1 /*
   2  * Copyright (c) 2003, 2018, 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 "asm/macroAssembler.hpp"
  27 #include "asm/macroAssembler.inline.hpp"
  28 #include "code/debugInfoRec.hpp"
  29 #include "code/icBuffer.hpp"
  30 #include "code/nativeInst.hpp"
  31 #include "code/vtableStubs.hpp"
  32 #include "gc/shared/gcLocker.hpp"
  33 #include "interpreter/interpreter.hpp"
  34 #include "logging/log.hpp"
  35 #include "memory/resourceArea.hpp"
  36 #include "oops/compiledICHolder.hpp"
  37 #include "runtime/safepointMechanism.hpp"
  38 #include "runtime/sharedRuntime.hpp"
  39 #include "runtime/vframeArray.hpp"
  40 #include "utilities/align.hpp"
  41 #include "vmreg_x86.inline.hpp"
  42 #ifdef COMPILER1
  43 #include "c1/c1_Runtime1.hpp"
  44 #endif
  45 #ifdef COMPILER2
  46 #include "opto/runtime.hpp"
  47 #endif
  48 #include "vm_version_x86.hpp"
  49 
  50 #define __ masm->
  51 
  52 const int StackAlignmentInSlots = StackAlignmentInBytes / VMRegImpl::stack_slot_size;
  53 
  54 class RegisterSaver {
  55   // Capture info about frame layout
  56 #define DEF_XMM_OFFS(regnum) xmm ## regnum ## _off = xmm_off + (regnum)*16/BytesPerInt, xmm ## regnum ## H_off
  57   enum layout {
  58                 fpu_state_off = 0,
  59                 fpu_state_end = fpu_state_off+FPUStateSizeInWords,
  60                 st0_off, st0H_off,
  61                 st1_off, st1H_off,
  62                 st2_off, st2H_off,
  63                 st3_off, st3H_off,
  64                 st4_off, st4H_off,
  65                 st5_off, st5H_off,
  66                 st6_off, st6H_off,
  67                 st7_off, st7H_off,
  68                 xmm_off,
  69                 DEF_XMM_OFFS(0),
  70                 DEF_XMM_OFFS(1),
  71                 DEF_XMM_OFFS(2),
  72                 DEF_XMM_OFFS(3),
  73                 DEF_XMM_OFFS(4),
  74                 DEF_XMM_OFFS(5),
  75                 DEF_XMM_OFFS(6),
  76                 DEF_XMM_OFFS(7),
  77                 flags_off = xmm7_off + 16/BytesPerInt + 1, // 16-byte stack alignment fill word
  78                 rdi_off,
  79                 rsi_off,
  80                 ignore_off,  // extra copy of rbp,
  81                 rsp_off,
  82                 rbx_off,
  83                 rdx_off,
  84                 rcx_off,
  85                 rax_off,
  86                 // The frame sender code expects that rbp will be in the "natural" place and
  87                 // will override any oopMap setting for it. We must therefore force the layout
  88                 // so that it agrees with the frame sender code.
  89                 rbp_off,
  90                 return_off,      // slot for return address
  91                 reg_save_size };
  92   enum { FPU_regs_live = flags_off - fpu_state_end };
  93 
  94   public:
  95 
  96   static OopMap* save_live_registers(MacroAssembler* masm, int additional_frame_words,
  97                                      int* total_frame_words, bool verify_fpu = true, bool save_vectors = false);
  98   static void restore_live_registers(MacroAssembler* masm, bool restore_vectors = false);
  99 
 100   static int rax_offset() { return rax_off; }
 101   static int rbx_offset() { return rbx_off; }
 102 
 103   // Offsets into the register save area
 104   // Used by deoptimization when it is managing result register
 105   // values on its own
 106 
 107   static int raxOffset(void) { return rax_off; }
 108   static int rdxOffset(void) { return rdx_off; }
 109   static int rbxOffset(void) { return rbx_off; }
 110   static int xmm0Offset(void) { return xmm0_off; }
 111   // This really returns a slot in the fp save area, which one is not important
 112   static int fpResultOffset(void) { return st0_off; }
 113 
 114   // During deoptimization only the result register need to be restored
 115   // all the other values have already been extracted.
 116 
 117   static void restore_result_registers(MacroAssembler* masm);
 118 
 119 };
 120 
 121 OopMap* RegisterSaver::save_live_registers(MacroAssembler* masm, int additional_frame_words,
 122                                            int* total_frame_words, bool verify_fpu, bool save_vectors) {
 123   int num_xmm_regs = XMMRegisterImpl::number_of_registers;
 124   int ymm_bytes = num_xmm_regs * 16;
 125   int zmm_bytes = num_xmm_regs * 32;
 126 #ifdef COMPILER2
 127   if (save_vectors) {
 128     assert(UseAVX > 0, "Vectors larger than 16 byte long are supported only with AVX");
 129     assert(MaxVectorSize <= 64, "Only up to 64 byte long vectors are supported");
 130     // Save upper half of YMM registers
 131     int vect_bytes = ymm_bytes;
 132     if (UseAVX > 2) {
 133       // Save upper half of ZMM registers as well
 134       vect_bytes += zmm_bytes;
 135     }
 136     additional_frame_words += vect_bytes / wordSize;
 137   }
 138 #else
 139   assert(!save_vectors, "vectors are generated only by C2");
 140 #endif
 141   int frame_size_in_bytes = (reg_save_size + additional_frame_words) * wordSize;
 142   int frame_words = frame_size_in_bytes / wordSize;
 143   *total_frame_words = frame_words;
 144 
 145   assert(FPUStateSizeInWords == 27, "update stack layout");
 146 
 147   // save registers, fpu state, and flags
 148   // We assume caller has already has return address slot on the stack
 149   // We push epb twice in this sequence because we want the real rbp,
 150   // to be under the return like a normal enter and we want to use pusha
 151   // We push by hand instead of using push.
 152   __ enter();
 153   __ pusha();
 154   __ pushf();
 155   __ subptr(rsp,FPU_regs_live*wordSize); // Push FPU registers space
 156   __ push_FPU_state();          // Save FPU state & init
 157 
 158   if (verify_fpu) {
 159     // Some stubs may have non standard FPU control word settings so
 160     // only check and reset the value when it required to be the
 161     // standard value.  The safepoint blob in particular can be used
 162     // in methods which are using the 24 bit control word for
 163     // optimized float math.
 164 
 165 #ifdef ASSERT
 166     // Make sure the control word has the expected value
 167     Label ok;
 168     __ cmpw(Address(rsp, 0), StubRoutines::fpu_cntrl_wrd_std());
 169     __ jccb(Assembler::equal, ok);
 170     __ stop("corrupted control word detected");
 171     __ bind(ok);
 172 #endif
 173 
 174     // Reset the control word to guard against exceptions being unmasked
 175     // since fstp_d can cause FPU stack underflow exceptions.  Write it
 176     // into the on stack copy and then reload that to make sure that the
 177     // current and future values are correct.
 178     __ movw(Address(rsp, 0), StubRoutines::fpu_cntrl_wrd_std());
 179   }
 180 
 181   __ frstor(Address(rsp, 0));
 182   if (!verify_fpu) {
 183     // Set the control word so that exceptions are masked for the
 184     // following code.
 185     __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
 186   }
 187 
 188   int off = st0_off;
 189   int delta = st1_off - off;
 190 
 191   // Save the FPU registers in de-opt-able form
 192   for (int n = 0; n < FloatRegisterImpl::number_of_registers; n++) {
 193     __ fstp_d(Address(rsp, off*wordSize));
 194     off += delta;
 195   }
 196 
 197   off = xmm0_off;
 198   delta = xmm1_off - off;
 199   if(UseSSE == 1) {
 200     // Save the XMM state
 201     for (int n = 0; n < num_xmm_regs; n++) {
 202       __ movflt(Address(rsp, off*wordSize), as_XMMRegister(n));
 203       off += delta;
 204     }
 205   } else if(UseSSE >= 2) {
 206     // Save whole 128bit (16 bytes) XMM registers
 207     for (int n = 0; n < num_xmm_regs; n++) {
 208       __ movdqu(Address(rsp, off*wordSize), as_XMMRegister(n));
 209       off += delta;
 210     }
 211   }
 212 
 213   if (save_vectors) {
 214     __ subptr(rsp, ymm_bytes);
 215     // Save upper half of YMM registers
 216     for (int n = 0; n < num_xmm_regs; n++) {
 217       __ vextractf128_high(Address(rsp, n*16), as_XMMRegister(n));
 218     }
 219     if (UseAVX > 2) {
 220       __ subptr(rsp, zmm_bytes);
 221       // Save upper half of ZMM registers
 222       for (int n = 0; n < num_xmm_regs; n++) {
 223         __ vextractf64x4_high(Address(rsp, n*32), as_XMMRegister(n));
 224       }
 225     }
 226   }
 227   __ vzeroupper();
 228 
 229   // Set an oopmap for the call site.  This oopmap will map all
 230   // oop-registers and debug-info registers as callee-saved.  This
 231   // will allow deoptimization at this safepoint to find all possible
 232   // debug-info recordings, as well as let GC find all oops.
 233 
 234   OopMapSet *oop_maps = new OopMapSet();
 235   OopMap* map =  new OopMap( frame_words, 0 );
 236 
 237 #define STACK_OFFSET(x) VMRegImpl::stack2reg((x) + additional_frame_words)
 238 #define NEXTREG(x) (x)->as_VMReg()->next()
 239 
 240   map->set_callee_saved(STACK_OFFSET(rax_off), rax->as_VMReg());
 241   map->set_callee_saved(STACK_OFFSET(rcx_off), rcx->as_VMReg());
 242   map->set_callee_saved(STACK_OFFSET(rdx_off), rdx->as_VMReg());
 243   map->set_callee_saved(STACK_OFFSET(rbx_off), rbx->as_VMReg());
 244   // rbp, location is known implicitly, no oopMap
 245   map->set_callee_saved(STACK_OFFSET(rsi_off), rsi->as_VMReg());
 246   map->set_callee_saved(STACK_OFFSET(rdi_off), rdi->as_VMReg());
 247   // %%% This is really a waste but we'll keep things as they were for now for the upper component
 248   off = st0_off;
 249   delta = st1_off - off;
 250   for (int n = 0; n < FloatRegisterImpl::number_of_registers; n++) {
 251     FloatRegister freg_name = as_FloatRegister(n);
 252     map->set_callee_saved(STACK_OFFSET(off), freg_name->as_VMReg());
 253     map->set_callee_saved(STACK_OFFSET(off+1), NEXTREG(freg_name));
 254     off += delta;
 255   }
 256   off = xmm0_off;
 257   delta = xmm1_off - off;
 258   for (int n = 0; n < num_xmm_regs; n++) {
 259     XMMRegister xmm_name = as_XMMRegister(n);
 260     map->set_callee_saved(STACK_OFFSET(off), xmm_name->as_VMReg());
 261     map->set_callee_saved(STACK_OFFSET(off+1), NEXTREG(xmm_name));
 262     off += delta;
 263   }
 264 #undef NEXTREG
 265 #undef STACK_OFFSET
 266 
 267   return map;
 268 }
 269 
 270 void RegisterSaver::restore_live_registers(MacroAssembler* masm, bool restore_vectors) {
 271   int num_xmm_regs = XMMRegisterImpl::number_of_registers;
 272   int ymm_bytes = num_xmm_regs * 16;
 273   int zmm_bytes = num_xmm_regs * 32;
 274   // Recover XMM & FPU state
 275   int additional_frame_bytes = 0;
 276 #ifdef COMPILER2
 277   if (restore_vectors) {
 278     assert(UseAVX > 0, "Vectors larger than 16 byte long are supported only with AVX");
 279     assert(MaxVectorSize <= 64, "Only up to 64 byte long vectors are supported");
 280     // Save upper half of YMM registers
 281     additional_frame_bytes = ymm_bytes;
 282     if (UseAVX > 2) {
 283       // Save upper half of ZMM registers as well
 284       additional_frame_bytes += zmm_bytes;
 285     }
 286   }
 287 #else
 288   assert(!restore_vectors, "vectors are generated only by C2");
 289 #endif
 290 
 291   int off = xmm0_off;
 292   int delta = xmm1_off - off;
 293 
 294   __ vzeroupper();
 295 
 296   if (UseSSE == 1) {
 297     // Restore XMM registers
 298     assert(additional_frame_bytes == 0, "");
 299     for (int n = 0; n < num_xmm_regs; n++) {
 300       __ movflt(as_XMMRegister(n), Address(rsp, off*wordSize));
 301       off += delta;
 302     }
 303   } else if (UseSSE >= 2) {
 304     // Restore whole 128bit (16 bytes) XMM registers. Do this before restoring YMM and
 305     // ZMM because the movdqu instruction zeros the upper part of the XMM register.
 306     for (int n = 0; n < num_xmm_regs; n++) {
 307       __ movdqu(as_XMMRegister(n), Address(rsp, off*wordSize+additional_frame_bytes));
 308       off += delta;
 309     }
 310   }
 311 
 312   if (restore_vectors) {
 313     if (UseAVX > 2) {
 314       // Restore upper half of ZMM registers.
 315       for (int n = 0; n < num_xmm_regs; n++) {
 316         __ vinsertf64x4_high(as_XMMRegister(n), Address(rsp, n*32));
 317       }
 318       __ addptr(rsp, zmm_bytes);
 319     }
 320     // Restore upper half of YMM registers.
 321     for (int n = 0; n < num_xmm_regs; n++) {
 322       __ vinsertf128_high(as_XMMRegister(n), Address(rsp, n*16));
 323     }
 324     __ addptr(rsp, ymm_bytes);
 325   }
 326 
 327   __ pop_FPU_state();
 328   __ addptr(rsp, FPU_regs_live*wordSize); // Pop FPU registers
 329 
 330   __ popf();
 331   __ popa();
 332   // Get the rbp, described implicitly by the frame sender code (no oopMap)
 333   __ pop(rbp);
 334 }
 335 
 336 void RegisterSaver::restore_result_registers(MacroAssembler* masm) {
 337 
 338   // Just restore result register. Only used by deoptimization. By
 339   // now any callee save register that needs to be restore to a c2
 340   // caller of the deoptee has been extracted into the vframeArray
 341   // and will be stuffed into the c2i adapter we create for later
 342   // restoration so only result registers need to be restored here.
 343   //
 344 
 345   __ frstor(Address(rsp, 0));      // Restore fpu state
 346 
 347   // Recover XMM & FPU state
 348   if( UseSSE == 1 ) {
 349     __ movflt(xmm0, Address(rsp, xmm0_off*wordSize));
 350   } else if( UseSSE >= 2 ) {
 351     __ movdbl(xmm0, Address(rsp, xmm0_off*wordSize));
 352   }
 353   __ movptr(rax, Address(rsp, rax_off*wordSize));
 354   __ movptr(rdx, Address(rsp, rdx_off*wordSize));
 355   // Pop all of the register save are off the stack except the return address
 356   __ addptr(rsp, return_off * wordSize);
 357 }
 358 
 359 // Is vector's size (in bytes) bigger than a size saved by default?
 360 // 16 bytes XMM registers are saved by default using SSE2 movdqu instructions.
 361 // Note, MaxVectorSize == 0 with UseSSE < 2 and vectors are not generated.
 362 bool SharedRuntime::is_wide_vector(int size) {
 363   return size > 16;
 364 }
 365 
 366 size_t SharedRuntime::trampoline_size() {
 367   return 16;
 368 }
 369 
 370 void SharedRuntime::generate_trampoline(MacroAssembler *masm, address destination) {
 371   __ jump(RuntimeAddress(destination));
 372 }
 373 
 374 // The java_calling_convention describes stack locations as ideal slots on
 375 // a frame with no abi restrictions. Since we must observe abi restrictions
 376 // (like the placement of the register window) the slots must be biased by
 377 // the following value.
 378 static int reg2offset_in(VMReg r) {
 379   // Account for saved rbp, and return address
 380   // This should really be in_preserve_stack_slots
 381   return (r->reg2stack() + 2) * VMRegImpl::stack_slot_size;
 382 }
 383 
 384 static int reg2offset_out(VMReg r) {
 385   return (r->reg2stack() + SharedRuntime::out_preserve_stack_slots()) * VMRegImpl::stack_slot_size;
 386 }
 387 
 388 // ---------------------------------------------------------------------------
 389 // Read the array of BasicTypes from a signature, and compute where the
 390 // arguments should go.  Values in the VMRegPair regs array refer to 4-byte
 391 // quantities.  Values less than SharedInfo::stack0 are registers, those above
 392 // refer to 4-byte stack slots.  All stack slots are based off of the stack pointer
 393 // as framesizes are fixed.
 394 // VMRegImpl::stack0 refers to the first slot 0(sp).
 395 // and VMRegImpl::stack0+1 refers to the memory word 4-byes higher.  Register
 396 // up to RegisterImpl::number_of_registers) are the 32-bit
 397 // integer registers.
 398 
 399 // Pass first two oop/int args in registers ECX and EDX.
 400 // Pass first two float/double args in registers XMM0 and XMM1.
 401 // Doubles have precedence, so if you pass a mix of floats and doubles
 402 // the doubles will grab the registers before the floats will.
 403 
 404 // Note: the INPUTS in sig_bt are in units of Java argument words, which are
 405 // either 32-bit or 64-bit depending on the build.  The OUTPUTS are in 32-bit
 406 // units regardless of build. Of course for i486 there is no 64 bit build
 407 
 408 
 409 // ---------------------------------------------------------------------------
 410 // The compiled Java calling convention.
 411 // Pass first two oop/int args in registers ECX and EDX.
 412 // Pass first two float/double args in registers XMM0 and XMM1.
 413 // Doubles have precedence, so if you pass a mix of floats and doubles
 414 // the doubles will grab the registers before the floats will.
 415 int SharedRuntime::java_calling_convention(const BasicType *sig_bt,
 416                                            VMRegPair *regs,
 417                                            int total_args_passed,
 418                                            int is_outgoing) {
 419   uint    stack = 0;          // Starting stack position for args on stack
 420 
 421 
 422   // Pass first two oop/int args in registers ECX and EDX.
 423   uint reg_arg0 = 9999;
 424   uint reg_arg1 = 9999;
 425 
 426   // Pass first two float/double args in registers XMM0 and XMM1.
 427   // Doubles have precedence, so if you pass a mix of floats and doubles
 428   // the doubles will grab the registers before the floats will.
 429   // CNC - TURNED OFF FOR non-SSE.
 430   //       On Intel we have to round all doubles (and most floats) at
 431   //       call sites by storing to the stack in any case.
 432   // UseSSE=0 ==> Don't Use ==> 9999+0
 433   // UseSSE=1 ==> Floats only ==> 9999+1
 434   // UseSSE>=2 ==> Floats or doubles ==> 9999+2
 435   enum { fltarg_dontuse = 9999+0, fltarg_float_only = 9999+1, fltarg_flt_dbl = 9999+2 };
 436   uint fargs = (UseSSE>=2) ? 2 : UseSSE;
 437   uint freg_arg0 = 9999+fargs;
 438   uint freg_arg1 = 9999+fargs;
 439 
 440   // Pass doubles & longs aligned on the stack.  First count stack slots for doubles
 441   int i;
 442   for( i = 0; i < total_args_passed; i++) {
 443     if( sig_bt[i] == T_DOUBLE ) {
 444       // first 2 doubles go in registers
 445       if( freg_arg0 == fltarg_flt_dbl ) freg_arg0 = i;
 446       else if( freg_arg1 == fltarg_flt_dbl ) freg_arg1 = i;
 447       else // Else double is passed low on the stack to be aligned.
 448         stack += 2;
 449     } else if( sig_bt[i] == T_LONG ) {
 450       stack += 2;
 451     }
 452   }
 453   int dstack = 0;             // Separate counter for placing doubles
 454 
 455   // Now pick where all else goes.
 456   for( i = 0; i < total_args_passed; i++) {
 457     // From the type and the argument number (count) compute the location
 458     switch( sig_bt[i] ) {
 459     case T_SHORT:
 460     case T_CHAR:
 461     case T_BYTE:
 462     case T_BOOLEAN:
 463     case T_INT:
 464     case T_ARRAY:
 465     case T_OBJECT:
 466     case T_ADDRESS:
 467       if( reg_arg0 == 9999 )  {
 468         reg_arg0 = i;
 469         regs[i].set1(rcx->as_VMReg());
 470       } else if( reg_arg1 == 9999 )  {
 471         reg_arg1 = i;
 472         regs[i].set1(rdx->as_VMReg());
 473       } else {
 474         regs[i].set1(VMRegImpl::stack2reg(stack++));
 475       }
 476       break;
 477     case T_FLOAT:
 478       if( freg_arg0 == fltarg_flt_dbl || freg_arg0 == fltarg_float_only ) {
 479         freg_arg0 = i;
 480         regs[i].set1(xmm0->as_VMReg());
 481       } else if( freg_arg1 == fltarg_flt_dbl || freg_arg1 == fltarg_float_only ) {
 482         freg_arg1 = i;
 483         regs[i].set1(xmm1->as_VMReg());
 484       } else {
 485         regs[i].set1(VMRegImpl::stack2reg(stack++));
 486       }
 487       break;
 488     case T_LONG:
 489       assert((i + 1) < total_args_passed && sig_bt[i+1] == T_VOID, "missing Half" );
 490       regs[i].set2(VMRegImpl::stack2reg(dstack));
 491       dstack += 2;
 492       break;
 493     case T_DOUBLE:
 494       assert((i + 1) < total_args_passed && sig_bt[i+1] == T_VOID, "missing Half" );
 495       if( freg_arg0 == (uint)i ) {
 496         regs[i].set2(xmm0->as_VMReg());
 497       } else if( freg_arg1 == (uint)i ) {
 498         regs[i].set2(xmm1->as_VMReg());
 499       } else {
 500         regs[i].set2(VMRegImpl::stack2reg(dstack));
 501         dstack += 2;
 502       }
 503       break;
 504     case T_VOID: regs[i].set_bad(); break;
 505       break;
 506     default:
 507       ShouldNotReachHere();
 508       break;
 509     }
 510   }
 511 
 512   // return value can be odd number of VMRegImpl stack slots make multiple of 2
 513   return align_up(stack, 2);
 514 }
 515 
 516 // Patch the callers callsite with entry to compiled code if it exists.
 517 static void patch_callers_callsite(MacroAssembler *masm) {
 518   Label L;
 519   __ cmpptr(Address(rbx, in_bytes(Method::code_offset())), (int32_t)NULL_WORD);
 520   __ jcc(Assembler::equal, L);
 521   // Schedule the branch target address early.
 522   // Call into the VM to patch the caller, then jump to compiled callee
 523   // rax, isn't live so capture return address while we easily can
 524   __ movptr(rax, Address(rsp, 0));
 525   __ pusha();
 526   __ pushf();
 527 
 528   if (UseSSE == 1) {
 529     __ subptr(rsp, 2*wordSize);
 530     __ movflt(Address(rsp, 0), xmm0);
 531     __ movflt(Address(rsp, wordSize), xmm1);
 532   }
 533   if (UseSSE >= 2) {
 534     __ subptr(rsp, 4*wordSize);
 535     __ movdbl(Address(rsp, 0), xmm0);
 536     __ movdbl(Address(rsp, 2*wordSize), xmm1);
 537   }
 538 #ifdef COMPILER2
 539   // C2 may leave the stack dirty if not in SSE2+ mode
 540   if (UseSSE >= 2) {
 541     __ verify_FPU(0, "c2i transition should have clean FPU stack");
 542   } else {
 543     __ empty_FPU_stack();
 544   }
 545 #endif /* COMPILER2 */
 546 
 547   // VM needs caller's callsite
 548   __ push(rax);
 549   // VM needs target method
 550   __ push(rbx);
 551   __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::fixup_callers_callsite)));
 552   __ addptr(rsp, 2*wordSize);
 553 
 554   if (UseSSE == 1) {
 555     __ movflt(xmm0, Address(rsp, 0));
 556     __ movflt(xmm1, Address(rsp, wordSize));
 557     __ addptr(rsp, 2*wordSize);
 558   }
 559   if (UseSSE >= 2) {
 560     __ movdbl(xmm0, Address(rsp, 0));
 561     __ movdbl(xmm1, Address(rsp, 2*wordSize));
 562     __ addptr(rsp, 4*wordSize);
 563   }
 564 
 565   __ popf();
 566   __ popa();
 567   __ bind(L);
 568 }
 569 
 570 
 571 static void move_c2i_double(MacroAssembler *masm, XMMRegister r, int st_off) {
 572   int next_off = st_off - Interpreter::stackElementSize;
 573   __ movdbl(Address(rsp, next_off), r);
 574 }
 575 
 576 static void gen_c2i_adapter(MacroAssembler *masm,
 577                             int total_args_passed,
 578                             int comp_args_on_stack,
 579                             const BasicType *sig_bt,
 580                             const VMRegPair *regs,
 581                             Label& skip_fixup) {
 582   // Before we get into the guts of the C2I adapter, see if we should be here
 583   // at all.  We've come from compiled code and are attempting to jump to the
 584   // interpreter, which means the caller made a static call to get here
 585   // (vcalls always get a compiled target if there is one).  Check for a
 586   // compiled target.  If there is one, we need to patch the caller's call.
 587   patch_callers_callsite(masm);
 588 
 589   __ bind(skip_fixup);
 590 
 591 #ifdef COMPILER2
 592   // C2 may leave the stack dirty if not in SSE2+ mode
 593   if (UseSSE >= 2) {
 594     __ verify_FPU(0, "c2i transition should have clean FPU stack");
 595   } else {
 596     __ empty_FPU_stack();
 597   }
 598 #endif /* COMPILER2 */
 599 
 600   // Since all args are passed on the stack, total_args_passed * interpreter_
 601   // stack_element_size  is the
 602   // space we need.
 603   int extraspace = total_args_passed * Interpreter::stackElementSize;
 604 
 605   // Get return address
 606   __ pop(rax);
 607 
 608   // set senderSP value
 609   __ movptr(rsi, rsp);
 610 
 611   __ subptr(rsp, extraspace);
 612 
 613   // Now write the args into the outgoing interpreter space
 614   for (int i = 0; i < total_args_passed; i++) {
 615     if (sig_bt[i] == T_VOID) {
 616       assert(i > 0 && (sig_bt[i-1] == T_LONG || sig_bt[i-1] == T_DOUBLE), "missing half");
 617       continue;
 618     }
 619 
 620     // st_off points to lowest address on stack.
 621     int st_off = ((total_args_passed - 1) - i) * Interpreter::stackElementSize;
 622     int next_off = st_off - Interpreter::stackElementSize;
 623 
 624     // Say 4 args:
 625     // i   st_off
 626     // 0   12 T_LONG
 627     // 1    8 T_VOID
 628     // 2    4 T_OBJECT
 629     // 3    0 T_BOOL
 630     VMReg r_1 = regs[i].first();
 631     VMReg r_2 = regs[i].second();
 632     if (!r_1->is_valid()) {
 633       assert(!r_2->is_valid(), "");
 634       continue;
 635     }
 636 
 637     if (r_1->is_stack()) {
 638       // memory to memory use fpu stack top
 639       int ld_off = r_1->reg2stack() * VMRegImpl::stack_slot_size + extraspace;
 640 
 641       if (!r_2->is_valid()) {
 642         __ movl(rdi, Address(rsp, ld_off));
 643         __ movptr(Address(rsp, st_off), rdi);
 644       } else {
 645 
 646         // ld_off == LSW, ld_off+VMRegImpl::stack_slot_size == MSW
 647         // st_off == MSW, st_off-wordSize == LSW
 648 
 649         __ movptr(rdi, Address(rsp, ld_off));
 650         __ movptr(Address(rsp, next_off), rdi);
 651 #ifndef _LP64
 652         __ movptr(rdi, Address(rsp, ld_off + wordSize));
 653         __ movptr(Address(rsp, st_off), rdi);
 654 #else
 655 #ifdef ASSERT
 656         // Overwrite the unused slot with known junk
 657         __ mov64(rax, CONST64(0xdeadffffdeadaaaa));
 658         __ movptr(Address(rsp, st_off), rax);
 659 #endif /* ASSERT */
 660 #endif // _LP64
 661       }
 662     } else if (r_1->is_Register()) {
 663       Register r = r_1->as_Register();
 664       if (!r_2->is_valid()) {
 665         __ movl(Address(rsp, st_off), r);
 666       } else {
 667         // long/double in gpr
 668         NOT_LP64(ShouldNotReachHere());
 669         // Two VMRegs can be T_OBJECT, T_ADDRESS, T_DOUBLE, T_LONG
 670         // T_DOUBLE and T_LONG use two slots in the interpreter
 671         if ( sig_bt[i] == T_LONG || sig_bt[i] == T_DOUBLE) {
 672           // long/double in gpr
 673 #ifdef ASSERT
 674           // Overwrite the unused slot with known junk
 675           LP64_ONLY(__ mov64(rax, CONST64(0xdeadffffdeadaaab)));
 676           __ movptr(Address(rsp, st_off), rax);
 677 #endif /* ASSERT */
 678           __ movptr(Address(rsp, next_off), r);
 679         } else {
 680           __ movptr(Address(rsp, st_off), r);
 681         }
 682       }
 683     } else {
 684       assert(r_1->is_XMMRegister(), "");
 685       if (!r_2->is_valid()) {
 686         __ movflt(Address(rsp, st_off), r_1->as_XMMRegister());
 687       } else {
 688         assert(sig_bt[i] == T_DOUBLE || sig_bt[i] == T_LONG, "wrong type");
 689         move_c2i_double(masm, r_1->as_XMMRegister(), st_off);
 690       }
 691     }
 692   }
 693 
 694   // Schedule the branch target address early.
 695   __ movptr(rcx, Address(rbx, in_bytes(Method::interpreter_entry_offset())));
 696   // And repush original return address
 697   __ push(rax);
 698   __ jmp(rcx);
 699 }
 700 
 701 
 702 static void move_i2c_double(MacroAssembler *masm, XMMRegister r, Register saved_sp, int ld_off) {
 703   int next_val_off = ld_off - Interpreter::stackElementSize;
 704   __ movdbl(r, Address(saved_sp, next_val_off));
 705 }
 706 
 707 static void range_check(MacroAssembler* masm, Register pc_reg, Register temp_reg,
 708                         address code_start, address code_end,
 709                         Label& L_ok) {
 710   Label L_fail;
 711   __ lea(temp_reg, ExternalAddress(code_start));
 712   __ cmpptr(pc_reg, temp_reg);
 713   __ jcc(Assembler::belowEqual, L_fail);
 714   __ lea(temp_reg, ExternalAddress(code_end));
 715   __ cmpptr(pc_reg, temp_reg);
 716   __ jcc(Assembler::below, L_ok);
 717   __ bind(L_fail);
 718 }
 719 
 720 void SharedRuntime::gen_i2c_adapter(MacroAssembler *masm,
 721                                     int total_args_passed,
 722                                     int comp_args_on_stack,
 723                                     const BasicType *sig_bt,
 724                                     const VMRegPair *regs) {
 725   // Note: rsi contains the senderSP on entry. We must preserve it since
 726   // we may do a i2c -> c2i transition if we lose a race where compiled
 727   // code goes non-entrant while we get args ready.
 728 
 729   // Adapters can be frameless because they do not require the caller
 730   // to perform additional cleanup work, such as correcting the stack pointer.
 731   // An i2c adapter is frameless because the *caller* frame, which is interpreted,
 732   // routinely repairs its own stack pointer (from interpreter_frame_last_sp),
 733   // even if a callee has modified the stack pointer.
 734   // A c2i adapter is frameless because the *callee* frame, which is interpreted,
 735   // routinely repairs its caller's stack pointer (from sender_sp, which is set
 736   // up via the senderSP register).
 737   // In other words, if *either* the caller or callee is interpreted, we can
 738   // get the stack pointer repaired after a call.
 739   // This is why c2i and i2c adapters cannot be indefinitely composed.
 740   // In particular, if a c2i adapter were to somehow call an i2c adapter,
 741   // both caller and callee would be compiled methods, and neither would
 742   // clean up the stack pointer changes performed by the two adapters.
 743   // If this happens, control eventually transfers back to the compiled
 744   // caller, but with an uncorrected stack, causing delayed havoc.
 745 
 746   // Pick up the return address
 747   __ movptr(rax, Address(rsp, 0));
 748 
 749   if (VerifyAdapterCalls &&
 750       (Interpreter::code() != NULL || StubRoutines::code1() != NULL)) {
 751     // So, let's test for cascading c2i/i2c adapters right now.
 752     //  assert(Interpreter::contains($return_addr) ||
 753     //         StubRoutines::contains($return_addr),
 754     //         "i2c adapter must return to an interpreter frame");
 755     __ block_comment("verify_i2c { ");
 756     Label L_ok;
 757     if (Interpreter::code() != NULL)
 758       range_check(masm, rax, rdi,
 759                   Interpreter::code()->code_start(), Interpreter::code()->code_end(),
 760                   L_ok);
 761     if (StubRoutines::code1() != NULL)
 762       range_check(masm, rax, rdi,
 763                   StubRoutines::code1()->code_begin(), StubRoutines::code1()->code_end(),
 764                   L_ok);
 765     if (StubRoutines::code2() != NULL)
 766       range_check(masm, rax, rdi,
 767                   StubRoutines::code2()->code_begin(), StubRoutines::code2()->code_end(),
 768                   L_ok);
 769     const char* msg = "i2c adapter must return to an interpreter frame";
 770     __ block_comment(msg);
 771     __ stop(msg);
 772     __ bind(L_ok);
 773     __ block_comment("} verify_i2ce ");
 774   }
 775 
 776   // Must preserve original SP for loading incoming arguments because
 777   // we need to align the outgoing SP for compiled code.
 778   __ movptr(rdi, rsp);
 779 
 780   // Cut-out for having no stack args.  Since up to 2 int/oop args are passed
 781   // in registers, we will occasionally have no stack args.
 782   int comp_words_on_stack = 0;
 783   if (comp_args_on_stack) {
 784     // Sig words on the stack are greater-than VMRegImpl::stack0.  Those in
 785     // registers are below.  By subtracting stack0, we either get a negative
 786     // number (all values in registers) or the maximum stack slot accessed.
 787     // int comp_args_on_stack = VMRegImpl::reg2stack(max_arg);
 788     // Convert 4-byte stack slots to words.
 789     comp_words_on_stack = align_up(comp_args_on_stack*4, wordSize)>>LogBytesPerWord;
 790     // Round up to miminum stack alignment, in wordSize
 791     comp_words_on_stack = align_up(comp_words_on_stack, 2);
 792     __ subptr(rsp, comp_words_on_stack * wordSize);
 793   }
 794 
 795   // Align the outgoing SP
 796   __ andptr(rsp, -(StackAlignmentInBytes));
 797 
 798   // push the return address on the stack (note that pushing, rather
 799   // than storing it, yields the correct frame alignment for the callee)
 800   __ push(rax);
 801 
 802   // Put saved SP in another register
 803   const Register saved_sp = rax;
 804   __ movptr(saved_sp, rdi);
 805 
 806 
 807   // Will jump to the compiled code just as if compiled code was doing it.
 808   // Pre-load the register-jump target early, to schedule it better.
 809   __ movptr(rdi, Address(rbx, in_bytes(Method::from_compiled_offset())));
 810 
 811   // Now generate the shuffle code.  Pick up all register args and move the
 812   // rest through the floating point stack top.
 813   for (int i = 0; i < total_args_passed; i++) {
 814     if (sig_bt[i] == T_VOID) {
 815       // Longs and doubles are passed in native word order, but misaligned
 816       // in the 32-bit build.
 817       assert(i > 0 && (sig_bt[i-1] == T_LONG || sig_bt[i-1] == T_DOUBLE), "missing half");
 818       continue;
 819     }
 820 
 821     // Pick up 0, 1 or 2 words from SP+offset.
 822 
 823     assert(!regs[i].second()->is_valid() || regs[i].first()->next() == regs[i].second(),
 824             "scrambled load targets?");
 825     // Load in argument order going down.
 826     int ld_off = (total_args_passed - i) * Interpreter::stackElementSize;
 827     // Point to interpreter value (vs. tag)
 828     int next_off = ld_off - Interpreter::stackElementSize;
 829     //
 830     //
 831     //
 832     VMReg r_1 = regs[i].first();
 833     VMReg r_2 = regs[i].second();
 834     if (!r_1->is_valid()) {
 835       assert(!r_2->is_valid(), "");
 836       continue;
 837     }
 838     if (r_1->is_stack()) {
 839       // Convert stack slot to an SP offset (+ wordSize to account for return address )
 840       int st_off = regs[i].first()->reg2stack()*VMRegImpl::stack_slot_size + wordSize;
 841 
 842       // We can use rsi as a temp here because compiled code doesn't need rsi as an input
 843       // and if we end up going thru a c2i because of a miss a reasonable value of rsi
 844       // we be generated.
 845       if (!r_2->is_valid()) {
 846         // __ fld_s(Address(saved_sp, ld_off));
 847         // __ fstp_s(Address(rsp, st_off));
 848         __ movl(rsi, Address(saved_sp, ld_off));
 849         __ movptr(Address(rsp, st_off), rsi);
 850       } else {
 851         // Interpreter local[n] == MSW, local[n+1] == LSW however locals
 852         // are accessed as negative so LSW is at LOW address
 853 
 854         // ld_off is MSW so get LSW
 855         // st_off is LSW (i.e. reg.first())
 856         // __ fld_d(Address(saved_sp, next_off));
 857         // __ fstp_d(Address(rsp, st_off));
 858         //
 859         // We are using two VMRegs. This can be either T_OBJECT, T_ADDRESS, T_LONG, or T_DOUBLE
 860         // the interpreter allocates two slots but only uses one for thr T_LONG or T_DOUBLE case
 861         // So we must adjust where to pick up the data to match the interpreter.
 862         //
 863         // Interpreter local[n] == MSW, local[n+1] == LSW however locals
 864         // are accessed as negative so LSW is at LOW address
 865 
 866         // ld_off is MSW so get LSW
 867         const int offset = (NOT_LP64(true ||) sig_bt[i]==T_LONG||sig_bt[i]==T_DOUBLE)?
 868                            next_off : ld_off;
 869         __ movptr(rsi, Address(saved_sp, offset));
 870         __ movptr(Address(rsp, st_off), rsi);
 871 #ifndef _LP64
 872         __ movptr(rsi, Address(saved_sp, ld_off));
 873         __ movptr(Address(rsp, st_off + wordSize), rsi);
 874 #endif // _LP64
 875       }
 876     } else if (r_1->is_Register()) {  // Register argument
 877       Register r = r_1->as_Register();
 878       assert(r != rax, "must be different");
 879       if (r_2->is_valid()) {
 880         //
 881         // We are using two VMRegs. This can be either T_OBJECT, T_ADDRESS, T_LONG, or T_DOUBLE
 882         // the interpreter allocates two slots but only uses one for thr T_LONG or T_DOUBLE case
 883         // So we must adjust where to pick up the data to match the interpreter.
 884 
 885         const int offset = (NOT_LP64(true ||) sig_bt[i]==T_LONG||sig_bt[i]==T_DOUBLE)?
 886                            next_off : ld_off;
 887 
 888         // this can be a misaligned move
 889         __ movptr(r, Address(saved_sp, offset));
 890 #ifndef _LP64
 891         assert(r_2->as_Register() != rax, "need another temporary register");
 892         // Remember r_1 is low address (and LSB on x86)
 893         // So r_2 gets loaded from high address regardless of the platform
 894         __ movptr(r_2->as_Register(), Address(saved_sp, ld_off));
 895 #endif // _LP64
 896       } else {
 897         __ movl(r, Address(saved_sp, ld_off));
 898       }
 899     } else {
 900       assert(r_1->is_XMMRegister(), "");
 901       if (!r_2->is_valid()) {
 902         __ movflt(r_1->as_XMMRegister(), Address(saved_sp, ld_off));
 903       } else {
 904         move_i2c_double(masm, r_1->as_XMMRegister(), saved_sp, ld_off);
 905       }
 906     }
 907   }
 908 
 909   // 6243940 We might end up in handle_wrong_method if
 910   // the callee is deoptimized as we race thru here. If that
 911   // happens we don't want to take a safepoint because the
 912   // caller frame will look interpreted and arguments are now
 913   // "compiled" so it is much better to make this transition
 914   // invisible to the stack walking code. Unfortunately if
 915   // we try and find the callee by normal means a safepoint
 916   // is possible. So we stash the desired callee in the thread
 917   // and the vm will find there should this case occur.
 918 
 919   __ get_thread(rax);
 920   __ movptr(Address(rax, JavaThread::callee_target_offset()), rbx);
 921 
 922   // move Method* to rax, in case we end up in an c2i adapter.
 923   // the c2i adapters expect Method* in rax, (c2) because c2's
 924   // resolve stubs return the result (the method) in rax,.
 925   // I'd love to fix this.
 926   __ mov(rax, rbx);
 927 
 928   __ jmp(rdi);
 929 }
 930 
 931 // ---------------------------------------------------------------
 932 AdapterHandlerEntry* SharedRuntime::generate_i2c2i_adapters(MacroAssembler *masm,
 933                                                             int total_args_passed,
 934                                                             int comp_args_on_stack,
 935                                                             const BasicType *sig_bt,
 936                                                             const VMRegPair *regs,
 937                                                             AdapterFingerPrint* fingerprint) {
 938   address i2c_entry = __ pc();
 939 
 940   gen_i2c_adapter(masm, total_args_passed, comp_args_on_stack, sig_bt, regs);
 941 
 942   // -------------------------------------------------------------------------
 943   // Generate a C2I adapter.  On entry we know rbx, holds the Method* during calls
 944   // to the interpreter.  The args start out packed in the compiled layout.  They
 945   // need to be unpacked into the interpreter layout.  This will almost always
 946   // require some stack space.  We grow the current (compiled) stack, then repack
 947   // the args.  We  finally end in a jump to the generic interpreter entry point.
 948   // On exit from the interpreter, the interpreter will restore our SP (lest the
 949   // compiled code, which relys solely on SP and not EBP, get sick).
 950 
 951   address c2i_unverified_entry = __ pc();
 952   Label skip_fixup;
 953 
 954   Register holder = rax;
 955   Register receiver = rcx;
 956   Register temp = rbx;
 957 
 958   {
 959 
 960     Label missed;
 961     __ movptr(temp, Address(receiver, oopDesc::klass_offset_in_bytes()));
 962     __ cmpptr(temp, Address(holder, CompiledICHolder::holder_klass_offset()));
 963     __ movptr(rbx, Address(holder, CompiledICHolder::holder_metadata_offset()));
 964     __ jcc(Assembler::notEqual, missed);
 965     // Method might have been compiled since the call site was patched to
 966     // interpreted if that is the case treat it as a miss so we can get
 967     // the call site corrected.
 968     __ cmpptr(Address(rbx, in_bytes(Method::code_offset())), (int32_t)NULL_WORD);
 969     __ jcc(Assembler::equal, skip_fixup);
 970 
 971     __ bind(missed);
 972     __ jump(RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
 973   }
 974 
 975   address c2i_entry = __ pc();
 976 
 977   gen_c2i_adapter(masm, total_args_passed, comp_args_on_stack, sig_bt, regs, skip_fixup);
 978 
 979   __ flush();
 980   return AdapterHandlerLibrary::new_entry(fingerprint, i2c_entry, c2i_entry, c2i_unverified_entry);
 981 }
 982 
 983 int SharedRuntime::c_calling_convention(const BasicType *sig_bt,
 984                                          VMRegPair *regs,
 985                                          VMRegPair *regs2,
 986                                          int total_args_passed) {
 987   assert(regs2 == NULL, "not needed on x86");
 988 // We return the amount of VMRegImpl stack slots we need to reserve for all
 989 // the arguments NOT counting out_preserve_stack_slots.
 990 
 991   uint    stack = 0;        // All arguments on stack
 992 
 993   for( int i = 0; i < total_args_passed; i++) {
 994     // From the type and the argument number (count) compute the location
 995     switch( sig_bt[i] ) {
 996     case T_BOOLEAN:
 997     case T_CHAR:
 998     case T_FLOAT:
 999     case T_BYTE:
1000     case T_SHORT:
1001     case T_INT:
1002     case T_OBJECT:
1003     case T_ARRAY:
1004     case T_ADDRESS:
1005     case T_METADATA:
1006       regs[i].set1(VMRegImpl::stack2reg(stack++));
1007       break;
1008     case T_LONG:
1009     case T_DOUBLE: // The stack numbering is reversed from Java
1010       // Since C arguments do not get reversed, the ordering for
1011       // doubles on the stack must be opposite the Java convention
1012       assert((i + 1) < total_args_passed && sig_bt[i+1] == T_VOID, "missing Half" );
1013       regs[i].set2(VMRegImpl::stack2reg(stack));
1014       stack += 2;
1015       break;
1016     case T_VOID: regs[i].set_bad(); break;
1017     default:
1018       ShouldNotReachHere();
1019       break;
1020     }
1021   }
1022   return stack;
1023 }
1024 
1025 // A simple move of integer like type
1026 static void simple_move32(MacroAssembler* masm, VMRegPair src, VMRegPair dst) {
1027   if (src.first()->is_stack()) {
1028     if (dst.first()->is_stack()) {
1029       // stack to stack
1030       // __ ld(FP, reg2offset(src.first()) + STACK_BIAS, L5);
1031       // __ st(L5, SP, reg2offset(dst.first()) + STACK_BIAS);
1032       __ movl2ptr(rax, Address(rbp, reg2offset_in(src.first())));
1033       __ movptr(Address(rsp, reg2offset_out(dst.first())), rax);
1034     } else {
1035       // stack to reg
1036       __ movl2ptr(dst.first()->as_Register(),  Address(rbp, reg2offset_in(src.first())));
1037     }
1038   } else if (dst.first()->is_stack()) {
1039     // reg to stack
1040     // no need to sign extend on 64bit
1041     __ movptr(Address(rsp, reg2offset_out(dst.first())), src.first()->as_Register());
1042   } else {
1043     if (dst.first() != src.first()) {
1044       __ mov(dst.first()->as_Register(), src.first()->as_Register());
1045     }
1046   }
1047 }
1048 
1049 // An oop arg. Must pass a handle not the oop itself
1050 static void object_move(MacroAssembler* masm,
1051                         OopMap* map,
1052                         int oop_handle_offset,
1053                         int framesize_in_slots,
1054                         VMRegPair src,
1055                         VMRegPair dst,
1056                         bool is_receiver,
1057                         int* receiver_offset) {
1058 
1059   // Because of the calling conventions we know that src can be a
1060   // register or a stack location. dst can only be a stack location.
1061 
1062   assert(dst.first()->is_stack(), "must be stack");
1063   // must pass a handle. First figure out the location we use as a handle
1064 
1065   if (src.first()->is_stack()) {
1066     // Oop is already on the stack as an argument
1067     Register rHandle = rax;
1068     Label nil;
1069     __ xorptr(rHandle, rHandle);
1070     __ cmpptr(Address(rbp, reg2offset_in(src.first())), (int32_t)NULL_WORD);
1071     __ jcc(Assembler::equal, nil);
1072     __ lea(rHandle, Address(rbp, reg2offset_in(src.first())));
1073     __ bind(nil);
1074     __ movptr(Address(rsp, reg2offset_out(dst.first())), rHandle);
1075 
1076     int offset_in_older_frame = src.first()->reg2stack() + SharedRuntime::out_preserve_stack_slots();
1077     map->set_oop(VMRegImpl::stack2reg(offset_in_older_frame + framesize_in_slots));
1078     if (is_receiver) {
1079       *receiver_offset = (offset_in_older_frame + framesize_in_slots) * VMRegImpl::stack_slot_size;
1080     }
1081   } else {
1082     // Oop is in an a register we must store it to the space we reserve
1083     // on the stack for oop_handles
1084     const Register rOop = src.first()->as_Register();
1085     const Register rHandle = rax;
1086     int oop_slot = (rOop == rcx ? 0 : 1) * VMRegImpl::slots_per_word + oop_handle_offset;
1087     int offset = oop_slot*VMRegImpl::stack_slot_size;
1088     Label skip;
1089     __ movptr(Address(rsp, offset), rOop);
1090     map->set_oop(VMRegImpl::stack2reg(oop_slot));
1091     __ xorptr(rHandle, rHandle);
1092     __ cmpptr(rOop, (int32_t)NULL_WORD);
1093     __ jcc(Assembler::equal, skip);
1094     __ lea(rHandle, Address(rsp, offset));
1095     __ bind(skip);
1096     // Store the handle parameter
1097     __ movptr(Address(rsp, reg2offset_out(dst.first())), rHandle);
1098     if (is_receiver) {
1099       *receiver_offset = offset;
1100     }
1101   }
1102 }
1103 
1104 // A float arg may have to do float reg int reg conversion
1105 static void float_move(MacroAssembler* masm, VMRegPair src, VMRegPair dst) {
1106   assert(!src.second()->is_valid() && !dst.second()->is_valid(), "bad float_move");
1107 
1108   // Because of the calling convention we know that src is either a stack location
1109   // or an xmm register. dst can only be a stack location.
1110 
1111   assert(dst.first()->is_stack() && ( src.first()->is_stack() || src.first()->is_XMMRegister()), "bad parameters");
1112 
1113   if (src.first()->is_stack()) {
1114     __ movl(rax, Address(rbp, reg2offset_in(src.first())));
1115     __ movptr(Address(rsp, reg2offset_out(dst.first())), rax);
1116   } else {
1117     // reg to stack
1118     __ movflt(Address(rsp, reg2offset_out(dst.first())), src.first()->as_XMMRegister());
1119   }
1120 }
1121 
1122 // A long move
1123 static void long_move(MacroAssembler* masm, VMRegPair src, VMRegPair dst) {
1124 
1125   // The only legal possibility for a long_move VMRegPair is:
1126   // 1: two stack slots (possibly unaligned)
1127   // as neither the java  or C calling convention will use registers
1128   // for longs.
1129 
1130   if (src.first()->is_stack() && dst.first()->is_stack()) {
1131     assert(src.second()->is_stack() && dst.second()->is_stack(), "must be all stack");
1132     __ movptr(rax, Address(rbp, reg2offset_in(src.first())));
1133     NOT_LP64(__ movptr(rbx, Address(rbp, reg2offset_in(src.second()))));
1134     __ movptr(Address(rsp, reg2offset_out(dst.first())), rax);
1135     NOT_LP64(__ movptr(Address(rsp, reg2offset_out(dst.second())), rbx));
1136   } else {
1137     ShouldNotReachHere();
1138   }
1139 }
1140 
1141 // A double move
1142 static void double_move(MacroAssembler* masm, VMRegPair src, VMRegPair dst) {
1143 
1144   // The only legal possibilities for a double_move VMRegPair are:
1145   // The painful thing here is that like long_move a VMRegPair might be
1146 
1147   // Because of the calling convention we know that src is either
1148   //   1: a single physical register (xmm registers only)
1149   //   2: two stack slots (possibly unaligned)
1150   // dst can only be a pair of stack slots.
1151 
1152   assert(dst.first()->is_stack() && (src.first()->is_XMMRegister() || src.first()->is_stack()), "bad args");
1153 
1154   if (src.first()->is_stack()) {
1155     // source is all stack
1156     __ movptr(rax, Address(rbp, reg2offset_in(src.first())));
1157     NOT_LP64(__ movptr(rbx, Address(rbp, reg2offset_in(src.second()))));
1158     __ movptr(Address(rsp, reg2offset_out(dst.first())), rax);
1159     NOT_LP64(__ movptr(Address(rsp, reg2offset_out(dst.second())), rbx));
1160   } else {
1161     // reg to stack
1162     // No worries about stack alignment
1163     __ movdbl(Address(rsp, reg2offset_out(dst.first())), src.first()->as_XMMRegister());
1164   }
1165 }
1166 
1167 
1168 void SharedRuntime::save_native_result(MacroAssembler *masm, BasicType ret_type, int frame_slots) {
1169   // We always ignore the frame_slots arg and just use the space just below frame pointer
1170   // which by this time is free to use
1171   switch (ret_type) {
1172   case T_FLOAT:
1173     __ fstp_s(Address(rbp, -wordSize));
1174     break;
1175   case T_DOUBLE:
1176     __ fstp_d(Address(rbp, -2*wordSize));
1177     break;
1178   case T_VOID:  break;
1179   case T_LONG:
1180     __ movptr(Address(rbp, -wordSize), rax);
1181     NOT_LP64(__ movptr(Address(rbp, -2*wordSize), rdx));
1182     break;
1183   default: {
1184     __ movptr(Address(rbp, -wordSize), rax);
1185     }
1186   }
1187 }
1188 
1189 void SharedRuntime::restore_native_result(MacroAssembler *masm, BasicType ret_type, int frame_slots) {
1190   // We always ignore the frame_slots arg and just use the space just below frame pointer
1191   // which by this time is free to use
1192   switch (ret_type) {
1193   case T_FLOAT:
1194     __ fld_s(Address(rbp, -wordSize));
1195     break;
1196   case T_DOUBLE:
1197     __ fld_d(Address(rbp, -2*wordSize));
1198     break;
1199   case T_LONG:
1200     __ movptr(rax, Address(rbp, -wordSize));
1201     NOT_LP64(__ movptr(rdx, Address(rbp, -2*wordSize)));
1202     break;
1203   case T_VOID:  break;
1204   default: {
1205     __ movptr(rax, Address(rbp, -wordSize));
1206     }
1207   }
1208 }
1209 
1210 
1211 static void save_or_restore_arguments(MacroAssembler* masm,
1212                                       const int stack_slots,
1213                                       const int total_in_args,
1214                                       const int arg_save_area,
1215                                       OopMap* map,
1216                                       VMRegPair* in_regs,
1217                                       BasicType* in_sig_bt) {
1218   // if map is non-NULL then the code should store the values,
1219   // otherwise it should load them.
1220   int handle_index = 0;
1221   // Save down double word first
1222   for ( int i = 0; i < total_in_args; i++) {
1223     if (in_regs[i].first()->is_XMMRegister() && in_sig_bt[i] == T_DOUBLE) {
1224       int slot = handle_index * VMRegImpl::slots_per_word + arg_save_area;
1225       int offset = slot * VMRegImpl::stack_slot_size;
1226       handle_index += 2;
1227       assert(handle_index <= stack_slots, "overflow");
1228       if (map != NULL) {
1229         __ movdbl(Address(rsp, offset), in_regs[i].first()->as_XMMRegister());
1230       } else {
1231         __ movdbl(in_regs[i].first()->as_XMMRegister(), Address(rsp, offset));
1232       }
1233     }
1234     if (in_regs[i].first()->is_Register() && in_sig_bt[i] == T_LONG) {
1235       int slot = handle_index * VMRegImpl::slots_per_word + arg_save_area;
1236       int offset = slot * VMRegImpl::stack_slot_size;
1237       handle_index += 2;
1238       assert(handle_index <= stack_slots, "overflow");
1239       if (map != NULL) {
1240         __ movl(Address(rsp, offset), in_regs[i].first()->as_Register());
1241         if (in_regs[i].second()->is_Register()) {
1242           __ movl(Address(rsp, offset + 4), in_regs[i].second()->as_Register());
1243         }
1244       } else {
1245         __ movl(in_regs[i].first()->as_Register(), Address(rsp, offset));
1246         if (in_regs[i].second()->is_Register()) {
1247           __ movl(in_regs[i].second()->as_Register(), Address(rsp, offset + 4));
1248         }
1249       }
1250     }
1251   }
1252   // Save or restore single word registers
1253   for ( int i = 0; i < total_in_args; i++) {
1254     if (in_regs[i].first()->is_Register()) {
1255       int slot = handle_index++ * VMRegImpl::slots_per_word + arg_save_area;
1256       int offset = slot * VMRegImpl::stack_slot_size;
1257       assert(handle_index <= stack_slots, "overflow");
1258       if (in_sig_bt[i] == T_ARRAY && map != NULL) {
1259         map->set_oop(VMRegImpl::stack2reg(slot));;
1260       }
1261 
1262       // Value is in an input register pass we must flush it to the stack
1263       const Register reg = in_regs[i].first()->as_Register();
1264       switch (in_sig_bt[i]) {
1265         case T_ARRAY:
1266           if (map != NULL) {
1267             __ movptr(Address(rsp, offset), reg);
1268           } else {
1269             __ movptr(reg, Address(rsp, offset));
1270           }
1271           break;
1272         case T_BOOLEAN:
1273         case T_CHAR:
1274         case T_BYTE:
1275         case T_SHORT:
1276         case T_INT:
1277           if (map != NULL) {
1278             __ movl(Address(rsp, offset), reg);
1279           } else {
1280             __ movl(reg, Address(rsp, offset));
1281           }
1282           break;
1283         case T_OBJECT:
1284         default: ShouldNotReachHere();
1285       }
1286     } else if (in_regs[i].first()->is_XMMRegister()) {
1287       if (in_sig_bt[i] == T_FLOAT) {
1288         int slot = handle_index++ * VMRegImpl::slots_per_word + arg_save_area;
1289         int offset = slot * VMRegImpl::stack_slot_size;
1290         assert(handle_index <= stack_slots, "overflow");
1291         if (map != NULL) {
1292           __ movflt(Address(rsp, offset), in_regs[i].first()->as_XMMRegister());
1293         } else {
1294           __ movflt(in_regs[i].first()->as_XMMRegister(), Address(rsp, offset));
1295         }
1296       }
1297     } else if (in_regs[i].first()->is_stack()) {
1298       if (in_sig_bt[i] == T_ARRAY && map != NULL) {
1299         int offset_in_older_frame = in_regs[i].first()->reg2stack() + SharedRuntime::out_preserve_stack_slots();
1300         map->set_oop(VMRegImpl::stack2reg(offset_in_older_frame + stack_slots));
1301       }
1302     }
1303   }
1304 }
1305 
1306 // Check GCLocker::needs_gc and enter the runtime if it's true.  This
1307 // keeps a new JNI critical region from starting until a GC has been
1308 // forced.  Save down any oops in registers and describe them in an
1309 // OopMap.
1310 static void check_needs_gc_for_critical_native(MacroAssembler* masm,
1311                                                Register thread,
1312                                                int stack_slots,
1313                                                int total_c_args,
1314                                                int total_in_args,
1315                                                int arg_save_area,
1316                                                OopMapSet* oop_maps,
1317                                                VMRegPair* in_regs,
1318                                                BasicType* in_sig_bt) {
1319   __ block_comment("check GCLocker::needs_gc");
1320   Label cont;
1321   __ cmp8(ExternalAddress((address)GCLocker::needs_gc_address()), false);
1322   __ jcc(Assembler::equal, cont);
1323 
1324   // Save down any incoming oops and call into the runtime to halt for a GC
1325 
1326   OopMap* map = new OopMap(stack_slots * 2, 0 /* arg_slots*/);
1327 
1328   save_or_restore_arguments(masm, stack_slots, total_in_args,
1329                             arg_save_area, map, in_regs, in_sig_bt);
1330 
1331   address the_pc = __ pc();
1332   oop_maps->add_gc_map( __ offset(), map);
1333   __ set_last_Java_frame(thread, rsp, noreg, the_pc);
1334 
1335   __ block_comment("block_for_jni_critical");
1336   __ push(thread);
1337   __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::block_for_jni_critical)));
1338   __ increment(rsp, wordSize);
1339 
1340   __ get_thread(thread);
1341   __ reset_last_Java_frame(thread, false);
1342 
1343   save_or_restore_arguments(masm, stack_slots, total_in_args,
1344                             arg_save_area, NULL, in_regs, in_sig_bt);
1345 
1346   __ bind(cont);
1347 #ifdef ASSERT
1348   if (StressCriticalJNINatives) {
1349     // Stress register saving
1350     OopMap* map = new OopMap(stack_slots * 2, 0 /* arg_slots*/);
1351     save_or_restore_arguments(masm, stack_slots, total_in_args,
1352                               arg_save_area, map, in_regs, in_sig_bt);
1353     // Destroy argument registers
1354     for (int i = 0; i < total_in_args - 1; i++) {
1355       if (in_regs[i].first()->is_Register()) {
1356         const Register reg = in_regs[i].first()->as_Register();
1357         __ xorptr(reg, reg);
1358       } else if (in_regs[i].first()->is_XMMRegister()) {
1359         __ xorpd(in_regs[i].first()->as_XMMRegister(), in_regs[i].first()->as_XMMRegister());
1360       } else if (in_regs[i].first()->is_FloatRegister()) {
1361         ShouldNotReachHere();
1362       } else if (in_regs[i].first()->is_stack()) {
1363         // Nothing to do
1364       } else {
1365         ShouldNotReachHere();
1366       }
1367       if (in_sig_bt[i] == T_LONG || in_sig_bt[i] == T_DOUBLE) {
1368         i++;
1369       }
1370     }
1371 
1372     save_or_restore_arguments(masm, stack_slots, total_in_args,
1373                               arg_save_area, NULL, in_regs, in_sig_bt);
1374   }
1375 #endif
1376 }
1377 
1378 // Unpack an array argument into a pointer to the body and the length
1379 // if the array is non-null, otherwise pass 0 for both.
1380 static void unpack_array_argument(MacroAssembler* masm, VMRegPair reg, BasicType in_elem_type, VMRegPair body_arg, VMRegPair length_arg) {
1381   Register tmp_reg = rax;
1382   assert(!body_arg.first()->is_Register() || body_arg.first()->as_Register() != tmp_reg,
1383          "possible collision");
1384   assert(!length_arg.first()->is_Register() || length_arg.first()->as_Register() != tmp_reg,
1385          "possible collision");
1386 
1387   // Pass the length, ptr pair
1388   Label is_null, done;
1389   VMRegPair tmp(tmp_reg->as_VMReg());
1390   if (reg.first()->is_stack()) {
1391     // Load the arg up from the stack
1392     simple_move32(masm, reg, tmp);
1393     reg = tmp;
1394   }
1395   __ testptr(reg.first()->as_Register(), reg.first()->as_Register());
1396   __ jccb(Assembler::equal, is_null);
1397   __ lea(tmp_reg, Address(reg.first()->as_Register(), arrayOopDesc::base_offset_in_bytes(in_elem_type)));
1398   simple_move32(masm, tmp, body_arg);
1399   // load the length relative to the body.
1400   __ movl(tmp_reg, Address(tmp_reg, arrayOopDesc::length_offset_in_bytes() -
1401                            arrayOopDesc::base_offset_in_bytes(in_elem_type)));
1402   simple_move32(masm, tmp, length_arg);
1403   __ jmpb(done);
1404   __ bind(is_null);
1405   // Pass zeros
1406   __ xorptr(tmp_reg, tmp_reg);
1407   simple_move32(masm, tmp, body_arg);
1408   simple_move32(masm, tmp, length_arg);
1409   __ bind(done);
1410 }
1411 
1412 static void verify_oop_args(MacroAssembler* masm,
1413                             const methodHandle& method,
1414                             const BasicType* sig_bt,
1415                             const VMRegPair* regs) {
1416   Register temp_reg = rbx;  // not part of any compiled calling seq
1417   if (VerifyOops) {
1418     for (int i = 0; i < method->size_of_parameters(); i++) {
1419       if (sig_bt[i] == T_OBJECT ||
1420           sig_bt[i] == T_ARRAY) {
1421         VMReg r = regs[i].first();
1422         assert(r->is_valid(), "bad oop arg");
1423         if (r->is_stack()) {
1424           __ movptr(temp_reg, Address(rsp, r->reg2stack() * VMRegImpl::stack_slot_size + wordSize));
1425           __ verify_oop(temp_reg);
1426         } else {
1427           __ verify_oop(r->as_Register());
1428         }
1429       }
1430     }
1431   }
1432 }
1433 
1434 static void gen_special_dispatch(MacroAssembler* masm,
1435                                  const methodHandle& method,
1436                                  const BasicType* sig_bt,
1437                                  const VMRegPair* regs) {
1438   verify_oop_args(masm, method, sig_bt, regs);
1439   vmIntrinsics::ID iid = method->intrinsic_id();
1440 
1441   // Now write the args into the outgoing interpreter space
1442   bool     has_receiver   = false;
1443   Register receiver_reg   = noreg;
1444   int      member_arg_pos = -1;
1445   Register member_reg     = noreg;
1446   int      ref_kind       = MethodHandles::signature_polymorphic_intrinsic_ref_kind(iid);
1447   if (ref_kind != 0) {
1448     member_arg_pos = method->size_of_parameters() - 1;  // trailing MemberName argument
1449     member_reg = rbx;  // known to be free at this point
1450     has_receiver = MethodHandles::ref_kind_has_receiver(ref_kind);
1451   } else if (iid == vmIntrinsics::_invokeBasic) {
1452     has_receiver = true;
1453   } else {
1454     fatal("unexpected intrinsic id %d", iid);
1455   }
1456 
1457   if (member_reg != noreg) {
1458     // Load the member_arg into register, if necessary.
1459     SharedRuntime::check_member_name_argument_is_last_argument(method, sig_bt, regs);
1460     VMReg r = regs[member_arg_pos].first();
1461     if (r->is_stack()) {
1462       __ movptr(member_reg, Address(rsp, r->reg2stack() * VMRegImpl::stack_slot_size + wordSize));
1463     } else {
1464       // no data motion is needed
1465       member_reg = r->as_Register();
1466     }
1467   }
1468 
1469   if (has_receiver) {
1470     // Make sure the receiver is loaded into a register.
1471     assert(method->size_of_parameters() > 0, "oob");
1472     assert(sig_bt[0] == T_OBJECT, "receiver argument must be an object");
1473     VMReg r = regs[0].first();
1474     assert(r->is_valid(), "bad receiver arg");
1475     if (r->is_stack()) {
1476       // Porting note:  This assumes that compiled calling conventions always
1477       // pass the receiver oop in a register.  If this is not true on some
1478       // platform, pick a temp and load the receiver from stack.
1479       fatal("receiver always in a register");
1480       receiver_reg = rcx;  // known to be free at this point
1481       __ movptr(receiver_reg, Address(rsp, r->reg2stack() * VMRegImpl::stack_slot_size + wordSize));
1482     } else {
1483       // no data motion is needed
1484       receiver_reg = r->as_Register();
1485     }
1486   }
1487 
1488   // Figure out which address we are really jumping to:
1489   MethodHandles::generate_method_handle_dispatch(masm, iid,
1490                                                  receiver_reg, member_reg, /*for_compiler_entry:*/ true);
1491 }
1492 
1493 // ---------------------------------------------------------------------------
1494 // Generate a native wrapper for a given method.  The method takes arguments
1495 // in the Java compiled code convention, marshals them to the native
1496 // convention (handlizes oops, etc), transitions to native, makes the call,
1497 // returns to java state (possibly blocking), unhandlizes any result and
1498 // returns.
1499 //
1500 // Critical native functions are a shorthand for the use of
1501 // GetPrimtiveArrayCritical and disallow the use of any other JNI
1502 // functions.  The wrapper is expected to unpack the arguments before
1503 // passing them to the callee and perform checks before and after the
1504 // native call to ensure that they GCLocker
1505 // lock_critical/unlock_critical semantics are followed.  Some other
1506 // parts of JNI setup are skipped like the tear down of the JNI handle
1507 // block and the check for pending exceptions it's impossible for them
1508 // to be thrown.
1509 //
1510 // They are roughly structured like this:
1511 //    if (GCLocker::needs_gc())
1512 //      SharedRuntime::block_for_jni_critical();
1513 //    tranistion to thread_in_native
1514 //    unpack arrray arguments and call native entry point
1515 //    check for safepoint in progress
1516 //    check if any thread suspend flags are set
1517 //      call into JVM and possible unlock the JNI critical
1518 //      if a GC was suppressed while in the critical native.
1519 //    transition back to thread_in_Java
1520 //    return to caller
1521 //
1522 nmethod* SharedRuntime::generate_native_wrapper(MacroAssembler* masm,
1523                                                 const methodHandle& method,
1524                                                 int compile_id,
1525                                                 BasicType* in_sig_bt,
1526                                                 VMRegPair* in_regs,
1527                                                 BasicType ret_type,
1528                                                 address critical_entry) {
1529   if (method->is_method_handle_intrinsic()) {
1530     vmIntrinsics::ID iid = method->intrinsic_id();
1531     intptr_t start = (intptr_t)__ pc();
1532     int vep_offset = ((intptr_t)__ pc()) - start;
1533     gen_special_dispatch(masm,
1534                          method,
1535                          in_sig_bt,
1536                          in_regs);
1537     int frame_complete = ((intptr_t)__ pc()) - start;  // not complete, period
1538     __ flush();
1539     int stack_slots = SharedRuntime::out_preserve_stack_slots();  // no out slots at all, actually
1540     return nmethod::new_native_nmethod(method,
1541                                        compile_id,
1542                                        masm->code(),
1543                                        vep_offset,
1544                                        frame_complete,
1545                                        stack_slots / VMRegImpl::slots_per_word,
1546                                        in_ByteSize(-1),
1547                                        in_ByteSize(-1),
1548                                        (OopMapSet*)NULL);
1549   }
1550   bool is_critical_native = true;
1551   address native_func = critical_entry;
1552   if (native_func == NULL) {
1553     native_func = method->native_function();
1554     is_critical_native = false;
1555   }
1556   assert(native_func != NULL, "must have function");
1557 
1558   // An OopMap for lock (and class if static)
1559   OopMapSet *oop_maps = new OopMapSet();
1560 
1561   // We have received a description of where all the java arg are located
1562   // on entry to the wrapper. We need to convert these args to where
1563   // the jni function will expect them. To figure out where they go
1564   // we convert the java signature to a C signature by inserting
1565   // the hidden arguments as arg[0] and possibly arg[1] (static method)
1566 
1567   const int total_in_args = method->size_of_parameters();
1568   int total_c_args = total_in_args;
1569   if (!is_critical_native) {
1570     total_c_args += 1;
1571     if (method->is_static()) {
1572       total_c_args++;
1573     }
1574   } else {
1575     for (int i = 0; i < total_in_args; i++) {
1576       if (in_sig_bt[i] == T_ARRAY) {
1577         total_c_args++;
1578       }
1579     }
1580   }
1581 
1582   BasicType* out_sig_bt = NEW_RESOURCE_ARRAY(BasicType, total_c_args);
1583   VMRegPair* out_regs   = NEW_RESOURCE_ARRAY(VMRegPair, total_c_args);
1584   BasicType* in_elem_bt = NULL;
1585 
1586   int argc = 0;
1587   if (!is_critical_native) {
1588     out_sig_bt[argc++] = T_ADDRESS;
1589     if (method->is_static()) {
1590       out_sig_bt[argc++] = T_OBJECT;
1591     }
1592 
1593     for (int i = 0; i < total_in_args ; i++ ) {
1594       out_sig_bt[argc++] = in_sig_bt[i];
1595     }
1596   } else {
1597     Thread* THREAD = Thread::current();
1598     in_elem_bt = NEW_RESOURCE_ARRAY(BasicType, total_in_args);
1599     SignatureStream ss(method->signature());
1600     for (int i = 0; i < total_in_args ; i++ ) {
1601       if (in_sig_bt[i] == T_ARRAY) {
1602         // Arrays are passed as int, elem* pair
1603         out_sig_bt[argc++] = T_INT;
1604         out_sig_bt[argc++] = T_ADDRESS;
1605         Symbol* atype = ss.as_symbol(CHECK_NULL);
1606         const char* at = atype->as_C_string();
1607         if (strlen(at) == 2) {
1608           assert(at[0] == '[', "must be");
1609           switch (at[1]) {
1610             case 'B': in_elem_bt[i]  = T_BYTE; break;
1611             case 'C': in_elem_bt[i]  = T_CHAR; break;
1612             case 'D': in_elem_bt[i]  = T_DOUBLE; break;
1613             case 'F': in_elem_bt[i]  = T_FLOAT; break;
1614             case 'I': in_elem_bt[i]  = T_INT; break;
1615             case 'J': in_elem_bt[i]  = T_LONG; break;
1616             case 'S': in_elem_bt[i]  = T_SHORT; break;
1617             case 'Z': in_elem_bt[i]  = T_BOOLEAN; break;
1618             default: ShouldNotReachHere();
1619           }
1620         }
1621       } else {
1622         out_sig_bt[argc++] = in_sig_bt[i];
1623         in_elem_bt[i] = T_VOID;
1624       }
1625       if (in_sig_bt[i] != T_VOID) {
1626         assert(in_sig_bt[i] == ss.type(), "must match");
1627         ss.next();
1628       }
1629     }
1630   }
1631 
1632   // Now figure out where the args must be stored and how much stack space
1633   // they require.
1634   int out_arg_slots;
1635   out_arg_slots = c_calling_convention(out_sig_bt, out_regs, NULL, total_c_args);
1636 
1637   // Compute framesize for the wrapper.  We need to handlize all oops in
1638   // registers a max of 2 on x86.
1639 
1640   // Calculate the total number of stack slots we will need.
1641 
1642   // First count the abi requirement plus all of the outgoing args
1643   int stack_slots = SharedRuntime::out_preserve_stack_slots() + out_arg_slots;
1644 
1645   // Now the space for the inbound oop handle area
1646   int total_save_slots = 2 * VMRegImpl::slots_per_word; // 2 arguments passed in registers
1647   if (is_critical_native) {
1648     // Critical natives may have to call out so they need a save area
1649     // for register arguments.
1650     int double_slots = 0;
1651     int single_slots = 0;
1652     for ( int i = 0; i < total_in_args; i++) {
1653       if (in_regs[i].first()->is_Register()) {
1654         const Register reg = in_regs[i].first()->as_Register();
1655         switch (in_sig_bt[i]) {
1656           case T_ARRAY:  // critical array (uses 2 slots on LP64)
1657           case T_BOOLEAN:
1658           case T_BYTE:
1659           case T_SHORT:
1660           case T_CHAR:
1661           case T_INT:  single_slots++; break;
1662           case T_LONG: double_slots++; break;
1663           default:  ShouldNotReachHere();
1664         }
1665       } else if (in_regs[i].first()->is_XMMRegister()) {
1666         switch (in_sig_bt[i]) {
1667           case T_FLOAT:  single_slots++; break;
1668           case T_DOUBLE: double_slots++; break;
1669           default:  ShouldNotReachHere();
1670         }
1671       } else if (in_regs[i].first()->is_FloatRegister()) {
1672         ShouldNotReachHere();
1673       }
1674     }
1675     total_save_slots = double_slots * 2 + single_slots;
1676     // align the save area
1677     if (double_slots != 0) {
1678       stack_slots = align_up(stack_slots, 2);
1679     }
1680   }
1681 
1682   int oop_handle_offset = stack_slots;
1683   stack_slots += total_save_slots;
1684 
1685   // Now any space we need for handlizing a klass if static method
1686 
1687   int klass_slot_offset = 0;
1688   int klass_offset = -1;
1689   int lock_slot_offset = 0;
1690   bool is_static = false;
1691 
1692   if (method->is_static()) {
1693     klass_slot_offset = stack_slots;
1694     stack_slots += VMRegImpl::slots_per_word;
1695     klass_offset = klass_slot_offset * VMRegImpl::stack_slot_size;
1696     is_static = true;
1697   }
1698 
1699   // Plus a lock if needed
1700 
1701   if (method->is_synchronized()) {
1702     lock_slot_offset = stack_slots;
1703     stack_slots += VMRegImpl::slots_per_word;
1704   }
1705 
1706   // Now a place (+2) to save return values or temp during shuffling
1707   // + 2 for return address (which we own) and saved rbp,
1708   stack_slots += 4;
1709 
1710   // Ok The space we have allocated will look like:
1711   //
1712   //
1713   // FP-> |                     |
1714   //      |---------------------|
1715   //      | 2 slots for moves   |
1716   //      |---------------------|
1717   //      | lock box (if sync)  |
1718   //      |---------------------| <- lock_slot_offset  (-lock_slot_rbp_offset)
1719   //      | klass (if static)   |
1720   //      |---------------------| <- klass_slot_offset
1721   //      | oopHandle area      |
1722   //      |---------------------| <- oop_handle_offset (a max of 2 registers)
1723   //      | outbound memory     |
1724   //      | based arguments     |
1725   //      |                     |
1726   //      |---------------------|
1727   //      |                     |
1728   // SP-> | out_preserved_slots |
1729   //
1730   //
1731   // ****************************************************************************
1732   // WARNING - on Windows Java Natives use pascal calling convention and pop the
1733   // arguments off of the stack after the jni call. Before the call we can use
1734   // instructions that are SP relative. After the jni call we switch to FP
1735   // relative instructions instead of re-adjusting the stack on windows.
1736   // ****************************************************************************
1737 
1738 
1739   // Now compute actual number of stack words we need rounding to make
1740   // stack properly aligned.
1741   stack_slots = align_up(stack_slots, StackAlignmentInSlots);
1742 
1743   int stack_size = stack_slots * VMRegImpl::stack_slot_size;
1744 
1745   intptr_t start = (intptr_t)__ pc();
1746 
1747   // First thing make an ic check to see if we should even be here
1748 
1749   // We are free to use all registers as temps without saving them and
1750   // restoring them except rbp. rbp is the only callee save register
1751   // as far as the interpreter and the compiler(s) are concerned.
1752 
1753 
1754   const Register ic_reg = rax;
1755   const Register receiver = rcx;
1756   Label hit;
1757   Label exception_pending;
1758 
1759   __ verify_oop(receiver);
1760   __ cmpptr(ic_reg, Address(receiver, oopDesc::klass_offset_in_bytes()));
1761   __ jcc(Assembler::equal, hit);
1762 
1763   __ jump(RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
1764 
1765   // verified entry must be aligned for code patching.
1766   // and the first 5 bytes must be in the same cache line
1767   // if we align at 8 then we will be sure 5 bytes are in the same line
1768   __ align(8);
1769 
1770   __ bind(hit);
1771 
1772   int vep_offset = ((intptr_t)__ pc()) - start;
1773 
1774 #ifdef COMPILER1
1775   // For Object.hashCode, System.identityHashCode try to pull hashCode from object header if available.
1776   if ((InlineObjectHash && method->intrinsic_id() == vmIntrinsics::_hashCode) || (method->intrinsic_id() == vmIntrinsics::_identityHashCode)) {
1777     inline_check_hashcode_from_object_header(masm, method, rcx /*obj_reg*/, rax /*result*/);
1778    }
1779 #endif // COMPILER1
1780 
1781   // The instruction at the verified entry point must be 5 bytes or longer
1782   // because it can be patched on the fly by make_non_entrant. The stack bang
1783   // instruction fits that requirement.
1784 
1785   // Generate stack overflow check
1786 
1787   if (UseStackBanging) {
1788     __ bang_stack_with_offset((int)JavaThread::stack_shadow_zone_size());
1789   } else {
1790     // need a 5 byte instruction to allow MT safe patching to non-entrant
1791     __ fat_nop();
1792   }
1793 
1794   // Generate a new frame for the wrapper.
1795   __ enter();
1796   // -2 because return address is already present and so is saved rbp
1797   __ subptr(rsp, stack_size - 2*wordSize);
1798 
1799   // Frame is now completed as far as size and linkage.
1800   int frame_complete = ((intptr_t)__ pc()) - start;
1801 
1802   if (UseRTMLocking) {
1803     // Abort RTM transaction before calling JNI
1804     // because critical section will be large and will be
1805     // aborted anyway. Also nmethod could be deoptimized.
1806     __ xabort(0);
1807   }
1808 
1809   // Calculate the difference between rsp and rbp,. We need to know it
1810   // after the native call because on windows Java Natives will pop
1811   // the arguments and it is painful to do rsp relative addressing
1812   // in a platform independent way. So after the call we switch to
1813   // rbp, relative addressing.
1814 
1815   int fp_adjustment = stack_size - 2*wordSize;
1816 
1817 #ifdef COMPILER2
1818   // C2 may leave the stack dirty if not in SSE2+ mode
1819   if (UseSSE >= 2) {
1820     __ verify_FPU(0, "c2i transition should have clean FPU stack");
1821   } else {
1822     __ empty_FPU_stack();
1823   }
1824 #endif /* COMPILER2 */
1825 
1826   // Compute the rbp, offset for any slots used after the jni call
1827 
1828   int lock_slot_rbp_offset = (lock_slot_offset*VMRegImpl::stack_slot_size) - fp_adjustment;
1829 
1830   // We use rdi as a thread pointer because it is callee save and
1831   // if we load it once it is usable thru the entire wrapper
1832   const Register thread = rdi;
1833 
1834   // We use rsi as the oop handle for the receiver/klass
1835   // It is callee save so it survives the call to native
1836 
1837   const Register oop_handle_reg = rsi;
1838 
1839   __ get_thread(thread);
1840 
1841   if (is_critical_native) {
1842     check_needs_gc_for_critical_native(masm, thread, stack_slots, total_c_args, total_in_args,
1843                                        oop_handle_offset, oop_maps, in_regs, in_sig_bt);
1844   }
1845 
1846   //
1847   // We immediately shuffle the arguments so that any vm call we have to
1848   // make from here on out (sync slow path, jvmti, etc.) we will have
1849   // captured the oops from our caller and have a valid oopMap for
1850   // them.
1851 
1852   // -----------------
1853   // The Grand Shuffle
1854   //
1855   // Natives require 1 or 2 extra arguments over the normal ones: the JNIEnv*
1856   // and, if static, the class mirror instead of a receiver.  This pretty much
1857   // guarantees that register layout will not match (and x86 doesn't use reg
1858   // parms though amd does).  Since the native abi doesn't use register args
1859   // and the java conventions does we don't have to worry about collisions.
1860   // All of our moved are reg->stack or stack->stack.
1861   // We ignore the extra arguments during the shuffle and handle them at the
1862   // last moment. The shuffle is described by the two calling convention
1863   // vectors we have in our possession. We simply walk the java vector to
1864   // get the source locations and the c vector to get the destinations.
1865 
1866   int c_arg = is_critical_native ? 0 : (method->is_static() ? 2 : 1 );
1867 
1868   // Record rsp-based slot for receiver on stack for non-static methods
1869   int receiver_offset = -1;
1870 
1871   // This is a trick. We double the stack slots so we can claim
1872   // the oops in the caller's frame. Since we are sure to have
1873   // more args than the caller doubling is enough to make
1874   // sure we can capture all the incoming oop args from the
1875   // caller.
1876   //
1877   OopMap* map = new OopMap(stack_slots * 2, 0 /* arg_slots*/);
1878 
1879   // Mark location of rbp,
1880   // map->set_callee_saved(VMRegImpl::stack2reg( stack_slots - 2), stack_slots * 2, 0, rbp->as_VMReg());
1881 
1882   // We know that we only have args in at most two integer registers (rcx, rdx). So rax, rbx
1883   // Are free to temporaries if we have to do  stack to steck moves.
1884   // All inbound args are referenced based on rbp, and all outbound args via rsp.
1885 
1886   for (int i = 0; i < total_in_args ; i++, c_arg++ ) {
1887     switch (in_sig_bt[i]) {
1888       case T_ARRAY:
1889         if (is_critical_native) {
1890           unpack_array_argument(masm, in_regs[i], in_elem_bt[i], out_regs[c_arg + 1], out_regs[c_arg]);
1891           c_arg++;
1892           break;
1893         }
1894       case T_OBJECT:
1895         assert(!is_critical_native, "no oop arguments");
1896         object_move(masm, map, oop_handle_offset, stack_slots, in_regs[i], out_regs[c_arg],
1897                     ((i == 0) && (!is_static)),
1898                     &receiver_offset);
1899         break;
1900       case T_VOID:
1901         break;
1902 
1903       case T_FLOAT:
1904         float_move(masm, in_regs[i], out_regs[c_arg]);
1905           break;
1906 
1907       case T_DOUBLE:
1908         assert( i + 1 < total_in_args &&
1909                 in_sig_bt[i + 1] == T_VOID &&
1910                 out_sig_bt[c_arg+1] == T_VOID, "bad arg list");
1911         double_move(masm, in_regs[i], out_regs[c_arg]);
1912         break;
1913 
1914       case T_LONG :
1915         long_move(masm, in_regs[i], out_regs[c_arg]);
1916         break;
1917 
1918       case T_ADDRESS: assert(false, "found T_ADDRESS in java args");
1919 
1920       default:
1921         simple_move32(masm, in_regs[i], out_regs[c_arg]);
1922     }
1923   }
1924 
1925   // Pre-load a static method's oop into rsi.  Used both by locking code and
1926   // the normal JNI call code.
1927   if (method->is_static() && !is_critical_native) {
1928 
1929     //  load opp into a register
1930     __ movoop(oop_handle_reg, JNIHandles::make_local(method->method_holder()->java_mirror()));
1931 
1932     // Now handlize the static class mirror it's known not-null.
1933     __ movptr(Address(rsp, klass_offset), oop_handle_reg);
1934     map->set_oop(VMRegImpl::stack2reg(klass_slot_offset));
1935 
1936     // Now get the handle
1937     __ lea(oop_handle_reg, Address(rsp, klass_offset));
1938     // store the klass handle as second argument
1939     __ movptr(Address(rsp, wordSize), oop_handle_reg);
1940   }
1941 
1942   // Change state to native (we save the return address in the thread, since it might not
1943   // be pushed on the stack when we do a a stack traversal). It is enough that the pc()
1944   // points into the right code segment. It does not have to be the correct return pc.
1945   // We use the same pc/oopMap repeatedly when we call out
1946 
1947   intptr_t the_pc = (intptr_t) __ pc();
1948   oop_maps->add_gc_map(the_pc - start, map);
1949 
1950   __ set_last_Java_frame(thread, rsp, noreg, (address)the_pc);
1951 
1952 
1953   // We have all of the arguments setup at this point. We must not touch any register
1954   // argument registers at this point (what if we save/restore them there are no oop?
1955 
1956   {
1957     SkipIfEqual skip_if(masm, &DTraceMethodProbes, 0);
1958     __ mov_metadata(rax, method());
1959     __ call_VM_leaf(
1960          CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
1961          thread, rax);
1962   }
1963 
1964   // RedefineClasses() tracing support for obsolete method entry
1965   if (log_is_enabled(Trace, redefine, class, obsolete)) {
1966     __ mov_metadata(rax, method());
1967     __ call_VM_leaf(
1968          CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
1969          thread, rax);
1970   }
1971 
1972   // These are register definitions we need for locking/unlocking
1973   const Register swap_reg = rax;  // Must use rax, for cmpxchg instruction
1974   const Register obj_reg  = rcx;  // Will contain the oop
1975   const Register lock_reg = rdx;  // Address of compiler lock object (BasicLock)
1976 
1977   Label slow_path_lock;
1978   Label lock_done;
1979 
1980   // Lock a synchronized method
1981   if (method->is_synchronized()) {
1982     assert(!is_critical_native, "unhandled");
1983 
1984 
1985     const int mark_word_offset = BasicLock::displaced_header_offset_in_bytes();
1986 
1987     // Get the handle (the 2nd argument)
1988     __ movptr(oop_handle_reg, Address(rsp, wordSize));
1989 
1990     // Get address of the box
1991 
1992     __ lea(lock_reg, Address(rbp, lock_slot_rbp_offset));
1993 
1994     // Load the oop from the handle
1995     __ movptr(obj_reg, Address(oop_handle_reg, 0));
1996 
1997     if (UseBiasedLocking) {
1998       // Note that oop_handle_reg is trashed during this call
1999       __ biased_locking_enter(lock_reg, obj_reg, swap_reg, oop_handle_reg, false, lock_done, &slow_path_lock);
2000     }
2001 
2002     // Load immediate 1 into swap_reg %rax,
2003     __ movptr(swap_reg, 1);
2004 
2005     // Load (object->mark() | 1) into swap_reg %rax,
2006     __ orptr(swap_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
2007 
2008     // Save (object->mark() | 1) into BasicLock's displaced header
2009     __ movptr(Address(lock_reg, mark_word_offset), swap_reg);
2010 
2011     if (os::is_MP()) {
2012       __ lock();
2013     }
2014 
2015     // src -> dest iff dest == rax, else rax, <- dest
2016     // *obj_reg = lock_reg iff *obj_reg == rax, else rax, = *(obj_reg)
2017     __ cmpxchgptr(lock_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
2018     __ jcc(Assembler::equal, lock_done);
2019 
2020     // Test if the oopMark is an obvious stack pointer, i.e.,
2021     //  1) (mark & 3) == 0, and
2022     //  2) rsp <= mark < mark + os::pagesize()
2023     // These 3 tests can be done by evaluating the following
2024     // expression: ((mark - rsp) & (3 - os::vm_page_size())),
2025     // assuming both stack pointer and pagesize have their
2026     // least significant 2 bits clear.
2027     // NOTE: the oopMark is in swap_reg %rax, as the result of cmpxchg
2028 
2029     __ subptr(swap_reg, rsp);
2030     __ andptr(swap_reg, 3 - os::vm_page_size());
2031 
2032     // Save the test result, for recursive case, the result is zero
2033     __ movptr(Address(lock_reg, mark_word_offset), swap_reg);
2034     __ jcc(Assembler::notEqual, slow_path_lock);
2035     // Slow path will re-enter here
2036     __ bind(lock_done);
2037 
2038     if (UseBiasedLocking) {
2039       // Re-fetch oop_handle_reg as we trashed it above
2040       __ movptr(oop_handle_reg, Address(rsp, wordSize));
2041     }
2042   }
2043 
2044 
2045   // Finally just about ready to make the JNI call
2046 
2047 
2048   // get JNIEnv* which is first argument to native
2049   if (!is_critical_native) {
2050     __ lea(rdx, Address(thread, in_bytes(JavaThread::jni_environment_offset())));
2051     __ movptr(Address(rsp, 0), rdx);
2052   }
2053 
2054   // Now set thread in native
2055   __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_native);
2056 
2057   __ call(RuntimeAddress(native_func));
2058 
2059   // Verify or restore cpu control state after JNI call
2060   __ restore_cpu_control_state_after_jni();
2061 
2062   // WARNING - on Windows Java Natives use pascal calling convention and pop the
2063   // arguments off of the stack. We could just re-adjust the stack pointer here
2064   // and continue to do SP relative addressing but we instead switch to FP
2065   // relative addressing.
2066 
2067   // Unpack native results.
2068   switch (ret_type) {
2069   case T_BOOLEAN: __ c2bool(rax);            break;
2070   case T_CHAR   : __ andptr(rax, 0xFFFF);    break;
2071   case T_BYTE   : __ sign_extend_byte (rax); break;
2072   case T_SHORT  : __ sign_extend_short(rax); break;
2073   case T_INT    : /* nothing to do */        break;
2074   case T_DOUBLE :
2075   case T_FLOAT  :
2076     // Result is in st0 we'll save as needed
2077     break;
2078   case T_ARRAY:                 // Really a handle
2079   case T_OBJECT:                // Really a handle
2080       break; // can't de-handlize until after safepoint check
2081   case T_VOID: break;
2082   case T_LONG: break;
2083   default       : ShouldNotReachHere();
2084   }
2085 
2086   // Switch thread to "native transition" state before reading the synchronization state.
2087   // This additional state is necessary because reading and testing the synchronization
2088   // state is not atomic w.r.t. GC, as this scenario demonstrates:
2089   //     Java thread A, in _thread_in_native state, loads _not_synchronized and is preempted.
2090   //     VM thread changes sync state to synchronizing and suspends threads for GC.
2091   //     Thread A is resumed to finish this native method, but doesn't block here since it
2092   //     didn't see any synchronization is progress, and escapes.
2093   __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_native_trans);
2094 
2095   if(os::is_MP()) {
2096     if (UseMembar) {
2097       // Force this write out before the read below
2098       __ membar(Assembler::Membar_mask_bits(
2099            Assembler::LoadLoad | Assembler::LoadStore |
2100            Assembler::StoreLoad | Assembler::StoreStore));
2101     } else {
2102       // Write serialization page so VM thread can do a pseudo remote membar.
2103       // We use the current thread pointer to calculate a thread specific
2104       // offset to write to within the page. This minimizes bus traffic
2105       // due to cache line collision.
2106       __ serialize_memory(thread, rcx);
2107     }
2108   }
2109 
2110   if (AlwaysRestoreFPU) {
2111     // Make sure the control word is correct.
2112     __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
2113   }
2114 
2115   Label after_transition;
2116 
2117   // check for safepoint operation in progress and/or pending suspend requests
2118   { Label Continue, slow_path;
2119 
2120     __ safepoint_poll(slow_path, thread, noreg);
2121 
2122     __ cmpl(Address(thread, JavaThread::suspend_flags_offset()), 0);
2123     __ jcc(Assembler::equal, Continue);
2124     __ bind(slow_path);
2125 
2126     // Don't use call_VM as it will see a possible pending exception and forward it
2127     // and never return here preventing us from clearing _last_native_pc down below.
2128     // Also can't use call_VM_leaf either as it will check to see if rsi & rdi are
2129     // preserved and correspond to the bcp/locals pointers. So we do a runtime call
2130     // by hand.
2131     //
2132     __ vzeroupper();
2133 
2134     save_native_result(masm, ret_type, stack_slots);
2135     __ push(thread);
2136     if (!is_critical_native) {
2137       __ call(RuntimeAddress(CAST_FROM_FN_PTR(address,
2138                                               JavaThread::check_special_condition_for_native_trans)));
2139     } else {
2140       __ call(RuntimeAddress(CAST_FROM_FN_PTR(address,
2141                                               JavaThread::check_special_condition_for_native_trans_and_transition)));
2142     }
2143     __ increment(rsp, wordSize);
2144     // Restore any method result value
2145     restore_native_result(masm, ret_type, stack_slots);
2146 
2147     if (is_critical_native) {
2148       // The call above performed the transition to thread_in_Java so
2149       // skip the transition logic below.
2150       __ jmpb(after_transition);
2151     }
2152 
2153     __ bind(Continue);
2154   }
2155 
2156   // change thread state
2157   __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_Java);
2158   __ bind(after_transition);
2159 
2160   Label reguard;
2161   Label reguard_done;
2162   __ cmpl(Address(thread, JavaThread::stack_guard_state_offset()), JavaThread::stack_guard_yellow_reserved_disabled);
2163   __ jcc(Assembler::equal, reguard);
2164 
2165   // slow path reguard  re-enters here
2166   __ bind(reguard_done);
2167 
2168   // Handle possible exception (will unlock if necessary)
2169 
2170   // native result if any is live
2171 
2172   // Unlock
2173   Label slow_path_unlock;
2174   Label unlock_done;
2175   if (method->is_synchronized()) {
2176 
2177     Label done;
2178 
2179     // Get locked oop from the handle we passed to jni
2180     __ movptr(obj_reg, Address(oop_handle_reg, 0));
2181 
2182     if (UseBiasedLocking) {
2183       __ biased_locking_exit(obj_reg, rbx, done);
2184     }
2185 
2186     // Simple recursive lock?
2187 
2188     __ cmpptr(Address(rbp, lock_slot_rbp_offset), (int32_t)NULL_WORD);
2189     __ jcc(Assembler::equal, done);
2190 
2191     // Must save rax, if if it is live now because cmpxchg must use it
2192     if (ret_type != T_FLOAT && ret_type != T_DOUBLE && ret_type != T_VOID) {
2193       save_native_result(masm, ret_type, stack_slots);
2194     }
2195 
2196     //  get old displaced header
2197     __ movptr(rbx, Address(rbp, lock_slot_rbp_offset));
2198 
2199     // get address of the stack lock
2200     __ lea(rax, Address(rbp, lock_slot_rbp_offset));
2201 
2202     // Atomic swap old header if oop still contains the stack lock
2203     if (os::is_MP()) {
2204     __ lock();
2205     }
2206 
2207     // src -> dest iff dest == rax, else rax, <- dest
2208     // *obj_reg = rbx, iff *obj_reg == rax, else rax, = *(obj_reg)
2209     __ cmpxchgptr(rbx, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
2210     __ jcc(Assembler::notEqual, slow_path_unlock);
2211 
2212     // slow path re-enters here
2213     __ bind(unlock_done);
2214     if (ret_type != T_FLOAT && ret_type != T_DOUBLE && ret_type != T_VOID) {
2215       restore_native_result(masm, ret_type, stack_slots);
2216     }
2217 
2218     __ bind(done);
2219 
2220   }
2221 
2222   {
2223     SkipIfEqual skip_if(masm, &DTraceMethodProbes, 0);
2224     // Tell dtrace about this method exit
2225     save_native_result(masm, ret_type, stack_slots);
2226     __ mov_metadata(rax, method());
2227     __ call_VM_leaf(
2228          CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
2229          thread, rax);
2230     restore_native_result(masm, ret_type, stack_slots);
2231   }
2232 
2233   // We can finally stop using that last_Java_frame we setup ages ago
2234 
2235   __ reset_last_Java_frame(thread, false);
2236 
2237   // Unbox oop result, e.g. JNIHandles::resolve value.
2238   if (ret_type == T_OBJECT || ret_type == T_ARRAY) {
2239     __ resolve_jobject(rax /* value */,
2240                        thread /* thread */,
2241                        rcx /* tmp */);
2242   }
2243 
2244   if (CheckJNICalls) {
2245     // clear_pending_jni_exception_check
2246     __ movptr(Address(thread, JavaThread::pending_jni_exception_check_fn_offset()), NULL_WORD);
2247   }
2248 
2249   if (!is_critical_native) {
2250     // reset handle block
2251     __ movptr(rcx, Address(thread, JavaThread::active_handles_offset()));
2252     __ movl(Address(rcx, JNIHandleBlock::top_offset_in_bytes()), NULL_WORD);
2253 
2254     // Any exception pending?
2255     __ cmpptr(Address(thread, in_bytes(Thread::pending_exception_offset())), (int32_t)NULL_WORD);
2256     __ jcc(Assembler::notEqual, exception_pending);
2257   }
2258 
2259   // no exception, we're almost done
2260 
2261   // check that only result value is on FPU stack
2262   __ verify_FPU(ret_type == T_FLOAT || ret_type == T_DOUBLE ? 1 : 0, "native_wrapper normal exit");
2263 
2264   // Fixup floating pointer results so that result looks like a return from a compiled method
2265   if (ret_type == T_FLOAT) {
2266     if (UseSSE >= 1) {
2267       // Pop st0 and store as float and reload into xmm register
2268       __ fstp_s(Address(rbp, -4));
2269       __ movflt(xmm0, Address(rbp, -4));
2270     }
2271   } else if (ret_type == T_DOUBLE) {
2272     if (UseSSE >= 2) {
2273       // Pop st0 and store as double and reload into xmm register
2274       __ fstp_d(Address(rbp, -8));
2275       __ movdbl(xmm0, Address(rbp, -8));
2276     }
2277   }
2278 
2279   // Return
2280 
2281   __ leave();
2282   __ ret(0);
2283 
2284   // Unexpected paths are out of line and go here
2285 
2286   // Slow path locking & unlocking
2287   if (method->is_synchronized()) {
2288 
2289     // BEGIN Slow path lock
2290 
2291     __ bind(slow_path_lock);
2292 
2293     // has last_Java_frame setup. No exceptions so do vanilla call not call_VM
2294     // args are (oop obj, BasicLock* lock, JavaThread* thread)
2295     __ push(thread);
2296     __ push(lock_reg);
2297     __ push(obj_reg);
2298     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::complete_monitor_locking_C)));
2299     __ addptr(rsp, 3*wordSize);
2300 
2301 #ifdef ASSERT
2302     { Label L;
2303     __ cmpptr(Address(thread, in_bytes(Thread::pending_exception_offset())), (int)NULL_WORD);
2304     __ jcc(Assembler::equal, L);
2305     __ stop("no pending exception allowed on exit from monitorenter");
2306     __ bind(L);
2307     }
2308 #endif
2309     __ jmp(lock_done);
2310 
2311     // END Slow path lock
2312 
2313     // BEGIN Slow path unlock
2314     __ bind(slow_path_unlock);
2315     __ vzeroupper();
2316     // Slow path unlock
2317 
2318     if (ret_type == T_FLOAT || ret_type == T_DOUBLE ) {
2319       save_native_result(masm, ret_type, stack_slots);
2320     }
2321     // Save pending exception around call to VM (which contains an EXCEPTION_MARK)
2322 
2323     __ pushptr(Address(thread, in_bytes(Thread::pending_exception_offset())));
2324     __ movptr(Address(thread, in_bytes(Thread::pending_exception_offset())), NULL_WORD);
2325 
2326 
2327     // should be a peal
2328     // +wordSize because of the push above
2329     // args are (oop obj, BasicLock* lock, JavaThread* thread)
2330     __ push(thread);
2331     __ lea(rax, Address(rbp, lock_slot_rbp_offset));
2332     __ push(rax);
2333 
2334     __ push(obj_reg);
2335     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::complete_monitor_unlocking_C)));
2336     __ addptr(rsp, 3*wordSize);
2337 #ifdef ASSERT
2338     {
2339       Label L;
2340       __ cmpptr(Address(thread, in_bytes(Thread::pending_exception_offset())), (int32_t)NULL_WORD);
2341       __ jcc(Assembler::equal, L);
2342       __ stop("no pending exception allowed on exit complete_monitor_unlocking_C");
2343       __ bind(L);
2344     }
2345 #endif /* ASSERT */
2346 
2347     __ popptr(Address(thread, in_bytes(Thread::pending_exception_offset())));
2348 
2349     if (ret_type == T_FLOAT || ret_type == T_DOUBLE ) {
2350       restore_native_result(masm, ret_type, stack_slots);
2351     }
2352     __ jmp(unlock_done);
2353     // END Slow path unlock
2354 
2355   }
2356 
2357   // SLOW PATH Reguard the stack if needed
2358 
2359   __ bind(reguard);
2360   __ vzeroupper();
2361   save_native_result(masm, ret_type, stack_slots);
2362   {
2363     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
2364   }
2365   restore_native_result(masm, ret_type, stack_slots);
2366   __ jmp(reguard_done);
2367 
2368 
2369   // BEGIN EXCEPTION PROCESSING
2370 
2371   if (!is_critical_native) {
2372     // Forward  the exception
2373     __ bind(exception_pending);
2374 
2375     // remove possible return value from FPU register stack
2376     __ empty_FPU_stack();
2377 
2378     // pop our frame
2379     __ leave();
2380     // and forward the exception
2381     __ jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
2382   }
2383 
2384   __ flush();
2385 
2386   nmethod *nm = nmethod::new_native_nmethod(method,
2387                                             compile_id,
2388                                             masm->code(),
2389                                             vep_offset,
2390                                             frame_complete,
2391                                             stack_slots / VMRegImpl::slots_per_word,
2392                                             (is_static ? in_ByteSize(klass_offset) : in_ByteSize(receiver_offset)),
2393                                             in_ByteSize(lock_slot_offset*VMRegImpl::stack_slot_size),
2394                                             oop_maps);
2395 
2396   if (is_critical_native) {
2397     nm->set_lazy_critical_native(true);
2398   }
2399 
2400   return nm;
2401 
2402 }
2403 
2404 // this function returns the adjust size (in number of words) to a c2i adapter
2405 // activation for use during deoptimization
2406 int Deoptimization::last_frame_adjust(int callee_parameters, int callee_locals ) {
2407   return (callee_locals - callee_parameters) * Interpreter::stackElementWords;
2408 }
2409 
2410 
2411 uint SharedRuntime::out_preserve_stack_slots() {
2412   return 0;
2413 }
2414 
2415 //------------------------------generate_deopt_blob----------------------------
2416 void SharedRuntime::generate_deopt_blob() {
2417   // allocate space for the code
2418   ResourceMark rm;
2419   // setup code generation tools
2420   // note: the buffer code size must account for StackShadowPages=50
2421   CodeBuffer   buffer("deopt_blob", 1536, 1024);
2422   MacroAssembler* masm = new MacroAssembler(&buffer);
2423   int frame_size_in_words;
2424   OopMap* map = NULL;
2425   // Account for the extra args we place on the stack
2426   // by the time we call fetch_unroll_info
2427   const int additional_words = 2; // deopt kind, thread
2428 
2429   OopMapSet *oop_maps = new OopMapSet();
2430 
2431   // -------------
2432   // This code enters when returning to a de-optimized nmethod.  A return
2433   // address has been pushed on the the stack, and return values are in
2434   // registers.
2435   // If we are doing a normal deopt then we were called from the patched
2436   // nmethod from the point we returned to the nmethod. So the return
2437   // address on the stack is wrong by NativeCall::instruction_size
2438   // We will adjust the value to it looks like we have the original return
2439   // address on the stack (like when we eagerly deoptimized).
2440   // In the case of an exception pending with deoptimized then we enter
2441   // with a return address on the stack that points after the call we patched
2442   // into the exception handler. We have the following register state:
2443   //    rax,: exception
2444   //    rbx,: exception handler
2445   //    rdx: throwing pc
2446   // So in this case we simply jam rdx into the useless return address and
2447   // the stack looks just like we want.
2448   //
2449   // At this point we need to de-opt.  We save the argument return
2450   // registers.  We call the first C routine, fetch_unroll_info().  This
2451   // routine captures the return values and returns a structure which
2452   // describes the current frame size and the sizes of all replacement frames.
2453   // The current frame is compiled code and may contain many inlined
2454   // functions, each with their own JVM state.  We pop the current frame, then
2455   // push all the new frames.  Then we call the C routine unpack_frames() to
2456   // populate these frames.  Finally unpack_frames() returns us the new target
2457   // address.  Notice that callee-save registers are BLOWN here; they have
2458   // already been captured in the vframeArray at the time the return PC was
2459   // patched.
2460   address start = __ pc();
2461   Label cont;
2462 
2463   // Prolog for non exception case!
2464 
2465   // Save everything in sight.
2466 
2467   map = RegisterSaver::save_live_registers(masm, additional_words, &frame_size_in_words, false);
2468   // Normal deoptimization
2469   __ push(Deoptimization::Unpack_deopt);
2470   __ jmp(cont);
2471 
2472   int reexecute_offset = __ pc() - start;
2473 
2474   // Reexecute case
2475   // return address is the pc describes what bci to do re-execute at
2476 
2477   // No need to update map as each call to save_live_registers will produce identical oopmap
2478   (void) RegisterSaver::save_live_registers(masm, additional_words, &frame_size_in_words, false);
2479 
2480   __ push(Deoptimization::Unpack_reexecute);
2481   __ jmp(cont);
2482 
2483   int exception_offset = __ pc() - start;
2484 
2485   // Prolog for exception case
2486 
2487   // all registers are dead at this entry point, except for rax, and
2488   // rdx which contain the exception oop and exception pc
2489   // respectively.  Set them in TLS and fall thru to the
2490   // unpack_with_exception_in_tls entry point.
2491 
2492   __ get_thread(rdi);
2493   __ movptr(Address(rdi, JavaThread::exception_pc_offset()), rdx);
2494   __ movptr(Address(rdi, JavaThread::exception_oop_offset()), rax);
2495 
2496   int exception_in_tls_offset = __ pc() - start;
2497 
2498   // new implementation because exception oop is now passed in JavaThread
2499 
2500   // Prolog for exception case
2501   // All registers must be preserved because they might be used by LinearScan
2502   // Exceptiop oop and throwing PC are passed in JavaThread
2503   // tos: stack at point of call to method that threw the exception (i.e. only
2504   // args are on the stack, no return address)
2505 
2506   // make room on stack for the return address
2507   // It will be patched later with the throwing pc. The correct value is not
2508   // available now because loading it from memory would destroy registers.
2509   __ push(0);
2510 
2511   // Save everything in sight.
2512 
2513   // No need to update map as each call to save_live_registers will produce identical oopmap
2514   (void) RegisterSaver::save_live_registers(masm, additional_words, &frame_size_in_words, false);
2515 
2516   // Now it is safe to overwrite any register
2517 
2518   // store the correct deoptimization type
2519   __ push(Deoptimization::Unpack_exception);
2520 
2521   // load throwing pc from JavaThread and patch it as the return address
2522   // of the current frame. Then clear the field in JavaThread
2523   __ get_thread(rdi);
2524   __ movptr(rdx, Address(rdi, JavaThread::exception_pc_offset()));
2525   __ movptr(Address(rbp, wordSize), rdx);
2526   __ movptr(Address(rdi, JavaThread::exception_pc_offset()), NULL_WORD);
2527 
2528 #ifdef ASSERT
2529   // verify that there is really an exception oop in JavaThread
2530   __ movptr(rax, Address(rdi, JavaThread::exception_oop_offset()));
2531   __ verify_oop(rax);
2532 
2533   // verify that there is no pending exception
2534   Label no_pending_exception;
2535   __ movptr(rax, Address(rdi, Thread::pending_exception_offset()));
2536   __ testptr(rax, rax);
2537   __ jcc(Assembler::zero, no_pending_exception);
2538   __ stop("must not have pending exception here");
2539   __ bind(no_pending_exception);
2540 #endif
2541 
2542   __ bind(cont);
2543 
2544   // Compiled code leaves the floating point stack dirty, empty it.
2545   __ empty_FPU_stack();
2546 
2547 
2548   // Call C code.  Need thread and this frame, but NOT official VM entry
2549   // crud.  We cannot block on this call, no GC can happen.
2550   __ get_thread(rcx);
2551   __ push(rcx);
2552   // fetch_unroll_info needs to call last_java_frame()
2553   __ set_last_Java_frame(rcx, noreg, noreg, NULL);
2554 
2555   __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, Deoptimization::fetch_unroll_info)));
2556 
2557   // Need to have an oopmap that tells fetch_unroll_info where to
2558   // find any register it might need.
2559 
2560   oop_maps->add_gc_map( __ pc()-start, map);
2561 
2562   // Discard args to fetch_unroll_info
2563   __ pop(rcx);
2564   __ pop(rcx);
2565 
2566   __ get_thread(rcx);
2567   __ reset_last_Java_frame(rcx, false);
2568 
2569   // Load UnrollBlock into EDI
2570   __ mov(rdi, rax);
2571 
2572   // Move the unpack kind to a safe place in the UnrollBlock because
2573   // we are very short of registers
2574 
2575   Address unpack_kind(rdi, Deoptimization::UnrollBlock::unpack_kind_offset_in_bytes());
2576   // retrieve the deopt kind from the UnrollBlock.
2577   __ movl(rax, unpack_kind);
2578 
2579    Label noException;
2580   __ cmpl(rax, Deoptimization::Unpack_exception);   // Was exception pending?
2581   __ jcc(Assembler::notEqual, noException);
2582   __ movptr(rax, Address(rcx, JavaThread::exception_oop_offset()));
2583   __ movptr(rdx, Address(rcx, JavaThread::exception_pc_offset()));
2584   __ movptr(Address(rcx, JavaThread::exception_oop_offset()), NULL_WORD);
2585   __ movptr(Address(rcx, JavaThread::exception_pc_offset()), NULL_WORD);
2586 
2587   __ verify_oop(rax);
2588 
2589   // Overwrite the result registers with the exception results.
2590   __ movptr(Address(rsp, RegisterSaver::raxOffset()*wordSize), rax);
2591   __ movptr(Address(rsp, RegisterSaver::rdxOffset()*wordSize), rdx);
2592 
2593   __ bind(noException);
2594 
2595   // Stack is back to only having register save data on the stack.
2596   // Now restore the result registers. Everything else is either dead or captured
2597   // in the vframeArray.
2598 
2599   RegisterSaver::restore_result_registers(masm);
2600 
2601   // Non standard control word may be leaked out through a safepoint blob, and we can
2602   // deopt at a poll point with the non standard control word. However, we should make
2603   // sure the control word is correct after restore_result_registers.
2604   __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
2605 
2606   // All of the register save area has been popped of the stack. Only the
2607   // return address remains.
2608 
2609   // Pop all the frames we must move/replace.
2610   //
2611   // Frame picture (youngest to oldest)
2612   // 1: self-frame (no frame link)
2613   // 2: deopting frame  (no frame link)
2614   // 3: caller of deopting frame (could be compiled/interpreted).
2615   //
2616   // Note: by leaving the return address of self-frame on the stack
2617   // and using the size of frame 2 to adjust the stack
2618   // when we are done the return to frame 3 will still be on the stack.
2619 
2620   // Pop deoptimized frame
2621   __ addptr(rsp, Address(rdi,Deoptimization::UnrollBlock::size_of_deoptimized_frame_offset_in_bytes()));
2622 
2623   // sp should be pointing at the return address to the caller (3)
2624 
2625   // Pick up the initial fp we should save
2626   // restore rbp before stack bang because if stack overflow is thrown it needs to be pushed (and preserved)
2627   __ movptr(rbp, Address(rdi, Deoptimization::UnrollBlock::initial_info_offset_in_bytes()));
2628 
2629 #ifdef ASSERT
2630   // Compilers generate code that bang the stack by as much as the
2631   // interpreter would need. So this stack banging should never
2632   // trigger a fault. Verify that it does not on non product builds.
2633   if (UseStackBanging) {
2634     __ movl(rbx, Address(rdi ,Deoptimization::UnrollBlock::total_frame_sizes_offset_in_bytes()));
2635     __ bang_stack_size(rbx, rcx);
2636   }
2637 #endif
2638 
2639   // Load array of frame pcs into ECX
2640   __ movptr(rcx,Address(rdi,Deoptimization::UnrollBlock::frame_pcs_offset_in_bytes()));
2641 
2642   __ pop(rsi); // trash the old pc
2643 
2644   // Load array of frame sizes into ESI
2645   __ movptr(rsi,Address(rdi,Deoptimization::UnrollBlock::frame_sizes_offset_in_bytes()));
2646 
2647   Address counter(rdi, Deoptimization::UnrollBlock::counter_temp_offset_in_bytes());
2648 
2649   __ movl(rbx, Address(rdi, Deoptimization::UnrollBlock::number_of_frames_offset_in_bytes()));
2650   __ movl(counter, rbx);
2651 
2652   // Now adjust the caller's stack to make up for the extra locals
2653   // but record the original sp so that we can save it in the skeletal interpreter
2654   // frame and the stack walking of interpreter_sender will get the unextended sp
2655   // value and not the "real" sp value.
2656 
2657   Address sp_temp(rdi, Deoptimization::UnrollBlock::sender_sp_temp_offset_in_bytes());
2658   __ movptr(sp_temp, rsp);
2659   __ movl2ptr(rbx, Address(rdi, Deoptimization::UnrollBlock::caller_adjustment_offset_in_bytes()));
2660   __ subptr(rsp, rbx);
2661 
2662   // Push interpreter frames in a loop
2663   Label loop;
2664   __ bind(loop);
2665   __ movptr(rbx, Address(rsi, 0));      // Load frame size
2666   __ subptr(rbx, 2*wordSize);           // we'll push pc and rbp, by hand
2667   __ pushptr(Address(rcx, 0));          // save return address
2668   __ enter();                           // save old & set new rbp,
2669   __ subptr(rsp, rbx);                  // Prolog!
2670   __ movptr(rbx, sp_temp);              // sender's sp
2671   // This value is corrected by layout_activation_impl
2672   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
2673   __ movptr(Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize), rbx); // Make it walkable
2674   __ movptr(sp_temp, rsp);              // pass to next frame
2675   __ addptr(rsi, wordSize);             // Bump array pointer (sizes)
2676   __ addptr(rcx, wordSize);             // Bump array pointer (pcs)
2677   __ decrementl(counter);             // decrement counter
2678   __ jcc(Assembler::notZero, loop);
2679   __ pushptr(Address(rcx, 0));          // save final return address
2680 
2681   // Re-push self-frame
2682   __ enter();                           // save old & set new rbp,
2683 
2684   //  Return address and rbp, are in place
2685   // We'll push additional args later. Just allocate a full sized
2686   // register save area
2687   __ subptr(rsp, (frame_size_in_words-additional_words - 2) * wordSize);
2688 
2689   // Restore frame locals after moving the frame
2690   __ movptr(Address(rsp, RegisterSaver::raxOffset()*wordSize), rax);
2691   __ movptr(Address(rsp, RegisterSaver::rdxOffset()*wordSize), rdx);
2692   __ fstp_d(Address(rsp, RegisterSaver::fpResultOffset()*wordSize));   // Pop float stack and store in local
2693   if( UseSSE>=2 ) __ movdbl(Address(rsp, RegisterSaver::xmm0Offset()*wordSize), xmm0);
2694   if( UseSSE==1 ) __ movflt(Address(rsp, RegisterSaver::xmm0Offset()*wordSize), xmm0);
2695 
2696   // Set up the args to unpack_frame
2697 
2698   __ pushl(unpack_kind);                     // get the unpack_kind value
2699   __ get_thread(rcx);
2700   __ push(rcx);
2701 
2702   // set last_Java_sp, last_Java_fp
2703   __ set_last_Java_frame(rcx, noreg, rbp, NULL);
2704 
2705   // Call C code.  Need thread but NOT official VM entry
2706   // crud.  We cannot block on this call, no GC can happen.  Call should
2707   // restore return values to their stack-slots with the new SP.
2708   __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, Deoptimization::unpack_frames)));
2709   // Set an oopmap for the call site
2710   oop_maps->add_gc_map( __ pc()-start, new OopMap( frame_size_in_words, 0 ));
2711 
2712   // rax, contains the return result type
2713   __ push(rax);
2714 
2715   __ get_thread(rcx);
2716   __ reset_last_Java_frame(rcx, false);
2717 
2718   // Collect return values
2719   __ movptr(rax,Address(rsp, (RegisterSaver::raxOffset() + additional_words + 1)*wordSize));
2720   __ movptr(rdx,Address(rsp, (RegisterSaver::rdxOffset() + additional_words + 1)*wordSize));
2721 
2722   // Clear floating point stack before returning to interpreter
2723   __ empty_FPU_stack();
2724 
2725   // Check if we should push the float or double return value.
2726   Label results_done, yes_double_value;
2727   __ cmpl(Address(rsp, 0), T_DOUBLE);
2728   __ jcc (Assembler::zero, yes_double_value);
2729   __ cmpl(Address(rsp, 0), T_FLOAT);
2730   __ jcc (Assembler::notZero, results_done);
2731 
2732   // return float value as expected by interpreter
2733   if( UseSSE>=1 ) __ movflt(xmm0, Address(rsp, (RegisterSaver::xmm0Offset() + additional_words + 1)*wordSize));
2734   else            __ fld_d(Address(rsp, (RegisterSaver::fpResultOffset() + additional_words + 1)*wordSize));
2735   __ jmp(results_done);
2736 
2737   // return double value as expected by interpreter
2738   __ bind(yes_double_value);
2739   if( UseSSE>=2 ) __ movdbl(xmm0, Address(rsp, (RegisterSaver::xmm0Offset() + additional_words + 1)*wordSize));
2740   else            __ fld_d(Address(rsp, (RegisterSaver::fpResultOffset() + additional_words + 1)*wordSize));
2741 
2742   __ bind(results_done);
2743 
2744   // Pop self-frame.
2745   __ leave();                              // Epilog!
2746 
2747   // Jump to interpreter
2748   __ ret(0);
2749 
2750   // -------------
2751   // make sure all code is generated
2752   masm->flush();
2753 
2754   _deopt_blob = DeoptimizationBlob::create( &buffer, oop_maps, 0, exception_offset, reexecute_offset, frame_size_in_words);
2755   _deopt_blob->set_unpack_with_exception_in_tls_offset(exception_in_tls_offset);
2756 }
2757 
2758 
2759 #ifdef COMPILER2
2760 //------------------------------generate_uncommon_trap_blob--------------------
2761 void SharedRuntime::generate_uncommon_trap_blob() {
2762   // allocate space for the code
2763   ResourceMark rm;
2764   // setup code generation tools
2765   CodeBuffer   buffer("uncommon_trap_blob", 512, 512);
2766   MacroAssembler* masm = new MacroAssembler(&buffer);
2767 
2768   enum frame_layout {
2769     arg0_off,      // thread                     sp + 0 // Arg location for
2770     arg1_off,      // unloaded_class_index       sp + 1 // calling C
2771     arg2_off,      // exec_mode                  sp + 2
2772     // The frame sender code expects that rbp will be in the "natural" place and
2773     // will override any oopMap setting for it. We must therefore force the layout
2774     // so that it agrees with the frame sender code.
2775     rbp_off,       // callee saved register      sp + 3
2776     return_off,    // slot for return address    sp + 4
2777     framesize
2778   };
2779 
2780   address start = __ pc();
2781 
2782   if (UseRTMLocking) {
2783     // Abort RTM transaction before possible nmethod deoptimization.
2784     __ xabort(0);
2785   }
2786 
2787   // Push self-frame.
2788   __ subptr(rsp, return_off*wordSize);     // Epilog!
2789 
2790   // rbp, is an implicitly saved callee saved register (i.e. the calling
2791   // convention will save restore it in prolog/epilog) Other than that
2792   // there are no callee save registers no that adapter frames are gone.
2793   __ movptr(Address(rsp, rbp_off*wordSize), rbp);
2794 
2795   // Clear the floating point exception stack
2796   __ empty_FPU_stack();
2797 
2798   // set last_Java_sp
2799   __ get_thread(rdx);
2800   __ set_last_Java_frame(rdx, noreg, noreg, NULL);
2801 
2802   // Call C code.  Need thread but NOT official VM entry
2803   // crud.  We cannot block on this call, no GC can happen.  Call should
2804   // capture callee-saved registers as well as return values.
2805   __ movptr(Address(rsp, arg0_off*wordSize), rdx);
2806   // argument already in ECX
2807   __ movl(Address(rsp, arg1_off*wordSize),rcx);
2808   __ movl(Address(rsp, arg2_off*wordSize), Deoptimization::Unpack_uncommon_trap);
2809   __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, Deoptimization::uncommon_trap)));
2810 
2811   // Set an oopmap for the call site
2812   OopMapSet *oop_maps = new OopMapSet();
2813   OopMap* map =  new OopMap( framesize, 0 );
2814   // No oopMap for rbp, it is known implicitly
2815 
2816   oop_maps->add_gc_map( __ pc()-start, map);
2817 
2818   __ get_thread(rcx);
2819 
2820   __ reset_last_Java_frame(rcx, false);
2821 
2822   // Load UnrollBlock into EDI
2823   __ movptr(rdi, rax);
2824 
2825 #ifdef ASSERT
2826   { Label L;
2827     __ cmpptr(Address(rdi, Deoptimization::UnrollBlock::unpack_kind_offset_in_bytes()),
2828             (int32_t)Deoptimization::Unpack_uncommon_trap);
2829     __ jcc(Assembler::equal, L);
2830     __ stop("SharedRuntime::generate_deopt_blob: expected Unpack_uncommon_trap");
2831     __ bind(L);
2832   }
2833 #endif
2834 
2835   // Pop all the frames we must move/replace.
2836   //
2837   // Frame picture (youngest to oldest)
2838   // 1: self-frame (no frame link)
2839   // 2: deopting frame  (no frame link)
2840   // 3: caller of deopting frame (could be compiled/interpreted).
2841 
2842   // Pop self-frame.  We have no frame, and must rely only on EAX and ESP.
2843   __ addptr(rsp,(framesize-1)*wordSize);     // Epilog!
2844 
2845   // Pop deoptimized frame
2846   __ movl2ptr(rcx, Address(rdi,Deoptimization::UnrollBlock::size_of_deoptimized_frame_offset_in_bytes()));
2847   __ addptr(rsp, rcx);
2848 
2849   // sp should be pointing at the return address to the caller (3)
2850 
2851   // Pick up the initial fp we should save
2852   // restore rbp before stack bang because if stack overflow is thrown it needs to be pushed (and preserved)
2853   __ movptr(rbp, Address(rdi, Deoptimization::UnrollBlock::initial_info_offset_in_bytes()));
2854 
2855 #ifdef ASSERT
2856   // Compilers generate code that bang the stack by as much as the
2857   // interpreter would need. So this stack banging should never
2858   // trigger a fault. Verify that it does not on non product builds.
2859   if (UseStackBanging) {
2860     __ movl(rbx, Address(rdi ,Deoptimization::UnrollBlock::total_frame_sizes_offset_in_bytes()));
2861     __ bang_stack_size(rbx, rcx);
2862   }
2863 #endif
2864 
2865   // Load array of frame pcs into ECX
2866   __ movl(rcx,Address(rdi,Deoptimization::UnrollBlock::frame_pcs_offset_in_bytes()));
2867 
2868   __ pop(rsi); // trash the pc
2869 
2870   // Load array of frame sizes into ESI
2871   __ movptr(rsi,Address(rdi,Deoptimization::UnrollBlock::frame_sizes_offset_in_bytes()));
2872 
2873   Address counter(rdi, Deoptimization::UnrollBlock::counter_temp_offset_in_bytes());
2874 
2875   __ movl(rbx, Address(rdi, Deoptimization::UnrollBlock::number_of_frames_offset_in_bytes()));
2876   __ movl(counter, rbx);
2877 
2878   // Now adjust the caller's stack to make up for the extra locals
2879   // but record the original sp so that we can save it in the skeletal interpreter
2880   // frame and the stack walking of interpreter_sender will get the unextended sp
2881   // value and not the "real" sp value.
2882 
2883   Address sp_temp(rdi, Deoptimization::UnrollBlock::sender_sp_temp_offset_in_bytes());
2884   __ movptr(sp_temp, rsp);
2885   __ movl(rbx, Address(rdi, Deoptimization::UnrollBlock::caller_adjustment_offset_in_bytes()));
2886   __ subptr(rsp, rbx);
2887 
2888   // Push interpreter frames in a loop
2889   Label loop;
2890   __ bind(loop);
2891   __ movptr(rbx, Address(rsi, 0));      // Load frame size
2892   __ subptr(rbx, 2*wordSize);           // we'll push pc and rbp, by hand
2893   __ pushptr(Address(rcx, 0));          // save return address
2894   __ enter();                           // save old & set new rbp,
2895   __ subptr(rsp, rbx);                  // Prolog!
2896   __ movptr(rbx, sp_temp);              // sender's sp
2897   // This value is corrected by layout_activation_impl
2898   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD );
2899   __ movptr(Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize), rbx); // Make it walkable
2900   __ movptr(sp_temp, rsp);              // pass to next frame
2901   __ addptr(rsi, wordSize);             // Bump array pointer (sizes)
2902   __ addptr(rcx, wordSize);             // Bump array pointer (pcs)
2903   __ decrementl(counter);             // decrement counter
2904   __ jcc(Assembler::notZero, loop);
2905   __ pushptr(Address(rcx, 0));            // save final return address
2906 
2907   // Re-push self-frame
2908   __ enter();                           // save old & set new rbp,
2909   __ subptr(rsp, (framesize-2) * wordSize);   // Prolog!
2910 
2911 
2912   // set last_Java_sp, last_Java_fp
2913   __ get_thread(rdi);
2914   __ set_last_Java_frame(rdi, noreg, rbp, NULL);
2915 
2916   // Call C code.  Need thread but NOT official VM entry
2917   // crud.  We cannot block on this call, no GC can happen.  Call should
2918   // restore return values to their stack-slots with the new SP.
2919   __ movptr(Address(rsp,arg0_off*wordSize),rdi);
2920   __ movl(Address(rsp,arg1_off*wordSize), Deoptimization::Unpack_uncommon_trap);
2921   __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, Deoptimization::unpack_frames)));
2922   // Set an oopmap for the call site
2923   oop_maps->add_gc_map( __ pc()-start, new OopMap( framesize, 0 ) );
2924 
2925   __ get_thread(rdi);
2926   __ reset_last_Java_frame(rdi, true);
2927 
2928   // Pop self-frame.
2929   __ leave();     // Epilog!
2930 
2931   // Jump to interpreter
2932   __ ret(0);
2933 
2934   // -------------
2935   // make sure all code is generated
2936   masm->flush();
2937 
2938    _uncommon_trap_blob = UncommonTrapBlob::create(&buffer, oop_maps, framesize);
2939 }
2940 #endif // COMPILER2
2941 
2942 //------------------------------generate_handler_blob------
2943 //
2944 // Generate a special Compile2Runtime blob that saves all registers,
2945 // setup oopmap, and calls safepoint code to stop the compiled code for
2946 // a safepoint.
2947 //
2948 SafepointBlob* SharedRuntime::generate_handler_blob(address call_ptr, int poll_type) {
2949 
2950   // Account for thread arg in our frame
2951   const int additional_words = 1;
2952   int frame_size_in_words;
2953 
2954   assert (StubRoutines::forward_exception_entry() != NULL, "must be generated before");
2955 
2956   ResourceMark rm;
2957   OopMapSet *oop_maps = new OopMapSet();
2958   OopMap* map;
2959 
2960   // allocate space for the code
2961   // setup code generation tools
2962   CodeBuffer   buffer("handler_blob", 1024, 512);
2963   MacroAssembler* masm = new MacroAssembler(&buffer);
2964 
2965   const Register java_thread = rdi; // callee-saved for VC++
2966   address start   = __ pc();
2967   address call_pc = NULL;
2968   bool cause_return = (poll_type == POLL_AT_RETURN);
2969   bool save_vectors = (poll_type == POLL_AT_VECTOR_LOOP);
2970 
2971   if (UseRTMLocking) {
2972     // Abort RTM transaction before calling runtime
2973     // because critical section will be large and will be
2974     // aborted anyway. Also nmethod could be deoptimized.
2975     __ xabort(0);
2976   }
2977 
2978   // If cause_return is true we are at a poll_return and there is
2979   // the return address on the stack to the caller on the nmethod
2980   // that is safepoint. We can leave this return on the stack and
2981   // effectively complete the return and safepoint in the caller.
2982   // Otherwise we push space for a return address that the safepoint
2983   // handler will install later to make the stack walking sensible.
2984   if (!cause_return)
2985     __ push(rbx);  // Make room for return address (or push it again)
2986 
2987   map = RegisterSaver::save_live_registers(masm, additional_words, &frame_size_in_words, false, save_vectors);
2988 
2989   // The following is basically a call_VM. However, we need the precise
2990   // address of the call in order to generate an oopmap. Hence, we do all the
2991   // work ourselves.
2992 
2993   // Push thread argument and setup last_Java_sp
2994   __ get_thread(java_thread);
2995   __ push(java_thread);
2996   __ set_last_Java_frame(java_thread, noreg, noreg, NULL);
2997 
2998   // if this was not a poll_return then we need to correct the return address now.
2999   if (!cause_return) {
3000     // Get the return pc saved by the signal handler and stash it in its appropriate place on the stack.
3001     // Additionally, rbx is a callee saved register and we can look at it later to determine
3002     // if someone changed the return address for us!
3003     __ movptr(rbx, Address(java_thread, JavaThread::saved_exception_pc_offset()));
3004     __ movptr(Address(rbp, wordSize), rbx);
3005   }
3006 
3007   // do the call
3008   __ call(RuntimeAddress(call_ptr));
3009 
3010   // Set an oopmap for the call site.  This oopmap will map all
3011   // oop-registers and debug-info registers as callee-saved.  This
3012   // will allow deoptimization at this safepoint to find all possible
3013   // debug-info recordings, as well as let GC find all oops.
3014 
3015   oop_maps->add_gc_map( __ pc() - start, map);
3016 
3017   // Discard arg
3018   __ pop(rcx);
3019 
3020   Label noException;
3021 
3022   // Clear last_Java_sp again
3023   __ get_thread(java_thread);
3024   __ reset_last_Java_frame(java_thread, false);
3025 
3026   __ cmpptr(Address(java_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
3027   __ jcc(Assembler::equal, noException);
3028 
3029   // Exception pending
3030   RegisterSaver::restore_live_registers(masm, save_vectors);
3031 
3032   __ jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
3033 
3034   __ bind(noException);
3035 
3036   Label no_adjust, bail, not_special;
3037   if (SafepointMechanism::uses_thread_local_poll() && !cause_return) {
3038     // If our stashed return pc was modified by the runtime we avoid touching it
3039     __ cmpptr(rbx, Address(rbp, wordSize));
3040     __ jccb(Assembler::notEqual, no_adjust);
3041 
3042     // Skip over the poll instruction.
3043     // See NativeInstruction::is_safepoint_poll()
3044     // Possible encodings:
3045     //      85 00       test   %eax,(%rax)
3046     //      85 01       test   %eax,(%rcx)
3047     //      85 02       test   %eax,(%rdx)
3048     //      85 03       test   %eax,(%rbx)
3049     //      85 06       test   %eax,(%rsi)
3050     //      85 07       test   %eax,(%rdi)
3051     //
3052     //      85 04 24    test   %eax,(%rsp)
3053     //      85 45 00    test   %eax,0x0(%rbp)
3054 
3055 #ifdef ASSERT
3056     __ movptr(rax, rbx); // remember where 0x85 should be, for verification below
3057 #endif
3058     // rsp/rbp base encoding takes 3 bytes with the following register values:
3059     // rsp 0x04
3060     // rbp 0x05
3061     __ movzbl(rcx, Address(rbx, 1));
3062     __ andptr(rcx, 0x07); // looking for 0x04 .. 0x05
3063     __ subptr(rcx, 4);    // looking for 0x00 .. 0x01
3064     __ cmpptr(rcx, 1);
3065     __ jcc(Assembler::above, not_special);
3066     __ addptr(rbx, 1);
3067     __ bind(not_special);
3068 #ifdef ASSERT
3069     // Verify the correct encoding of the poll we're about to skip.
3070     __ cmpb(Address(rax, 0), NativeTstRegMem::instruction_code_memXregl);
3071     __ jcc(Assembler::notEqual, bail);
3072     // Mask out the modrm bits
3073     __ testb(Address(rax, 1), NativeTstRegMem::modrm_mask);
3074     // rax encodes to 0, so if the bits are nonzero it's incorrect
3075     __ jcc(Assembler::notZero, bail);
3076 #endif
3077     // Adjust return pc forward to step over the safepoint poll instruction
3078     __ addptr(rbx, 2);
3079     __ movptr(Address(rbp, wordSize), rbx);
3080   }
3081 
3082   __ bind(no_adjust);
3083   // Normal exit, register restoring and exit
3084   RegisterSaver::restore_live_registers(masm, save_vectors);
3085 
3086   __ ret(0);
3087 
3088 #ifdef ASSERT
3089   __ bind(bail);
3090   __ stop("Attempting to adjust pc to skip safepoint poll but the return point is not what we expected");
3091 #endif
3092 
3093   // make sure all code is generated
3094   masm->flush();
3095 
3096   // Fill-out other meta info
3097   return SafepointBlob::create(&buffer, oop_maps, frame_size_in_words);
3098 }
3099 
3100 //
3101 // generate_resolve_blob - call resolution (static/virtual/opt-virtual/ic-miss
3102 //
3103 // Generate a stub that calls into vm to find out the proper destination
3104 // of a java call. All the argument registers are live at this point
3105 // but since this is generic code we don't know what they are and the caller
3106 // must do any gc of the args.
3107 //
3108 RuntimeStub* SharedRuntime::generate_resolve_blob(address destination, const char* name) {
3109   assert (StubRoutines::forward_exception_entry() != NULL, "must be generated before");
3110 
3111   // allocate space for the code
3112   ResourceMark rm;
3113 
3114   CodeBuffer buffer(name, 1000, 512);
3115   MacroAssembler* masm                = new MacroAssembler(&buffer);
3116 
3117   int frame_size_words;
3118   enum frame_layout {
3119                 thread_off,
3120                 extra_words };
3121 
3122   OopMapSet *oop_maps = new OopMapSet();
3123   OopMap* map = NULL;
3124 
3125   int start = __ offset();
3126 
3127   map = RegisterSaver::save_live_registers(masm, extra_words, &frame_size_words);
3128 
3129   int frame_complete = __ offset();
3130 
3131   const Register thread = rdi;
3132   __ get_thread(rdi);
3133 
3134   __ push(thread);
3135   __ set_last_Java_frame(thread, noreg, rbp, NULL);
3136 
3137   __ call(RuntimeAddress(destination));
3138 
3139 
3140   // Set an oopmap for the call site.
3141   // We need this not only for callee-saved registers, but also for volatile
3142   // registers that the compiler might be keeping live across a safepoint.
3143 
3144   oop_maps->add_gc_map( __ offset() - start, map);
3145 
3146   // rax, contains the address we are going to jump to assuming no exception got installed
3147 
3148   __ addptr(rsp, wordSize);
3149 
3150   // clear last_Java_sp
3151   __ reset_last_Java_frame(thread, true);
3152   // check for pending exceptions
3153   Label pending;
3154   __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
3155   __ jcc(Assembler::notEqual, pending);
3156 
3157   // get the returned Method*
3158   __ get_vm_result_2(rbx, thread);
3159   __ movptr(Address(rsp, RegisterSaver::rbx_offset() * wordSize), rbx);
3160 
3161   __ movptr(Address(rsp, RegisterSaver::rax_offset() * wordSize), rax);
3162 
3163   RegisterSaver::restore_live_registers(masm);
3164 
3165   // We are back the the original state on entry and ready to go.
3166 
3167   __ jmp(rax);
3168 
3169   // Pending exception after the safepoint
3170 
3171   __ bind(pending);
3172 
3173   RegisterSaver::restore_live_registers(masm);
3174 
3175   // exception pending => remove activation and forward to exception handler
3176 
3177   __ get_thread(thread);
3178   __ movptr(Address(thread, JavaThread::vm_result_offset()), NULL_WORD);
3179   __ movptr(rax, Address(thread, Thread::pending_exception_offset()));
3180   __ jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
3181 
3182   // -------------
3183   // make sure all code is generated
3184   masm->flush();
3185 
3186   // return the  blob
3187   // frame_size_words or bytes??
3188   return RuntimeStub::new_runtime_stub(name, &buffer, frame_complete, frame_size_words, oop_maps, true);
3189 }