1 /*
   2  * Copyright (c) 1997, 2011, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 #include "precompiled.hpp"
  26 #include "interpreter/interpreter.hpp"
  27 #include "memory/allocation.inline.hpp"
  28 #include "prims/methodHandles.hpp"
  29 
  30 #define __ _masm->
  31 
  32 #ifdef PRODUCT
  33 #define BLOCK_COMMENT(str) /* nothing */
  34 #else
  35 #define BLOCK_COMMENT(str) __ block_comment(str)
  36 #endif
  37 
  38 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
  39 
  40 address MethodHandleEntry::start_compiled_entry(MacroAssembler* _masm,
  41                                                 address interpreted_entry) {
  42   // Just before the actual machine code entry point, allocate space
  43   // for a MethodHandleEntry::Data record, so that we can manage everything
  44   // from one base pointer.
  45   __ align(wordSize);
  46   address target = __ pc() + sizeof(Data);
  47   while (__ pc() < target) {
  48     __ nop();
  49     __ align(wordSize);
  50   }
  51 
  52   MethodHandleEntry* me = (MethodHandleEntry*) __ pc();
  53   me->set_end_address(__ pc());         // set a temporary end_address
  54   me->set_from_interpreted_entry(interpreted_entry);
  55   me->set_type_checking_entry(NULL);
  56 
  57   return (address) me;
  58 }
  59 
  60 MethodHandleEntry* MethodHandleEntry::finish_compiled_entry(MacroAssembler* _masm,
  61                                                 address start_addr) {
  62   MethodHandleEntry* me = (MethodHandleEntry*) start_addr;
  63   assert(me->end_address() == start_addr, "valid ME");
  64 
  65   // Fill in the real end_address:
  66   __ align(wordSize);
  67   me->set_end_address(__ pc());
  68 
  69   return me;
  70 }
  71 
  72 // stack walking support
  73 
  74 frame MethodHandles::ricochet_frame_sender(const frame& fr, RegisterMap *map) {
  75   RicochetFrame* f = RicochetFrame::from_frame(fr);
  76   if (map->update_map())
  77     frame::update_map_with_saved_link(map, &f->_sender_link);
  78   return frame(f->extended_sender_sp(), f->exact_sender_sp(), f->sender_link(), f->sender_pc());
  79 }
  80 
  81 void MethodHandles::ricochet_frame_oops_do(const frame& fr, OopClosure* blk, const RegisterMap* reg_map) {
  82   RicochetFrame* f = RicochetFrame::from_frame(fr);
  83 
  84   // pick up the argument type descriptor:
  85   Thread* thread = Thread::current();
  86   Handle cookie(thread, f->compute_saved_args_layout(true, true));
  87 
  88   // process fixed part
  89   blk->do_oop((oop*)f->saved_target_addr());
  90   blk->do_oop((oop*)f->saved_args_layout_addr());
  91 
  92   // process variable arguments:
  93   if (cookie.is_null())  return;  // no arguments to describe
  94 
  95   // the cookie is actually the invokeExact method for my target
  96   // his argument signature is what I'm interested in
  97   assert(cookie->is_method(), "");
  98   methodHandle invoker(thread, methodOop(cookie()));
  99   assert(invoker->name() == vmSymbols::invokeExact_name(), "must be this kind of method");
 100   assert(!invoker->is_static(), "must have MH argument");
 101   int slot_count = invoker->size_of_parameters();
 102   assert(slot_count >= 1, "must include 'this'");
 103   intptr_t* base = f->saved_args_base();
 104   intptr_t* retval = NULL;
 105   if (f->has_return_value_slot())
 106     retval = f->return_value_slot_addr();
 107   int slot_num = slot_count;
 108   intptr_t* loc = &base[slot_num -= 1];
 109   //blk->do_oop((oop*) loc);   // original target, which is irrelevant
 110   int arg_num = 0;
 111   for (SignatureStream ss(invoker->signature()); !ss.is_done(); ss.next()) {
 112     if (ss.at_return_type())  continue;
 113     BasicType ptype = ss.type();
 114     if (ptype == T_ARRAY)  ptype = T_OBJECT; // fold all refs to T_OBJECT
 115     assert(ptype >= T_BOOLEAN && ptype <= T_OBJECT, "not array or void");
 116     loc = &base[slot_num -= type2size[ptype]];
 117     bool is_oop = (ptype == T_OBJECT && loc != retval);
 118     if (is_oop)  blk->do_oop((oop*)loc);
 119     arg_num += 1;
 120   }
 121   assert(slot_num == 0, "must have processed all the arguments");
 122 }
 123 
 124 oop MethodHandles::RicochetFrame::compute_saved_args_layout(bool read_cache, bool write_cache) {
 125   oop cookie = NULL;
 126   if (read_cache) {
 127     cookie = saved_args_layout();
 128     if (cookie != NULL)  return cookie;
 129   }
 130   oop target = saved_target();
 131   oop mtype  = java_lang_invoke_MethodHandle::type(target);
 132   oop mtform = java_lang_invoke_MethodType::form(mtype);
 133   cookie = java_lang_invoke_MethodTypeForm::vmlayout(mtform);
 134   if (write_cache)  {
 135     (*saved_args_layout_addr()) = cookie;
 136   }
 137   return cookie;
 138 }
 139 
 140 void MethodHandles::RicochetFrame::generate_ricochet_blob(MacroAssembler* _masm,
 141                                                           // output params:
 142                                                           int* bounce_offset,
 143                                                           int* exception_offset,
 144                                                           int* frame_size_in_words) {
 145   (*frame_size_in_words) = RicochetFrame::frame_size_in_bytes() / wordSize;
 146 
 147   address start = __ pc();
 148 
 149 #ifdef ASSERT
 150   __ hlt(); __ hlt(); __ hlt();
 151   // here's a hint of something special:
 152   __ push(MAGIC_NUMBER_1);
 153   __ push(MAGIC_NUMBER_2);
 154 #endif //ASSERT
 155   __ hlt();  // not reached
 156 
 157   // A return PC has just been popped from the stack.
 158   // Return values are in registers.
 159   // The ebp points into the RicochetFrame, which contains
 160   // a cleanup continuation we must return to.
 161 
 162   (*bounce_offset) = __ pc() - start;
 163   BLOCK_COMMENT("ricochet_blob.bounce");
 164 
 165   if (VerifyMethodHandles)  RicochetFrame::verify_clean(_masm);
 166   trace_method_handle(_masm, "return/ricochet_blob.bounce");
 167 
 168   __ jmp(frame_address(continuation_offset_in_bytes()));
 169   __ hlt();
 170   DEBUG_ONLY(__ push(MAGIC_NUMBER_2));
 171 
 172   (*exception_offset) = __ pc() - start;
 173   BLOCK_COMMENT("ricochet_blob.exception");
 174 
 175   // compare this to Interpreter::rethrow_exception_entry, which is parallel code
 176   // for example, see TemplateInterpreterGenerator::generate_throw_exception
 177   // Live registers in:
 178   //   rax: exception
 179   //   rdx: return address/pc that threw exception (ignored, always equal to bounce addr)
 180   __ verify_oop(rax);
 181 
 182   // no need to empty_FPU_stack or reinit_heapbase, since caller frame will do the same if needed
 183 
 184   // Take down the frame.
 185 
 186   // Cf. InterpreterMacroAssembler::remove_activation.
 187   leave_ricochet_frame(_masm, /*rcx_recv=*/ noreg,
 188                        saved_last_sp_register(),
 189                        /*sender_pc_reg=*/ rdx);
 190 
 191   // In between activations - previous activation type unknown yet
 192   // compute continuation point - the continuation point expects the
 193   // following registers set up:
 194   //
 195   // rax: exception
 196   // rdx: return address/pc that threw exception
 197   // rsp: expression stack of caller
 198   // rbp: ebp of caller
 199   __ push(rax);                                  // save exception
 200   __ push(rdx);                                  // save return address
 201   Register thread_reg = LP64_ONLY(r15_thread) NOT_LP64(rdi);
 202   NOT_LP64(__ get_thread(thread_reg));
 203   __ call_VM_leaf(CAST_FROM_FN_PTR(address,
 204                                    SharedRuntime::exception_handler_for_return_address),
 205                   thread_reg, rdx);
 206   __ mov(rbx, rax);                              // save exception handler
 207   __ pop(rdx);                                   // restore return address
 208   __ pop(rax);                                   // restore exception
 209   __ jmp(rbx);                                   // jump to exception
 210                                                  // handler of caller
 211 }
 212 
 213 void MethodHandles::RicochetFrame::enter_ricochet_frame(MacroAssembler* _masm,
 214                                                         Register rcx_recv,
 215                                                         Register rax_argv,
 216                                                         address return_handler,
 217                                                         Register rbx_temp) {
 218   const Register saved_last_sp = saved_last_sp_register();
 219   Address rcx_mh_vmtarget(    rcx_recv, java_lang_invoke_MethodHandle::vmtarget_offset_in_bytes() );
 220   Address rcx_amh_conversion( rcx_recv, java_lang_invoke_AdapterMethodHandle::conversion_offset_in_bytes() );
 221 
 222   // Push the RicochetFrame a word at a time.
 223   // This creates something similar to an interpreter frame.
 224   // Cf. TemplateInterpreterGenerator::generate_fixed_frame.
 225   BLOCK_COMMENT("push RicochetFrame {");
 226   DEBUG_ONLY(int rfo = (int) sizeof(RicochetFrame));
 227   assert((rfo -= wordSize) == RicochetFrame::sender_pc_offset_in_bytes(), "");
 228 #define RF_FIELD(push_value, name)                                      \
 229   { push_value;                                                         \
 230     assert((rfo -= wordSize) == RicochetFrame::name##_offset_in_bytes(), ""); }
 231   RF_FIELD(__ push(rbp),                   sender_link);
 232   RF_FIELD(__ push(saved_last_sp),         exact_sender_sp);  // rsi/r13
 233   RF_FIELD(__ pushptr(rcx_amh_conversion), conversion);
 234   RF_FIELD(__ push(rax_argv),              saved_args_base);   // can be updated if args are shifted
 235   RF_FIELD(__ push((int32_t) NULL_WORD),   saved_args_layout); // cache for GC layout cookie
 236   if (UseCompressedOops) {
 237     __ load_heap_oop(rbx_temp, rcx_mh_vmtarget);
 238     RF_FIELD(__ push(rbx_temp),            saved_target);
 239   } else {
 240     RF_FIELD(__ pushptr(rcx_mh_vmtarget),  saved_target);
 241   }
 242   __ lea(rbx_temp, ExternalAddress(return_handler));
 243   RF_FIELD(__ push(rbx_temp),              continuation);
 244 #undef RF_FIELD
 245   assert(rfo == 0, "fully initialized the RicochetFrame");
 246   // compute new frame pointer:
 247   __ lea(rbp, Address(rsp, RicochetFrame::sender_link_offset_in_bytes()));
 248   // Push guard word #1 in debug mode.
 249   DEBUG_ONLY(__ push((int32_t) RicochetFrame::MAGIC_NUMBER_1));
 250   // For debugging, leave behind an indication of which stub built this frame.
 251   DEBUG_ONLY({ Label L; __ call(L, relocInfo::none); __ bind(L); });
 252   BLOCK_COMMENT("} RicochetFrame");
 253 }
 254 
 255 void MethodHandles::RicochetFrame::leave_ricochet_frame(MacroAssembler* _masm,
 256                                                         Register rcx_recv,
 257                                                         Register new_sp_reg,
 258                                                         Register sender_pc_reg) {
 259   assert_different_registers(rcx_recv, new_sp_reg, sender_pc_reg);
 260   const Register saved_last_sp = saved_last_sp_register();
 261   // Take down the frame.
 262   // Cf. InterpreterMacroAssembler::remove_activation.
 263   BLOCK_COMMENT("end_ricochet_frame {");
 264   // TO DO: If (exact_sender_sp - extended_sender_sp) > THRESH, compact the frame down.
 265   // This will keep stack in bounds even with unlimited tailcalls, each with an adapter.
 266   if (rcx_recv->is_valid())
 267     __ movptr(rcx_recv,    RicochetFrame::frame_address(RicochetFrame::saved_target_offset_in_bytes()));
 268   __ movptr(sender_pc_reg, RicochetFrame::frame_address(RicochetFrame::sender_pc_offset_in_bytes()));
 269   __ movptr(saved_last_sp, RicochetFrame::frame_address(RicochetFrame::exact_sender_sp_offset_in_bytes()));
 270   __ movptr(rbp,           RicochetFrame::frame_address(RicochetFrame::sender_link_offset_in_bytes()));
 271   __ mov(rsp, new_sp_reg);
 272   BLOCK_COMMENT("} end_ricochet_frame");
 273 }
 274 
 275 // Emit code to verify that RBP is pointing at a valid ricochet frame.
 276 #ifdef ASSERT
 277 enum {
 278   ARG_LIMIT = 255, SLOP = 4,
 279   // use this parameter for checking for garbage stack movements:
 280   UNREASONABLE_STACK_MOVE = (ARG_LIMIT + SLOP)
 281   // the slop defends against false alarms due to fencepost errors
 282 };
 283 
 284 void MethodHandles::RicochetFrame::verify_clean(MacroAssembler* _masm) {
 285   // The stack should look like this:
 286   //    ... keep1 | dest=42 | keep2 | RF | magic | handler | magic | recursive args |
 287   // Check various invariants.
 288   verify_offsets();
 289 
 290   Register rdi_temp = rdi;
 291   Register rcx_temp = rcx;
 292   { __ push(rdi_temp); __ push(rcx_temp); }
 293 #define UNPUSH_TEMPS \
 294   { __ pop(rcx_temp);  __ pop(rdi_temp); }
 295 
 296   Address magic_number_1_addr  = RicochetFrame::frame_address(RicochetFrame::magic_number_1_offset_in_bytes());
 297   Address magic_number_2_addr  = RicochetFrame::frame_address(RicochetFrame::magic_number_2_offset_in_bytes());
 298   Address continuation_addr    = RicochetFrame::frame_address(RicochetFrame::continuation_offset_in_bytes());
 299   Address conversion_addr      = RicochetFrame::frame_address(RicochetFrame::conversion_offset_in_bytes());
 300   Address saved_args_base_addr = RicochetFrame::frame_address(RicochetFrame::saved_args_base_offset_in_bytes());
 301 
 302   Label L_bad, L_ok;
 303   BLOCK_COMMENT("verify_clean {");
 304   // Magic numbers must check out:
 305   __ cmpptr(magic_number_1_addr, (int32_t) MAGIC_NUMBER_1);
 306   __ jcc(Assembler::notEqual, L_bad);
 307   __ cmpptr(magic_number_2_addr, (int32_t) MAGIC_NUMBER_2);
 308   __ jcc(Assembler::notEqual, L_bad);
 309 
 310   // Arguments pointer must look reasonable:
 311   __ movptr(rcx_temp, saved_args_base_addr);
 312   __ cmpptr(rcx_temp, rbp);
 313   __ jcc(Assembler::below, L_bad);
 314   __ subptr(rcx_temp, UNREASONABLE_STACK_MOVE * Interpreter::stackElementSize);
 315   __ cmpptr(rcx_temp, rbp);
 316   __ jcc(Assembler::above, L_bad);
 317 
 318   load_conversion_dest_type(_masm, rdi_temp, conversion_addr);
 319   __ cmpl(rdi_temp, T_VOID);
 320   __ jcc(Assembler::equal, L_ok);
 321   __ movptr(rcx_temp, saved_args_base_addr);
 322   load_conversion_vminfo(_masm, rdi_temp, conversion_addr);
 323   __ cmpptr(Address(rcx_temp, rdi_temp, Interpreter::stackElementScale()),
 324             (int32_t) RETURN_VALUE_PLACEHOLDER);
 325   __ jcc(Assembler::equal, L_ok);
 326   __ BIND(L_bad);
 327   UNPUSH_TEMPS;
 328   __ stop("damaged ricochet frame");
 329   __ BIND(L_ok);
 330   UNPUSH_TEMPS;
 331   BLOCK_COMMENT("} verify_clean");
 332 
 333 #undef UNPUSH_TEMPS
 334 
 335 }
 336 #endif //ASSERT
 337 
 338 void MethodHandles::load_klass_from_Class(MacroAssembler* _masm, Register klass_reg) {
 339   if (VerifyMethodHandles)
 340     verify_klass(_masm, klass_reg, SystemDictionaryHandles::Class_klass(),
 341                  "AMH argument is a Class");
 342   __ load_heap_oop(klass_reg, Address(klass_reg, java_lang_Class::klass_offset_in_bytes()));
 343 }
 344 
 345 void MethodHandles::load_conversion_vminfo(MacroAssembler* _masm, Register reg, Address conversion_field_addr) {
 346   int bits   = BitsPerByte;
 347   int offset = (CONV_VMINFO_SHIFT / bits);
 348   int shift  = (CONV_VMINFO_SHIFT % bits);
 349   __ load_unsigned_byte(reg, conversion_field_addr.plus_disp(offset));
 350   assert(CONV_VMINFO_MASK == right_n_bits(bits - shift), "else change type of previous load");
 351   assert(shift == 0, "no shift needed");
 352 }
 353 
 354 void MethodHandles::load_conversion_dest_type(MacroAssembler* _masm, Register reg, Address conversion_field_addr) {
 355   int bits   = BitsPerByte;
 356   int offset = (CONV_DEST_TYPE_SHIFT / bits);
 357   int shift  = (CONV_DEST_TYPE_SHIFT % bits);
 358   __ load_unsigned_byte(reg, conversion_field_addr.plus_disp(offset));
 359   assert(CONV_TYPE_MASK == right_n_bits(bits - shift), "else change type of previous load");
 360   __ shrl(reg, shift);
 361   DEBUG_ONLY(int conv_type_bits = (int) exact_log2(CONV_TYPE_MASK+1));
 362   assert((shift + conv_type_bits) == bits, "left justified in byte");
 363 }
 364 
 365 void MethodHandles::load_stack_move(MacroAssembler* _masm,
 366                                     Register rdi_stack_move,
 367                                     Register rcx_amh,
 368                                     bool might_be_negative) {
 369   BLOCK_COMMENT("load_stack_move {");
 370   Address rcx_amh_conversion(rcx_amh, java_lang_invoke_AdapterMethodHandle::conversion_offset_in_bytes());
 371   __ movl(rdi_stack_move, rcx_amh_conversion);
 372   __ sarl(rdi_stack_move, CONV_STACK_MOVE_SHIFT);
 373 #ifdef _LP64
 374   if (might_be_negative) {
 375     // clean high bits of stack motion register (was loaded as an int)
 376     __ movslq(rdi_stack_move, rdi_stack_move);
 377   }
 378 #endif //_LP64
 379   if (VerifyMethodHandles) {
 380     Label L_ok, L_bad;
 381     int32_t stack_move_limit = 0x4000;  // extra-large
 382     __ cmpptr(rdi_stack_move, stack_move_limit);
 383     __ jcc(Assembler::greaterEqual, L_bad);
 384     __ cmpptr(rdi_stack_move, -stack_move_limit);
 385     __ jcc(Assembler::greater, L_ok);
 386     __ bind(L_bad);
 387     __ stop("load_stack_move of garbage value");
 388     __ BIND(L_ok);
 389   }
 390   BLOCK_COMMENT("} load_stack_move");
 391 }
 392 
 393 #ifdef ASSERT
 394 void MethodHandles::RicochetFrame::verify_offsets() {
 395   // Check compatibility of this struct with the more generally used offsets of class frame:
 396   int ebp_off = sender_link_offset_in_bytes();  // offset from struct base to local rbp value
 397   assert(ebp_off + wordSize*frame::interpreter_frame_method_offset      == saved_args_base_offset_in_bytes(), "");
 398   assert(ebp_off + wordSize*frame::interpreter_frame_last_sp_offset     == conversion_offset_in_bytes(), "");
 399   assert(ebp_off + wordSize*frame::interpreter_frame_sender_sp_offset   == exact_sender_sp_offset_in_bytes(), "");
 400   // These last two have to be exact:
 401   assert(ebp_off + wordSize*frame::link_offset                          == sender_link_offset_in_bytes(), "");
 402   assert(ebp_off + wordSize*frame::return_addr_offset                   == sender_pc_offset_in_bytes(), "");
 403 }
 404 
 405 void MethodHandles::RicochetFrame::verify() const {
 406   verify_offsets();
 407   assert(magic_number_1() == MAGIC_NUMBER_1, "");
 408   assert(magic_number_2() == MAGIC_NUMBER_2, "");
 409   if (!Universe::heap()->is_gc_active()) {
 410     if (saved_args_layout() != NULL) {
 411       assert(saved_args_layout()->is_method(), "must be valid oop");
 412     }
 413     if (saved_target() != NULL) {
 414       assert(java_lang_invoke_MethodHandle::is_instance(saved_target()), "checking frame value");
 415     }
 416   }
 417   int conv_op = adapter_conversion_op(conversion());
 418   assert(conv_op == java_lang_invoke_AdapterMethodHandle::OP_COLLECT_ARGS ||
 419          conv_op == java_lang_invoke_AdapterMethodHandle::OP_FOLD_ARGS ||
 420          conv_op == java_lang_invoke_AdapterMethodHandle::OP_PRIM_TO_REF,
 421          "must be a sane conversion");
 422   if (has_return_value_slot()) {
 423     assert(*return_value_slot_addr() == RETURN_VALUE_PLACEHOLDER, "");
 424   }
 425 }
 426 #endif //PRODUCT
 427 
 428 #ifdef ASSERT
 429 void MethodHandles::verify_argslot(MacroAssembler* _masm,
 430                                    Register argslot_reg,
 431                                    const char* error_message) {
 432   // Verify that argslot lies within (rsp, rbp].
 433   Label L_ok, L_bad;
 434   BLOCK_COMMENT("verify_argslot {");
 435   __ cmpptr(argslot_reg, rbp);
 436   __ jccb(Assembler::above, L_bad);
 437   __ cmpptr(rsp, argslot_reg);
 438   __ jccb(Assembler::below, L_ok);
 439   __ bind(L_bad);
 440   __ stop(error_message);
 441   __ BIND(L_ok);
 442   BLOCK_COMMENT("} verify_argslot");
 443 }
 444 
 445 void MethodHandles::verify_argslots(MacroAssembler* _masm,
 446                                     RegisterOrConstant arg_slots,
 447                                     Register arg_slot_base_reg,
 448                                     bool negate_argslots,
 449                                     const char* error_message) {
 450   // Verify that [argslot..argslot+size) lies within (rsp, rbp).
 451   Label L_ok, L_bad;
 452   Register rdi_temp = rdi;
 453   BLOCK_COMMENT("verify_argslots {");
 454   __ push(rdi_temp);
 455   if (negate_argslots) {
 456     if (arg_slots.is_constant()) {
 457       arg_slots = -1 * arg_slots.as_constant();
 458     } else {
 459       __ movptr(rdi_temp, arg_slots);
 460       __ negptr(rdi_temp);
 461       arg_slots = rdi_temp;
 462     }
 463   }
 464   __ lea(rdi_temp, Address(arg_slot_base_reg, arg_slots, Interpreter::stackElementScale()));
 465   __ cmpptr(rdi_temp, rbp);
 466   __ pop(rdi_temp);
 467   __ jcc(Assembler::above, L_bad);
 468   __ cmpptr(rsp, arg_slot_base_reg);
 469   __ jcc(Assembler::below, L_ok);
 470   __ bind(L_bad);
 471   __ stop(error_message);
 472   __ BIND(L_ok);
 473   BLOCK_COMMENT("} verify_argslots");
 474 }
 475 
 476 // Make sure that arg_slots has the same sign as the given direction.
 477 // If (and only if) arg_slots is a assembly-time constant, also allow it to be zero.
 478 void MethodHandles::verify_stack_move(MacroAssembler* _masm,
 479                                       RegisterOrConstant arg_slots, int direction) {
 480   bool allow_zero = arg_slots.is_constant();
 481   if (direction == 0) { direction = +1; allow_zero = true; }
 482   assert(stack_move_unit() == -1, "else add extra checks here");
 483   if (arg_slots.is_register()) {
 484     Label L_ok, L_bad;
 485     BLOCK_COMMENT("verify_stack_move {");
 486     // testl(arg_slots.as_register(), -stack_move_unit() - 1);  // no need
 487     // jcc(Assembler::notZero, L_bad);
 488     __ cmpptr(arg_slots.as_register(), (int32_t) NULL_WORD);
 489     if (direction > 0) {
 490       __ jcc(allow_zero ? Assembler::less : Assembler::lessEqual, L_bad);
 491       __ cmpptr(arg_slots.as_register(), (int32_t) UNREASONABLE_STACK_MOVE);
 492       __ jcc(Assembler::less, L_ok);
 493     } else {
 494       __ jcc(allow_zero ? Assembler::greater : Assembler::greaterEqual, L_bad);
 495       __ cmpptr(arg_slots.as_register(), (int32_t) -UNREASONABLE_STACK_MOVE);
 496       __ jcc(Assembler::greater, L_ok);
 497     }
 498     __ bind(L_bad);
 499     if (direction > 0)
 500       __ stop("assert arg_slots > 0");
 501     else
 502       __ stop("assert arg_slots < 0");
 503     __ BIND(L_ok);
 504     BLOCK_COMMENT("} verify_stack_move");
 505   } else {
 506     intptr_t size = arg_slots.as_constant();
 507     if (direction < 0)  size = -size;
 508     assert(size >= 0, "correct direction of constant move");
 509     assert(size < UNREASONABLE_STACK_MOVE, "reasonable size of constant move");
 510   }
 511 }
 512 
 513 void MethodHandles::verify_klass(MacroAssembler* _masm,
 514                                  Register obj, KlassHandle klass,
 515                                  const char* error_message) {
 516   oop* klass_addr = klass.raw_value();
 517   assert(klass_addr >= SystemDictionaryHandles::Object_klass().raw_value() &&
 518          klass_addr <= SystemDictionaryHandles::Long_klass().raw_value(),
 519          "must be one of the SystemDictionaryHandles");
 520   Register temp = rdi;
 521   Label L_ok, L_bad;
 522   BLOCK_COMMENT("verify_klass {");
 523   __ verify_oop(obj);
 524   __ testptr(obj, obj);
 525   __ jcc(Assembler::zero, L_bad);
 526   __ push(temp);
 527   __ load_klass(temp, obj);
 528   __ cmpptr(temp, ExternalAddress((address) klass_addr));
 529   __ jcc(Assembler::equal, L_ok);
 530   intptr_t super_check_offset = klass->super_check_offset();
 531   __ movptr(temp, Address(temp, super_check_offset));
 532   __ cmpptr(temp, ExternalAddress((address) klass_addr));
 533   __ jcc(Assembler::equal, L_ok);
 534   __ pop(temp);
 535   __ bind(L_bad);
 536   __ stop(error_message);
 537   __ BIND(L_ok);
 538   __ pop(temp);
 539   BLOCK_COMMENT("} verify_klass");
 540 }
 541 #endif //ASSERT
 542 
 543 // Code generation
 544 address MethodHandles::generate_method_handle_interpreter_entry(MacroAssembler* _masm) {
 545   // rbx: methodOop
 546   // rcx: receiver method handle (must load from sp[MethodTypeForm.vmslots])
 547   // rsi/r13: sender SP (must preserve; see prepare_to_jump_from_interpreted)
 548   // rdx, rdi: garbage temp, blown away
 549 
 550   Register rbx_method = rbx;
 551   Register rcx_recv   = rcx;
 552   Register rax_mtype  = rax;
 553   Register rdx_temp   = rdx;
 554   Register rdi_temp   = rdi;
 555 
 556   // emit WrongMethodType path first, to enable jccb back-branch from main path
 557   Label wrong_method_type;
 558   __ bind(wrong_method_type);
 559   Label invoke_generic_slow_path;
 560   assert(methodOopDesc::intrinsic_id_size_in_bytes() == sizeof(u1), "");;
 561   __ cmpb(Address(rbx_method, methodOopDesc::intrinsic_id_offset_in_bytes()), (int) vmIntrinsics::_invokeExact);
 562   __ jcc(Assembler::notEqual, invoke_generic_slow_path);
 563   __ push(rax_mtype);       // required mtype
 564   __ push(rcx_recv);        // bad mh (1st stacked argument)
 565   __ jump(ExternalAddress(Interpreter::throw_WrongMethodType_entry()));
 566 
 567   // here's where control starts out:
 568   __ align(CodeEntryAlignment);
 569   address entry_point = __ pc();
 570 
 571   // fetch the MethodType from the method handle into rax (the 'check' register)
 572   // FIXME: Interpreter should transmit pre-popped stack pointer, to locate base of arg list.
 573   // This would simplify several touchy bits of code.
 574   // See 6984712: JSR 292 method handle calls need a clean argument base pointer
 575   {
 576     Register tem = rbx_method;
 577     for (jint* pchase = methodOopDesc::method_type_offsets_chain(); (*pchase) != -1; pchase++) {
 578       __ movptr(rax_mtype, Address(tem, *pchase));
 579       tem = rax_mtype;          // in case there is another indirection
 580     }
 581   }
 582 
 583   // given the MethodType, find out where the MH argument is buried
 584   __ load_heap_oop(rdx_temp, Address(rax_mtype, __ delayed_value(java_lang_invoke_MethodType::form_offset_in_bytes, rdi_temp)));
 585   Register rdx_vmslots = rdx_temp;
 586   __ movl(rdx_vmslots, Address(rdx_temp, __ delayed_value(java_lang_invoke_MethodTypeForm::vmslots_offset_in_bytes, rdi_temp)));
 587   Address mh_receiver_slot_addr = __ argument_address(rdx_vmslots);
 588   __ movptr(rcx_recv, mh_receiver_slot_addr);
 589 
 590   trace_method_handle(_masm, "invokeExact");
 591 
 592   __ check_method_handle_type(rax_mtype, rcx_recv, rdi_temp, wrong_method_type);
 593 
 594   // Nobody uses the MH receiver slot after this.  Make sure.
 595   DEBUG_ONLY(__ movptr(mh_receiver_slot_addr, (int32_t)0x999999));
 596 
 597   __ jump_to_method_handle_entry(rcx_recv, rdi_temp);
 598 
 599   // for invokeGeneric (only), apply argument and result conversions on the fly
 600   __ bind(invoke_generic_slow_path);
 601 #ifdef ASSERT
 602   if (VerifyMethodHandles) {
 603     Label L;
 604     __ cmpb(Address(rbx_method, methodOopDesc::intrinsic_id_offset_in_bytes()), (int) vmIntrinsics::_invokeGeneric);
 605     __ jcc(Assembler::equal, L);
 606     __ stop("bad methodOop::intrinsic_id");
 607     __ bind(L);
 608   }
 609 #endif //ASSERT
 610   Register rbx_temp = rbx_method;  // don't need it now
 611 
 612   // make room on the stack for another pointer:
 613   Register rcx_argslot = rcx_recv;
 614   __ lea(rcx_argslot, __ argument_address(rdx_vmslots, 1));
 615   insert_arg_slots(_masm, 2 * stack_move_unit(),
 616                    rcx_argslot, rbx_temp, rdx_temp);
 617 
 618   // load up an adapter from the calling type (Java weaves this)
 619   Register rdx_adapter = rdx_temp;
 620   __ load_heap_oop(rdx_temp,    Address(rax_mtype, __ delayed_value(java_lang_invoke_MethodType::form_offset_in_bytes,               rdi_temp)));
 621   __ load_heap_oop(rdx_adapter, Address(rdx_temp,  __ delayed_value(java_lang_invoke_MethodTypeForm::genericInvoker_offset_in_bytes, rdi_temp)));
 622   __ verify_oop(rdx_adapter);
 623   __ movptr(Address(rcx_argslot, 1 * Interpreter::stackElementSize), rdx_adapter);
 624   // As a trusted first argument, pass the type being called, so the adapter knows
 625   // the actual types of the arguments and return values.
 626   // (Generic invokers are shared among form-families of method-type.)
 627   __ movptr(Address(rcx_argslot, 0 * Interpreter::stackElementSize), rax_mtype);
 628   // FIXME: assert that rdx_adapter is of the right method-type.
 629   __ mov(rcx, rdx_adapter);
 630   trace_method_handle(_masm, "invokeGeneric");
 631   __ jump_to_method_handle_entry(rcx, rdi_temp);
 632 
 633   return entry_point;
 634 }
 635 
 636 // Workaround for C++ overloading nastiness on '0' for RegisterOrConstant.
 637 static RegisterOrConstant constant(int value) {
 638   return RegisterOrConstant(value);
 639 }
 640 
 641 // Helper to insert argument slots into the stack.
 642 // arg_slots must be a multiple of stack_move_unit() and < 0
 643 // rax_argslot is decremented to point to the new (shifted) location of the argslot
 644 // But, rdx_temp ends up holding the original value of rax_argslot.
 645 void MethodHandles::insert_arg_slots(MacroAssembler* _masm,
 646                                      RegisterOrConstant arg_slots,
 647                                      Register rax_argslot,
 648                                      Register rbx_temp, Register rdx_temp) {
 649   // allow constant zero
 650   if (arg_slots.is_constant() && arg_slots.as_constant() == 0)
 651     return;
 652   assert_different_registers(rax_argslot, rbx_temp, rdx_temp,
 653                              (!arg_slots.is_register() ? rsp : arg_slots.as_register()));
 654   if (VerifyMethodHandles)
 655     verify_argslot(_masm, rax_argslot, "insertion point must fall within current frame");
 656   if (VerifyMethodHandles)
 657     verify_stack_move(_masm, arg_slots, -1);
 658 
 659   // Make space on the stack for the inserted argument(s).
 660   // Then pull down everything shallower than rax_argslot.
 661   // The stacked return address gets pulled down with everything else.
 662   // That is, copy [rsp, argslot) downward by -size words.  In pseudo-code:
 663   //   rsp -= size;
 664   //   for (rdx = rsp + size; rdx < argslot; rdx++)
 665   //     rdx[-size] = rdx[0]
 666   //   argslot -= size;
 667   BLOCK_COMMENT("insert_arg_slots {");
 668   __ mov(rdx_temp, rsp);                        // source pointer for copy
 669   __ lea(rsp, Address(rsp, arg_slots, Interpreter::stackElementScale()));
 670   {
 671     Label loop;
 672     __ BIND(loop);
 673     // pull one word down each time through the loop
 674     __ movptr(rbx_temp, Address(rdx_temp, 0));
 675     __ movptr(Address(rdx_temp, arg_slots, Interpreter::stackElementScale()), rbx_temp);
 676     __ addptr(rdx_temp, wordSize);
 677     __ cmpptr(rdx_temp, rax_argslot);
 678     __ jcc(Assembler::below, loop);
 679   }
 680 
 681   // Now move the argslot down, to point to the opened-up space.
 682   __ lea(rax_argslot, Address(rax_argslot, arg_slots, Interpreter::stackElementScale()));
 683   BLOCK_COMMENT("} insert_arg_slots");
 684 }
 685 
 686 // Helper to remove argument slots from the stack.
 687 // arg_slots must be a multiple of stack_move_unit() and > 0
 688 void MethodHandles::remove_arg_slots(MacroAssembler* _masm,
 689                                      RegisterOrConstant arg_slots,
 690                                      Register rax_argslot,
 691                                      Register rbx_temp, Register rdx_temp) {
 692   // allow constant zero
 693   if (arg_slots.is_constant() && arg_slots.as_constant() == 0)
 694     return;
 695   assert_different_registers(rax_argslot, rbx_temp, rdx_temp,
 696                              (!arg_slots.is_register() ? rsp : arg_slots.as_register()));
 697   if (VerifyMethodHandles)
 698     verify_argslots(_masm, arg_slots, rax_argslot, false,
 699                     "deleted argument(s) must fall within current frame");
 700   if (VerifyMethodHandles)
 701     verify_stack_move(_masm, arg_slots, +1);
 702 
 703   BLOCK_COMMENT("remove_arg_slots {");
 704   // Pull up everything shallower than rax_argslot.
 705   // Then remove the excess space on the stack.
 706   // The stacked return address gets pulled up with everything else.
 707   // That is, copy [rsp, argslot) upward by size words.  In pseudo-code:
 708   //   for (rdx = argslot-1; rdx >= rsp; --rdx)
 709   //     rdx[size] = rdx[0]
 710   //   argslot += size;
 711   //   rsp += size;
 712   __ lea(rdx_temp, Address(rax_argslot, -wordSize)); // source pointer for copy
 713   {
 714     Label loop;
 715     __ BIND(loop);
 716     // pull one word up each time through the loop
 717     __ movptr(rbx_temp, Address(rdx_temp, 0));
 718     __ movptr(Address(rdx_temp, arg_slots, Interpreter::stackElementScale()), rbx_temp);
 719     __ addptr(rdx_temp, -wordSize);
 720     __ cmpptr(rdx_temp, rsp);
 721     __ jcc(Assembler::aboveEqual, loop);
 722   }
 723 
 724   // Now move the argslot up, to point to the just-copied block.
 725   __ lea(rsp, Address(rsp, arg_slots, Interpreter::stackElementScale()));
 726   // And adjust the argslot address to point at the deletion point.
 727   __ lea(rax_argslot, Address(rax_argslot, arg_slots, Interpreter::stackElementScale()));
 728   BLOCK_COMMENT("} remove_arg_slots");
 729 }
 730 
 731 // Helper to copy argument slots to the top of the stack.
 732 // The sequence starts with rax_argslot and is counted by slot_count
 733 // slot_count must be a multiple of stack_move_unit() and >= 0
 734 // This function blows the temps but does not change rax_argslot.
 735 void MethodHandles::push_arg_slots(MacroAssembler* _masm,
 736                                    Register rax_argslot,
 737                                    RegisterOrConstant slot_count,
 738                                    int skip_words_count,
 739                                    Register rbx_temp, Register rdx_temp) {
 740   assert_different_registers(rax_argslot, rbx_temp, rdx_temp,
 741                              (!slot_count.is_register() ? rbp : slot_count.as_register()),
 742                              rsp);
 743   assert(Interpreter::stackElementSize == wordSize, "else change this code");
 744 
 745   if (VerifyMethodHandles)
 746     verify_stack_move(_masm, slot_count, 0);
 747 
 748   // allow constant zero
 749   if (slot_count.is_constant() && slot_count.as_constant() == 0)
 750     return;
 751 
 752   BLOCK_COMMENT("push_arg_slots {");
 753 
 754   Register rbx_top = rbx_temp;
 755 
 756   // There is at most 1 word to carry down with the TOS.
 757   switch (skip_words_count) {
 758   case 1: __ pop(rdx_temp); break;
 759   case 0:                   break;
 760   default: ShouldNotReachHere();
 761   }
 762 
 763   if (slot_count.is_constant()) {
 764     for (int i = slot_count.as_constant() - 1; i >= 0; i--) {
 765       __ pushptr(Address(rax_argslot, i * wordSize));
 766     }
 767   } else {
 768     Label L_plural, L_loop, L_break;
 769     // Emit code to dynamically check for the common cases, zero and one slot.
 770     __ cmpl(slot_count.as_register(), (int32_t) 1);
 771     __ jccb(Assembler::greater, L_plural);
 772     __ jccb(Assembler::less, L_break);
 773     __ pushptr(Address(rax_argslot, 0));
 774     __ jmpb(L_break);
 775     __ BIND(L_plural);
 776 
 777     // Loop for 2 or more:
 778     //   rbx = &rax[slot_count]
 779     //   while (rbx > rax)  *(--rsp) = *(--rbx)
 780     __ lea(rbx_top, Address(rax_argslot, slot_count, Address::times_ptr));
 781     __ BIND(L_loop);
 782     __ subptr(rbx_top, wordSize);
 783     __ pushptr(Address(rbx_top, 0));
 784     __ cmpptr(rbx_top, rax_argslot);
 785     __ jcc(Assembler::above, L_loop);
 786     __ bind(L_break);
 787   }
 788   switch (skip_words_count) {
 789   case 1: __ push(rdx_temp); break;
 790   case 0:                    break;
 791   default: ShouldNotReachHere();
 792   }
 793   BLOCK_COMMENT("} push_arg_slots");
 794 }
 795 
 796 // in-place movement; no change to rsp
 797 // blows rax_temp, rdx_temp
 798 void MethodHandles::move_arg_slots_up(MacroAssembler* _masm,
 799                                       Register rbx_bottom,  // invariant
 800                                       Address  top_addr,     // can use rax_temp
 801                                       RegisterOrConstant positive_distance_in_slots,
 802                                       Register rax_temp, Register rdx_temp) {
 803   BLOCK_COMMENT("move_arg_slots_up {");
 804   assert_different_registers(rbx_bottom,
 805                              rax_temp, rdx_temp,
 806                              positive_distance_in_slots.register_or_noreg());
 807   Label L_loop, L_break;
 808   Register rax_top = rax_temp;
 809   if (!top_addr.is_same_address(Address(rax_top, 0)))
 810     __ lea(rax_top, top_addr);
 811   // Detect empty (or broken) loop:
 812 #ifdef ASSERT
 813   if (VerifyMethodHandles) {
 814     // Verify that &bottom < &top (non-empty interval)
 815     Label L_ok, L_bad;
 816     if (positive_distance_in_slots.is_register()) {
 817       __ cmpptr(positive_distance_in_slots.as_register(), (int32_t) 0);
 818       __ jcc(Assembler::lessEqual, L_bad);
 819     }
 820     __ cmpptr(rbx_bottom, rax_top);
 821     __ jcc(Assembler::below, L_ok);
 822     __ bind(L_bad);
 823     __ stop("valid bounds (copy up)");
 824     __ BIND(L_ok);
 825   }
 826 #endif
 827   __ cmpptr(rbx_bottom, rax_top);
 828   __ jccb(Assembler::aboveEqual, L_break);
 829   // work rax down to rbx, copying contiguous data upwards
 830   // In pseudo-code:
 831   //   [rbx, rax) = &[bottom, top)
 832   //   while (--rax >= rbx) *(rax + distance) = *(rax + 0), rax--;
 833   __ BIND(L_loop);
 834   __ subptr(rax_top, wordSize);
 835   __ movptr(rdx_temp, Address(rax_top, 0));
 836   __ movptr(          Address(rax_top, positive_distance_in_slots, Address::times_ptr), rdx_temp);
 837   __ cmpptr(rax_top, rbx_bottom);
 838   __ jcc(Assembler::above, L_loop);
 839   assert(Interpreter::stackElementSize == wordSize, "else change loop");
 840   __ bind(L_break);
 841   BLOCK_COMMENT("} move_arg_slots_up");
 842 }
 843 
 844 // in-place movement; no change to rsp
 845 // blows rax_temp, rdx_temp
 846 void MethodHandles::move_arg_slots_down(MacroAssembler* _masm,
 847                                         Address  bottom_addr,  // can use rax_temp
 848                                         Register rbx_top,      // invariant
 849                                         RegisterOrConstant negative_distance_in_slots,
 850                                         Register rax_temp, Register rdx_temp) {
 851   BLOCK_COMMENT("move_arg_slots_down {");
 852   assert_different_registers(rbx_top,
 853                              negative_distance_in_slots.register_or_noreg(),
 854                              rax_temp, rdx_temp);
 855   Label L_loop, L_break;
 856   Register rax_bottom = rax_temp;
 857   if (!bottom_addr.is_same_address(Address(rax_bottom, 0)))
 858     __ lea(rax_bottom, bottom_addr);
 859   // Detect empty (or broken) loop:
 860 #ifdef ASSERT
 861   assert(!negative_distance_in_slots.is_constant() || negative_distance_in_slots.as_constant() < 0, "");
 862   if (VerifyMethodHandles) {
 863     // Verify that &bottom < &top (non-empty interval)
 864     Label L_ok, L_bad;
 865     if (negative_distance_in_slots.is_register()) {
 866       __ cmpptr(negative_distance_in_slots.as_register(), (int32_t) 0);
 867       __ jcc(Assembler::greaterEqual, L_bad);
 868     }
 869     __ cmpptr(rax_bottom, rbx_top);
 870     __ jcc(Assembler::below, L_ok);
 871     __ bind(L_bad);
 872     __ stop("valid bounds (copy down)");
 873     __ BIND(L_ok);
 874   }
 875 #endif
 876   __ cmpptr(rax_bottom, rbx_top);
 877   __ jccb(Assembler::aboveEqual, L_break);
 878   // work rax up to rbx, copying contiguous data downwards
 879   // In pseudo-code:
 880   //   [rax, rbx) = &[bottom, top)
 881   //   while (rax < rbx) *(rax - distance) = *(rax + 0), rax++;
 882   __ BIND(L_loop);
 883   __ movptr(rdx_temp, Address(rax_bottom, 0));
 884   __ movptr(          Address(rax_bottom, negative_distance_in_slots, Address::times_ptr), rdx_temp);
 885   __ addptr(rax_bottom, wordSize);
 886   __ cmpptr(rax_bottom, rbx_top);
 887   __ jcc(Assembler::below, L_loop);
 888   assert(Interpreter::stackElementSize == wordSize, "else change loop");
 889   __ bind(L_break);
 890   BLOCK_COMMENT("} move_arg_slots_down");
 891 }
 892 
 893 // Copy from a field or array element to a stacked argument slot.
 894 // is_element (ignored) says whether caller is loading an array element instead of an instance field.
 895 void MethodHandles::move_typed_arg(MacroAssembler* _masm,
 896                                    BasicType type, bool is_element,
 897                                    Address slot_dest, Address value_src,
 898                                    Register rbx_temp, Register rdx_temp) {
 899   BLOCK_COMMENT(!is_element ? "move_typed_arg {" : "move_typed_arg { (array element)");
 900   if (type == T_OBJECT || type == T_ARRAY) {
 901     __ load_heap_oop(rbx_temp, value_src);
 902     __ movptr(slot_dest, rbx_temp);
 903   } else if (type != T_VOID) {
 904     int  arg_size      = type2aelembytes(type);
 905     bool arg_is_signed = is_signed_subword_type(type);
 906     int  slot_size     = (arg_size > wordSize) ? arg_size : wordSize;
 907     __ load_sized_value(  rdx_temp,  value_src, arg_size, arg_is_signed, rbx_temp);
 908     __ store_sized_value( slot_dest, rdx_temp,  slot_size,               rbx_temp);
 909   }
 910   BLOCK_COMMENT("} move_typed_arg");
 911 }
 912 
 913 void MethodHandles::move_return_value(MacroAssembler* _masm, BasicType type,
 914                                       Address return_slot) {
 915   BLOCK_COMMENT("move_return_value {");
 916   // Old versions of the JVM must clean the FPU stack after every return.
 917 #ifndef _LP64
 918 #ifdef COMPILER2
 919   // The FPU stack is clean if UseSSE >= 2 but must be cleaned in other cases
 920   if ((type == T_FLOAT && UseSSE < 1) || (type == T_DOUBLE && UseSSE < 2)) {
 921     for (int i = 1; i < 8; i++) {
 922         __ ffree(i);
 923     }
 924   } else if (UseSSE < 2) {
 925     __ empty_FPU_stack();
 926   }
 927 #endif //COMPILER2
 928 #endif //!_LP64
 929 
 930   // Look at the type and pull the value out of the corresponding register.
 931   if (type == T_VOID) {
 932     // nothing to do
 933   } else if (type == T_OBJECT) {
 934     __ movptr(return_slot, rax);
 935   } else if (type == T_INT || is_subword_type(type)) {
 936     // write the whole word, even if only 32 bits is significant
 937     __ movptr(return_slot, rax);
 938   } else if (type == T_LONG) {
 939     // store the value by parts
 940     // Note: We assume longs are continguous (if misaligned) on the interpreter stack.
 941     __ store_sized_value(return_slot, rax, BytesPerLong, rdx);
 942   } else if (NOT_LP64((type == T_FLOAT  && UseSSE < 1) ||
 943                       (type == T_DOUBLE && UseSSE < 2) ||)
 944              false) {
 945     // Use old x86 FPU registers:
 946     if (type == T_FLOAT)
 947       __ fstp_s(return_slot);
 948     else
 949       __ fstp_d(return_slot);
 950   } else if (type == T_FLOAT) {
 951     __ movflt(return_slot, xmm0);
 952   } else if (type == T_DOUBLE) {
 953     __ movdbl(return_slot, xmm0);
 954   } else {
 955     ShouldNotReachHere();
 956   }
 957   BLOCK_COMMENT("} move_return_value");
 958 }
 959 
 960 
 961 #ifndef PRODUCT
 962 extern "C" void print_method_handle(oop mh);
 963 void trace_method_handle_stub(const char* adaptername,
 964                               oop mh,
 965                               intptr_t* saved_regs,
 966                               intptr_t* entry_sp,
 967                               intptr_t* saved_sp,
 968                               intptr_t* saved_bp) {
 969   // called as a leaf from native code: do not block the JVM!
 970   bool has_mh = (strstr(adaptername, "return/") == NULL);  // return adapters don't have rcx_mh
 971   intptr_t* last_sp = (intptr_t*) saved_bp[frame::interpreter_frame_last_sp_offset];
 972   intptr_t* base_sp = last_sp;
 973   typedef MethodHandles::RicochetFrame RicochetFrame;
 974   RicochetFrame* rfp = (RicochetFrame*)((address)saved_bp - RicochetFrame::sender_link_offset_in_bytes());
 975   if (!UseRicochetFrames || Universe::heap()->is_in((address) rfp->saved_args_base())) {
 976     // Probably an interpreter frame.
 977     base_sp = (intptr_t*) saved_bp[frame::interpreter_frame_monitor_block_top_offset];
 978   }
 979   intptr_t    mh_reg = (intptr_t)mh;
 980   const char* mh_reg_name = "rcx_mh";
 981   if (!has_mh)  mh_reg_name = "rcx";
 982   tty->print_cr("MH %s %s="PTR_FORMAT" sp=("PTR_FORMAT"+"INTX_FORMAT") stack_size="INTX_FORMAT" bp="PTR_FORMAT,
 983                 adaptername, mh_reg_name, mh_reg,
 984                 (intptr_t)entry_sp, (intptr_t)(saved_sp - entry_sp), (intptr_t)(base_sp - last_sp), (intptr_t)saved_bp);
 985   if (Verbose) {
 986     tty->print(" reg dump: ");
 987     int saved_regs_count = (entry_sp-1) - saved_regs;
 988     // 32 bit: rdi rsi rbp rsp; rbx rdx rcx (*) rax
 989     int i;
 990     for (i = 0; i <= saved_regs_count; i++) {
 991       if (i > 0 && i % 4 == 0 && i != saved_regs_count) {
 992         tty->cr();
 993         tty->print("   + dump: ");
 994       }
 995       tty->print(" %d: "PTR_FORMAT, i, saved_regs[i]);
 996     }
 997     tty->cr();
 998     if (last_sp != saved_sp && last_sp != NULL)
 999       tty->print_cr("*** last_sp="PTR_FORMAT, (intptr_t)last_sp);
1000     int stack_dump_count = 16;
1001     if (stack_dump_count < (int)(saved_bp + 2 - saved_sp))
1002       stack_dump_count = (int)(saved_bp + 2 - saved_sp);
1003     if (stack_dump_count > 64)  stack_dump_count = 48;
1004     for (i = 0; i < stack_dump_count; i += 4) {
1005       tty->print_cr(" dump at SP[%d] "PTR_FORMAT": "PTR_FORMAT" "PTR_FORMAT" "PTR_FORMAT" "PTR_FORMAT,
1006                     i, (intptr_t) &entry_sp[i+0], entry_sp[i+0], entry_sp[i+1], entry_sp[i+2], entry_sp[i+3]);
1007     }
1008     if (has_mh)
1009       print_method_handle(mh);
1010   }
1011 }
1012 
1013 // The stub wraps the arguments in a struct on the stack to avoid
1014 // dealing with the different calling conventions for passing 6
1015 // arguments.
1016 struct MethodHandleStubArguments {
1017   const char* adaptername;
1018   oopDesc* mh;
1019   intptr_t* saved_regs;
1020   intptr_t* entry_sp;
1021   intptr_t* saved_sp;
1022   intptr_t* saved_bp;
1023 };
1024 void trace_method_handle_stub_wrapper(MethodHandleStubArguments* args) {
1025   trace_method_handle_stub(args->adaptername,
1026                            args->mh,
1027                            args->saved_regs,
1028                            args->entry_sp,
1029                            args->saved_sp,
1030                            args->saved_bp);
1031 }
1032 
1033 void MethodHandles::trace_method_handle(MacroAssembler* _masm, const char* adaptername) {
1034   if (!TraceMethodHandles)  return;
1035   BLOCK_COMMENT("trace_method_handle {");
1036   __ push(rax);
1037   __ lea(rax, Address(rsp, wordSize * NOT_LP64(6) LP64_ONLY(14))); // entry_sp  __ pusha();
1038   __ pusha();
1039   __ mov(rbx, rsp);
1040   __ enter();
1041   // incoming state:
1042   // rcx: method handle
1043   // r13 or rsi: saved sp
1044   // To avoid calling convention issues, build a record on the stack and pass the pointer to that instead.
1045   __ push(rbp);               // saved_bp
1046   __ push(rsi);               // saved_sp
1047   __ push(rax);               // entry_sp
1048   __ push(rbx);               // pusha saved_regs
1049   __ push(rcx);               // mh
1050   __ push(rcx);               // adaptername
1051   __ movptr(Address(rsp, 0), (intptr_t) adaptername);
1052   __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, trace_method_handle_stub_wrapper), rsp);
1053   __ leave();
1054   __ popa();
1055   __ pop(rax);
1056   BLOCK_COMMENT("} trace_method_handle");
1057 }
1058 #endif //PRODUCT
1059 
1060 // which conversion op types are implemented here?
1061 int MethodHandles::adapter_conversion_ops_supported_mask() {
1062   return ((1<<java_lang_invoke_AdapterMethodHandle::OP_RETYPE_ONLY)
1063          |(1<<java_lang_invoke_AdapterMethodHandle::OP_RETYPE_RAW)
1064          |(1<<java_lang_invoke_AdapterMethodHandle::OP_CHECK_CAST)
1065          |(1<<java_lang_invoke_AdapterMethodHandle::OP_PRIM_TO_PRIM)
1066          |(1<<java_lang_invoke_AdapterMethodHandle::OP_REF_TO_PRIM)
1067           //OP_PRIM_TO_REF is below...
1068          |(1<<java_lang_invoke_AdapterMethodHandle::OP_SWAP_ARGS)
1069          |(1<<java_lang_invoke_AdapterMethodHandle::OP_ROT_ARGS)
1070          |(1<<java_lang_invoke_AdapterMethodHandle::OP_DUP_ARGS)
1071          |(1<<java_lang_invoke_AdapterMethodHandle::OP_DROP_ARGS)
1072           //OP_COLLECT_ARGS is below...
1073          |(1<<java_lang_invoke_AdapterMethodHandle::OP_SPREAD_ARGS)
1074          |(!UseRicochetFrames ? 0 :
1075            java_lang_invoke_MethodTypeForm::vmlayout_offset_in_bytes() <= 0 ? 0 :
1076            ((1<<java_lang_invoke_AdapterMethodHandle::OP_PRIM_TO_REF)
1077            |(1<<java_lang_invoke_AdapterMethodHandle::OP_COLLECT_ARGS)
1078            |(1<<java_lang_invoke_AdapterMethodHandle::OP_FOLD_ARGS)
1079             ))
1080          );
1081 }
1082 
1083 //------------------------------------------------------------------------------
1084 // MethodHandles::generate_method_handle_stub
1085 //
1086 // Generate an "entry" field for a method handle.
1087 // This determines how the method handle will respond to calls.
1088 void MethodHandles::generate_method_handle_stub(MacroAssembler* _masm, MethodHandles::EntryKind ek) {
1089   MethodHandles::EntryKind ek_orig = ek_original_kind(ek);
1090 
1091   // Here is the register state during an interpreted call,
1092   // as set up by generate_method_handle_interpreter_entry():
1093   // - rbx: garbage temp (was MethodHandle.invoke methodOop, unused)
1094   // - rcx: receiver method handle
1095   // - rax: method handle type (only used by the check_mtype entry point)
1096   // - rsi/r13: sender SP (must preserve; see prepare_to_jump_from_interpreted)
1097   // - rdx: garbage temp, can blow away
1098 
1099   const Register rcx_recv    = rcx;
1100   const Register rax_argslot = rax;
1101   const Register rbx_temp    = rbx;
1102   const Register rdx_temp    = rdx;
1103   const Register rdi_temp    = rdi;
1104 
1105   // This guy is set up by prepare_to_jump_from_interpreted (from interpreted calls)
1106   // and gen_c2i_adapter (from compiled calls):
1107   const Register saved_last_sp = saved_last_sp_register();
1108 
1109   // Argument registers for _raise_exception.
1110   // 32-bit: Pass first two oop/int args in registers ECX and EDX.
1111   const Register rarg0_code     = LP64_ONLY(j_rarg0) NOT_LP64(rcx);
1112   const Register rarg1_actual   = LP64_ONLY(j_rarg1) NOT_LP64(rdx);
1113   const Register rarg2_required = LP64_ONLY(j_rarg2) NOT_LP64(rdi);
1114   assert_different_registers(rarg0_code, rarg1_actual, rarg2_required, saved_last_sp);
1115 
1116   guarantee(java_lang_invoke_MethodHandle::vmentry_offset_in_bytes() != 0, "must have offsets");
1117 
1118   // some handy addresses
1119   Address rbx_method_fie(     rbx,      methodOopDesc::from_interpreted_offset() );
1120   Address rbx_method_fce(     rbx,      methodOopDesc::from_compiled_offset() );
1121 
1122   Address rcx_mh_vmtarget(    rcx_recv, java_lang_invoke_MethodHandle::vmtarget_offset_in_bytes() );
1123   Address rcx_dmh_vmindex(    rcx_recv, java_lang_invoke_DirectMethodHandle::vmindex_offset_in_bytes() );
1124 
1125   Address rcx_bmh_vmargslot(  rcx_recv, java_lang_invoke_BoundMethodHandle::vmargslot_offset_in_bytes() );
1126   Address rcx_bmh_argument(   rcx_recv, java_lang_invoke_BoundMethodHandle::argument_offset_in_bytes() );
1127 
1128   Address rcx_amh_vmargslot(  rcx_recv, java_lang_invoke_AdapterMethodHandle::vmargslot_offset_in_bytes() );
1129   Address rcx_amh_argument(   rcx_recv, java_lang_invoke_AdapterMethodHandle::argument_offset_in_bytes() );
1130   Address rcx_amh_conversion( rcx_recv, java_lang_invoke_AdapterMethodHandle::conversion_offset_in_bytes() );
1131   Address vmarg;                // __ argument_address(vmargslot)
1132 
1133   const int java_mirror_offset = klassOopDesc::klass_part_offset_in_bytes() + Klass::java_mirror_offset_in_bytes();
1134 
1135   if (have_entry(ek)) {
1136     __ nop();                   // empty stubs make SG sick
1137     return;
1138   }
1139 
1140 #ifdef ASSERT
1141   __ push((int32_t) 0xEEEEEEEE);
1142   __ push((int32_t) (intptr_t) entry_name(ek));
1143   LP64_ONLY(__ push((int32_t) high((intptr_t) entry_name(ek))));
1144   __ push((int32_t) 0x33333333);
1145 #endif //ASSERT
1146 
1147   address interp_entry = __ pc();
1148 
1149   trace_method_handle(_masm, entry_name(ek));
1150 
1151   BLOCK_COMMENT(err_msg("Entry %s {", entry_name(ek)));
1152 
1153   switch ((int) ek) {
1154   case _raise_exception:
1155     {
1156       // Not a real MH entry, but rather shared code for raising an
1157       // exception.  Since we use the compiled entry, arguments are
1158       // expected in compiler argument registers.
1159       assert(raise_exception_method(), "must be set");
1160       assert(raise_exception_method()->from_compiled_entry(), "method must be linked");
1161 
1162       const Register rdi_pc = rax;
1163       __ pop(rdi_pc);  // caller PC
1164       __ mov(rsp, saved_last_sp);  // cut the stack back to where the caller started
1165 
1166       Register rbx_method = rbx_temp;
1167       Label L_no_method;
1168       // FIXME: fill in _raise_exception_method with a suitable java.lang.invoke method
1169       __ movptr(rbx_method, ExternalAddress((address) &_raise_exception_method));
1170       __ testptr(rbx_method, rbx_method);
1171       __ jccb(Assembler::zero, L_no_method);
1172 
1173       const int jobject_oop_offset = 0;
1174       __ movptr(rbx_method, Address(rbx_method, jobject_oop_offset));  // dereference the jobject
1175       __ testptr(rbx_method, rbx_method);
1176       __ jccb(Assembler::zero, L_no_method);
1177       __ verify_oop(rbx_method);
1178 
1179       NOT_LP64(__ push(rarg2_required));
1180       __ push(rdi_pc);         // restore caller PC
1181       __ jmp(rbx_method_fce);  // jump to compiled entry
1182 
1183       // Do something that is at least causes a valid throw from the interpreter.
1184       __ bind(L_no_method);
1185       __ push(rarg2_required);
1186       __ push(rarg1_actual);
1187       __ jump(ExternalAddress(Interpreter::throw_WrongMethodType_entry()));
1188     }
1189     break;
1190 
1191   case _invokestatic_mh:
1192   case _invokespecial_mh:
1193     {
1194       Register rbx_method = rbx_temp;
1195       __ load_heap_oop(rbx_method, rcx_mh_vmtarget); // target is a methodOop
1196       __ verify_oop(rbx_method);
1197       // same as TemplateTable::invokestatic or invokespecial,
1198       // minus the CP setup and profiling:
1199       if (ek == _invokespecial_mh) {
1200         // Must load & check the first argument before entering the target method.
1201         __ load_method_handle_vmslots(rax_argslot, rcx_recv, rdx_temp);
1202         __ movptr(rcx_recv, __ argument_address(rax_argslot, -1));
1203         __ null_check(rcx_recv);
1204         __ verify_oop(rcx_recv);
1205       }
1206       __ jmp(rbx_method_fie);
1207     }
1208     break;
1209 
1210   case _invokevirtual_mh:
1211     {
1212       // same as TemplateTable::invokevirtual,
1213       // minus the CP setup and profiling:
1214 
1215       // pick out the vtable index and receiver offset from the MH,
1216       // and then we can discard it:
1217       __ load_method_handle_vmslots(rax_argslot, rcx_recv, rdx_temp);
1218       Register rbx_index = rbx_temp;
1219       __ movl(rbx_index, rcx_dmh_vmindex);
1220       // Note:  The verifier allows us to ignore rcx_mh_vmtarget.
1221       __ movptr(rcx_recv, __ argument_address(rax_argslot, -1));
1222       __ null_check(rcx_recv, oopDesc::klass_offset_in_bytes());
1223 
1224       // get receiver klass
1225       Register rax_klass = rax_argslot;
1226       __ load_klass(rax_klass, rcx_recv);
1227       __ verify_oop(rax_klass);
1228 
1229       // get target methodOop & entry point
1230       const int base = instanceKlass::vtable_start_offset() * wordSize;
1231       assert(vtableEntry::size() * wordSize == wordSize, "adjust the scaling in the code below");
1232       Address vtable_entry_addr(rax_klass,
1233                                 rbx_index, Address::times_ptr,
1234                                 base + vtableEntry::method_offset_in_bytes());
1235       Register rbx_method = rbx_temp;
1236       __ movptr(rbx_method, vtable_entry_addr);
1237 
1238       __ verify_oop(rbx_method);
1239       __ jmp(rbx_method_fie);
1240     }
1241     break;
1242 
1243   case _invokeinterface_mh:
1244     {
1245       // same as TemplateTable::invokeinterface,
1246       // minus the CP setup and profiling:
1247 
1248       // pick out the interface and itable index from the MH.
1249       __ load_method_handle_vmslots(rax_argslot, rcx_recv, rdx_temp);
1250       Register rdx_intf  = rdx_temp;
1251       Register rbx_index = rbx_temp;
1252       __ load_heap_oop(rdx_intf, rcx_mh_vmtarget);
1253       __ movl(rbx_index, rcx_dmh_vmindex);
1254       __ movptr(rcx_recv, __ argument_address(rax_argslot, -1));
1255       __ null_check(rcx_recv, oopDesc::klass_offset_in_bytes());
1256 
1257       // get receiver klass
1258       Register rax_klass = rax_argslot;
1259       __ load_klass(rax_klass, rcx_recv);
1260       __ verify_oop(rax_klass);
1261 
1262       Register rbx_method = rbx_index;
1263 
1264       // get interface klass
1265       Label no_such_interface;
1266       __ verify_oop(rdx_intf);
1267       __ lookup_interface_method(rax_klass, rdx_intf,
1268                                  // note: next two args must be the same:
1269                                  rbx_index, rbx_method,
1270                                  rdi_temp,
1271                                  no_such_interface);
1272 
1273       __ verify_oop(rbx_method);
1274       __ jmp(rbx_method_fie);
1275       __ hlt();
1276 
1277       __ bind(no_such_interface);
1278       // Throw an exception.
1279       // For historical reasons, it will be IncompatibleClassChangeError.
1280       __ mov(rbx_temp, rcx_recv);  // rarg2_required might be RCX
1281       assert_different_registers(rarg2_required, rbx_temp);
1282       __ movptr(rarg2_required, Address(rdx_intf, java_mirror_offset));  // required interface
1283       __ mov(   rarg1_actual,   rbx_temp);                               // bad receiver
1284       __ movl(  rarg0_code,     (int) Bytecodes::_invokeinterface);      // who is complaining?
1285       __ jump(ExternalAddress(from_interpreted_entry(_raise_exception)));
1286     }
1287     break;
1288 
1289   case _bound_ref_mh:
1290   case _bound_int_mh:
1291   case _bound_long_mh:
1292   case _bound_ref_direct_mh:
1293   case _bound_int_direct_mh:
1294   case _bound_long_direct_mh:
1295     {
1296       const bool direct_to_method = (ek >= _bound_ref_direct_mh);
1297       BasicType arg_type  = ek_bound_mh_arg_type(ek);
1298       int       arg_slots = type2size[arg_type];
1299 
1300       // make room for the new argument:
1301       __ movl(rax_argslot, rcx_bmh_vmargslot);
1302       __ lea(rax_argslot, __ argument_address(rax_argslot));
1303 
1304       insert_arg_slots(_masm, arg_slots * stack_move_unit(), rax_argslot, rbx_temp, rdx_temp);
1305 
1306       // store bound argument into the new stack slot:
1307       __ load_heap_oop(rbx_temp, rcx_bmh_argument);
1308       if (arg_type == T_OBJECT) {
1309         __ movptr(Address(rax_argslot, 0), rbx_temp);
1310       } else {
1311         Address prim_value_addr(rbx_temp, java_lang_boxing_object::value_offset_in_bytes(arg_type));
1312         move_typed_arg(_masm, arg_type, false,
1313                        Address(rax_argslot, 0),
1314                        prim_value_addr,
1315                        rbx_temp, rdx_temp);
1316       }
1317 
1318       if (direct_to_method) {
1319         Register rbx_method = rbx_temp;
1320         __ load_heap_oop(rbx_method, rcx_mh_vmtarget);
1321         __ verify_oop(rbx_method);
1322         __ jmp(rbx_method_fie);
1323       } else {
1324         __ load_heap_oop(rcx_recv, rcx_mh_vmtarget);
1325         __ verify_oop(rcx_recv);
1326         __ jump_to_method_handle_entry(rcx_recv, rdx_temp);
1327       }
1328     }
1329     break;
1330 
1331   case _adapter_retype_only:
1332   case _adapter_retype_raw:
1333     // immediately jump to the next MH layer:
1334     __ load_heap_oop(rcx_recv, rcx_mh_vmtarget);
1335     __ verify_oop(rcx_recv);
1336     __ jump_to_method_handle_entry(rcx_recv, rdx_temp);
1337     // This is OK when all parameter types widen.
1338     // It is also OK when a return type narrows.
1339     break;
1340 
1341   case _adapter_check_cast:
1342     {
1343       // temps:
1344       Register rbx_klass = rbx_temp; // interesting AMH data
1345 
1346       // check a reference argument before jumping to the next layer of MH:
1347       __ movl(rax_argslot, rcx_amh_vmargslot);
1348       vmarg = __ argument_address(rax_argslot);
1349 
1350       // What class are we casting to?
1351       __ load_heap_oop(rbx_klass, rcx_amh_argument); // this is a Class object!
1352       load_klass_from_Class(_masm, rbx_klass);
1353 
1354       Label done;
1355       __ movptr(rdx_temp, vmarg);
1356       __ testptr(rdx_temp, rdx_temp);
1357       __ jcc(Assembler::zero, done);         // no cast if null
1358       __ load_klass(rdx_temp, rdx_temp);
1359 
1360       // live at this point:
1361       // - rbx_klass:  klass required by the target method
1362       // - rdx_temp:   argument klass to test
1363       // - rcx_recv:   adapter method handle
1364       __ check_klass_subtype(rdx_temp, rbx_klass, rax_argslot, done);
1365 
1366       // If we get here, the type check failed!
1367       // Call the wrong_method_type stub, passing the failing argument type in rax.
1368       Register rax_mtype = rax_argslot;
1369       __ movl(rax_argslot, rcx_amh_vmargslot);  // reload argslot field
1370       __ movptr(rdx_temp, vmarg);
1371 
1372       assert_different_registers(rarg2_required, rdx_temp);
1373       __ load_heap_oop(rarg2_required, rcx_amh_argument);             // required class
1374       __ mov(          rarg1_actual,   rdx_temp);                     // bad object
1375       __ movl(         rarg0_code,     (int) Bytecodes::_checkcast);  // who is complaining?
1376       __ jump(ExternalAddress(from_interpreted_entry(_raise_exception)));
1377 
1378       __ bind(done);
1379       // get the new MH:
1380       __ load_heap_oop(rcx_recv, rcx_mh_vmtarget);
1381       __ jump_to_method_handle_entry(rcx_recv, rdx_temp);
1382     }
1383     break;
1384 
1385   case _adapter_prim_to_prim:
1386   case _adapter_ref_to_prim:
1387   case _adapter_prim_to_ref:
1388     // handled completely by optimized cases
1389     __ stop("init_AdapterMethodHandle should not issue this");
1390     break;
1391 
1392   case _adapter_opt_i2i:        // optimized subcase of adapt_prim_to_prim
1393 //case _adapter_opt_f2i:        // optimized subcase of adapt_prim_to_prim
1394   case _adapter_opt_l2i:        // optimized subcase of adapt_prim_to_prim
1395   case _adapter_opt_unboxi:     // optimized subcase of adapt_ref_to_prim
1396     {
1397       // perform an in-place conversion to int or an int subword
1398       __ movl(rax_argslot, rcx_amh_vmargslot);
1399       vmarg = __ argument_address(rax_argslot);
1400 
1401       switch (ek) {
1402       case _adapter_opt_i2i:
1403         __ movl(rdx_temp, vmarg);
1404         break;
1405       case _adapter_opt_l2i:
1406         {
1407           // just delete the extra slot; on a little-endian machine we keep the first
1408           __ lea(rax_argslot, __ argument_address(rax_argslot, 1));
1409           remove_arg_slots(_masm, -stack_move_unit(),
1410                            rax_argslot, rbx_temp, rdx_temp);
1411           vmarg = Address(rax_argslot, -Interpreter::stackElementSize);
1412           __ movl(rdx_temp, vmarg);
1413         }
1414         break;
1415       case _adapter_opt_unboxi:
1416         {
1417           // Load the value up from the heap.
1418           __ movptr(rdx_temp, vmarg);
1419           int value_offset = java_lang_boxing_object::value_offset_in_bytes(T_INT);
1420 #ifdef ASSERT
1421           for (int bt = T_BOOLEAN; bt < T_INT; bt++) {
1422             if (is_subword_type(BasicType(bt)))
1423               assert(value_offset == java_lang_boxing_object::value_offset_in_bytes(BasicType(bt)), "");
1424           }
1425 #endif
1426           __ null_check(rdx_temp, value_offset);
1427           __ movl(rdx_temp, Address(rdx_temp, value_offset));
1428           // We load this as a word.  Because we are little-endian,
1429           // the low bits will be correct, but the high bits may need cleaning.
1430           // The vminfo will guide us to clean those bits.
1431         }
1432         break;
1433       default:
1434         ShouldNotReachHere();
1435       }
1436 
1437       // Do the requested conversion and store the value.
1438       Register rbx_vminfo = rbx_temp;
1439       load_conversion_vminfo(_masm, rbx_vminfo, rcx_amh_conversion);
1440 
1441       // get the new MH:
1442       __ load_heap_oop(rcx_recv, rcx_mh_vmtarget);
1443       // (now we are done with the old MH)
1444 
1445       // original 32-bit vmdata word must be of this form:
1446       //    | MBZ:6 | signBitCount:8 | srcDstTypes:8 | conversionOp:8 |
1447       __ xchgptr(rcx, rbx_vminfo);                // free rcx for shifts
1448       __ shll(rdx_temp /*, rcx*/);
1449       Label zero_extend, done;
1450       __ testl(rcx, CONV_VMINFO_SIGN_FLAG);
1451       __ jccb(Assembler::zero, zero_extend);
1452 
1453       // this path is taken for int->byte, int->short
1454       __ sarl(rdx_temp /*, rcx*/);
1455       __ jmpb(done);
1456 
1457       __ bind(zero_extend);
1458       // this is taken for int->char
1459       __ shrl(rdx_temp /*, rcx*/);
1460 
1461       __ bind(done);
1462       __ movl(vmarg, rdx_temp);  // Store the value.
1463       __ xchgptr(rcx, rbx_vminfo);                // restore rcx_recv
1464 
1465       __ jump_to_method_handle_entry(rcx_recv, rdx_temp);
1466     }
1467     break;
1468 
1469   case _adapter_opt_i2l:        // optimized subcase of adapt_prim_to_prim
1470   case _adapter_opt_unboxl:     // optimized subcase of adapt_ref_to_prim
1471     {
1472       // perform an in-place int-to-long or ref-to-long conversion
1473       __ movl(rax_argslot, rcx_amh_vmargslot);
1474 
1475       // on a little-endian machine we keep the first slot and add another after
1476       __ lea(rax_argslot, __ argument_address(rax_argslot, 1));
1477       insert_arg_slots(_masm, stack_move_unit(),
1478                        rax_argslot, rbx_temp, rdx_temp);
1479       Address vmarg1(rax_argslot, -Interpreter::stackElementSize);
1480       Address vmarg2 = vmarg1.plus_disp(Interpreter::stackElementSize);
1481 
1482       switch (ek) {
1483       case _adapter_opt_i2l:
1484         {
1485 #ifdef _LP64
1486           __ movslq(rdx_temp, vmarg1);  // Load sign-extended
1487           __ movq(vmarg1, rdx_temp);    // Store into first slot
1488 #else
1489           __ movl(rdx_temp, vmarg1);
1490           __ sarl(rdx_temp, BitsPerInt - 1);  // __ extend_sign()
1491           __ movl(vmarg2, rdx_temp); // store second word
1492 #endif
1493         }
1494         break;
1495       case _adapter_opt_unboxl:
1496         {
1497           // Load the value up from the heap.
1498           __ movptr(rdx_temp, vmarg1);
1499           int value_offset = java_lang_boxing_object::value_offset_in_bytes(T_LONG);
1500           assert(value_offset == java_lang_boxing_object::value_offset_in_bytes(T_DOUBLE), "");
1501           __ null_check(rdx_temp, value_offset);
1502 #ifdef _LP64
1503           __ movq(rbx_temp, Address(rdx_temp, value_offset));
1504           __ movq(vmarg1, rbx_temp);
1505 #else
1506           __ movl(rbx_temp, Address(rdx_temp, value_offset + 0*BytesPerInt));
1507           __ movl(rdx_temp, Address(rdx_temp, value_offset + 1*BytesPerInt));
1508           __ movl(vmarg1, rbx_temp);
1509           __ movl(vmarg2, rdx_temp);
1510 #endif
1511         }
1512         break;
1513       default:
1514         ShouldNotReachHere();
1515       }
1516 
1517       __ load_heap_oop(rcx_recv, rcx_mh_vmtarget);
1518       __ jump_to_method_handle_entry(rcx_recv, rdx_temp);
1519     }
1520     break;
1521 
1522   case _adapter_opt_f2d:        // optimized subcase of adapt_prim_to_prim
1523   case _adapter_opt_d2f:        // optimized subcase of adapt_prim_to_prim
1524     {
1525       // perform an in-place floating primitive conversion
1526       __ movl(rax_argslot, rcx_amh_vmargslot);
1527       __ lea(rax_argslot, __ argument_address(rax_argslot, 1));
1528       if (ek == _adapter_opt_f2d) {
1529         insert_arg_slots(_masm, stack_move_unit(),
1530                          rax_argslot, rbx_temp, rdx_temp);
1531       }
1532       Address vmarg(rax_argslot, -Interpreter::stackElementSize);
1533 
1534 #ifdef _LP64
1535       if (ek == _adapter_opt_f2d) {
1536         __ movflt(xmm0, vmarg);
1537         __ cvtss2sd(xmm0, xmm0);
1538         __ movdbl(vmarg, xmm0);
1539       } else {
1540         __ movdbl(xmm0, vmarg);
1541         __ cvtsd2ss(xmm0, xmm0);
1542         __ movflt(vmarg, xmm0);
1543       }
1544 #else //_LP64
1545       if (ek == _adapter_opt_f2d) {
1546         __ fld_s(vmarg);        // load float to ST0
1547         __ fstp_d(vmarg);       // store double
1548       } else {
1549         __ fld_d(vmarg);        // load double to ST0
1550         __ fstp_s(vmarg);       // store single
1551       }
1552 #endif //_LP64
1553 
1554       if (ek == _adapter_opt_d2f) {
1555         remove_arg_slots(_masm, -stack_move_unit(),
1556                          rax_argslot, rbx_temp, rdx_temp);
1557       }
1558 
1559       __ load_heap_oop(rcx_recv, rcx_mh_vmtarget);
1560       __ jump_to_method_handle_entry(rcx_recv, rdx_temp);
1561     }
1562     break;
1563 
1564   case _adapter_swap_args:
1565   case _adapter_rot_args:
1566     // handled completely by optimized cases
1567     __ stop("init_AdapterMethodHandle should not issue this");
1568     break;
1569 
1570   case _adapter_opt_swap_1:
1571   case _adapter_opt_swap_2:
1572   case _adapter_opt_rot_1_up:
1573   case _adapter_opt_rot_1_down:
1574   case _adapter_opt_rot_2_up:
1575   case _adapter_opt_rot_2_down:
1576     {
1577       int swap_slots = ek_adapter_opt_swap_slots(ek);
1578       int rotate     = ek_adapter_opt_swap_mode(ek);
1579 
1580       // 'argslot' is the position of the first argument to swap
1581       __ movl(rax_argslot, rcx_amh_vmargslot);
1582       __ lea(rax_argslot, __ argument_address(rax_argslot));
1583 
1584       // 'vminfo' is the second
1585       Register rbx_destslot = rbx_temp;
1586       load_conversion_vminfo(_masm, rbx_destslot, rcx_amh_conversion);
1587       __ lea(rbx_destslot, __ argument_address(rbx_destslot));
1588       if (VerifyMethodHandles)
1589         verify_argslot(_masm, rbx_destslot, "swap point must fall within current frame");
1590 
1591       assert(Interpreter::stackElementSize == wordSize, "else rethink use of wordSize here");
1592       if (!rotate) {
1593         // simple swap
1594         for (int i = 0; i < swap_slots; i++) {
1595           __ movptr(rdi_temp, Address(rax_argslot,  i * wordSize));
1596           __ movptr(rdx_temp, Address(rbx_destslot, i * wordSize));
1597           __ movptr(Address(rax_argslot,  i * wordSize), rdx_temp);
1598           __ movptr(Address(rbx_destslot, i * wordSize), rdi_temp);
1599         }
1600       } else {
1601         // A rotate is actually pair of moves, with an "odd slot" (or pair)
1602         // changing place with a series of other slots.
1603         // First, push the "odd slot", which is going to get overwritten
1604         for (int i = swap_slots - 1; i >= 0; i--) {
1605           // handle one with rdi_temp instead of a push:
1606           if (i == 0)  __ movptr(rdi_temp, Address(rax_argslot, i * wordSize));
1607           else         __ pushptr(         Address(rax_argslot, i * wordSize));
1608         }
1609         if (rotate > 0) {
1610           // Here is rotate > 0:
1611           // (low mem)                                          (high mem)
1612           //     | dest:     more_slots...     | arg: odd_slot :arg+1 |
1613           // =>
1614           //     | dest: odd_slot | dest+1: more_slots...      :arg+1 |
1615           // work argslot down to destslot, copying contiguous data upwards
1616           // pseudo-code:
1617           //   rax = src_addr - swap_bytes
1618           //   rbx = dest_addr
1619           //   while (rax >= rbx) *(rax + swap_bytes) = *(rax + 0), rax--;
1620           move_arg_slots_up(_masm,
1621                             rbx_destslot,
1622                             Address(rax_argslot, 0),
1623                             swap_slots,
1624                             rax_argslot, rdx_temp);
1625         } else {
1626           // Here is the other direction, rotate < 0:
1627           // (low mem)                                          (high mem)
1628           //     | arg: odd_slot | arg+1: more_slots...       :dest+1 |
1629           // =>
1630           //     | arg:    more_slots...     | dest: odd_slot :dest+1 |
1631           // work argslot up to destslot, copying contiguous data downwards
1632           // pseudo-code:
1633           //   rax = src_addr + swap_bytes
1634           //   rbx = dest_addr
1635           //   while (rax <= rbx) *(rax - swap_bytes) = *(rax + 0), rax++;
1636           // dest_slot denotes an exclusive upper limit
1637           int limit_bias = OP_ROT_ARGS_DOWN_LIMIT_BIAS;
1638           if (limit_bias != 0)
1639             __ addptr(rbx_destslot, - limit_bias * wordSize);
1640           move_arg_slots_down(_masm,
1641                               Address(rax_argslot, swap_slots * wordSize),
1642                               rbx_destslot,
1643                               -swap_slots,
1644                               rax_argslot, rdx_temp);
1645           __ subptr(rbx_destslot, swap_slots * wordSize);
1646         }
1647         // pop the original first chunk into the destination slot, now free
1648         for (int i = 0; i < swap_slots; i++) {
1649           if (i == 0)  __ movptr(Address(rbx_destslot, i * wordSize), rdi_temp);
1650           else         __ popptr(Address(rbx_destslot, i * wordSize));
1651         }
1652       }
1653 
1654       __ load_heap_oop(rcx_recv, rcx_mh_vmtarget);
1655       __ jump_to_method_handle_entry(rcx_recv, rdx_temp);
1656     }
1657     break;
1658 
1659   case _adapter_dup_args:
1660     {
1661       // 'argslot' is the position of the first argument to duplicate
1662       __ movl(rax_argslot, rcx_amh_vmargslot);
1663       __ lea(rax_argslot, __ argument_address(rax_argslot));
1664 
1665       // 'stack_move' is negative number of words to duplicate
1666       Register rdi_stack_move = rdi_temp;
1667       load_stack_move(_masm, rdi_stack_move, rcx_recv, true);
1668 
1669       if (VerifyMethodHandles) {
1670         verify_argslots(_masm, rdi_stack_move, rax_argslot, true,
1671                         "copied argument(s) must fall within current frame");
1672       }
1673 
1674       // insert location is always the bottom of the argument list:
1675       Address insert_location = __ argument_address(constant(0));
1676       int pre_arg_words = insert_location.disp() / wordSize;   // return PC is pushed
1677       assert(insert_location.base() == rsp, "");
1678 
1679       __ negl(rdi_stack_move);
1680       push_arg_slots(_masm, rax_argslot, rdi_stack_move,
1681                      pre_arg_words, rbx_temp, rdx_temp);
1682 
1683       __ load_heap_oop(rcx_recv, rcx_mh_vmtarget);
1684       __ jump_to_method_handle_entry(rcx_recv, rdx_temp);
1685     }
1686     break;
1687 
1688   case _adapter_drop_args:
1689     {
1690       // 'argslot' is the position of the first argument to nuke
1691       __ movl(rax_argslot, rcx_amh_vmargslot);
1692       __ lea(rax_argslot, __ argument_address(rax_argslot));
1693 
1694       // (must do previous push after argslot address is taken)
1695 
1696       // 'stack_move' is number of words to drop
1697       Register rdi_stack_move = rdi_temp;
1698       load_stack_move(_masm, rdi_stack_move, rcx_recv, false);
1699       remove_arg_slots(_masm, rdi_stack_move,
1700                        rax_argslot, rbx_temp, rdx_temp);
1701 
1702       __ load_heap_oop(rcx_recv, rcx_mh_vmtarget);
1703       __ jump_to_method_handle_entry(rcx_recv, rdx_temp);
1704     }
1705     break;
1706 
1707   case _adapter_collect_args:
1708   case _adapter_fold_args:
1709   case _adapter_spread_args:
1710     // handled completely by optimized cases
1711     __ stop("init_AdapterMethodHandle should not issue this");
1712     break;
1713 
1714   case _adapter_opt_collect_ref:
1715   case _adapter_opt_collect_int:
1716   case _adapter_opt_collect_long:
1717   case _adapter_opt_collect_float:
1718   case _adapter_opt_collect_double:
1719   case _adapter_opt_collect_void:
1720   case _adapter_opt_collect_0_ref:
1721   case _adapter_opt_collect_1_ref:
1722   case _adapter_opt_collect_2_ref:
1723   case _adapter_opt_collect_3_ref:
1724   case _adapter_opt_collect_4_ref:
1725   case _adapter_opt_collect_5_ref:
1726   case _adapter_opt_filter_S0_ref:
1727   case _adapter_opt_filter_S1_ref:
1728   case _adapter_opt_filter_S2_ref:
1729   case _adapter_opt_filter_S3_ref:
1730   case _adapter_opt_filter_S4_ref:
1731   case _adapter_opt_filter_S5_ref:
1732   case _adapter_opt_collect_2_S0_ref:
1733   case _adapter_opt_collect_2_S1_ref:
1734   case _adapter_opt_collect_2_S2_ref:
1735   case _adapter_opt_collect_2_S3_ref:
1736   case _adapter_opt_collect_2_S4_ref:
1737   case _adapter_opt_collect_2_S5_ref:
1738   case _adapter_opt_fold_ref:
1739   case _adapter_opt_fold_int:
1740   case _adapter_opt_fold_long:
1741   case _adapter_opt_fold_float:
1742   case _adapter_opt_fold_double:
1743   case _adapter_opt_fold_void:
1744   case _adapter_opt_fold_1_ref:
1745   case _adapter_opt_fold_2_ref:
1746   case _adapter_opt_fold_3_ref:
1747   case _adapter_opt_fold_4_ref:
1748   case _adapter_opt_fold_5_ref:
1749     {
1750       // Given a fresh incoming stack frame, build a new ricochet frame.
1751       // On entry, TOS points at a return PC, and RBP is the callers frame ptr.
1752       // RSI/R13 has the caller's exact stack pointer, which we must also preserve.
1753       // RCX contains an AdapterMethodHandle of the indicated kind.
1754 
1755       // Relevant AMH fields:
1756       // amh.vmargslot:
1757       //   points to the trailing edge of the arguments
1758       //   to filter, collect, or fold.  For a boxing operation,
1759       //   it points just after the single primitive value.
1760       // amh.argument:
1761       //   recursively called MH, on |collect| arguments
1762       // amh.vmtarget:
1763       //   final destination MH, on return value, etc.
1764       // amh.conversion.dest:
1765       //   tells what is the type of the return value
1766       //   (not needed here, since dest is also derived from ek)
1767       // amh.conversion.vminfo:
1768       //   points to the trailing edge of the return value
1769       //   when the vmtarget is to be called; this is
1770       //   equal to vmargslot + (retained ? |collect| : 0)
1771 
1772       // Pass 0 or more argument slots to the recursive target.
1773       int collect_count_constant = ek_adapter_opt_collect_count(ek);
1774 
1775       // The collected arguments are copied from the saved argument list:
1776       int collect_slot_constant = ek_adapter_opt_collect_slot(ek);
1777 
1778       assert(ek_orig == _adapter_collect_args ||
1779              ek_orig == _adapter_fold_args, "");
1780       bool retain_original_args = (ek_orig == _adapter_fold_args);
1781 
1782       // The return value is replaced (or inserted) at the 'vminfo' argslot.
1783       // Sometimes we can compute this statically.
1784       int dest_slot_constant = -1;
1785       if (!retain_original_args)
1786         dest_slot_constant = collect_slot_constant;
1787       else if (collect_slot_constant >= 0 && collect_count_constant >= 0)
1788         // We are preserving all the arguments, and the return value is prepended,
1789         // so the return slot is to the left (above) the |collect| sequence.
1790         dest_slot_constant = collect_slot_constant + collect_count_constant;
1791 
1792       // Replace all those slots by the result of the recursive call.
1793       // The result type can be one of ref, int, long, float, double, void.
1794       // In the case of void, nothing is pushed on the stack after return.
1795       BasicType dest = ek_adapter_opt_collect_type(ek);
1796       assert(dest == type2wfield[dest], "dest is a stack slot type");
1797       int dest_count = type2size[dest];
1798       assert(dest_count == 1 || dest_count == 2 || (dest_count == 0 && dest == T_VOID), "dest has a size");
1799 
1800       // Choose a return continuation.
1801       EntryKind ek_ret = _adapter_opt_return_any;
1802       if (dest != T_CONFLICT && OptimizeMethodHandles) {
1803         switch (dest) {
1804         case T_INT    : ek_ret = _adapter_opt_return_int;     break;
1805         case T_LONG   : ek_ret = _adapter_opt_return_long;    break;
1806         case T_FLOAT  : ek_ret = _adapter_opt_return_float;   break;
1807         case T_DOUBLE : ek_ret = _adapter_opt_return_double;  break;
1808         case T_OBJECT : ek_ret = _adapter_opt_return_ref;     break;
1809         case T_VOID   : ek_ret = _adapter_opt_return_void;    break;
1810         default       : ShouldNotReachHere();
1811         }
1812         if (dest == T_OBJECT && dest_slot_constant >= 0) {
1813           EntryKind ek_try = EntryKind(_adapter_opt_return_S0_ref + dest_slot_constant);
1814           if (ek_try <= _adapter_opt_return_LAST &&
1815               ek_adapter_opt_return_slot(ek_try) == dest_slot_constant) {
1816             ek_ret = ek_try;
1817           }
1818         }
1819         assert(ek_adapter_opt_return_type(ek_ret) == dest, "");
1820       }
1821 
1822       // Already pushed:  ... keep1 | collect | keep2 | sender_pc |
1823       // push(sender_pc);
1824 
1825       // Compute argument base:
1826       Register rax_argv = rax_argslot;
1827       __ lea(rax_argv, __ argument_address(constant(0)));
1828 
1829       // Push a few extra argument words, if we need them to store the return value.
1830       {
1831         int extra_slots = 0;
1832         if (retain_original_args) {
1833           extra_slots = dest_count;
1834         } else if (collect_count_constant == -1) {
1835           extra_slots = dest_count;  // collect_count might be zero; be generous
1836         } else if (dest_count > collect_count_constant) {
1837           extra_slots = (dest_count - collect_count_constant);
1838         } else {
1839           // else we know we have enough dead space in |collect| to repurpose for return values
1840         }
1841         DEBUG_ONLY(extra_slots += 1);
1842         if (extra_slots > 0) {
1843           __ pop(rbx_temp);   // return value
1844           __ subptr(rsp, (extra_slots * Interpreter::stackElementSize));
1845           // Push guard word #2 in debug mode.
1846           DEBUG_ONLY(__ movptr(Address(rsp, 0), (int32_t) RicochetFrame::MAGIC_NUMBER_2));
1847           __ push(rbx_temp);
1848         }
1849       }
1850 
1851       RicochetFrame::enter_ricochet_frame(_masm, rcx_recv, rax_argv,
1852                                           entry(ek_ret)->from_interpreted_entry(), rbx_temp);
1853 
1854       // Now pushed:  ... keep1 | collect | keep2 | RF |
1855       // some handy frame slots:
1856       Address exact_sender_sp_addr = RicochetFrame::frame_address(RicochetFrame::exact_sender_sp_offset_in_bytes());
1857       Address conversion_addr      = RicochetFrame::frame_address(RicochetFrame::conversion_offset_in_bytes());
1858       Address saved_args_base_addr = RicochetFrame::frame_address(RicochetFrame::saved_args_base_offset_in_bytes());
1859 
1860 #ifdef ASSERT
1861       if (VerifyMethodHandles && dest != T_CONFLICT) {
1862         BLOCK_COMMENT("verify AMH.conv.dest");
1863         load_conversion_dest_type(_masm, rbx_temp, conversion_addr);
1864         Label L_dest_ok;
1865         __ cmpl(rbx_temp, (int) dest);
1866         __ jcc(Assembler::equal, L_dest_ok);
1867         if (dest == T_INT) {
1868           for (int bt = T_BOOLEAN; bt < T_INT; bt++) {
1869             if (is_subword_type(BasicType(bt))) {
1870               __ cmpl(rbx_temp, (int) bt);
1871               __ jcc(Assembler::equal, L_dest_ok);
1872             }
1873           }
1874         }
1875         __ stop("bad dest in AMH.conv");
1876         __ BIND(L_dest_ok);
1877       }
1878 #endif //ASSERT
1879 
1880       // Find out where the original copy of the recursive argument sequence begins.
1881       Register rax_coll = rax_argv;
1882       {
1883         RegisterOrConstant collect_slot = collect_slot_constant;
1884         if (collect_slot_constant == -1) {
1885           __ movl(rdi_temp, rcx_amh_vmargslot);
1886           collect_slot = rdi_temp;
1887         }
1888         if (collect_slot_constant != 0)
1889           __ lea(rax_coll, Address(rax_argv, collect_slot, Interpreter::stackElementScale()));
1890         // rax_coll now points at the trailing edge of |collect| and leading edge of |keep2|
1891       }
1892 
1893       // Replace the old AMH with the recursive MH.  (No going back now.)
1894       // In the case of a boxing call, the recursive call is to a 'boxer' method,
1895       // such as Integer.valueOf or Long.valueOf.  In the case of a filter
1896       // or collect call, it will take one or more arguments, transform them,
1897       // and return some result, to store back into argument_base[vminfo].
1898       __ load_heap_oop(rcx_recv, rcx_amh_argument);
1899       if (VerifyMethodHandles)  verify_method_handle(_masm, rcx_recv);
1900 
1901       // Push a space for the recursively called MH first:
1902       __ push((int32_t)NULL_WORD);
1903 
1904       // Calculate |collect|, the number of arguments we are collecting.
1905       Register rdi_collect_count = rdi_temp;
1906       RegisterOrConstant collect_count;
1907       if (collect_count_constant >= 0) {
1908         collect_count = collect_count_constant;
1909       } else {
1910         __ load_method_handle_vmslots(rdi_collect_count, rcx_recv, rdx_temp);
1911         collect_count = rdi_collect_count;
1912       }
1913 #ifdef ASSERT
1914       if (VerifyMethodHandles && collect_count_constant >= 0) {
1915         __ load_method_handle_vmslots(rbx_temp, rcx_recv, rdx_temp);
1916         Label L_count_ok;
1917         __ cmpl(rbx_temp, collect_count_constant);
1918         __ jcc(Assembler::equal, L_count_ok);
1919         __ stop("bad vminfo in AMH.conv");
1920         __ BIND(L_count_ok);
1921       }
1922 #endif //ASSERT
1923 
1924       // copy |collect| slots directly to TOS:
1925       push_arg_slots(_masm, rax_coll, collect_count, 0, rbx_temp, rdx_temp);
1926       // Now pushed:  ... keep1 | collect | keep2 | RF... | collect |
1927       // rax_coll still points at the trailing edge of |collect| and leading edge of |keep2|
1928 
1929       // If necessary, adjust the saved arguments to make room for the eventual return value.
1930       // Normal adjustment:  ... keep1 | +dest+ | -collect- | keep2 | RF... | collect |
1931       // If retaining args:  ... keep1 | +dest+ |  collect  | keep2 | RF... | collect |
1932       // In the non-retaining case, this might move keep2 either up or down.
1933       // We don't have to copy the whole | RF... collect | complex,
1934       // but we must adjust RF.saved_args_base.
1935       // Also, from now on, we will forget about the original copy of |collect|.
1936       // If we are retaining it, we will treat it as part of |keep2|.
1937       // For clarity we will define |keep3| = |collect|keep2| or |keep2|.
1938 
1939       BLOCK_COMMENT("adjust trailing arguments {");
1940       // Compare the sizes of |+dest+| and |-collect-|, which are opposed opening and closing movements.
1941       int                open_count  = dest_count;
1942       RegisterOrConstant close_count = collect_count_constant;
1943       Register rdi_close_count = rdi_collect_count;
1944       if (retain_original_args) {
1945         close_count = constant(0);
1946       } else if (collect_count_constant == -1) {
1947         close_count = rdi_collect_count;
1948       }
1949 
1950       // How many slots need moving?  This is simply dest_slot (0 => no |keep3|).
1951       RegisterOrConstant keep3_count;
1952       Register rsi_keep3_count = rsi;  // can repair from RF.exact_sender_sp
1953       if (dest_slot_constant >= 0) {
1954         keep3_count = dest_slot_constant;
1955       } else  {
1956         load_conversion_vminfo(_masm, rsi_keep3_count, conversion_addr);
1957         keep3_count = rsi_keep3_count;
1958       }
1959 #ifdef ASSERT
1960       if (VerifyMethodHandles && dest_slot_constant >= 0) {
1961         load_conversion_vminfo(_masm, rbx_temp, conversion_addr);
1962         Label L_vminfo_ok;
1963         __ cmpl(rbx_temp, dest_slot_constant);
1964         __ jcc(Assembler::equal, L_vminfo_ok);
1965         __ stop("bad vminfo in AMH.conv");
1966         __ BIND(L_vminfo_ok);
1967       }
1968 #endif //ASSERT
1969 
1970       // tasks remaining:
1971       bool move_keep3 = (!keep3_count.is_constant() || keep3_count.as_constant() != 0);
1972       bool stomp_dest = (NOT_DEBUG(dest == T_OBJECT) DEBUG_ONLY(dest_count != 0));
1973       bool fix_arg_base = (!close_count.is_constant() || open_count != close_count.as_constant());
1974 
1975       if (stomp_dest | fix_arg_base) {
1976         // we will probably need an updated rax_argv value
1977         if (collect_slot_constant >= 0) {
1978           // rax_coll already holds the leading edge of |keep2|, so tweak it
1979           assert(rax_coll == rax_argv, "elided a move");
1980           if (collect_slot_constant != 0)
1981             __ subptr(rax_argv, collect_slot_constant * Interpreter::stackElementSize);
1982         } else {
1983           // Just reload from RF.saved_args_base.
1984           __ movptr(rax_argv, saved_args_base_addr);
1985         }
1986       }
1987 
1988       // Old and new argument locations (based at slot 0).
1989       // Net shift (&new_argv - &old_argv) is (close_count - open_count).
1990       bool zero_open_count = (open_count == 0);  // remember this bit of info
1991       if (move_keep3 && fix_arg_base) {
1992         // It will be easier to have everything in one register:
1993         if (close_count.is_register()) {
1994           // Deduct open_count from close_count register to get a clean +/- value.
1995           __ subptr(close_count.as_register(), open_count);
1996         } else {
1997           close_count = close_count.as_constant() - open_count;
1998         }
1999         open_count = 0;
2000       }
2001       Address old_argv(rax_argv, 0);
2002       Address new_argv(rax_argv, close_count,  Interpreter::stackElementScale(),
2003                                 - open_count * Interpreter::stackElementSize);
2004 
2005       // First decide if any actual data are to be moved.
2006       // We can skip if (a) |keep3| is empty, or (b) the argument list size didn't change.
2007       // (As it happens, all movements involve an argument list size change.)
2008 
2009       // If there are variable parameters, use dynamic checks to skip around the whole mess.
2010       Label L_done;
2011       if (!keep3_count.is_constant()) {
2012         __ testl(keep3_count.as_register(), keep3_count.as_register());
2013         __ jcc(Assembler::zero, L_done);
2014       }
2015       if (!close_count.is_constant()) {
2016         __ cmpl(close_count.as_register(), open_count);
2017         __ jcc(Assembler::equal, L_done);
2018       }
2019 
2020       if (move_keep3 && fix_arg_base) {
2021         bool emit_move_down = false, emit_move_up = false, emit_guard = false;
2022         if (!close_count.is_constant()) {
2023           emit_move_down = emit_guard = !zero_open_count;
2024           emit_move_up   = true;
2025         } else if (open_count != close_count.as_constant()) {
2026           emit_move_down = (open_count > close_count.as_constant());
2027           emit_move_up   = !emit_move_down;
2028         }
2029         Label L_move_up;
2030         if (emit_guard) {
2031           __ cmpl(close_count.as_register(), open_count);
2032           __ jcc(Assembler::greater, L_move_up);
2033         }
2034 
2035         if (emit_move_down) {
2036           // Move arguments down if |+dest+| > |-collect-|
2037           // (This is rare, except when arguments are retained.)
2038           // This opens space for the return value.
2039           if (keep3_count.is_constant()) {
2040             for (int i = 0; i < keep3_count.as_constant(); i++) {
2041               __ movptr(rdx_temp, old_argv.plus_disp(i * Interpreter::stackElementSize));
2042               __ movptr(          new_argv.plus_disp(i * Interpreter::stackElementSize), rdx_temp);
2043             }
2044           } else {
2045             Register rbx_argv_top = rbx_temp;
2046             __ lea(rbx_argv_top, old_argv.plus_disp(keep3_count, Interpreter::stackElementScale()));
2047             move_arg_slots_down(_masm,
2048                                 old_argv,     // beginning of old argv
2049                                 rbx_argv_top, // end of old argv
2050                                 close_count,  // distance to move down (must be negative)
2051                                 rax_argv, rdx_temp);
2052             // Used argv as an iteration variable; reload from RF.saved_args_base.
2053             __ movptr(rax_argv, saved_args_base_addr);
2054           }
2055         }
2056 
2057         if (emit_guard) {
2058           __ jmp(L_done);  // assumes emit_move_up is true also
2059           __ BIND(L_move_up);
2060         }
2061 
2062         if (emit_move_up) {
2063 
2064           // Move arguments up if |+dest+| < |-collect-|
2065           // (This is usual, except when |keep3| is empty.)
2066           // This closes up the space occupied by the now-deleted collect values.
2067           if (keep3_count.is_constant()) {
2068             for (int i = keep3_count.as_constant() - 1; i >= 0; i--) {
2069               __ movptr(rdx_temp, old_argv.plus_disp(i * Interpreter::stackElementSize));
2070               __ movptr(          new_argv.plus_disp(i * Interpreter::stackElementSize), rdx_temp);
2071             }
2072           } else {
2073             Address argv_top = old_argv.plus_disp(keep3_count, Interpreter::stackElementScale());
2074             move_arg_slots_up(_masm,
2075                               rax_argv,     // beginning of old argv
2076                               argv_top,     // end of old argv
2077                               close_count,  // distance to move up (must be positive)
2078                               rbx_temp, rdx_temp);
2079           }
2080         }
2081       }
2082       __ BIND(L_done);
2083 
2084       if (fix_arg_base) {
2085         // adjust RF.saved_args_base by adding (close_count - open_count)
2086         if (!new_argv.is_same_address(Address(rax_argv, 0)))
2087           __ lea(rax_argv, new_argv);
2088         __ movptr(saved_args_base_addr, rax_argv);
2089       }
2090 
2091       if (stomp_dest) {
2092         // Stomp the return slot, so it doesn't hold garbage.
2093         // This isn't strictly necessary, but it may help detect bugs.
2094         int forty_two = RicochetFrame::RETURN_VALUE_PLACEHOLDER;
2095         __ movptr(Address(rax_argv, keep3_count, Address::times_ptr),
2096                   (int32_t) forty_two);
2097         // uses rsi_keep3_count
2098       }
2099       BLOCK_COMMENT("} adjust trailing arguments");
2100 
2101       BLOCK_COMMENT("do_recursive_call");
2102       __ mov(saved_last_sp, rsp);    // set rsi/r13 for callee
2103       __ pushptr(ExternalAddress(SharedRuntime::ricochet_blob()->bounce_addr()).addr());
2104       // The globally unique bounce address has two purposes:
2105       // 1. It helps the JVM recognize this frame (frame::is_ricochet_frame).
2106       // 2. When returned to, it cuts back the stack and redirects control flow
2107       //    to the return handler.
2108       // The return handler will further cut back the stack when it takes
2109       // down the RF.  Perhaps there is a way to streamline this further.
2110 
2111       // State during recursive call:
2112       // ... keep1 | dest | dest=42 | keep3 | RF... | collect | bounce_pc |
2113       __ jump_to_method_handle_entry(rcx_recv, rdx_temp);
2114 
2115       break;
2116     }
2117 
2118   case _adapter_opt_return_ref:
2119   case _adapter_opt_return_int:
2120   case _adapter_opt_return_long:
2121   case _adapter_opt_return_float:
2122   case _adapter_opt_return_double:
2123   case _adapter_opt_return_void:
2124   case _adapter_opt_return_S0_ref:
2125   case _adapter_opt_return_S1_ref:
2126   case _adapter_opt_return_S2_ref:
2127   case _adapter_opt_return_S3_ref:
2128   case _adapter_opt_return_S4_ref:
2129   case _adapter_opt_return_S5_ref:
2130     {
2131       BasicType dest_type_constant = ek_adapter_opt_return_type(ek);
2132       int       dest_slot_constant = ek_adapter_opt_return_slot(ek);
2133 
2134       if (VerifyMethodHandles)  RicochetFrame::verify_clean(_masm);
2135 
2136       if (dest_slot_constant == -1) {
2137         // The current stub is a general handler for this dest_type.
2138         // It can be called from _adapter_opt_return_any below.
2139         // Stash the address in a little table.
2140         assert((dest_type_constant & CONV_TYPE_MASK) == dest_type_constant, "oob");
2141         address return_handler = __ pc();
2142         _adapter_return_handlers[dest_type_constant] = return_handler;
2143         if (dest_type_constant == T_INT) {
2144           // do the subword types too
2145           for (int bt = T_BOOLEAN; bt < T_INT; bt++) {
2146             if (is_subword_type(BasicType(bt)) &&
2147                 _adapter_return_handlers[bt] == NULL) {
2148               _adapter_return_handlers[bt] = return_handler;
2149             }
2150           }
2151         }
2152       }
2153 
2154       Register rbx_arg_base = rbx_temp;
2155       assert_different_registers(rax, rdx,  // possibly live return value registers
2156                                  rdi_temp, rbx_arg_base);
2157 
2158       Address conversion_addr      = RicochetFrame::frame_address(RicochetFrame::conversion_offset_in_bytes());
2159       Address saved_args_base_addr = RicochetFrame::frame_address(RicochetFrame::saved_args_base_offset_in_bytes());
2160 
2161       __ movptr(rbx_arg_base, saved_args_base_addr);
2162       RegisterOrConstant dest_slot = dest_slot_constant;
2163       if (dest_slot_constant == -1) {
2164         load_conversion_vminfo(_masm, rdi_temp, conversion_addr);
2165         dest_slot = rdi_temp;
2166       }
2167       // Store the result back into the argslot.
2168       // This code uses the interpreter calling sequence, in which the return value
2169       // is usually left in the TOS register, as defined by InterpreterMacroAssembler::pop.
2170       // There are certain irregularities with floating point values, which can be seen
2171       // in TemplateInterpreterGenerator::generate_return_entry_for.
2172       move_return_value(_masm, dest_type_constant, Address(rbx_arg_base, dest_slot, Interpreter::stackElementScale()));
2173 
2174       RicochetFrame::leave_ricochet_frame(_masm, rcx_recv, rbx_arg_base, rdx_temp);
2175       __ push(rdx_temp);  // repush the return PC
2176 
2177       // Load the final target and go.
2178       if (VerifyMethodHandles)  verify_method_handle(_masm, rcx_recv);
2179       __ jump_to_method_handle_entry(rcx_recv, rdx_temp);
2180       __ hlt(); // --------------------
2181       break;
2182     }
2183 
2184   case _adapter_opt_return_any:
2185     {
2186       if (VerifyMethodHandles)  RicochetFrame::verify_clean(_masm);
2187       Register rdi_conv = rdi_temp;
2188       assert_different_registers(rax, rdx,  // possibly live return value registers
2189                                  rdi_conv, rbx_temp);
2190 
2191       Address conversion_addr = RicochetFrame::frame_address(RicochetFrame::conversion_offset_in_bytes());
2192       load_conversion_dest_type(_masm, rdi_conv, conversion_addr);
2193       __ lea(rbx_temp, ExternalAddress((address) &_adapter_return_handlers[0]));
2194       __ movptr(rbx_temp, Address(rbx_temp, rdi_conv, Address::times_ptr));
2195 
2196 #ifdef ASSERT
2197       { Label L_badconv;
2198         __ testptr(rbx_temp, rbx_temp);
2199         __ jccb(Assembler::zero, L_badconv);
2200         __ jmp(rbx_temp);
2201         __ bind(L_badconv);
2202         __ stop("bad method handle return");
2203       }
2204 #else //ASSERT
2205       __ jmp(rbx_temp);
2206 #endif //ASSERT
2207       break;
2208     }
2209 
2210   case _adapter_opt_spread_0:
2211   case _adapter_opt_spread_1_ref:
2212   case _adapter_opt_spread_2_ref:
2213   case _adapter_opt_spread_3_ref:
2214   case _adapter_opt_spread_4_ref:
2215   case _adapter_opt_spread_5_ref:
2216   case _adapter_opt_spread_ref:
2217   case _adapter_opt_spread_byte:
2218   case _adapter_opt_spread_char:
2219   case _adapter_opt_spread_short:
2220   case _adapter_opt_spread_int:
2221   case _adapter_opt_spread_long:
2222   case _adapter_opt_spread_float:
2223   case _adapter_opt_spread_double:
2224     {
2225       // spread an array out into a group of arguments
2226       int length_constant = ek_adapter_opt_spread_count(ek);
2227       bool length_can_be_zero = (length_constant == 0);
2228       if (length_constant < 0) {
2229         // some adapters with variable length must handle the zero case
2230         if (!OptimizeMethodHandles ||
2231             ek_adapter_opt_spread_type(ek) != T_OBJECT)
2232           length_can_be_zero = true;
2233       }
2234 
2235       // find the address of the array argument
2236       __ movl(rax_argslot, rcx_amh_vmargslot);
2237       __ lea(rax_argslot, __ argument_address(rax_argslot));
2238 
2239       // grab another temp
2240       Register rsi_temp = rsi;
2241       { if (rsi_temp == saved_last_sp)  __ push(saved_last_sp); }
2242       // (preceding push must be done after argslot address is taken!)
2243 #define UNPUSH_RSI \
2244       { if (rsi_temp == saved_last_sp)  __ pop(saved_last_sp); }
2245 
2246       // arx_argslot points both to the array and to the first output arg
2247       vmarg = Address(rax_argslot, 0);
2248 
2249       // Get the array value.
2250       Register  rsi_array       = rsi_temp;
2251       Register  rdx_array_klass = rdx_temp;
2252       BasicType elem_type = ek_adapter_opt_spread_type(ek);
2253       int       elem_slots = type2size[elem_type];  // 1 or 2
2254       int       array_slots = 1;  // array is always a T_OBJECT
2255       int       length_offset   = arrayOopDesc::length_offset_in_bytes();
2256       int       elem0_offset    = arrayOopDesc::base_offset_in_bytes(elem_type);
2257       __ movptr(rsi_array, vmarg);
2258 
2259       Label L_array_is_empty, L_insert_arg_space, L_copy_args, L_args_done;
2260       if (length_can_be_zero) {
2261         // handle the null pointer case, if zero is allowed
2262         Label L_skip;
2263         if (length_constant < 0) {
2264           load_conversion_vminfo(_masm, rbx_temp, rcx_amh_conversion);
2265           __ testl(rbx_temp, rbx_temp);
2266           __ jcc(Assembler::notZero, L_skip);
2267         }
2268         __ testptr(rsi_array, rsi_array);
2269         __ jcc(Assembler::zero, L_array_is_empty);
2270         __ bind(L_skip);
2271       }
2272       __ null_check(rsi_array, oopDesc::klass_offset_in_bytes());
2273       __ load_klass(rdx_array_klass, rsi_array);
2274 
2275       // Check the array type.
2276       Register rbx_klass = rbx_temp;
2277       __ load_heap_oop(rbx_klass, rcx_amh_argument); // this is a Class object!
2278       load_klass_from_Class(_masm, rbx_klass);
2279 
2280       Label ok_array_klass, bad_array_klass, bad_array_length;
2281       __ check_klass_subtype(rdx_array_klass, rbx_klass, rdi_temp, ok_array_klass);
2282       // If we get here, the type check failed!
2283       __ jmp(bad_array_klass);
2284       __ BIND(ok_array_klass);
2285 
2286       // Check length.
2287       if (length_constant >= 0) {
2288         __ cmpl(Address(rsi_array, length_offset), length_constant);
2289       } else {
2290         Register rbx_vminfo = rbx_temp;
2291         load_conversion_vminfo(_masm, rbx_vminfo, rcx_amh_conversion);
2292         __ cmpl(rbx_vminfo, Address(rsi_array, length_offset));
2293       }
2294       __ jcc(Assembler::notEqual, bad_array_length);
2295 
2296       Register rdx_argslot_limit = rdx_temp;
2297 
2298       // Array length checks out.  Now insert any required stack slots.
2299       if (length_constant == -1) {
2300         // Form a pointer to the end of the affected region.
2301         __ lea(rdx_argslot_limit, Address(rax_argslot, Interpreter::stackElementSize));
2302         // 'stack_move' is negative number of words to insert
2303         // This number already accounts for elem_slots.
2304         Register rdi_stack_move = rdi_temp;
2305         load_stack_move(_masm, rdi_stack_move, rcx_recv, true);
2306         __ cmpptr(rdi_stack_move, 0);
2307         assert(stack_move_unit() < 0, "else change this comparison");
2308         __ jcc(Assembler::less, L_insert_arg_space);
2309         __ jcc(Assembler::equal, L_copy_args);
2310         // single argument case, with no array movement
2311         __ BIND(L_array_is_empty);
2312         remove_arg_slots(_masm, -stack_move_unit() * array_slots,
2313                          rax_argslot, rbx_temp, rdx_temp);
2314         __ jmp(L_args_done);  // no spreading to do
2315         __ BIND(L_insert_arg_space);
2316         // come here in the usual case, stack_move < 0 (2 or more spread arguments)
2317         Register rsi_temp = rsi_array;  // spill this
2318         insert_arg_slots(_masm, rdi_stack_move,
2319                          rax_argslot, rbx_temp, rsi_temp);
2320         // reload the array since rsi was killed
2321         // reload from rdx_argslot_limit since rax_argslot is now decremented
2322         __ movptr(rsi_array, Address(rdx_argslot_limit, -Interpreter::stackElementSize));
2323       } else if (length_constant >= 1) {
2324         int new_slots = (length_constant * elem_slots) - array_slots;
2325         insert_arg_slots(_masm, new_slots * stack_move_unit(),
2326                          rax_argslot, rbx_temp, rdx_temp);
2327       } else if (length_constant == 0) {
2328         __ BIND(L_array_is_empty);
2329         remove_arg_slots(_masm, -stack_move_unit() * array_slots,
2330                          rax_argslot, rbx_temp, rdx_temp);
2331       } else {
2332         ShouldNotReachHere();
2333       }
2334 
2335       // Copy from the array to the new slots.
2336       // Note: Stack change code preserves integrity of rax_argslot pointer.
2337       // So even after slot insertions, rax_argslot still points to first argument.
2338       // Beware:  Arguments that are shallow on the stack are deep in the array,
2339       // and vice versa.  So a downward-growing stack (the usual) has to be copied
2340       // elementwise in reverse order from the source array.
2341       __ BIND(L_copy_args);
2342       if (length_constant == -1) {
2343         // [rax_argslot, rdx_argslot_limit) is the area we are inserting into.
2344         // Array element [0] goes at rdx_argslot_limit[-wordSize].
2345         Register rsi_source = rsi_array;
2346         __ lea(rsi_source, Address(rsi_array, elem0_offset));
2347         Register rdx_fill_ptr = rdx_argslot_limit;
2348         Label loop;
2349         __ BIND(loop);
2350         __ addptr(rdx_fill_ptr, -Interpreter::stackElementSize * elem_slots);
2351         move_typed_arg(_masm, elem_type, true,
2352                        Address(rdx_fill_ptr, 0), Address(rsi_source, 0),
2353                        rbx_temp, rdi_temp);
2354         __ addptr(rsi_source, type2aelembytes(elem_type));
2355         __ cmpptr(rdx_fill_ptr, rax_argslot);
2356         __ jcc(Assembler::above, loop);
2357       } else if (length_constant == 0) {
2358         // nothing to copy
2359       } else {
2360         int elem_offset = elem0_offset;
2361         int slot_offset = length_constant * Interpreter::stackElementSize;
2362         for (int index = 0; index < length_constant; index++) {
2363           slot_offset -= Interpreter::stackElementSize * elem_slots;  // fill backward
2364           move_typed_arg(_masm, elem_type, true,
2365                          Address(rax_argslot, slot_offset), Address(rsi_array, elem_offset),
2366                          rbx_temp, rdi_temp);
2367           elem_offset += type2aelembytes(elem_type);
2368         }
2369       }
2370       __ BIND(L_args_done);
2371 
2372       // Arguments are spread.  Move to next method handle.
2373       UNPUSH_RSI;
2374       __ load_heap_oop(rcx_recv, rcx_mh_vmtarget);
2375       __ jump_to_method_handle_entry(rcx_recv, rdx_temp);
2376 
2377       __ bind(bad_array_klass);
2378       UNPUSH_RSI;
2379       assert(!vmarg.uses(rarg2_required), "must be different registers");
2380       __ load_heap_oop( rarg2_required, Address(rdx_array_klass, java_mirror_offset));  // required type
2381       __ movptr(        rarg1_actual,   vmarg);                                         // bad array
2382       __ movl(          rarg0_code,     (int) Bytecodes::_aaload);                      // who is complaining?
2383       __ jump(ExternalAddress(from_interpreted_entry(_raise_exception)));
2384 
2385       __ bind(bad_array_length);
2386       UNPUSH_RSI;
2387       assert(!vmarg.uses(rarg2_required), "must be different registers");
2388       __ mov(    rarg2_required, rcx_recv);                       // AMH requiring a certain length
2389       __ movptr( rarg1_actual,   vmarg);                          // bad array
2390       __ movl(   rarg0_code,     (int) Bytecodes::_arraylength);  // who is complaining?
2391       __ jump(ExternalAddress(from_interpreted_entry(_raise_exception)));
2392 #undef UNPUSH_RSI
2393 
2394       break;
2395     }
2396 
2397   default:
2398     // do not require all platforms to recognize all adapter types
2399     __ nop();
2400     return;
2401   }
2402   BLOCK_COMMENT(err_msg("} Entry %s", entry_name(ek)));
2403   __ hlt();
2404 
2405   address me_cookie = MethodHandleEntry::start_compiled_entry(_masm, interp_entry);
2406   __ unimplemented(entry_name(ek)); // %%% FIXME: NYI
2407 
2408   init_entry(ek, MethodHandleEntry::finish_compiled_entry(_masm, me_cookie));
2409 }