klass(), "MTForm only");
+  assert(_vmlayout_offset > 0, "");
+  return mtform->obj_field(_vmlayout_offset);
+}
+
+oop java_lang_invoke_MethodTypeForm::init_vmlayout(oop mtform, oop cookie) {
+  assert(mtform->klass() == SystemDictionary::MethodTypeForm_klass(), "MTForm only");
+  oop previous = vmlayout(mtform);
+  if (previous != NULL) {
+    return previous;  // someone else beat us to it
+  }
+  HeapWord* cookie_addr = (HeapWord*) mtform->obj_field_addr<oop>(_vmlayout_offset);
+  OrderAccess::storestore();  // make sure our copy is fully committed
+  previous = oopDesc::atomic_compare_exchange_oop(cookie, cookie_addr, previous);
+  if (previous != NULL) {
+    return previous;  // someone else beat us to it
+  }
+  return cookie;
+}
+
 oop java_lang_invoke_MethodTypeForm::erasedType(oop mtform) {
   assert(mtform->klass() == SystemDictionary::MethodTypeForm_klass(), "MTForm only");
   return mtform->obj_field(_erasedType_offset);
--- a/src/share/vm/classfile/javaClasses.hpp	Fri May 06 12:12:29 2011 -0700
+++ b/src/share/vm/classfile/javaClasses.hpp	Fri May 06 16:33:13 2011 -0700
@@ -949,18 +949,19 @@
     OP_CHECK_CAST    = 0x2, // ref-to-ref conversion; requires a Class argument
     OP_PRIM_TO_PRIM  = 0x3, // converts from one primitive to another
     OP_REF_TO_PRIM   = 0x4, // unboxes a wrapper to produce a primitive
-    OP_PRIM_TO_REF   = 0x5, // boxes a primitive into a wrapper (NYI)
+    OP_PRIM_TO_REF   = 0x5, // boxes a primitive into a wrapper
     OP_SWAP_ARGS     = 0x6, // swap arguments (vminfo is 2nd arg)
     OP_ROT_ARGS      = 0x7, // rotate arguments (vminfo is displaced arg)
     OP_DUP_ARGS      = 0x8, // duplicates one or more arguments (at TOS)
     OP_DROP_ARGS     = 0x9, // remove one or more argument slots
-    OP_COLLECT_ARGS  = 0xA, // combine one or more arguments into a varargs (NYI)
+    OP_COLLECT_ARGS  = 0xA, // combine arguments using an auxiliary function
     OP_SPREAD_ARGS   = 0xB, // expand in place a varargs array (of known size)
-    OP_FLYBY         = 0xC, // operate first on reified argument list (NYI)
-    OP_RICOCHET      = 0xD, // run an adapter chain on the return value (NYI)
+    OP_FOLD_ARGS     = 0xC, // combine but do not remove arguments; prepend result
+    //OP_UNUSED_13   = 0xD, // unused code, perhaps for reified argument lists
     CONV_OP_LIMIT    = 0xE, // limit of CONV_OP enumeration
 
     CONV_OP_MASK     = 0xF00, // this nybble contains the conversion op field
+    CONV_TYPE_MASK   = 0x0F,  // fits T_ADDRESS and below
     CONV_VMINFO_MASK = 0x0FF, // LSB is reserved for JVM use
     CONV_VMINFO_SHIFT     =  0, // position of bits in CONV_VMINFO_MASK
     CONV_OP_SHIFT         =  8, // position of bits in CONV_OP_MASK
@@ -1089,6 +1090,7 @@
 
  private:
   static int _vmslots_offset;           // number of argument slots needed
+  static int _vmlayout_offset;          // object describing internal calling sequence
   static int _erasedType_offset;        // erasedType = canonical MethodType
   static int _genericInvoker_offset;    // genericInvoker = adapter for invokeGeneric
 
@@ -1100,8 +1102,12 @@
   static oop            erasedType(oop mtform);
   static oop            genericInvoker(oop mtform);
 
+  static oop            vmlayout(oop mtform);
+  static oop       init_vmlayout(oop mtform, oop cookie);
+
   // Accessors for code generation:
   static int vmslots_offset_in_bytes()          { return _vmslots_offset; }
+  static int vmlayout_offset_in_bytes()         { return _vmlayout_offset; }
   static int erasedType_offset_in_bytes()       { return _erasedType_offset; }
   static int genericInvoker_offset_in_bytes()   { return _genericInvoker_offset; }
 };
--- a/src/share/vm/classfile/systemDictionary.cpp	Fri May 06 12:12:29 2011 -0700
+++ b/src/share/vm/classfile/systemDictionary.cpp	Fri May 06 16:33:13 2011 -0700
@@ -2362,8 +2362,15 @@
       spe = invoke_method_table()->find_entry(index, hash, signature, name_id);
       if (spe == NULL)
         spe = invoke_method_table()->add_entry(index, hash, signature, name_id);
-      if (spe->property_oop() == NULL)
+      if (spe->property_oop() == NULL) {
         spe->set_property_oop(m());
+        // Link m to his method type, if it is suitably generic.
+        oop mtform = java_lang_invoke_MethodType::form(mt());
+        if (mtform != NULL && mt() == java_lang_invoke_MethodTypeForm::erasedType(mtform)
+            && java_lang_invoke_MethodTypeForm::vmlayout_offset_in_bytes() > 0) {
+          java_lang_invoke_MethodTypeForm::init_vmlayout(mtform, m());
+        }
+      }
     } else {
       non_cached_result = m;
     }
--- a/src/share/vm/classfile/vmSymbols.hpp	Fri May 06 12:12:29 2011 -0700
+++ b/src/share/vm/classfile/vmSymbols.hpp	Fri May 06 16:33:13 2011 -0700
@@ -341,6 +341,7 @@
   template(vmtarget_name,                             "vmtarget")                                 \
   template(vmentry_name,                              "vmentry")                                  \
   template(vmslots_name,                              "vmslots")                                  \
+  template(vmlayout_name,                             "vmlayout")                                 \
   template(vmindex_name,                              "vmindex")                                  \
   template(vmargslot_name,                            "vmargslot")                                \
   template(flags_name,                                "flags")                                    \
@@ -393,6 +394,7 @@
   template(void_signature,                            "V")                                        \
   template(byte_array_signature,                      "[B")                                       \
   template(char_array_signature,                      "[C")                                       \
+  template(int_array_signature,                       "[I")                                       \
   template(object_void_signature,                     "(Ljava/lang/Object;)V")                    \
   template(object_int_signature,                      "(Ljava/lang/Object;)I")                    \
   template(object_boolean_signature,                  "(Ljava/lang/Object;)Z")                    \
--- a/src/share/vm/code/codeBlob.cpp	Fri May 06 12:12:29 2011 -0700
+++ b/src/share/vm/code/codeBlob.cpp	Fri May 06 16:33:13 2011 -0700
@@ -152,6 +152,32 @@
 }
 
 
+void CodeBlob::trace_new_stub(CodeBlob* stub, const char* name1, const char* name2) {
+  // Do not hold the CodeCache lock during name formatting.
+  assert(!CodeCache_lock->owned_by_self(), "release CodeCache before registering the stub");
+
+  if (stub != NULL) {
+    char stub_id[256];
+    assert(strlen(name1) + strlen(name2) < sizeof(stub_id), "");
+    jio_snprintf(stub_id, sizeof(stub_id), "%s%s", name1, name2);
+    if (PrintStubCode) {
+      tty->print_cr("Decoding %s " INTPTR_FORMAT, stub_id, (intptr_t) stub);
+      Disassembler::decode(stub->code_begin(), stub->code_end());
+    }
+    Forte::register_stub(stub_id, stub->code_begin(), stub->code_end());
+
+    if (JvmtiExport::should_post_dynamic_code_generated()) {
+      const char* stub_name = name2;
+      if (name2[0] == '\0')  stub_name = name1;
+      JvmtiExport::post_dynamic_code_generated(stub_name, stub->code_begin(), stub->code_end());
+    }
+  }
+
+  // Track memory usage statistic after releasing CodeCache_lock
+  MemoryService::track_code_cache_memory_usage();
+}
+
+
 void CodeBlob::flush() {
   if (_oop_maps) {
     FREE_C_HEAP_ARRAY(unsigned char, _oop_maps);
@@ -312,23 +338,7 @@
     stub = new (size) RuntimeStub(stub_name, cb, size, frame_complete, frame_size, oop_maps, caller_must_gc_arguments);
   }
 
-  // Do not hold the CodeCache lock during name formatting.
-  if (stub != NULL) {
-    char stub_id[256];
-    jio_snprintf(stub_id, sizeof(stub_id), "RuntimeStub - %s", stub_name);
-    if (PrintStubCode) {
-      tty->print_cr("Decoding %s " INTPTR_FORMAT, stub_id, stub);
-      Disassembler::decode(stub->code_begin(), stub->code_end());
-    }
-    Forte::register_stub(stub_id, stub->code_begin(), stub->code_end());
-
-    if (JvmtiExport::should_post_dynamic_code_generated()) {
-      JvmtiExport::post_dynamic_code_generated(stub_name, stub->code_begin(), stub->code_end());
-    }
-  }
-
-  // Track memory usage statistic after releasing CodeCache_lock
-  MemoryService::track_code_cache_memory_usage();
+  trace_new_stub(stub, "RuntimeStub - ", stub_name);
 
   return stub;
 }
@@ -340,6 +350,50 @@
   return p;
 }
 
+// operator new shared by all singletons:
+void* SingletonBlob::operator new(size_t s, unsigned size) {
+  void* p = CodeCache::allocate(size);
+  if (!p) fatal("Initial size of CodeCache is too small");
+  return p;
+}
+
+
+//----------------------------------------------------------------------------------------------------
+// Implementation of RicochetBlob
+
+RicochetBlob::RicochetBlob(
+  CodeBuffer* cb,
+  int         size,
+  int         bounce_offset,
+  int         exception_offset,
+  int         frame_size
+)
+: SingletonBlob("RicochetBlob", cb, sizeof(RicochetBlob), size, frame_size, (OopMapSet*) NULL)
+{
+  _bounce_offset = bounce_offset;
+  _exception_offset = exception_offset;
+}
+
+
+RicochetBlob* RicochetBlob::create(
+  CodeBuffer* cb,
+  int         bounce_offset,
+  int         exception_offset,
+  int         frame_size)
+{
+  RicochetBlob* blob = NULL;
+  ThreadInVMfromUnknown __tiv;  // get to VM state in case we block on CodeCache_lock
+  {
+    MutexLockerEx mu(CodeCache_lock, Mutex::_no_safepoint_check_flag);
+    unsigned int size = allocation_size(cb, sizeof(RicochetBlob));
+    blob = new (size) RicochetBlob(cb, size, bounce_offset, exception_offset, frame_size);
+  }
+
+  trace_new_stub(blob, "RicochetBlob");
+
+  return blob;
+}
+
 
 //----------------------------------------------------------------------------------------------------
 // Implementation of DeoptimizationBlob
@@ -386,34 +440,12 @@
                                          frame_size);
   }
 
-  // Do not hold the CodeCache lock during name formatting.
-  if (blob != NULL) {
-    char blob_id[256];
-    jio_snprintf(blob_id, sizeof(blob_id), "DeoptimizationBlob@" PTR_FORMAT, blob->code_begin());
-    if (PrintStubCode) {
-      tty->print_cr("Decoding %s " INTPTR_FORMAT, blob_id, blob);
-      Disassembler::decode(blob->code_begin(), blob->code_end());
-    }
-    Forte::register_stub(blob_id, blob->code_begin(), blob->code_end());
-
-    if (JvmtiExport::should_post_dynamic_code_generated()) {
-      JvmtiExport::post_dynamic_code_generated("DeoptimizationBlob", blob->code_begin(), blob->code_end());
-    }
-  }
-
-  // Track memory usage statistic after releasing CodeCache_lock
-  MemoryService::track_code_cache_memory_usage();
+  trace_new_stub(blob, "DeoptimizationBlob");
 
   return blob;
 }
 
 
-void* DeoptimizationBlob::operator new(size_t s, unsigned size) {
-  void* p = CodeCache::allocate(size);
-  if (!p) fatal("Initial size of CodeCache is too small");
-  return p;
-}
-
 //----------------------------------------------------------------------------------------------------
 // Implementation of UncommonTrapBlob
 
@@ -441,33 +473,12 @@
     blob = new (size) UncommonTrapBlob(cb, size, oop_maps, frame_size);
   }
 
-  // Do not hold the CodeCache lock during name formatting.
-  if (blob != NULL) {
-    char blob_id[256];
-    jio_snprintf(blob_id, sizeof(blob_id), "UncommonTrapBlob@" PTR_FORMAT, blob->code_begin());
-    if (PrintStubCode) {
-      tty->print_cr("Decoding %s " INTPTR_FORMAT, blob_id, blob);
-      Disassembler::decode(blob->code_begin(), blob->code_end());
-    }
-    Forte::register_stub(blob_id, blob->code_begin(), blob->code_end());
-
-    if (JvmtiExport::should_post_dynamic_code_generated()) {
-      JvmtiExport::post_dynamic_code_generated("UncommonTrapBlob", blob->code_begin(), blob->code_end());
-    }
-  }
-
-  // Track memory usage statistic after releasing CodeCache_lock
-  MemoryService::track_code_cache_memory_usage();
+  trace_new_stub(blob, "UncommonTrapBlob");
 
   return blob;
 }
 
 
-void* UncommonTrapBlob::operator new(size_t s, unsigned size) {
-  void* p = CodeCache::allocate(size);
-  if (!p) fatal("Initial size of CodeCache is too small");
-  return p;
-}
 #endif // COMPILER2
 
 
@@ -498,33 +509,12 @@
     blob = new (size) ExceptionBlob(cb, size, oop_maps, frame_size);
   }
 
-  // We do not need to hold the CodeCache lock during name formatting
-  if (blob != NULL) {
-    char blob_id[256];
-    jio_snprintf(blob_id, sizeof(blob_id), "ExceptionBlob@" PTR_FORMAT, blob->code_begin());
-    if (PrintStubCode) {
-      tty->print_cr("Decoding %s " INTPTR_FORMAT, blob_id, blob);
-      Disassembler::decode(blob->code_begin(), blob->code_end());
-    }
-    Forte::register_stub(blob_id, blob->code_begin(), blob->code_end());
-
-    if (JvmtiExport::should_post_dynamic_code_generated()) {
-      JvmtiExport::post_dynamic_code_generated("ExceptionBlob", blob->code_begin(), blob->code_end());
-    }
-  }
-
-  // Track memory usage statistic after releasing CodeCache_lock
-  MemoryService::track_code_cache_memory_usage();
+  trace_new_stub(blob, "ExceptionBlob");
 
   return blob;
 }
 
 
-void* ExceptionBlob::operator new(size_t s, unsigned size) {
-  void* p = CodeCache::allocate(size);
-  if (!p) fatal("Initial size of CodeCache is too small");
-  return p;
-}
 #endif // COMPILER2
 
 
@@ -554,35 +544,12 @@
     blob = new (size) SafepointBlob(cb, size, oop_maps, frame_size);
   }
 
-  // We do not need to hold the CodeCache lock during name formatting.
-  if (blob != NULL) {
-    char blob_id[256];
-    jio_snprintf(blob_id, sizeof(blob_id), "SafepointBlob@" PTR_FORMAT, blob->code_begin());
-    if (PrintStubCode) {
-      tty->print_cr("Decoding %s " INTPTR_FORMAT, blob_id, blob);
-      Disassembler::decode(blob->code_begin(), blob->code_end());
-    }
-    Forte::register_stub(blob_id, blob->code_begin(), blob->code_end());
-
-    if (JvmtiExport::should_post_dynamic_code_generated()) {
-      JvmtiExport::post_dynamic_code_generated("SafepointBlob", blob->code_begin(), blob->code_end());
-    }
-  }
-
-  // Track memory usage statistic after releasing CodeCache_lock
-  MemoryService::track_code_cache_memory_usage();
+  trace_new_stub(blob, "SafepointBlob");
 
   return blob;
 }
 
 
-void* SafepointBlob::operator new(size_t s, unsigned size) {
-  void* p = CodeCache::allocate(size);
-  if (!p) fatal("Initial size of CodeCache is too small");
-  return p;
-}
-
-
 //----------------------------------------------------------------------------------------------------
 // Verification and printing
 
--- a/src/share/vm/code/codeBlob.hpp	Fri May 06 12:12:29 2011 -0700
+++ b/src/share/vm/code/codeBlob.hpp	Fri May 06 16:33:13 2011 -0700
@@ -35,6 +35,7 @@
 // Suptypes are:
 //   nmethod            : Compiled Java methods (include method that calls to native code)
 //   RuntimeStub        : Call to VM runtime methods
+//   RicochetBlob       : Used for blocking MethodHandle adapters
 //   DeoptimizationBlob : Used for deoptimizatation
 //   ExceptionBlob      : Used for stack unrolling
 //   SafepointBlob      : Used to handle illegal instruction exceptions
@@ -95,12 +96,13 @@
   void flush();
 
   // Typing
-  virtual bool is_buffer_blob() const                 { return false; }
-  virtual bool is_nmethod() const                     { return false; }
-  virtual bool is_runtime_stub() const                { return false; }
-  virtual bool is_deoptimization_stub() const         { return false; }
-  virtual bool is_uncommon_trap_stub() const          { return false; }
-  virtual bool is_exception_stub() const              { return false; }
+  virtual bool is_buffer_blob() const            { return false; }
+  virtual bool is_nmethod() const                { return false; }
+  virtual bool is_runtime_stub() const           { return false; }
+  virtual bool is_ricochet_stub() const          { return false; }
+  virtual bool is_deoptimization_stub() const    { return false; }
+  virtual bool is_uncommon_trap_stub() const     { return false; }
+  virtual bool is_exception_stub() const         { return false; }
   virtual bool is_safepoint_stub() const              { return false; }
   virtual bool is_adapter_blob() const                { return false; }
   virtual bool is_method_handles_adapter_blob() const { return false; }
@@ -182,6 +184,9 @@
   virtual void print_on(outputStream* st) const;
   virtual void print_value_on(outputStream* st) const;
 
+  // Deal with Disassembler, VTune, Forte, JvmtiExport, MemoryService.
+  static void trace_new_stub(CodeBlob* blob, const char* name1, const char* name2 = "");
+
   // Print the comment associated with offset on stream, if there is one
   virtual void print_block_comment(outputStream* stream, address block_begin) {
     intptr_t offset = (intptr_t)(block_begin - code_begin());
@@ -318,7 +323,11 @@
 
 class SingletonBlob: public CodeBlob {
   friend class VMStructs;
-  public:
+
+ protected:
+  void* operator new(size_t s, unsigned size);
+
+ public:
    SingletonBlob(
      const char* name,
      CodeBuffer* cb,
@@ -341,6 +350,50 @@
 
 
 //----------------------------------------------------------------------------------------------------
+// RicochetBlob
+// Holds an arbitrary argument list indefinitely while Java code executes recursively.
+
+class RicochetBlob: public SingletonBlob {
+  friend class VMStructs;
+ private:
+
+  int _bounce_offset;
+  int _exception_offset;
+
+  // Creation support
+  RicochetBlob(
+    CodeBuffer* cb,
+    int         size,
+    int         bounce_offset,
+    int         exception_offset,
+    int         frame_size
+  );
+
+ public:
+  // Creation
+  static RicochetBlob* create(
+    CodeBuffer* cb,
+    int         bounce_offset,
+    int         exception_offset,
+    int         frame_size
+  );
+
+  // Typing
+  bool is_ricochet_stub() const { return true; }
+
+  // GC for args
+  void preserve_callee_argument_oops(frame fr, const RegisterMap *reg_map, OopClosure* f) { /* Nothing to do */ }
+
+  address bounce_addr() const           { return code_begin() + _bounce_offset; }
+  address exception_addr() const        { return code_begin() + _exception_offset; }
+  bool returns_to_bounce_addr(address pc) const {
+    address bounce_pc = bounce_addr();
+    return (pc == bounce_pc || (pc + frame::pc_return_offset) == bounce_pc);
+  }
+};
+
+
+//----------------------------------------------------------------------------------------------------
 // DeoptimizationBlob
 
 class DeoptimizationBlob: public SingletonBlob {
@@ -363,8 +416,6 @@
     int         frame_size
   );
 
-  void* operator new(size_t s, unsigned size);
-
  public:
   // Creation
   static DeoptimizationBlob* create(
@@ -378,7 +429,6 @@
 
   // Typing
   bool is_deoptimization_stub() const { return true; }
-  const DeoptimizationBlob *as_deoptimization_stub() const { return this; }
   bool exception_address_is_unpack_entry(address pc) const {
     address unpack_pc = unpack();
     return (pc == unpack_pc || (pc + frame::pc_return_offset) == unpack_pc);
@@ -426,8 +476,6 @@
     int         frame_size
   );
 
-  void* operator new(size_t s, unsigned size);
-
  public:
   // Creation
   static UncommonTrapBlob* create(
@@ -458,8 +506,6 @@
     int         frame_size
   );
 
-  void* operator new(size_t s, unsigned size);
-
  public:
   // Creation
   static ExceptionBlob* create(
@@ -491,8 +537,6 @@
     int         frame_size
   );
 
-  void* operator new(size_t s, unsigned size);
-
  public:
   // Creation
   static SafepointBlob* create(
--- a/src/share/vm/code/codeCache.cpp	Fri May 06 12:12:29 2011 -0700
+++ b/src/share/vm/code/codeCache.cpp	Fri May 06 16:33:13 2011 -0700
@@ -796,6 +796,7 @@
   int nmethodCount = 0;
   int runtimeStubCount = 0;
   int adapterCount = 0;
+  int ricochetStubCount = 0;
   int deoptimizationStubCount = 0;
   int uncommonTrapStubCount = 0;
   int bufferBlobCount = 0;
@@ -840,6 +841,8 @@
       }
     } else if (cb->is_runtime_stub()) {
       runtimeStubCount++;
+    } else if (cb->is_ricochet_stub()) {
+      ricochetStubCount++;
     } else if (cb->is_deoptimization_stub()) {
       deoptimizationStubCount++;
     } else if (cb->is_uncommon_trap_stub()) {
@@ -876,6 +879,7 @@
   tty->print_cr("runtime_stubs: %d",runtimeStubCount);
   tty->print_cr("adapters: %d",adapterCount);
   tty->print_cr("buffer blobs: %d",bufferBlobCount);
+  tty->print_cr("ricochet_stubs: %d",ricochetStubCount);
   tty->print_cr("deoptimization_stubs: %d",deoptimizationStubCount);
   tty->print_cr("uncommon_traps: %d",uncommonTrapStubCount);
   tty->print_cr("\nnmethod size distribution (non-zombie java)");
--- a/src/share/vm/compiler/disassembler.cpp	Fri May 06 12:12:29 2011 -0700
+++ b/src/share/vm/compiler/disassembler.cpp	Fri May 06 16:33:13 2011 -0700
@@ -283,10 +283,10 @@
         st->print("Stub::%s", desc->name());
         if (desc->begin() != adr)
           st->print("%+d 0x%p",adr - desc->begin(), adr);
-        else if (WizardMode) st->print(" " INTPTR_FORMAT, adr);
+        else if (WizardMode) st->print(" " PTR_FORMAT, adr);
         return;
       }
-      st->print("Stub::<unknown> " INTPTR_FORMAT, adr);
+      st->print("Stub::<unknown> " PTR_FORMAT, adr);
       return;
     }
 
@@ -314,8 +314,8 @@
     }
   }
 
-  // Fall through to a simple numeral.
-  st->print(INTPTR_FORMAT, (intptr_t)adr);
+  // Fall through to a simple (hexadecimal) numeral.
+  st->print(PTR_FORMAT, adr);
 }
 
 void decode_env::print_insn_labels() {
@@ -326,7 +326,7 @@
     cb->print_block_comment(st, p);
   }
   if (_print_pc) {
-    st->print("  " INTPTR_FORMAT ": ", (intptr_t) p);
+    st->print("  " PTR_FORMAT ": ", p);
   }
 }
 
@@ -432,7 +432,7 @@
 void Disassembler::decode(CodeBlob* cb, outputStream* st) {
   if (!load_library())  return;
   decode_env env(cb, st);
-  env.output()->print_cr("Decoding CodeBlob " INTPTR_FORMAT, cb);
+  env.output()->print_cr("Decoding CodeBlob " PTR_FORMAT, cb);
   env.decode_instructions(cb->code_begin(), cb->code_end());
 }
 
@@ -446,7 +446,7 @@
 void Disassembler::decode(nmethod* nm, outputStream* st) {
   if (!load_library())  return;
   decode_env env(nm, st);
-  env.output()->print_cr("Decoding compiled method " INTPTR_FORMAT ":", nm);
+  env.output()->print_cr("Decoding compiled method " PTR_FORMAT ":", nm);
   env.output()->print_cr("Code:");
 
 #ifdef SHARK
@@ -478,9 +478,9 @@
     int offset = 0;
     for (address p = nm->consts_begin(); p < nm->consts_end(); p += 4, offset += 4) {
       if ((offset % 8) == 0) {
-        env.output()->print_cr("  " INTPTR_FORMAT " (offset: %4d): " PTR32_FORMAT "   " PTR64_FORMAT, (intptr_t) p, offset, *((int32_t*) p), *((int64_t*) p));
+        env.output()->print_cr("  " PTR_FORMAT " (offset: %4d): " PTR32_FORMAT "   " PTR64_FORMAT, p, offset, *((int32_t*) p), *((int64_t*) p));
       } else {
-        env.output()->print_cr("  " INTPTR_FORMAT " (offset: %4d): " PTR32_FORMAT,                    (intptr_t) p, offset, *((int32_t*) p));
+        env.output()->print_cr("  " PTR_FORMAT " (offset: %4d): " PTR32_FORMAT,                    p, offset, *((int32_t*) p));
       }
     }
   }
--- a/src/share/vm/prims/jvmtiTagMap.cpp	Fri May 06 12:12:29 2011 -0700
+++ b/src/share/vm/prims/jvmtiTagMap.cpp	Fri May 06 16:33:13 2011 -0700
@@ -3158,6 +3158,9 @@
         if (fr->is_entry_frame()) {
           last_entry_frame = fr;
         }
+        if (fr->is_ricochet_frame()) {
+          fr->oops_ricochet_do(blk, vf->register_map());
+        }
       }
 
       vf = vf->sender();
--- a/src/share/vm/prims/methodHandleWalk.cpp	Fri May 06 12:12:29 2011 -0700
+++ b/src/share/vm/prims/methodHandleWalk.cpp	Fri May 06 16:33:13 2011 -0700
@@ -409,6 +409,11 @@
         break;
       }
 
+      case java_lang_invoke_AdapterMethodHandle::OP_FOLD_ARGS: { //NYI, may GC
+        lose("unimplemented", CHECK_(empty));
+        break;
+      }
+
       case java_lang_invoke_AdapterMethodHandle::OP_SPREAD_ARGS: {
         klassOop array_klass_oop = NULL;
         BasicType array_type = java_lang_Class::as_BasicType(chain().adapter_arg_oop(),
@@ -452,9 +457,18 @@
                     Bytecodes::_invokestatic, false, 3, &arglist[0], CHECK_(empty));
 
         // Spread out the array elements.
-        Bytecodes::Code aload_op = Bytecodes::_aaload;
-        if (element_type != T_OBJECT) {
-          lose("primitive array NYI", CHECK_(empty));
+        Bytecodes::Code aload_op = Bytecodes::_nop;
+        switch (element_type) {
+        case T_INT:       aload_op = Bytecodes::_iaload; break;
+        case T_LONG:      aload_op = Bytecodes::_laload; break;
+        case T_FLOAT:     aload_op = Bytecodes::_faload; break;
+        case T_DOUBLE:    aload_op = Bytecodes::_daload; break;
+        case T_OBJECT:    aload_op = Bytecodes::_aaload; break;
+        case T_BOOLEAN:   // fall through:
+        case T_BYTE:      aload_op = Bytecodes::_baload; break;
+        case T_CHAR:      aload_op = Bytecodes::_caload; break;
+        case T_SHORT:     aload_op = Bytecodes::_saload; break;
+        default:          lose("primitive array NYI", CHECK_(empty));
         }
         int ap = arg_slot;
         for (int i = 0; i < spread_length; i++) {
@@ -467,11 +481,6 @@
         break;
       }
 
-      case java_lang_invoke_AdapterMethodHandle::OP_FLYBY: //NYI, runs Java code
-      case java_lang_invoke_AdapterMethodHandle::OP_RICOCHET: //NYI, runs Java code
-        lose("unimplemented", CHECK_(empty));
-        break;
-
       default:
         lose("bad adapter conversion", CHECK_(empty));
         break;
--- a/src/share/vm/prims/methodHandles.cpp	Fri May 06 12:12:29 2011 -0700
+++ b/src/share/vm/prims/methodHandles.cpp	Fri May 06 16:33:13 2011 -0700
@@ -66,8 +66,8 @@
   "adapter_drop_args",
   "adapter_collect_args",
   "adapter_spread_args",
-  "adapter_flyby",
-  "adapter_ricochet",
+  "adapter_fold_args",
+  "adapter_unused_13",
 
   // optimized adapter types:
   "adapter_swap_args/1",
@@ -83,9 +83,76 @@
   "adapter_prim_to_prim/f2d",
   "adapter_ref_to_prim/unboxi",
   "adapter_ref_to_prim/unboxl",
-  "adapter_spread_args/0",
-  "adapter_spread_args/1",
-  "adapter_spread_args/more",
+
+  // return value handlers for collect/filter/fold adapters:
+  "return/ref",
+  "return/int",
+  "return/long",
+  "return/float",
+  "return/double",
+  "return/void",
+  "return/S0/ref",
+  "return/S1/ref",
+  "return/S2/ref",
+  "return/S3/ref",
+  "return/S4/ref",
+  "return/S5/ref",
+  "return/any",
+
+  // spreading (array length cases 0, 1, ...)
+  "adapter_spread/0",
+  "adapter_spread/1/ref",
+  "adapter_spread/2/ref",
+  "adapter_spread/3/ref",
+  "adapter_spread/4/ref",
+  "adapter_spread/5/ref",
+  "adapter_spread/ref",
+  "adapter_spread/byte",
+  "adapter_spread/char",
+  "adapter_spread/short",
+  "adapter_spread/int",
+  "adapter_spread/long",
+  "adapter_spread/float",
+  "adapter_spread/double",
+
+  // blocking filter/collect conversions:
+  "adapter_collect/ref",
+  "adapter_collect/int",
+  "adapter_collect/long",
+  "adapter_collect/float",
+  "adapter_collect/double",
+  "adapter_collect/void",
+  "adapter_collect/0/ref",
+  "adapter_collect/1/ref",
+  "adapter_collect/2/ref",
+  "adapter_collect/3/ref",
+  "adapter_collect/4/ref",
+  "adapter_collect/5/ref",
+  "adapter_filter/S0/ref",
+  "adapter_filter/S1/ref",
+  "adapter_filter/S2/ref",
+  "adapter_filter/S3/ref",
+  "adapter_filter/S4/ref",
+  "adapter_filter/S5/ref",
+  "adapter_collect/2/S0/ref",
+  "adapter_collect/2/S1/ref",
+  "adapter_collect/2/S2/ref",
+  "adapter_collect/2/S3/ref",
+  "adapter_collect/2/S4/ref",
+  "adapter_collect/2/S5/ref",
+
+  // blocking fold conversions:
+  "adapter_fold/ref",
+  "adapter_fold/int",
+  "adapter_fold/long",
+  "adapter_fold/float",
+  "adapter_fold/double",
+  "adapter_fold/void",
+  "adapter_fold/1/ref",
+  "adapter_fold/2/ref",
+  "adapter_fold/3/ref",
+  "adapter_fold/4/ref",
+  "adapter_fold/5/ref",
 
   NULL
 };
@@ -96,13 +163,23 @@
 
 jobject MethodHandles::_raise_exception_method;
 
+address MethodHandles::_adapter_return_handlers[CONV_TYPE_MASK+1];
+
 #ifdef ASSERT
 bool MethodHandles::spot_check_entry_names() {
   assert(!strcmp(entry_name(_invokestatic_mh), "invokestatic"), "");
   assert(!strcmp(entry_name(_bound_ref_mh), "bound_ref"), "");
   assert(!strcmp(entry_name(_adapter_retype_only), "adapter_retype_only"), "");
-  assert(!strcmp(entry_name(_adapter_ricochet), "adapter_ricochet"), "");
+  assert(!strcmp(entry_name(_adapter_fold_args), "adapter_fold_args"), "");
   assert(!strcmp(entry_name(_adapter_opt_unboxi), "adapter_ref_to_prim/unboxi"), "");
+  assert(!strcmp(entry_name(_adapter_opt_spread_char), "adapter_spread/char"), "");
+  assert(!strcmp(entry_name(_adapter_opt_spread_double), "adapter_spread/double"), "");
+  assert(!strcmp(entry_name(_adapter_opt_collect_int), "adapter_collect/int"), "");
+  assert(!strcmp(entry_name(_adapter_opt_collect_0_ref), "adapter_collect/0/ref"), "");
+  assert(!strcmp(entry_name(_adapter_opt_collect_2_S3_ref), "adapter_collect/2/S3/ref"), "");
+  assert(!strcmp(entry_name(_adapter_opt_filter_S5_ref), "adapter_filter/S5/ref"), "");
+  assert(!strcmp(entry_name(_adapter_opt_fold_3_ref), "adapter_fold/3/ref"), "");
+  assert(!strcmp(entry_name(_adapter_opt_fold_void), "adapter_fold/void"), "");
   return true;
 }
 #endif
@@ -112,6 +189,9 @@
 // MethodHandles::generate_adapters
 //
 void MethodHandles::generate_adapters() {
+#ifdef TARGET_ARCH_NYI_6939861
+  if (FLAG_IS_DEFAULT(UseRicochetFrames))  UseRicochetFrames = false;
+#endif
   if (!EnableInvokeDynamic || SystemDictionary::MethodHandle_klass() == NULL)  return;
 
   assert(_adapter_code == NULL, "generate only once");
@@ -126,7 +206,6 @@
   g.generate();
 }
 
-
 //------------------------------------------------------------------------------
 // MethodHandlesAdapterGenerator::generate
 //
@@ -135,12 +214,62 @@
   for (MethodHandles::EntryKind ek = MethodHandles::_EK_FIRST;
        ek < MethodHandles::_EK_LIMIT;
        ek = MethodHandles::EntryKind(1 + (int)ek)) {
-    StubCodeMark mark(this, "MethodHandle", MethodHandles::entry_name(ek));
-    MethodHandles::generate_method_handle_stub(_masm, ek);
+    if (MethodHandles::ek_supported(ek)) {
+      StubCodeMark mark(this, "MethodHandle", MethodHandles::entry_name(ek));
+      MethodHandles::generate_method_handle_stub(_masm, ek);
+    }
   }
 }
 
 
+#ifdef TARGET_ARCH_NYI_6939861
+// these defs belong in methodHandles_<arch>.cpp
+frame MethodHandles::ricochet_frame_sender(const frame& fr, RegisterMap *map) {
+  ShouldNotCallThis();
+  return fr;
+}
+void MethodHandles::ricochet_frame_oops_do(const frame& fr, OopClosure* f, const RegisterMap* reg_map) {
+  ShouldNotCallThis();
+}
+#endif //TARGET_ARCH_NYI_6939861
+
+
+//------------------------------------------------------------------------------
+// MethodHandles::ek_supported
+//
+bool MethodHandles::ek_supported(MethodHandles::EntryKind ek) {
+  MethodHandles::EntryKind ek_orig = MethodHandles::ek_original_kind(ek);
+  switch (ek_orig) {
+  case _adapter_unused_13:
+    return false;  // not defined yet
+  case _adapter_prim_to_ref:
+    return UseRicochetFrames && conv_op_supported(java_lang_invoke_AdapterMethodHandle::OP_PRIM_TO_REF);
+  case _adapter_collect_args:
+    return UseRicochetFrames && conv_op_supported(java_lang_invoke_AdapterMethodHandle::OP_COLLECT_ARGS);
+  case _adapter_fold_args:
+    return UseRicochetFrames && conv_op_supported(java_lang_invoke_AdapterMethodHandle::OP_FOLD_ARGS);
+  case _adapter_opt_return_any:
+    return UseRicochetFrames;
+#ifdef TARGET_ARCH_NYI_6939861
+  // ports before 6939861 supported only three kinds of spread ops
+  case _adapter_spread_args:
+    // restrict spreads to three kinds:
+    switch (ek) {
+    case _adapter_opt_spread_0:
+    case _adapter_opt_spread_1:
+    case _adapter_opt_spread_more:
+      break;
+    default:
+      return false;
+      break;
+    }
+    break;
+#endif //TARGET_ARCH_NYI_6939861
+  }
+  return true;
+}
+
+
 void MethodHandles::set_enabled(bool z) {
   if (_enabled != z) {
     guarantee(z && EnableInvokeDynamic, "can only enable once, and only if -XX:+EnableInvokeDynamic");
@@ -1564,6 +1693,8 @@
   if (m->is_abstract()) { THROW(vmSymbols::java_lang_AbstractMethodError()); }
 
   java_lang_invoke_MethodHandle::init_vmslots(mh());
+  int vmargslot = m->size_of_parameters() - 1;
+  assert(java_lang_invoke_BoundMethodHandle::vmargslot(mh()) == vmargslot, "");
 
   if (VerifyMethodHandles) {
     verify_BoundMethodHandle_with_receiver(mh, m, CHECK);
@@ -1642,14 +1773,9 @@
     DEBUG_ONLY(int this_pushes = decode_MethodHandle_stack_pushes(mh()));
     if (direct_to_method) {
       assert(this_pushes == slots_pushed, "BMH pushes one or two stack slots");
-      assert(slots_pushed <= MethodHandlePushLimit, "");
     } else {
       int target_pushes = decode_MethodHandle_stack_pushes(target());
       assert(this_pushes == slots_pushed + target_pushes, "BMH stack motion must be correct");
-      // do not blow the stack; use a Java-based adapter if this limit is exceeded
-      // FIXME
-      // if (slots_pushed + target_pushes > MethodHandlePushLimit)
-      //   err = "too many bound parameters";
     }
   }
 
@@ -1672,10 +1798,11 @@
   }
 
   java_lang_invoke_MethodHandle::init_vmslots(mh());
+  int argslot = java_lang_invoke_BoundMethodHandle::vmargslot(mh());
 
   if (VerifyMethodHandles) {
     int insert_after = argnum - 1;
-    verify_vmargslot(mh, insert_after, java_lang_invoke_BoundMethodHandle::vmargslot(mh()), CHECK);
+    verify_vmargslot(mh, insert_after, argslot, CHECK);
     verify_vmslots(mh, CHECK);
   }
 
@@ -1769,6 +1896,7 @@
   Handle target(THREAD,    java_lang_invoke_AdapterMethodHandle::vmtarget(mh()));
   Handle src_mtype(THREAD, java_lang_invoke_MethodHandle::type(mh()));
   Handle dst_mtype(THREAD, java_lang_invoke_MethodHandle::type(target()));
+  Handle arg_mtype;
 
   const char* err = NULL;
 
@@ -1777,25 +1905,29 @@
     switch (ek) {
     case _adapter_check_cast:     // target type of cast
     case _adapter_ref_to_prim:    // wrapper type from which to unbox
-    case _adapter_prim_to_ref:    // wrapper type to box into
-    case _adapter_collect_args:   // array type to collect into
     case _adapter_spread_args:    // array type to spread from
       if (!java_lang_Class::is_instance(argument())
           || java_lang_Class::is_primitive(argument()))
         { err = "adapter requires argument of type java.lang.Class"; break; }
-      if (ek == _adapter_collect_args ||
-          ek == _adapter_spread_args) {
+      if (ek == _adapter_spread_args) {
         // Make sure it is a suitable collection type.  (Array, for now.)
         Klass* ak = Klass::cast(java_lang_Class::as_klassOop(argument()));
-        if (!ak->oop_is_objArray()) {
-          { err = "adapter requires argument of type java.lang.Class<Object[]>"; break; }
-        }
+        if (!ak->oop_is_array())
+          { err = "spread adapter requires argument representing an array class"; break; }
+        BasicType et = arrayKlass::cast(ak->as_klassOop())->element_type();
+        if (et != dest && stack_move <= 0)
+          { err = "spread adapter requires array class argument of correct type"; break; }
       }
       break;
-    case _adapter_flyby:
-    case _adapter_ricochet:
+    case _adapter_prim_to_ref:    // boxer MH to use
+    case _adapter_collect_args:   // method handle which collects the args
+    case _adapter_fold_args:      // method handle which collects the args
+      if (!UseRicochetFrames) {
+        { err = "box/collect/fold operators are not supported"; break; }
+      }
       if (!java_lang_invoke_MethodHandle::is_instance(argument()))
         { err = "MethodHandle adapter argument required"; break; }
+      arg_mtype = Handle(THREAD, java_lang_invoke_MethodHandle::type(argument()));
       break;
     default:
       if (argument.not_null())
@@ -1806,6 +1938,7 @@
 
   if (err == NULL) {
     // Check that the src/dest types are supplied if needed.
+    // Also check relevant parameter or return types.
     switch (ek) {
     case _adapter_check_cast:
       if (src != T_OBJECT || dest != T_OBJECT) {
@@ -1828,8 +1961,7 @@
       }
       break;
     case _adapter_prim_to_ref:
-      if (!is_java_primitive(src) || dest != T_OBJECT
-          || argument() != Klass::cast(SystemDictionary::box_klass(src))->java_mirror()) {
+      if (!is_java_primitive(src) || dest != T_OBJECT) {
         err = "adapter requires primitive src conversion subfield"; break;
       }
       break;
@@ -1840,14 +1972,12 @@
           err = "adapter requires src/dest conversion subfields for swap"; break;
         }
         int swap_size = type2size[src];
-        oop src_mtype  = java_lang_invoke_AdapterMethodHandle::type(mh());
-        oop dest_mtype = java_lang_invoke_AdapterMethodHandle::type(target());
-        int slot_limit = java_lang_invoke_AdapterMethodHandle::vmslots(target());
+        int slot_limit = java_lang_invoke_MethodHandle::vmslots(target());
         int src_slot   = argslot;
         int dest_slot  = vminfo;
         bool rotate_up = (src_slot > dest_slot); // upward rotation
         int src_arg    = argnum;
-        int dest_arg   = argument_slot_to_argnum(dest_mtype, dest_slot);
+        int dest_arg   = argument_slot_to_argnum(dst_mtype(), dest_slot);
         verify_vmargslot(mh, dest_arg, dest_slot, CHECK);
         if (!(dest_slot >= src_slot + swap_size) &&
             !(src_slot >= dest_slot + swap_size)) {
@@ -1855,8 +1985,8 @@
         } else if (ek == _adapter_swap_args && !(src_slot > dest_slot)) {
           err = "source of swap must be deeper in stack";
         } else if (ek == _adapter_swap_args) {
-          err = check_argument_type_change(java_lang_invoke_MethodType::ptype(src_mtype, dest_arg),
-                                           java_lang_invoke_MethodType::ptype(dest_mtype, src_arg),
+          err = check_argument_type_change(java_lang_invoke_MethodType::ptype(src_mtype(), dest_arg),
+                                           java_lang_invoke_MethodType::ptype(dst_mtype(), src_arg),
                                            dest_arg);
         } else if (ek == _adapter_rot_args) {
           if (rotate_up) {
@@ -1864,8 +1994,8 @@
             // rotate up: [dest_slot..src_slot-ss] --> [dest_slot+ss..src_slot]
             // that is:   [src_arg+1..dest_arg] --> [src_arg..dest_arg-1]
             for (int i = src_arg+1; i <= dest_arg && err == NULL; i++) {
-              err = check_argument_type_change(java_lang_invoke_MethodType::ptype(src_mtype, i),
-                                               java_lang_invoke_MethodType::ptype(dest_mtype, i-1),
+              err = check_argument_type_change(java_lang_invoke_MethodType::ptype(src_mtype(), i),
+                                               java_lang_invoke_MethodType::ptype(dst_mtype(), i-1),
                                                i);
             }
           } else { // rotate down
@@ -1873,28 +2003,54 @@
             // rotate down: [src_slot+ss..dest_slot] --> [src_slot..dest_slot-ss]
             // that is:     [dest_arg..src_arg-1] --> [dst_arg+1..src_arg]
             for (int i = dest_arg; i <= src_arg-1 && err == NULL; i++) {
-              err = check_argument_type_change(java_lang_invoke_MethodType::ptype(src_mtype, i),
-                                               java_lang_invoke_MethodType::ptype(dest_mtype, i+1),
+              err = check_argument_type_change(java_lang_invoke_MethodType::ptype(src_mtype(), i),
+                                               java_lang_invoke_MethodType::ptype(dst_mtype(), i+1),
                                                i);
             }
           }
         }
         if (err == NULL)
-          err = check_argument_type_change(java_lang_invoke_MethodType::ptype(src_mtype, src_arg),
-                                           java_lang_invoke_MethodType::ptype(dest_mtype, dest_arg),
+          err = check_argument_type_change(java_lang_invoke_MethodType::ptype(src_mtype(), src_arg),
+                                           java_lang_invoke_MethodType::ptype(dst_mtype(), dest_arg),
                                            src_arg);
       }
       break;
+    case _adapter_spread_args:
     case _adapter_collect_args:
-    case _adapter_spread_args:
+    case _adapter_fold_args:
       {
-        BasicType coll_type = (ek == _adapter_collect_args) ? dest : src;
-        BasicType elem_type = (ek == _adapter_collect_args) ? src : dest;
-        if (coll_type != T_OBJECT || elem_type != T_OBJECT) {
-          err = "adapter requires src/dest subfields"; break;
-          // later:
-          // - consider making coll be a primitive array
-          // - consider making coll be a heterogeneous collection
+        bool is_spread = (ek == _adapter_spread_args);
+        bool is_fold   = (ek == _adapter_fold_args);
+        BasicType coll_type = is_spread ? src : dest;
+        BasicType elem_type = is_spread ? dest : src;
+        // coll_type is type of args in collected form (or T_VOID if none)
+        // elem_type is common type of args in spread form (or T_VOID if missing or heterogeneous)
+        if (coll_type == 0 || elem_type == 0) {
+          err = "adapter requires src/dest subfields for spread or collect"; break;
+        }
+        if (is_spread && coll_type != T_OBJECT) {
+          err = "spread adapter requires object type for argument bundle"; break;
+        }
+        Handle spread_mtype = (is_spread ? dst_mtype : src_mtype);
+        int spread_slot = argslot;
+        int spread_arg  = argnum;
+        int slots_pushed = stack_move / stack_move_unit();
+        int coll_slot_count = type2size[coll_type];
+        int spread_slot_count = (is_spread ? slots_pushed : -slots_pushed) + coll_slot_count;
+        if (is_fold)  spread_slot_count = argument_slot_count(arg_mtype());
+        if (!is_spread) {
+          int init_slots = argument_slot_count(src_mtype());
+          int coll_slots = argument_slot_count(arg_mtype());
+          if (spread_slot_count > init_slots ||
+              spread_slot_count != coll_slots) {
+            err = "collect adapter has inconsistent arg counts"; break;
+          }
+          int next_slots = argument_slot_count(dst_mtype());
+          int unchanged_slots_in  = (init_slots - spread_slot_count);
+          int unchanged_slots_out = (next_slots - coll_slot_count - (is_fold ? spread_slot_count : 0));
+          if (unchanged_slots_in != unchanged_slots_out) {
+            err = "collect adapter continuation has inconsistent arg counts"; break;
+          }
         }
       }
       break;
@@ -1929,8 +2085,9 @@
       }
       break;
     case _adapter_collect_args:
-      if (slots_pushed > 1) {
-        err = "adapter requires conversion subfield slots_pushed <= 1";
+    case _adapter_fold_args:
+      if (slots_pushed > 2) {
+        err = "adapter requires conversion subfield slots_pushed <= 2";
       }
       break;
     case _adapter_spread_args:
@@ -1950,32 +2107,36 @@
   }
 
   if (err == NULL) {
-    // Make sure this adapter does not push too deeply.
+    // Make sure this adapter's stack pushing is accurately recorded.
     int slots_pushed = stack_move / stack_move_unit();
     int this_vmslots = java_lang_invoke_MethodHandle::vmslots(mh());
     int target_vmslots = java_lang_invoke_MethodHandle::vmslots(target());
+    int target_pushes = decode_MethodHandle_stack_pushes(target());
     if (slots_pushed != (target_vmslots - this_vmslots)) {
       err = "stack_move inconsistent with previous and current MethodType vmslots";
-    } else if (slots_pushed > 0)  {
-      // verify stack_move against MethodHandlePushLimit
-      int target_pushes = decode_MethodHandle_stack_pushes(target());
-      // do not blow the stack; use a Java-based adapter if this limit is exceeded
-      if (slots_pushed + target_pushes > MethodHandlePushLimit) {
-        err = "adapter pushes too many parameters";
+    } else {
+      int this_pushes = decode_MethodHandle_stack_pushes(mh());
+      if (slots_pushed + target_pushes != this_pushes) {
+        if (this_pushes == 0)
+          err = "adapter push count not initialized";
+        else
+          err = "adapter push count is wrong";
       }
     }
 
     // While we're at it, check that the stack motion decoder works:
-    DEBUG_ONLY(int target_pushes = decode_MethodHandle_stack_pushes(target()));
     DEBUG_ONLY(int this_pushes = decode_MethodHandle_stack_pushes(mh()));
     assert(this_pushes == slots_pushed + target_pushes, "AMH stack motion must be correct");
   }
 
   if (err == NULL && vminfo != 0) {
     switch (ek) {
-      case _adapter_swap_args:
-      case _adapter_rot_args:
-        break;                // OK
+    case _adapter_swap_args:
+    case _adapter_rot_args:
+    case _adapter_prim_to_ref:
+    case _adapter_collect_args:
+    case _adapter_fold_args:
+      break;                // OK
     default:
       err = "vminfo subfield is reserved to the JVM";
     }
@@ -2026,6 +2187,7 @@
   // adjust the adapter code to the internal EntryKind enumeration:
   EntryKind ek_orig = adapter_entry_kind(conv_op);
   EntryKind ek_opt  = ek_orig;  // may be optimized
+  EntryKind ek_try;             // temp
 
   // Finalize the vmtarget field (Java initialized it to null).
   if (!java_lang_invoke_MethodHandle::is_instance(target())) {
@@ -2034,17 +2196,23 @@
   }
   java_lang_invoke_AdapterMethodHandle::set_vmtarget(mh(), target());
 
-  if (VerifyMethodHandles) {
-    verify_AdapterMethodHandle(mh, argnum, CHECK);
-  }
-
   int stack_move = adapter_conversion_stack_move(conversion);
   BasicType src  = adapter_conversion_src_type(conversion);
   BasicType dest = adapter_conversion_dest_type(conversion);
   int vminfo     = adapter_conversion_vminfo(conversion); // should be zero
 
+  int slots_pushed = stack_move / stack_move_unit();
+
+  if (VerifyMethodHandles) {
+    verify_AdapterMethodHandle(mh, argnum, CHECK);
+  }
+
   const char* err = NULL;
 
+  if (!conv_op_supported(conv_op)) {
+    err = "adapter not yet implemented in the JVM";
+  }
+
   // Now it's time to finish the case analysis and pick a MethodHandleEntry.
   switch (ek_orig) {
   case _adapter_retype_only:
@@ -2077,7 +2245,7 @@
         }
         break;
       case 1 *4+ 2:
-        if (src == T_INT && dest == T_LONG) {
+        if ((src == T_INT || is_subword_type(src)) && dest == T_LONG) {
           ek_opt = _adapter_opt_i2l;
         } else if (src == T_FLOAT && dest == T_DOUBLE) {
           ek_opt = _adapter_opt_f2d;
@@ -2110,7 +2278,44 @@
     break;
 
   case _adapter_prim_to_ref:
-    goto throw_not_impl;        // allocates, hence could block
+    {
+      assert(UseRicochetFrames, "else don't come here");
+      // vminfo will be the location to insert the return value
+      vminfo = argslot;
+      ek_opt = _adapter_opt_collect_ref;
+      ensure_vmlayout_field(target, CHECK);
+      if (!OptimizeMethodHandles)  break;
+      switch (type2size[src]) {
+      case 1:
+        ek_try = EntryKind(_adapter_opt_filter_S0_ref + argslot);
+        if (ek_try < _adapter_opt_collect_LAST &&
+            ek_adapter_opt_collect_slot(ek_try) == argslot) {
+          assert(ek_adapter_opt_collect_count(ek_try) == 1 &&
+                 ek_adapter_opt_collect_type(ek_try) == T_OBJECT, "");
+          ek_opt = ek_try;
+          break;
+        }
+        // else downgrade to variable slot:
+        ek_opt = _adapter_opt_collect_1_ref;
+        break;
+      case 2:
+        ek_try = EntryKind(_adapter_opt_collect_2_S0_ref + argslot);
+        if (ek_try < _adapter_opt_collect_LAST &&
+            ek_adapter_opt_collect_slot(ek_try) == argslot) {
+          assert(ek_adapter_opt_collect_count(ek_try) == 2 &&
+                 ek_adapter_opt_collect_type(ek_try) == T_OBJECT, "");
+          ek_opt = ek_try;
+          break;
+        }
+        // else downgrade to variable slot:
+        ek_opt = _adapter_opt_collect_2_ref;
+        break;
+      default:
+        assert(false, "");
+        break;
+      }
+    }
+    break;
 
   case _adapter_swap_args:
   case _adapter_rot_args:
@@ -2136,29 +2341,180 @@
     }
     break;
 
-  case _adapter_collect_args:
-    goto throw_not_impl;        // allocates, hence could block
-
   case _adapter_spread_args:
     {
+#ifdef TARGET_ARCH_NYI_6939861
+      // ports before 6939861 supported only three kinds of spread ops
+      if (!UseRicochetFrames) {
+        int array_size   = slots_pushed + 1;
+        assert(array_size >= 0, "");
+        vminfo = array_size;
+        switch (array_size) {
+        case 0:   ek_opt = _adapter_opt_spread_0;       break;
+        case 1:   ek_opt = _adapter_opt_spread_1;       break;
+        default:  ek_opt = _adapter_opt_spread_more;    break;
+        }
+        break;
+      }
+#endif //TARGET_ARCH_NYI_6939861
       // vminfo will be the required length of the array
-      int slots_pushed = stack_move / stack_move_unit();
-      int array_size   = slots_pushed + 1;
-      assert(array_size >= 0, "");
+      int array_size = (slots_pushed + 1) / (type2size[dest] == 2 ? 2 : 1);
       vminfo = array_size;
-      switch (array_size) {
-      case 0:   ek_opt = _adapter_opt_spread_0;       break;
-      case 1:   ek_opt = _adapter_opt_spread_1;       break;
-      default:  ek_opt = _adapter_opt_spread_more;    break;
+      // general case
+      switch (dest) {
+      case T_BOOLEAN : // fall through to T_BYTE:
+      case T_BYTE    : ek_opt = _adapter_opt_spread_byte;    break;
+      case T_CHAR    : ek_opt = _adapter_opt_spread_char;    break;
+      case T_SHORT   : ek_opt = _adapter_opt_spread_short;   break;
+      case T_INT     : ek_opt = _adapter_opt_spread_int;     break;
+      case T_LONG    : ek_opt = _adapter_opt_spread_long;    break;
+      case T_FLOAT   : ek_opt = _adapter_opt_spread_float;   break;
+      case T_DOUBLE  : ek_opt = _adapter_opt_spread_double;  break;
+      case T_OBJECT  : ek_opt = _adapter_opt_spread_ref;     break;
+      case T_VOID    : if (array_size != 0)  goto throw_not_impl;
+                       ek_opt = _adapter_opt_spread_ref;     break;
+      default        : goto throw_not_impl;
       }
-      if ((vminfo & CONV_VMINFO_MASK) != vminfo)
-        goto throw_not_impl;    // overflow
+      assert(array_size == 0 ||  // it doesn't matter what the spreader is
+             (ek_adapter_opt_spread_count(ek_opt) == -1 &&
+              (ek_adapter_opt_spread_type(ek_opt) == dest ||
+               (ek_adapter_opt_spread_type(ek_opt) == T_BYTE && dest == T_BOOLEAN))),
+             err_msg("dest=%d ek_opt=%d", dest, ek_opt));
+
+      if (array_size <= 0) {
+        // since the general case does not handle length 0, this case is required:
+        ek_opt = _adapter_opt_spread_0;
+        break;
+      }
+      if (dest == T_OBJECT) {
+        ek_try = EntryKind(_adapter_opt_spread_1_ref - 1 + array_size);
+        if (ek_try < _adapter_opt_spread_LAST &&
+            ek_adapter_opt_spread_count(ek_try) == array_size) {
+          assert(ek_adapter_opt_spread_type(ek_try) == dest, "");
+          ek_opt = ek_try;
+          break;
+        }
+      }
+      break;
     }
     break;
 
-  case _adapter_flyby:
-  case _adapter_ricochet:
-    goto throw_not_impl;        // runs Java code, hence could block
+  case _adapter_collect_args:
+    {
+      assert(UseRicochetFrames, "else don't come here");
+      int elem_slots = argument_slot_count(
+                           java_lang_invoke_MethodHandle::type(
+                               java_lang_invoke_AdapterMethodHandle::argument(mh()) ) );
+      // vminfo will be the location to insert the return value
+      vminfo = argslot;
+      ensure_vmlayout_field(target, CHECK);
+
+      // general case:
+      switch (dest) {
+      default       : if (!is_subword_type(dest))  goto throw_not_impl;
+                    // else fall through:
+      case T_INT    : ek_opt = _adapter_opt_collect_int;     break;
+      case T_LONG   : ek_opt = _adapter_opt_collect_long;    break;
+      case T_FLOAT  : ek_opt = _adapter_opt_collect_float;   break;
+      case T_DOUBLE : ek_opt = _adapter_opt_collect_double;  break;
+      case T_OBJECT : ek_opt = _adapter_opt_collect_ref;     break;
+      case T_VOID   : ek_opt = _adapter_opt_collect_void;    break;
+      }
+      assert(ek_adapter_opt_collect_slot(ek_opt) == -1 &&
+             ek_adapter_opt_collect_count(ek_opt) == -1 &&
+             (ek_adapter_opt_collect_type(ek_opt) == dest ||
+              ek_adapter_opt_collect_type(ek_opt) == T_INT && is_subword_type(dest)),
+             "");
+
+      if (dest == T_OBJECT && elem_slots == 1 && OptimizeMethodHandles) {
+        // filter operation on a ref
+        ek_try = EntryKind(_adapter_opt_filter_S0_ref + argslot);
+        if (ek_try < _adapter_opt_collect_LAST &&
+            ek_adapter_opt_collect_slot(ek_try) == argslot) {
+          assert(ek_adapter_opt_collect_count(ek_try) == elem_slots &&
+                 ek_adapter_opt_collect_type(ek_try) == dest, "");
+          ek_opt = ek_try;
+          break;
+        }
+        ek_opt = _adapter_opt_collect_1_ref;
+        break;
+      }
+
+      if (dest == T_OBJECT && elem_slots == 2 && OptimizeMethodHandles) {
+        // filter of two arguments
+        ek_try = EntryKind(_adapter_opt_collect_2_S0_ref + argslot);
+        if (ek_try < _adapter_opt_collect_LAST &&
+            ek_adapter_opt_collect_slot(ek_try) == argslot) {
+          assert(ek_adapter_opt_collect_count(ek_try) == elem_slots &&
+                 ek_adapter_opt_collect_type(ek_try) == dest, "");
+          ek_opt = ek_try;
+          break;
+        }
+        ek_opt = _adapter_opt_collect_2_ref;
+        break;
+      }
+
+      if (dest == T_OBJECT && OptimizeMethodHandles) {
+        // try to use a fixed length adapter
+        ek_try = EntryKind(_adapter_opt_collect_0_ref + elem_slots);
+        if (ek_try < _adapter_opt_collect_LAST &&
+            ek_adapter_opt_collect_count(ek_try) == elem_slots) {
+          assert(ek_adapter_opt_collect_slot(ek_try) == -1 &&
+                 ek_adapter_opt_collect_type(ek_try) == dest, "");
+          ek_opt = ek_try;
+          break;
+        }
+      }
+
+      break;
+    }
+
+  case _adapter_fold_args:
+    {
+      assert(UseRicochetFrames, "else don't come here");
+      int elem_slots = argument_slot_count(
+                           java_lang_invoke_MethodHandle::type(
+                               java_lang_invoke_AdapterMethodHandle::argument(mh()) ) );
+      // vminfo will be the location to insert the return value
+      vminfo = argslot + elem_slots;
+      ensure_vmlayout_field(target, CHECK);
+
+      switch (dest) {
+      default       : if (!is_subword_type(dest))  goto throw_not_impl;
+                    // else fall through:
+      case T_INT    : ek_opt = _adapter_opt_fold_int;     break;
+      case T_LONG   : ek_opt = _adapter_opt_fold_long;    break;
+      case T_FLOAT  : ek_opt = _adapter_opt_fold_float;   break;
+      case T_DOUBLE : ek_opt = _adapter_opt_fold_double;  break;
+      case T_OBJECT : ek_opt = _adapter_opt_fold_ref;     break;
+      case T_VOID   : ek_opt = _adapter_opt_fold_void;    break;
+      }
+      assert(ek_adapter_opt_collect_slot(ek_opt) == -1 &&
+             ek_adapter_opt_collect_count(ek_opt) == -1 &&
+             (ek_adapter_opt_collect_type(ek_opt) == dest ||
+              ek_adapter_opt_collect_type(ek_opt) == T_INT && is_subword_type(dest)),
+             "");
+
+      if (dest == T_OBJECT && elem_slots == 0 && OptimizeMethodHandles) {
+        // if there are no args, just pretend it's a collect
+        ek_opt = _adapter_opt_collect_0_ref;
+        break;
+      }
+
+      if (dest == T_OBJECT && OptimizeMethodHandles) {
+        // try to use a fixed length adapter
+        ek_try = EntryKind(_adapter_opt_fold_1_ref - 1 + elem_slots);
+        if (ek_try < _adapter_opt_fold_LAST &&
+            ek_adapter_opt_collect_count(ek_try) == elem_slots) {
+          assert(ek_adapter_opt_collect_slot(ek_try) == -1 &&
+                 ek_adapter_opt_collect_type(ek_try) == dest, "");
+          ek_opt = ek_try;
+          break;
+        }
+      }
+
+      break;
+    }
 
   default:
     // should have failed much earlier; must be a missing case here
@@ -2166,11 +2522,20 @@
     // and fall through:
 
   throw_not_impl:
-    // FIXME: these adapters are NYI
-    err = "adapter not yet implemented in the JVM";
+    if (err == NULL)
+      err = "unknown adapter type";
     break;
   }
 
+  if (err != NULL && (vminfo & CONV_VMINFO_MASK) != vminfo) {
+    // should not happen, since vminfo is used to encode arg/slot indexes < 255
+    err = "vminfo overflow";
+  }
+
+  if (err != NULL && !have_entry(ek_opt)) {
+    err = "adapter stub for this kind of method handle is missing";
+  }
+
   if (err != NULL) {
     throw_InternalError_for_bad_conversion(conversion, err, THREAD);
     return;
@@ -2190,6 +2555,26 @@
   // Java code can publish it in global data structures.
 }
 
+void MethodHandles::ensure_vmlayout_field(Handle target, TRAPS) {
+  Handle mtype(THREAD, java_lang_invoke_MethodHandle::type(target()));
+  Handle mtform(THREAD, java_lang_invoke_MethodType::form(mtype()));
+  if (mtform.is_null()) { THROW(vmSymbols::java_lang_InternalError()); }
+  if (java_lang_invoke_MethodTypeForm::vmlayout_offset_in_bytes() > 0) {
+    if (java_lang_invoke_MethodTypeForm::vmlayout(mtform()) == NULL) {
+      // fill it in
+      Handle erased_mtype(THREAD, java_lang_invoke_MethodTypeForm::erasedType(mtform()));
+      TempNewSymbol erased_signature
+        = java_lang_invoke_MethodType::as_signature(erased_mtype(), /*intern:*/true, CHECK);
+      methodOop cookie
+        = SystemDictionary::find_method_handle_invoke(vmSymbols::invokeExact_name(),
+                                                      erased_signature,
+                                                      SystemDictionaryHandles::Object_klass(),
+                                                      THREAD);
+      java_lang_invoke_MethodTypeForm::init_vmlayout(mtform(), cookie);
+    }
+  }
+}
+
 //
 // Here are the native methods on sun.invoke.MethodHandleImpl.
 // They are the private interface between this JVM and the HotSpot-specific
@@ -2360,8 +2745,10 @@
 
 #ifndef PRODUCT
 #define EACH_NAMED_CON(template) \
-    template(MethodHandles,GC_JVM_PUSH_LIMIT) \
-    template(MethodHandles,GC_JVM_STACK_MOVE_UNIT) \
+  /* hold back this one until JDK stabilizes */ \
+  /* template(MethodHandles,GC_JVM_PUSH_LIMIT) */  \
+  /* hold back this one until JDK stabilizes */ \
+  /* template(MethodHandles,GC_JVM_STACK_MOVE_UNIT) */ \
     template(MethodHandles,ETF_HANDLE_OR_METHOD_NAME) \
     template(MethodHandles,ETF_DIRECT_HANDLE) \
     template(MethodHandles,ETF_METHOD_NAME) \
@@ -2385,9 +2772,8 @@
     template(java_lang_invoke_AdapterMethodHandle,OP_DROP_ARGS) \
     template(java_lang_invoke_AdapterMethodHandle,OP_COLLECT_ARGS) \
     template(java_lang_invoke_AdapterMethodHandle,OP_SPREAD_ARGS) \
-    template(java_lang_invoke_AdapterMethodHandle,OP_FLYBY) \
-    template(java_lang_invoke_AdapterMethodHandle,OP_RICOCHET) \
-    template(java_lang_invoke_AdapterMethodHandle,CONV_OP_LIMIT) \
+      /* hold back this one until JDK stabilizes */ \
+      /*template(java_lang_invoke_AdapterMethodHandle,CONV_OP_LIMIT)*/  \
     template(java_lang_invoke_AdapterMethodHandle,CONV_OP_MASK) \
     template(java_lang_invoke_AdapterMethodHandle,CONV_VMINFO_MASK) \
     template(java_lang_invoke_AdapterMethodHandle,CONV_VMINFO_SHIFT) \
--- a/src/share/vm/prims/methodHandles.hpp	Fri May 06 12:12:29 2011 -0700
+++ b/src/share/vm/prims/methodHandles.hpp	Fri May 06 16:33:13 2011 -0700
@@ -66,8 +66,8 @@
     _adapter_drop_args     = _adapter_mh_first + java_lang_invoke_AdapterMethodHandle::OP_DROP_ARGS,
     _adapter_collect_args  = _adapter_mh_first + java_lang_invoke_AdapterMethodHandle::OP_COLLECT_ARGS,
     _adapter_spread_args   = _adapter_mh_first + java_lang_invoke_AdapterMethodHandle::OP_SPREAD_ARGS,
-    _adapter_flyby         = _adapter_mh_first + java_lang_invoke_AdapterMethodHandle::OP_FLYBY,
-    _adapter_ricochet      = _adapter_mh_first + java_lang_invoke_AdapterMethodHandle::OP_RICOCHET,
+    _adapter_fold_args     = _adapter_mh_first + java_lang_invoke_AdapterMethodHandle::OP_FOLD_ARGS,
+    _adapter_unused_13     = _adapter_mh_first + 13,  //hole in the CONV_OP enumeration
     _adapter_mh_last       = _adapter_mh_first + java_lang_invoke_AdapterMethodHandle::CONV_OP_LIMIT - 1,
 
     // Optimized adapter types
@@ -93,10 +93,99 @@
     _adapter_opt_unboxi,
     _adapter_opt_unboxl,
 
-    // spreading (array length cases 0, 1, >=2)
-    _adapter_opt_spread_0,
-    _adapter_opt_spread_1,
-    _adapter_opt_spread_more,
+    // %% Maybe tame the following with a VM_SYMBOLS_DO type macro?
+
+    // how a blocking adapter returns (platform-dependent)
+    _adapter_opt_return_ref,
+    _adapter_opt_return_int,
+    _adapter_opt_return_long,
+    _adapter_opt_return_float,
+    _adapter_opt_return_double,
+    _adapter_opt_return_void,
+    _adapter_opt_return_S0_ref,  // return ref to S=0 (last slot)
+    _adapter_opt_return_S1_ref,  // return ref to S=1 (2nd-to-last slot)
+    _adapter_opt_return_S2_ref,
+    _adapter_opt_return_S3_ref,
+    _adapter_opt_return_S4_ref,
+    _adapter_opt_return_S5_ref,
+    _adapter_opt_return_any,     // dynamically select r/i/l/f/d
+    _adapter_opt_return_FIRST = _adapter_opt_return_ref,
+    _adapter_opt_return_LAST  = _adapter_opt_return_any,
+
+    // spreading (array length cases 0, 1, ...)
+    _adapter_opt_spread_0,       // spread empty array to N=0 arguments
+    _adapter_opt_spread_1_ref,   // spread Object[] to N=1 argument
+    _adapter_opt_spread_2_ref,   // spread Object[] to N=2 arguments
+    _adapter_opt_spread_3_ref,   // spread Object[] to N=3 arguments
+    _adapter_opt_spread_4_ref,   // spread Object[] to N=4 arguments
+    _adapter_opt_spread_5_ref,   // spread Object[] to N=5 arguments
+    _adapter_opt_spread_ref,     // spread Object[] to N arguments
+    _adapter_opt_spread_byte,    // spread byte[] or boolean[] to N arguments
+    _adapter_opt_spread_char,    // spread char[], etc., to N arguments
+    _adapter_opt_spread_short,   // spread short[], etc., to N arguments
+    _adapter_opt_spread_int,     // spread int[], short[], etc., to N arguments
+    _adapter_opt_spread_long,    // spread long[] to N arguments
+    _adapter_opt_spread_float,   // spread float[] to N arguments
+    _adapter_opt_spread_double,  // spread double[] to N arguments
+    _adapter_opt_spread_FIRST = _adapter_opt_spread_0,
+    _adapter_opt_spread_LAST  = _adapter_opt_spread_double,
+
+    // blocking filter/collect conversions
+    // These collect N arguments and replace them (at slot S) by a return value
+    // which is passed to the final target, along with the unaffected arguments.
+    // collect_{N}_{T} collects N arguments at any position into a T value
+    // collect_{N}_S{S}_{T} collects N arguments at slot S into a T value
+    // collect_{T} collects any number of arguments at any position
+    // filter_S{S}_{T} is the same as collect_1_S{S}_{T} (a unary collection)
+    // (collect_2 is also usable as a filter, with long or double arguments)
+    _adapter_opt_collect_ref,    // combine N arguments, replace with a reference
+    _adapter_opt_collect_int,    // combine N arguments, replace with an int, short, etc.
+    _adapter_opt_collect_long,   // combine N arguments, replace with a long
+    _adapter_opt_collect_float,  // combine N arguments, replace with a float
+    _adapter_opt_collect_double, // combine N arguments, replace with a double
+    _adapter_opt_collect_void,   // combine N arguments, replace with nothing
+    // if there is a small fixed number to push, do so without a loop:
+    _adapter_opt_collect_0_ref,  // collect N=0 arguments, insert a reference
+    _adapter_opt_collect_1_ref,  // collect N=1 argument, replace with a reference
+    _adapter_opt_collect_2_ref,  // combine N=2 arguments, replace with a reference
+    _adapter_opt_collect_3_ref,  // combine N=3 arguments, replace with a reference
+    _adapter_opt_collect_4_ref,  // combine N=4 arguments, replace with a reference
+    _adapter_opt_collect_5_ref,  // combine N=5 arguments, replace with a reference
+    // filters are an important special case because they never move arguments:
+    _adapter_opt_filter_S0_ref,  // filter N=1 argument at S=0, replace with a reference
+    _adapter_opt_filter_S1_ref,  // filter N=1 argument at S=1, replace with a reference
+    _adapter_opt_filter_S2_ref,  // filter N=1 argument at S=2, replace with a reference
+    _adapter_opt_filter_S3_ref,  // filter N=1 argument at S=3, replace with a reference
+    _adapter_opt_filter_S4_ref,  // filter N=1 argument at S=4, replace with a reference
+    _adapter_opt_filter_S5_ref,  // filter N=1 argument at S=5, replace with a reference
+    // these move arguments, but they are important for boxing
+    _adapter_opt_collect_2_S0_ref,  // combine last N=2 arguments, replace with a reference
+    _adapter_opt_collect_2_S1_ref,  // combine N=2 arguments at S=1, replace with a reference
+    _adapter_opt_collect_2_S2_ref,  // combine N=2 arguments at S=2, replace with a reference
+    _adapter_opt_collect_2_S3_ref,  // combine N=2 arguments at S=3, replace with a reference
+    _adapter_opt_collect_2_S4_ref,  // combine N=2 arguments at S=4, replace with a reference
+    _adapter_opt_collect_2_S5_ref,  // combine N=2 arguments at S=5, replace with a reference
+    _adapter_opt_collect_FIRST = _adapter_opt_collect_ref,
+    _adapter_opt_collect_LAST  = _adapter_opt_collect_2_S5_ref,
+
+    // blocking folding conversions
+    // these are like collects, but retain all the N arguments for the final target
+    //_adapter_opt_fold_0_ref,   // same as _adapter_opt_collect_0_ref
+    // fold_{N}_{T} processes N arguments at any position into a T value, which it inserts
+    // fold_{T} processes any number of arguments at any position
+    _adapter_opt_fold_ref,       // process N arguments, prepend a reference
+    _adapter_opt_fold_int,       // process N arguments, prepend an int, short, etc.
+    _adapter_opt_fold_long,      // process N arguments, prepend a long
+    _adapter_opt_fold_float,     // process N arguments, prepend a float
+    _adapter_opt_fold_double,    // process N arguments, prepend a double
+    _adapter_opt_fold_void,      // process N arguments, but leave the list unchanged
+    _adapter_opt_fold_1_ref,     // process N=1 argument, prepend a reference
+    _adapter_opt_fold_2_ref,     // process N=2 arguments, prepend a reference
+    _adapter_opt_fold_3_ref,     // process N=3 arguments, prepend a reference
+    _adapter_opt_fold_4_ref,     // process N=4 arguments, prepend a reference
+    _adapter_opt_fold_5_ref,     // process N=5 arguments, prepend a reference
+    _adapter_opt_fold_FIRST = _adapter_opt_fold_ref,
+    _adapter_opt_fold_LAST  = _adapter_opt_fold_5_ref,
 
     _EK_LIMIT,
     _EK_FIRST = 0
@@ -110,6 +199,7 @@
   enum {  // import java_lang_invoke_AdapterMethodHandle::CONV_OP_*
     CONV_OP_LIMIT         = java_lang_invoke_AdapterMethodHandle::CONV_OP_LIMIT,
     CONV_OP_MASK          = java_lang_invoke_AdapterMethodHandle::CONV_OP_MASK,
+    CONV_TYPE_MASK        = java_lang_invoke_AdapterMethodHandle::CONV_TYPE_MASK,
     CONV_VMINFO_MASK      = java_lang_invoke_AdapterMethodHandle::CONV_VMINFO_MASK,
     CONV_VMINFO_SHIFT     = java_lang_invoke_AdapterMethodHandle::CONV_VMINFO_SHIFT,
     CONV_OP_SHIFT         = java_lang_invoke_AdapterMethodHandle::CONV_OP_SHIFT,
@@ -123,6 +213,7 @@
   static MethodHandleEntry* _entries[_EK_LIMIT];
   static const char*        _entry_names[_EK_LIMIT+1];
   static jobject            _raise_exception_method;
+  static address            _adapter_return_handlers[CONV_TYPE_MASK+1];
 
   // Adapters.
   static MethodHandlesAdapterBlob* _adapter_code;
@@ -147,39 +238,195 @@
   }
 
   // Some adapter helper functions.
-  static void get_ek_bound_mh_info(EntryKind ek, BasicType& arg_type, int& arg_mask, int& arg_slots) {
+  static EntryKind ek_original_kind(EntryKind ek) {
+    if (ek <= _adapter_mh_last)  return ek;
     switch (ek) {
-    case _bound_int_mh        : // fall-thru
-    case _bound_int_direct_mh : arg_type = T_INT;    arg_mask = _INSERT_INT_MASK;  break;
-    case _bound_long_mh       : // fall-thru
-    case _bound_long_direct_mh: arg_type = T_LONG;   arg_mask = _INSERT_LONG_MASK; break;
-    case _bound_ref_mh        : // fall-thru
-    case _bound_ref_direct_mh : arg_type = T_OBJECT; arg_mask = _INSERT_REF_MASK;  break;
-    default: ShouldNotReachHere();
+    case _adapter_opt_swap_1:
+    case _adapter_opt_swap_2:
+      return _adapter_swap_args;
+    case _adapter_opt_rot_1_up:
+    case _adapter_opt_rot_1_down:
+    case _adapter_opt_rot_2_up:
+    case _adapter_opt_rot_2_down:
+      return _adapter_rot_args;
+    case _adapter_opt_i2i:
+    case _adapter_opt_l2i:
+    case _adapter_opt_d2f:
+    case _adapter_opt_i2l:
+    case _adapter_opt_f2d:
+      return _adapter_prim_to_prim;
+    case _adapter_opt_unboxi:
+    case _adapter_opt_unboxl:
+      return _adapter_ref_to_prim;
     }
-    arg_slots = type2size[arg_type];
+    if (ek >= _adapter_opt_spread_FIRST && ek <= _adapter_opt_spread_LAST)
+      return _adapter_spread_args;
+    if (ek >= _adapter_opt_collect_FIRST && ek <= _adapter_opt_collect_LAST)
+      return _adapter_collect_args;
+    if (ek >= _adapter_opt_fold_FIRST && ek <= _adapter_opt_fold_LAST)
+      return _adapter_fold_args;
+    if (ek >= _adapter_opt_return_FIRST && ek <= _adapter_opt_return_LAST)
+      return _adapter_opt_return_any;
+    assert(false, "oob");
+    return _EK_LIMIT;
   }
 
-  static void get_ek_adapter_opt_swap_rot_info(EntryKind ek, int& swap_bytes, int& rotate) {
-    int swap_slots = 0;
+  static bool ek_supported(MethodHandles::EntryKind ek);
+
+  static BasicType ek_bound_mh_arg_type(EntryKind ek) {
     switch (ek) {
-    case _adapter_opt_swap_1:     swap_slots = 1; rotate =  0; break;
-    case _adapter_opt_swap_2:     swap_slots = 2; rotate =  0; break;
-    case _adapter_opt_rot_1_up:   swap_slots = 1; rotate =  1; break;
-    case _adapter_opt_rot_1_down: swap_slots = 1; rotate = -1; break;
-    case _adapter_opt_rot_2_up:   swap_slots = 2; rotate =  1; break;
-    case _adapter_opt_rot_2_down: swap_slots = 2; rotate = -1; break;
-    default: ShouldNotReachHere();
+    case _bound_int_mh         : // fall-thru
+    case _bound_int_direct_mh  : return T_INT;
+    case _bound_long_mh        : // fall-thru
+    case _bound_long_direct_mh : return T_LONG;
+    default                    : return T_OBJECT;
     }
-    // Return the size of the stack slots to move in bytes.
-    swap_bytes = swap_slots * Interpreter::stackElementSize;
   }
 
-  static int get_ek_adapter_opt_spread_info(EntryKind ek) {
+  static int ek_adapter_opt_swap_slots(EntryKind ek) {
     switch (ek) {
-    case _adapter_opt_spread_0: return  0;
-    case _adapter_opt_spread_1: return  1;
-    default                   : return -1;
+    case _adapter_opt_swap_1        : return  1;
+    case _adapter_opt_swap_2        : return  2;
+    case _adapter_opt_rot_1_up      : return  1;
+    case _adapter_opt_rot_1_down    : return  1;
+    case _adapter_opt_rot_2_up      : return  2;
+    case _adapter_opt_rot_2_down    : return  2;
+    default : ShouldNotReachHere();   return -1;
+    }
+  }
+
+  static int ek_adapter_opt_swap_mode(EntryKind ek) {
+    switch (ek) {
+    case _adapter_opt_swap_1       : return  0;
+    case _adapter_opt_swap_2       : return  0;
+    case _adapter_opt_rot_1_up     : return  1;
+    case _adapter_opt_rot_1_down   : return -1;
+    case _adapter_opt_rot_2_up     : return  1;
+    case _adapter_opt_rot_2_down   : return -1;
+    default : ShouldNotReachHere();  return  0;
+    }
+  }
+
+  static int ek_adapter_opt_collect_count(EntryKind ek) {
+    assert(ek >= _adapter_opt_collect_FIRST && ek <= _adapter_opt_collect_LAST ||
+           ek >= _adapter_opt_fold_FIRST    && ek <= _adapter_opt_fold_LAST, "");
+    switch (ek) {
+    case _adapter_opt_collect_0_ref    : return  0;
+    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_fold_1_ref       :
+    case _adapter_opt_collect_1_ref    : return  1;
+    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_2_ref       :
+    case _adapter_opt_collect_2_ref    : return  2;
+    case _adapter_opt_fold_3_ref       :
+    case _adapter_opt_collect_3_ref    : return  3;
+    case _adapter_opt_fold_4_ref       :
+    case _adapter_opt_collect_4_ref    : return  4;
+    case _adapter_opt_fold_5_ref       :
+    case _adapter_opt_collect_5_ref    : return  5;
+    default                            : return -1;  // sentinel value for "variable"
+    }
+  }
+
+  static int ek_adapter_opt_collect_slot(EntryKind ek) {
+    assert(ek >= _adapter_opt_collect_FIRST && ek <= _adapter_opt_collect_LAST ||
+           ek >= _adapter_opt_fold_FIRST    && ek <= _adapter_opt_fold_LAST, "");
+    switch (ek) {
+    case _adapter_opt_collect_2_S0_ref  :
+    case _adapter_opt_filter_S0_ref     : return 0;
+    case _adapter_opt_collect_2_S1_ref  :
+    case _adapter_opt_filter_S1_ref     : return 1;
+    case _adapter_opt_collect_2_S2_ref  :
+    case _adapter_opt_filter_S2_ref     : return 2;
+    case _adapter_opt_collect_2_S3_ref  :
+    case _adapter_opt_filter_S3_ref     : return 3;
+    case _adapter_opt_collect_2_S4_ref  :
+    case _adapter_opt_filter_S4_ref     : return 4;
+    case _adapter_opt_collect_2_S5_ref  :
+    case _adapter_opt_filter_S5_ref     : return 5;
+    default                             : return -1;  // sentinel value for "variable"
+    }
+  }
+
+  static BasicType ek_adapter_opt_collect_type(EntryKind ek) {
+    assert(ek >= _adapter_opt_collect_FIRST && ek <= _adapter_opt_collect_LAST ||
+           ek >= _adapter_opt_fold_FIRST    && ek <= _adapter_opt_fold_LAST, "");
+    switch (ek) {
+    case _adapter_opt_fold_int          :
+    case _adapter_opt_collect_int       : return T_INT;
+    case _adapter_opt_fold_long         :
+    case _adapter_opt_collect_long      : return T_LONG;
+    case _adapter_opt_fold_float        :
+    case _adapter_opt_collect_float     : return T_FLOAT;
+    case _adapter_opt_fold_double       :
+    case _adapter_opt_collect_double    : return T_DOUBLE;
+    case _adapter_opt_fold_void         :
+    case _adapter_opt_collect_void      : return T_VOID;
+    default                             : return T_OBJECT;
+    }
+  }
+
+  static int ek_adapter_opt_return_slot(EntryKind ek) {
+    assert(ek >= _adapter_opt_return_FIRST && ek <= _adapter_opt_return_LAST, "");
+    switch (ek) {
+    case _adapter_opt_return_S0_ref : return 0;
+    case _adapter_opt_return_S1_ref : return 1;
+    case _adapter_opt_return_S2_ref : return 2;
+    case _adapter_opt_return_S3_ref : return 3;
+    case _adapter_opt_return_S4_ref : return 4;
+    case _adapter_opt_return_S5_ref : return 5;
+    default                         : return -1;  // sentinel value for "variable"
+    }
+  }
+
+  static BasicType ek_adapter_opt_return_type(EntryKind ek) {
+    assert(ek >= _adapter_opt_return_FIRST && ek <= _adapter_opt_return_LAST, "");
+    switch (ek) {
+    case _adapter_opt_return_int    : return T_INT;
+    case _adapter_opt_return_long   : return T_LONG;
+    case _adapter_opt_return_float  : return T_FLOAT;
+    case _adapter_opt_return_double : return T_DOUBLE;
+    case _adapter_opt_return_void   : return T_VOID;
+    case _adapter_opt_return_any    : return T_CONFLICT;  // sentinel value for "variable"
+    default                         : return T_OBJECT;
+    }
+  }
+
+  static int ek_adapter_opt_spread_count(EntryKind ek) {
+    assert(ek >= _adapter_opt_spread_FIRST && ek <= _adapter_opt_spread_LAST, "");
+    switch (ek) {
+    case _adapter_opt_spread_0     : return  0;
+    case _adapter_opt_spread_1_ref : return  1;
+    case _adapter_opt_spread_2_ref : return  2;
+    case _adapter_opt_spread_3_ref : return  3;
+    case _adapter_opt_spread_4_ref : return  4;
+    case _adapter_opt_spread_5_ref : return  5;
+    default                        : return -1;  // sentinel value for "variable"
+    }
+  }
+
+  static BasicType ek_adapter_opt_spread_type(EntryKind ek) {
+    assert(ek >= _adapter_opt_spread_FIRST && ek <= _adapter_opt_spread_LAST, "");
+    switch (ek) {
+    // (there is no _adapter_opt_spread_boolean; we use byte)
+    case _adapter_opt_spread_byte   : return T_BYTE;
+    case _adapter_opt_spread_char   : return T_CHAR;
+    case _adapter_opt_spread_short  : return T_SHORT;
+    case _adapter_opt_spread_int    : return T_INT;
+    case _adapter_opt_spread_long   : return T_LONG;
+    case _adapter_opt_spread_float  : return T_FLOAT;
+    case _adapter_opt_spread_double : return T_DOUBLE;
+    default                         : return T_OBJECT;
     }
   }
 
@@ -228,12 +475,21 @@
   // Bit mask of conversion_op values.  May vary by platform.
   static int adapter_conversion_ops_supported_mask();
 
+  static bool conv_op_supported(int conv_op) {
+    assert(conv_op_valid(conv_op), "");
+    return ((adapter_conversion_ops_supported_mask() & nth_bit(conv_op)) != 0);
+  }
+
   // Offset in words that the interpreter stack pointer moves when an argument is pushed.
   // The stack_move value must always be a multiple of this.
   static int stack_move_unit() {
     return frame::interpreter_frame_expression_stack_direction() * Interpreter::stackElementWords;
   }
 
+  // Adapter frame traversal.  (Implementation-specific.)
+  static frame ricochet_frame_sender(const frame& fr, RegisterMap* reg_map);
+  static void ricochet_frame_oops_do(const frame& fr, OopClosure* blk, const RegisterMap* reg_map);
+
   enum { CONV_VMINFO_SIGN_FLAG = 0x80 };
   // Shift values for prim-to-prim conversions.
   static int adapter_prim_to_prim_subword_vminfo(BasicType dest) {
@@ -429,6 +685,7 @@
 
   // Fill in the fields of an AdapterMethodHandle mh.  (MH.type must be pre-filled.)
   static void init_AdapterMethodHandle(Handle mh, Handle target, int argnum, TRAPS);
+  static void ensure_vmlayout_field(Handle target, TRAPS);
 
 #ifdef ASSERT
   static bool spot_check_entry_names();
@@ -448,12 +705,54 @@
     return same_basic_type_for_arguments(src, dst, raw, true);
   }
 
-  enum {                        // arg_mask values
+  static Symbol* convert_to_signature(oop type_str, bool polymorphic, TRAPS);
+
+#ifdef TARGET_ARCH_x86
+# include "methodHandles_x86.hpp"
+#endif
+#ifdef TARGET_ARCH_sparc
+#define TARGET_ARCH_NYI_6939861 1 //FIXME
+//# include "methodHandles_sparc.hpp"
+#endif
+#ifdef TARGET_ARCH_zero
+#define TARGET_ARCH_NYI_6939861 1 //FIXME
+//# include "methodHandles_zero.hpp"
+#endif
+#ifdef TARGET_ARCH_arm
+#define TARGET_ARCH_NYI_6939861 1 //FIXME
+//# include "methodHandles_arm.hpp"
+#endif
+#ifdef TARGET_ARCH_ppc
+#define TARGET_ARCH_NYI_6939861 1 //FIXME
+//# include "methodHandles_ppc.hpp"
+#endif
+
+#ifdef TARGET_ARCH_NYI_6939861
+  // Here are some backward compatible declarations until the 6939861 ports are updated.
+  #define _adapter_flyby    (_EK_LIMIT + 10)
+  #define _adapter_ricochet (_EK_LIMIT + 11)
+  #define _adapter_opt_spread_1    _adapter_opt_spread_1_ref
+  #define _adapter_opt_spread_more _adapter_opt_spread_ref
+  enum {
     _INSERT_NO_MASK   = -1,
     _INSERT_REF_MASK  = 0,
     _INSERT_INT_MASK  = 1,
     _INSERT_LONG_MASK = 3
   };
+  static void get_ek_bound_mh_info(EntryKind ek, BasicType& arg_type, int& arg_mask, int& arg_slots) {
+    arg_type = ek_bound_mh_arg_type(ek);
+    arg_mask = 0;
+    arg_slots = type2size[arg_type];;
+  }
+  static void get_ek_adapter_opt_swap_rot_info(EntryKind ek, int& swap_bytes, int& rotate) {
+    int swap_slots = ek_adapter_opt_swap_slots(ek);
+    rotate = ek_adapter_opt_swap_mode(ek);
+    swap_bytes = swap_slots * Interpreter::stackElementSize;
+  }
+  static int get_ek_adapter_opt_spread_info(EntryKind ek) {
+    return ek_adapter_opt_spread_count(ek);
+  }
+
   static void insert_arg_slots(MacroAssembler* _masm,
                                RegisterOrConstant arg_slots,
                                int arg_mask,
@@ -466,8 +765,7 @@
                                Register temp_reg, Register temp2_reg, Register temp3_reg = noreg);
 
   static void trace_method_handle(MacroAssembler* _masm, const char* adaptername) PRODUCT_RETURN;
-
-  static Symbol* convert_to_signature(oop type_str, bool polymorphic, TRAPS);
+#endif //TARGET_ARCH_NYI_6939861
 };
 
 
--- a/src/share/vm/runtime/frame.cpp	Fri May 06 12:12:29 2011 -0700
+++ b/src/share/vm/runtime/frame.cpp	Fri May 06 16:33:13 2011 -0700
@@ -33,6 +33,7 @@
 #include "oops/methodOop.hpp"
 #include "oops/oop.inline.hpp"
 #include "oops/oop.inline2.hpp"
+#include "prims/methodHandles.hpp"
 #include "runtime/frame.inline.hpp"
 #include "runtime/handles.inline.hpp"
 #include "runtime/javaCalls.hpp"
@@ -169,6 +170,11 @@
 }
 
 // type testers
+bool frame::is_ricochet_frame() const {
+  RicochetBlob* rcb = SharedRuntime::ricochet_blob();
+  return (_cb == rcb && rcb != NULL && rcb->returns_to_bounce_addr(_pc));
+}
+
 bool frame::is_deoptimized_frame() const {
   assert(_deopt_state != unknown, "not answerable");
   return _deopt_state == is_deoptimized;
@@ -341,12 +347,18 @@
 
 frame frame::real_sender(RegisterMap* map) const {
   frame result = sender(map);
-  while (result.is_runtime_frame()) {
+  while (result.is_runtime_frame() ||
+         result.is_ricochet_frame()) {
     result = result.sender(map);
   }
   return result;
 }
 
+frame frame::sender_for_ricochet_frame(RegisterMap* map) const {
+  assert(is_ricochet_frame(), "");
+  return MethodHandles::ricochet_frame_sender(*this, map);
+}
+
 // Note: called by profiler - NOT for current thread
 frame frame::profile_find_Java_sender_frame(JavaThread *thread) {
 // If we don't recognize this frame, walk back up the stack until we do
@@ -529,6 +541,7 @@
 const char* frame::print_name() const {
   if (is_native_frame())      return "Native";
   if (is_interpreted_frame()) return "Interpreted";
+  if (is_ricochet_frame())    return "Ricochet";
   if (is_compiled_frame()) {
     if (is_deoptimized_frame()) return "Deoptimized";
     return "Compiled";
@@ -715,6 +728,8 @@
       st->print("v  ~RuntimeStub::%s", ((RuntimeStub *)_cb)->name());
     } else if (_cb->is_deoptimization_stub()) {
       st->print("v  ~DeoptimizationBlob");
+    } else if (_cb->is_ricochet_stub()) {
+      st->print("v  ~RichochetBlob");
     } else if (_cb->is_exception_stub()) {
       st->print("v  ~ExceptionBlob");
     } else if (_cb->is_safepoint_stub()) {
@@ -978,6 +993,9 @@
 
 void frame::oops_code_blob_do(OopClosure* f, CodeBlobClosure* cf, const RegisterMap* reg_map) {
   assert(_cb != NULL, "sanity check");
+  if (_cb == SharedRuntime::ricochet_blob()) {
+    oops_ricochet_do(f, reg_map);
+  }
   if (_cb->oop_maps() != NULL) {
     OopMapSet::oops_do(this, reg_map, f);
 
@@ -996,6 +1014,11 @@
     cf->do_code_blob(_cb);
 }
 
+void frame::oops_ricochet_do(OopClosure* f, const RegisterMap* map) {
+  assert(is_ricochet_frame(), "");
+  MethodHandles::ricochet_frame_oops_do(*this, f, map);
+}
+
 class CompiledArgumentOopFinder: public SignatureInfo {
  protected:
   OopClosure*     _f;
--- a/src/share/vm/runtime/frame.hpp	Fri May 06 12:12:29 2011 -0700
+++ b/src/share/vm/runtime/frame.hpp	Fri May 06 16:33:13 2011 -0700
@@ -135,6 +135,7 @@
   bool is_interpreted_frame()    const;
   bool is_java_frame()           const;
   bool is_entry_frame()          const;             // Java frame called from C?
+  bool is_ricochet_frame()       const;
   bool is_native_frame()         const;
   bool is_runtime_frame()        const;
   bool is_compiled_frame()       const;
@@ -175,6 +176,7 @@
   // Helper methods for better factored code in frame::sender
   frame sender_for_compiled_frame(RegisterMap* map) const;
   frame sender_for_entry_frame(RegisterMap* map) const;
+  frame sender_for_ricochet_frame(RegisterMap* map) const;
   frame sender_for_interpreter_frame(RegisterMap* map) const;
   frame sender_for_native_frame(RegisterMap* map) const;
 
@@ -400,6 +402,7 @@
   // Oops-do's
   void oops_compiled_arguments_do(Symbol* signature, bool has_receiver, const RegisterMap* reg_map, OopClosure* f);
   void oops_interpreted_do(OopClosure* f, const RegisterMap* map, bool query_oop_map_cache = true);
+  void oops_ricochet_do(OopClosure* f, const RegisterMap* map);
 
  private:
   void oops_interpreted_arguments_do(Symbol* signature, bool has_receiver, OopClosure* f);
--- a/src/share/vm/runtime/globals.hpp	Fri May 06 12:12:29 2011 -0700
+++ b/src/share/vm/runtime/globals.hpp	Fri May 06 16:33:13 2011 -0700
@@ -3708,6 +3708,10 @@
   diagnostic(bool, OptimizeMethodHandles, true,                             \
           "when constructing method handles, try to improve them")          \
                                                                             \
+  diagnostic(bool, UseRicochetFrames, true,                                 \
+          "use ricochet stack frames for method handle combination, "       \
+          "if the platform supports them")                                  \
+                                                                            \
   experimental(bool, TrustFinalNonStaticFields, false,                      \
           "trust final non-static declarations for constant folding")       \
                                                                             \
--- a/src/share/vm/runtime/sharedRuntime.cpp	Fri May 06 12:12:29 2011 -0700
+++ b/src/share/vm/runtime/sharedRuntime.cpp	Fri May 06 16:33:13 2011 -0700
@@ -88,6 +88,8 @@
 HS_DTRACE_PROBE_DECL7(hotspot, method__return, int,
                       char*, int, char*, int, char*, int);
 
+RicochetBlob*      SharedRuntime::_ricochet_blob = NULL;
+
 // Implementation of SharedRuntime
 
 #ifndef PRODUCT
@@ -460,6 +462,10 @@
   if (Interpreter::contains(return_address)) {
     return Interpreter::rethrow_exception_entry();
   }
+  // Ricochet frame unwind code
+  if (SharedRuntime::ricochet_blob() != NULL && SharedRuntime::ricochet_blob()->returns_to_bounce_addr(return_address)) {
+    return SharedRuntime::ricochet_blob()->exception_addr();
+  }
 
   guarantee(blob == NULL || !blob->is_runtime_stub(), "caller should have skipped stub");
   guarantee(!VtableStubs::contains(return_address), "NULL exceptions in vtables should have been handled already!");
@@ -1174,6 +1180,7 @@
   assert(stub_frame.is_runtime_frame(), "sanity check");
   frame caller_frame = stub_frame.sender(&reg_map);
   assert(!caller_frame.is_interpreted_frame() && !caller_frame.is_entry_frame(), "unexpected frame");
+  assert(!caller_frame.is_ricochet_frame(), "unexpected frame");
 #endif /* ASSERT */
 
   methodHandle callee_method;
@@ -1222,6 +1229,7 @@
 
   if (caller_frame.is_interpreted_frame() ||
       caller_frame.is_entry_frame()       ||
+      caller_frame.is_ricochet_frame()    ||
       is_mh_invoke_via_adapter) {
     methodOop callee = thread->callee_target();
     guarantee(callee != NULL && callee->is_method(), "bad handshake");
--- a/src/share/vm/runtime/sharedRuntime.hpp	Fri May 06 12:12:29 2011 -0700
+++ b/src/share/vm/runtime/sharedRuntime.hpp	Fri May 06 16:33:13 2011 -0700
@@ -58,6 +58,8 @@
   static RuntimeStub* _resolve_virtual_call_blob;
   static RuntimeStub* _resolve_static_call_blob;
 
+  static RicochetBlob* _ricochet_blob;
+
   static SafepointBlob* _polling_page_safepoint_handler_blob;
   static SafepointBlob* _polling_page_return_handler_blob;
 #ifdef COMPILER2
@@ -213,6 +215,16 @@
     return _resolve_static_call_blob->entry_point();
   }
 
+  static RicochetBlob* ricochet_blob() {
+#ifdef X86
+    // Currently only implemented on x86
+    assert(!EnableInvokeDynamic || _ricochet_blob != NULL, "oops");
+#endif
+    return _ricochet_blob;
+  }
+
+  static void generate_ricochet_blob();
+
   static SafepointBlob* polling_page_return_handler_blob()     { return _polling_page_return_handler_blob; }
   static SafepointBlob* polling_page_safepoint_handler_blob()  { return _polling_page_safepoint_handler_blob; }
 
--- a/src/share/vm/services/heapDumper.cpp	Fri May 06 12:12:29 2011 -0700
+++ b/src/share/vm/services/heapDumper.cpp	Fri May 06 16:33:13 2011 -0700
@@ -1649,6 +1649,9 @@
         if (fr->is_entry_frame()) {
           last_entry_frame = fr;
         }
+        if (fr->is_ricochet_frame()) {
+          fr->oops_ricochet_do(&blk, vf->register_map());
+        }
       }
       vf = vf->sender();
     }