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