1 /*
   2  * Copyright (c) 2003, 2024, 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/compiledIC.hpp"
  29 #include "code/debugInfoRec.hpp"
  30 #include "code/nativeInst.hpp"
  31 #include "code/vtableStubs.hpp"
  32 #include "compiler/oopMap.hpp"
  33 #include "gc/shared/gcLocker.hpp"
  34 #include "gc/shared/barrierSet.hpp"
  35 #include "gc/shared/barrierSetAssembler.hpp"
  36 #include "interpreter/interpreter.hpp"
  37 #include "logging/log.hpp"
  38 #include "memory/resourceArea.hpp"
  39 #include "oops/klass.inline.hpp"
  40 #include "prims/methodHandles.hpp"
  41 #include "runtime/jniHandles.hpp"
  42 #include "runtime/safepointMechanism.hpp"
  43 #include "runtime/sharedRuntime.hpp"
  44 #include "runtime/signature.hpp"
  45 #include "runtime/stubRoutines.hpp"
  46 #include "runtime/vframeArray.hpp"
  47 #include "runtime/vm_version.hpp"
  48 #include "utilities/align.hpp"
  49 #include "utilities/globalDefinitions.hpp"
  50 #include "vmreg_x86.inline.hpp"
  51 #ifdef COMPILER1
  52 #include "c1/c1_Runtime1.hpp"
  53 #endif
  54 #ifdef COMPILER2
  55 #include "opto/runtime.hpp"
  56 #endif
  57 
  58 #define __ masm->
  59 
  60 const int StackAlignmentInSlots = StackAlignmentInBytes / VMRegImpl::stack_slot_size;
  61 
  62 class RegisterSaver {
  63   // Capture info about frame layout
  64 #define DEF_XMM_OFFS(regnum) xmm ## regnum ## _off = xmm_off + (regnum)*16/BytesPerInt, xmm ## regnum ## H_off
  65   enum layout {
  66                 fpu_state_off = 0,
  67                 fpu_state_end = fpu_state_off+FPUStateSizeInWords,
  68                 st0_off, st0H_off,
  69                 st1_off, st1H_off,
  70                 st2_off, st2H_off,
  71                 st3_off, st3H_off,
  72                 st4_off, st4H_off,
  73                 st5_off, st5H_off,
  74                 st6_off, st6H_off,
  75                 st7_off, st7H_off,
  76                 xmm_off,
  77                 DEF_XMM_OFFS(0),
  78                 DEF_XMM_OFFS(1),
  79                 DEF_XMM_OFFS(2),
  80                 DEF_XMM_OFFS(3),
  81                 DEF_XMM_OFFS(4),
  82                 DEF_XMM_OFFS(5),
  83                 DEF_XMM_OFFS(6),
  84                 DEF_XMM_OFFS(7),
  85                 flags_off = xmm7_off + 16/BytesPerInt + 1, // 16-byte stack alignment fill word
  86                 rdi_off,
  87                 rsi_off,
  88                 ignore_off,  // extra copy of rbp,
  89                 rsp_off,
  90                 rbx_off,
  91                 rdx_off,
  92                 rcx_off,
  93                 rax_off,
  94                 // The frame sender code expects that rbp will be in the "natural" place and
  95                 // will override any oopMap setting for it. We must therefore force the layout
  96                 // so that it agrees with the frame sender code.
  97                 rbp_off,
  98                 return_off,      // slot for return address
  99                 reg_save_size };
 100   enum { FPU_regs_live = flags_off - fpu_state_end };
 101 
 102   public:
 103 
 104   static OopMap* save_live_registers(MacroAssembler* masm, int additional_frame_words,
 105                                      int* total_frame_words, bool verify_fpu = true, bool save_vectors = false);
 106   static void restore_live_registers(MacroAssembler* masm, bool restore_vectors = false);
 107 
 108   static int rax_offset() { return rax_off; }
 109   static int rbx_offset() { return rbx_off; }
 110 
 111   // Offsets into the register save area
 112   // Used by deoptimization when it is managing result register
 113   // values on its own
 114 
 115   static int raxOffset(void) { return rax_off; }
 116   static int rdxOffset(void) { return rdx_off; }
 117   static int rbxOffset(void) { return rbx_off; }
 118   static int xmm0Offset(void) { return xmm0_off; }
 119   // This really returns a slot in the fp save area, which one is not important
 120   static int fpResultOffset(void) { return st0_off; }
 121 
 122   // During deoptimization only the result register need to be restored
 123   // all the other values have already been extracted.
 124 
 125   static void restore_result_registers(MacroAssembler* masm);
 126 
 127 };
 128 
 129 OopMap* RegisterSaver::save_live_registers(MacroAssembler* masm, int additional_frame_words,
 130                                            int* total_frame_words, bool verify_fpu, bool save_vectors) {
 131   int num_xmm_regs = XMMRegister::number_of_registers;
 132   int ymm_bytes = num_xmm_regs * 16;
 133   int zmm_bytes = num_xmm_regs * 32;
 134 #ifdef COMPILER2
 135   int opmask_state_bytes = KRegister::number_of_registers * 8;
 136   if (save_vectors) {
 137     assert(UseAVX > 0, "Vectors larger than 16 byte long are supported only with AVX");
 138     assert(MaxVectorSize <= 64, "Only up to 64 byte long vectors are supported");
 139     // Save upper half of YMM registers
 140     int vect_bytes = ymm_bytes;
 141     if (UseAVX > 2) {
 142       // Save upper half of ZMM registers as well
 143       vect_bytes += zmm_bytes;
 144       additional_frame_words += opmask_state_bytes / wordSize;
 145     }
 146     additional_frame_words += vect_bytes / wordSize;
 147   }
 148 #else
 149   assert(!save_vectors, "vectors are generated only by C2");
 150 #endif
 151   int frame_size_in_bytes = (reg_save_size + additional_frame_words) * wordSize;
 152   int frame_words = frame_size_in_bytes / wordSize;
 153   *total_frame_words = frame_words;
 154 
 155   assert(FPUStateSizeInWords == 27, "update stack layout");
 156 
 157   // save registers, fpu state, and flags
 158   // We assume caller has already has return address slot on the stack
 159   // We push epb twice in this sequence because we want the real rbp,
 160   // to be under the return like a normal enter and we want to use pusha
 161   // We push by hand instead of using push.
 162   __ enter();
 163   __ pusha();
 164   __ pushf();
 165   __ subptr(rsp,FPU_regs_live*wordSize); // Push FPU registers space
 166   __ push_FPU_state();          // Save FPU state & init
 167 
 168   if (verify_fpu) {
 169     // Some stubs may have non standard FPU control word settings so
 170     // only check and reset the value when it required to be the
 171     // standard value.  The safepoint blob in particular can be used
 172     // in methods which are using the 24 bit control word for
 173     // optimized float math.
 174 
 175 #ifdef ASSERT
 176     // Make sure the control word has the expected value
 177     Label ok;
 178     __ cmpw(Address(rsp, 0), StubRoutines::x86::fpu_cntrl_wrd_std());
 179     __ jccb(Assembler::equal, ok);
 180     __ stop("corrupted control word detected");
 181     __ bind(ok);
 182 #endif
 183 
 184     // Reset the control word to guard against exceptions being unmasked
 185     // since fstp_d can cause FPU stack underflow exceptions.  Write it
 186     // into the on stack copy and then reload that to make sure that the
 187     // current and future values are correct.
 188     __ movw(Address(rsp, 0), StubRoutines::x86::fpu_cntrl_wrd_std());
 189   }
 190 
 191   __ frstor(Address(rsp, 0));
 192   if (!verify_fpu) {
 193     // Set the control word so that exceptions are masked for the
 194     // following code.
 195     __ fldcw(ExternalAddress(StubRoutines::x86::addr_fpu_cntrl_wrd_std()));
 196   }
 197 
 198   int off = st0_off;
 199   int delta = st1_off - off;
 200 
 201   // Save the FPU registers in de-opt-able form
 202   for (int n = 0; n < FloatRegister::number_of_registers; n++) {
 203     __ fstp_d(Address(rsp, off*wordSize));
 204     off += delta;
 205   }
 206 
 207   off = xmm0_off;
 208   delta = xmm1_off - off;
 209   if(UseSSE == 1) {
 210     // Save the XMM state
 211     for (int n = 0; n < num_xmm_regs; n++) {
 212       __ movflt(Address(rsp, off*wordSize), as_XMMRegister(n));
 213       off += delta;
 214     }
 215   } else if(UseSSE >= 2) {
 216     // Save whole 128bit (16 bytes) XMM registers
 217     for (int n = 0; n < num_xmm_regs; n++) {
 218       __ movdqu(Address(rsp, off*wordSize), as_XMMRegister(n));
 219       off += delta;
 220     }
 221   }
 222 
 223 #ifdef COMPILER2
 224   if (save_vectors) {
 225     __ subptr(rsp, ymm_bytes);
 226     // Save upper half of YMM registers
 227     for (int n = 0; n < num_xmm_regs; n++) {
 228       __ vextractf128_high(Address(rsp, n*16), as_XMMRegister(n));
 229     }
 230     if (UseAVX > 2) {
 231       __ subptr(rsp, zmm_bytes);
 232       // Save upper half of ZMM registers
 233       for (int n = 0; n < num_xmm_regs; n++) {
 234         __ vextractf64x4_high(Address(rsp, n*32), as_XMMRegister(n));
 235       }
 236       __ subptr(rsp, opmask_state_bytes);
 237       // Save opmask registers
 238       for (int n = 0; n < KRegister::number_of_registers; n++) {
 239         __ kmov(Address(rsp, n*8), as_KRegister(n));
 240       }
 241     }
 242   }
 243 #else
 244   assert(!save_vectors, "vectors are generated only by C2");
 245 #endif
 246 
 247   __ vzeroupper();
 248 
 249   // Set an oopmap for the call site.  This oopmap will map all
 250   // oop-registers and debug-info registers as callee-saved.  This
 251   // will allow deoptimization at this safepoint to find all possible
 252   // debug-info recordings, as well as let GC find all oops.
 253 
 254   OopMapSet *oop_maps = new OopMapSet();
 255   OopMap* map =  new OopMap( frame_words, 0 );
 256 
 257 #define STACK_OFFSET(x) VMRegImpl::stack2reg((x) + additional_frame_words)
 258 #define NEXTREG(x) (x)->as_VMReg()->next()
 259 
 260   map->set_callee_saved(STACK_OFFSET(rax_off), rax->as_VMReg());
 261   map->set_callee_saved(STACK_OFFSET(rcx_off), rcx->as_VMReg());
 262   map->set_callee_saved(STACK_OFFSET(rdx_off), rdx->as_VMReg());
 263   map->set_callee_saved(STACK_OFFSET(rbx_off), rbx->as_VMReg());
 264   // rbp, location is known implicitly, no oopMap
 265   map->set_callee_saved(STACK_OFFSET(rsi_off), rsi->as_VMReg());
 266   map->set_callee_saved(STACK_OFFSET(rdi_off), rdi->as_VMReg());
 267 
 268   // %%% This is really a waste but we'll keep things as they were for now for the upper component
 269   off = st0_off;
 270   delta = st1_off - off;
 271   for (int n = 0; n < FloatRegister::number_of_registers; n++) {
 272     FloatRegister freg_name = as_FloatRegister(n);
 273     map->set_callee_saved(STACK_OFFSET(off), freg_name->as_VMReg());
 274     map->set_callee_saved(STACK_OFFSET(off+1), NEXTREG(freg_name));
 275     off += delta;
 276   }
 277   off = xmm0_off;
 278   delta = xmm1_off - off;
 279   for (int n = 0; n < num_xmm_regs; n++) {
 280     XMMRegister xmm_name = as_XMMRegister(n);
 281     map->set_callee_saved(STACK_OFFSET(off), xmm_name->as_VMReg());
 282     map->set_callee_saved(STACK_OFFSET(off+1), NEXTREG(xmm_name));
 283     off += delta;
 284   }
 285 #undef NEXTREG
 286 #undef STACK_OFFSET
 287 
 288   return map;
 289 }
 290 
 291 void RegisterSaver::restore_live_registers(MacroAssembler* masm, bool restore_vectors) {
 292   int opmask_state_bytes = 0;
 293   int additional_frame_bytes = 0;
 294   int num_xmm_regs = XMMRegister::number_of_registers;
 295   int ymm_bytes = num_xmm_regs * 16;
 296   int zmm_bytes = num_xmm_regs * 32;
 297   // Recover XMM & FPU state
 298 #ifdef COMPILER2
 299   if (restore_vectors) {
 300     assert(UseAVX > 0, "Vectors larger than 16 byte long are supported only with AVX");
 301     assert(MaxVectorSize <= 64, "Only up to 64 byte long vectors are supported");
 302     // Save upper half of YMM registers
 303     additional_frame_bytes = ymm_bytes;
 304     if (UseAVX > 2) {
 305       // Save upper half of ZMM registers as well
 306       additional_frame_bytes += zmm_bytes;
 307       opmask_state_bytes = KRegister::number_of_registers * 8;
 308       additional_frame_bytes += opmask_state_bytes;
 309     }
 310   }
 311 #else
 312   assert(!restore_vectors, "vectors are generated only by C2");
 313 #endif
 314 
 315   int off = xmm0_off;
 316   int delta = xmm1_off - off;
 317 
 318   __ vzeroupper();
 319 
 320   if (UseSSE == 1) {
 321     // Restore XMM registers
 322     assert(additional_frame_bytes == 0, "");
 323     for (int n = 0; n < num_xmm_regs; n++) {
 324       __ movflt(as_XMMRegister(n), Address(rsp, off*wordSize));
 325       off += delta;
 326     }
 327   } else if (UseSSE >= 2) {
 328     // Restore whole 128bit (16 bytes) XMM registers. Do this before restoring YMM and
 329     // ZMM because the movdqu instruction zeros the upper part of the XMM register.
 330     for (int n = 0; n < num_xmm_regs; n++) {
 331       __ movdqu(as_XMMRegister(n), Address(rsp, off*wordSize+additional_frame_bytes));
 332       off += delta;
 333     }
 334   }
 335 
 336   if (restore_vectors) {
 337     off = additional_frame_bytes - ymm_bytes;
 338     // Restore upper half of YMM registers.
 339     for (int n = 0; n < num_xmm_regs; n++) {
 340       __ vinsertf128_high(as_XMMRegister(n), Address(rsp, n*16+off));
 341     }
 342     if (UseAVX > 2) {
 343       // Restore upper half of ZMM registers.
 344       off = opmask_state_bytes;
 345       for (int n = 0; n < num_xmm_regs; n++) {
 346         __ vinsertf64x4_high(as_XMMRegister(n), Address(rsp, n*32+off));
 347       }
 348       for (int n = 0; n < KRegister::number_of_registers; n++) {
 349         __ kmov(as_KRegister(n), Address(rsp, n*8));
 350       }
 351     }
 352     __ addptr(rsp, additional_frame_bytes);
 353   }
 354 
 355   __ pop_FPU_state();
 356   __ addptr(rsp, FPU_regs_live*wordSize); // Pop FPU registers
 357 
 358   __ popf();
 359   __ popa();
 360   // Get the rbp, described implicitly by the frame sender code (no oopMap)
 361   __ pop(rbp);
 362 }
 363 
 364 void RegisterSaver::restore_result_registers(MacroAssembler* masm) {
 365 
 366   // Just restore result register. Only used by deoptimization. By
 367   // now any callee save register that needs to be restore to a c2
 368   // caller of the deoptee has been extracted into the vframeArray
 369   // and will be stuffed into the c2i adapter we create for later
 370   // restoration so only result registers need to be restored here.
 371   //
 372 
 373   __ frstor(Address(rsp, 0));      // Restore fpu state
 374 
 375   // Recover XMM & FPU state
 376   if( UseSSE == 1 ) {
 377     __ movflt(xmm0, Address(rsp, xmm0_off*wordSize));
 378   } else if( UseSSE >= 2 ) {
 379     __ movdbl(xmm0, Address(rsp, xmm0_off*wordSize));
 380   }
 381   __ movptr(rax, Address(rsp, rax_off*wordSize));
 382   __ movptr(rdx, Address(rsp, rdx_off*wordSize));
 383   // Pop all of the register save are off the stack except the return address
 384   __ addptr(rsp, return_off * wordSize);
 385 }
 386 
 387 // Is vector's size (in bytes) bigger than a size saved by default?
 388 // 16 bytes XMM registers are saved by default using SSE2 movdqu instructions.
 389 // Note, MaxVectorSize == 0 with UseSSE < 2 and vectors are not generated.
 390 bool SharedRuntime::is_wide_vector(int size) {
 391   return size > 16;
 392 }
 393 
 394 // The java_calling_convention describes stack locations as ideal slots on
 395 // a frame with no abi restrictions. Since we must observe abi restrictions
 396 // (like the placement of the register window) the slots must be biased by
 397 // the following value.
 398 static int reg2offset_in(VMReg r) {
 399   // Account for saved rbp, and return address
 400   // This should really be in_preserve_stack_slots
 401   return (r->reg2stack() + 2) * VMRegImpl::stack_slot_size;
 402 }
 403 
 404 static int reg2offset_out(VMReg r) {
 405   return (r->reg2stack() + SharedRuntime::out_preserve_stack_slots()) * VMRegImpl::stack_slot_size;
 406 }
 407 
 408 // ---------------------------------------------------------------------------
 409 // Read the array of BasicTypes from a signature, and compute where the
 410 // arguments should go.  Values in the VMRegPair regs array refer to 4-byte
 411 // quantities.  Values less than SharedInfo::stack0 are registers, those above
 412 // refer to 4-byte stack slots.  All stack slots are based off of the stack pointer
 413 // as framesizes are fixed.
 414 // VMRegImpl::stack0 refers to the first slot 0(sp).
 415 // and VMRegImpl::stack0+1 refers to the memory word 4-byes higher.
 416 // Register up to Register::number_of_registers are the 32-bit
 417 // integer registers.
 418 
 419 // Pass first two oop/int args in registers ECX and EDX.
 420 // Pass first two float/double args in registers XMM0 and XMM1.
 421 // Doubles have precedence, so if you pass a mix of floats and doubles
 422 // the doubles will grab the registers before the floats will.
 423 
 424 // Note: the INPUTS in sig_bt are in units of Java argument words, which are
 425 // either 32-bit or 64-bit depending on the build.  The OUTPUTS are in 32-bit
 426 // units regardless of build. Of course for i486 there is no 64 bit build
 427 
 428 
 429 // ---------------------------------------------------------------------------
 430 // The compiled Java calling convention.
 431 // Pass first two oop/int args in registers ECX and EDX.
 432 // Pass first two float/double args in registers XMM0 and XMM1.
 433 // Doubles have precedence, so if you pass a mix of floats and doubles
 434 // the doubles will grab the registers before the floats will.
 435 int SharedRuntime::java_calling_convention(const BasicType *sig_bt,
 436                                            VMRegPair *regs,
 437                                            int total_args_passed) {
 438   uint    stack = 0;          // Starting stack position for args on stack
 439 
 440 
 441   // Pass first two oop/int args in registers ECX and EDX.
 442   uint reg_arg0 = 9999;
 443   uint reg_arg1 = 9999;
 444 
 445   // Pass first two float/double args in registers XMM0 and XMM1.
 446   // Doubles have precedence, so if you pass a mix of floats and doubles
 447   // the doubles will grab the registers before the floats will.
 448   // CNC - TURNED OFF FOR non-SSE.
 449   //       On Intel we have to round all doubles (and most floats) at
 450   //       call sites by storing to the stack in any case.
 451   // UseSSE=0 ==> Don't Use ==> 9999+0
 452   // UseSSE=1 ==> Floats only ==> 9999+1
 453   // UseSSE>=2 ==> Floats or doubles ==> 9999+2
 454   enum { fltarg_dontuse = 9999+0, fltarg_float_only = 9999+1, fltarg_flt_dbl = 9999+2 };
 455   uint fargs = (UseSSE>=2) ? 2 : UseSSE;
 456   uint freg_arg0 = 9999+fargs;
 457   uint freg_arg1 = 9999+fargs;
 458 
 459   // Pass doubles & longs aligned on the stack.  First count stack slots for doubles
 460   int i;
 461   for( i = 0; i < total_args_passed; i++) {
 462     if( sig_bt[i] == T_DOUBLE ) {
 463       // first 2 doubles go in registers
 464       if( freg_arg0 == fltarg_flt_dbl ) freg_arg0 = i;
 465       else if( freg_arg1 == fltarg_flt_dbl ) freg_arg1 = i;
 466       else // Else double is passed low on the stack to be aligned.
 467         stack += 2;
 468     } else if( sig_bt[i] == T_LONG ) {
 469       stack += 2;
 470     }
 471   }
 472   int dstack = 0;             // Separate counter for placing doubles
 473 
 474   // Now pick where all else goes.
 475   for( i = 0; i < total_args_passed; i++) {
 476     // From the type and the argument number (count) compute the location
 477     switch( sig_bt[i] ) {
 478     case T_SHORT:
 479     case T_CHAR:
 480     case T_BYTE:
 481     case T_BOOLEAN:
 482     case T_INT:
 483     case T_ARRAY:
 484     case T_OBJECT:
 485     case T_ADDRESS:
 486       if( reg_arg0 == 9999 )  {
 487         reg_arg0 = i;
 488         regs[i].set1(rcx->as_VMReg());
 489       } else if( reg_arg1 == 9999 )  {
 490         reg_arg1 = i;
 491         regs[i].set1(rdx->as_VMReg());
 492       } else {
 493         regs[i].set1(VMRegImpl::stack2reg(stack++));
 494       }
 495       break;
 496     case T_FLOAT:
 497       if( freg_arg0 == fltarg_flt_dbl || freg_arg0 == fltarg_float_only ) {
 498         freg_arg0 = i;
 499         regs[i].set1(xmm0->as_VMReg());
 500       } else if( freg_arg1 == fltarg_flt_dbl || freg_arg1 == fltarg_float_only ) {
 501         freg_arg1 = i;
 502         regs[i].set1(xmm1->as_VMReg());
 503       } else {
 504         regs[i].set1(VMRegImpl::stack2reg(stack++));
 505       }
 506       break;
 507     case T_LONG:
 508       assert((i + 1) < total_args_passed && sig_bt[i+1] == T_VOID, "missing Half" );
 509       regs[i].set2(VMRegImpl::stack2reg(dstack));
 510       dstack += 2;
 511       break;
 512     case T_DOUBLE:
 513       assert((i + 1) < total_args_passed && sig_bt[i+1] == T_VOID, "missing Half" );
 514       if( freg_arg0 == (uint)i ) {
 515         regs[i].set2(xmm0->as_VMReg());
 516       } else if( freg_arg1 == (uint)i ) {
 517         regs[i].set2(xmm1->as_VMReg());
 518       } else {
 519         regs[i].set2(VMRegImpl::stack2reg(dstack));
 520         dstack += 2;
 521       }
 522       break;
 523     case T_VOID: regs[i].set_bad(); break;
 524       break;
 525     default:
 526       ShouldNotReachHere();
 527       break;
 528     }
 529   }
 530 
 531   return stack;
 532 }
 533 
 534 // Patch the callers callsite with entry to compiled code if it exists.
 535 static void patch_callers_callsite(MacroAssembler *masm) {
 536   Label L;
 537   __ cmpptr(Address(rbx, in_bytes(Method::code_offset())), NULL_WORD);
 538   __ jcc(Assembler::equal, L);
 539   // Schedule the branch target address early.
 540   // Call into the VM to patch the caller, then jump to compiled callee
 541   // rax, isn't live so capture return address while we easily can
 542   __ movptr(rax, Address(rsp, 0));
 543   __ pusha();
 544   __ pushf();
 545 
 546   if (UseSSE == 1) {
 547     __ subptr(rsp, 2*wordSize);
 548     __ movflt(Address(rsp, 0), xmm0);
 549     __ movflt(Address(rsp, wordSize), xmm1);
 550   }
 551   if (UseSSE >= 2) {
 552     __ subptr(rsp, 4*wordSize);
 553     __ movdbl(Address(rsp, 0), xmm0);
 554     __ movdbl(Address(rsp, 2*wordSize), xmm1);
 555   }
 556 #ifdef COMPILER2
 557   // C2 may leave the stack dirty if not in SSE2+ mode
 558   if (UseSSE >= 2) {
 559     __ verify_FPU(0, "c2i transition should have clean FPU stack");
 560   } else {
 561     __ empty_FPU_stack();
 562   }
 563 #endif /* COMPILER2 */
 564 
 565   // VM needs caller's callsite
 566   __ push(rax);
 567   // VM needs target method
 568   __ push(rbx);
 569   __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::fixup_callers_callsite)));
 570   __ addptr(rsp, 2*wordSize);
 571 
 572   if (UseSSE == 1) {
 573     __ movflt(xmm0, Address(rsp, 0));
 574     __ movflt(xmm1, Address(rsp, wordSize));
 575     __ addptr(rsp, 2*wordSize);
 576   }
 577   if (UseSSE >= 2) {
 578     __ movdbl(xmm0, Address(rsp, 0));
 579     __ movdbl(xmm1, Address(rsp, 2*wordSize));
 580     __ addptr(rsp, 4*wordSize);
 581   }
 582 
 583   __ popf();
 584   __ popa();
 585   __ bind(L);
 586 }
 587 
 588 
 589 static void move_c2i_double(MacroAssembler *masm, XMMRegister r, int st_off) {
 590   int next_off = st_off - Interpreter::stackElementSize;
 591   __ movdbl(Address(rsp, next_off), r);
 592 }
 593 
 594 static void gen_c2i_adapter(MacroAssembler *masm,
 595                             int total_args_passed,
 596                             int comp_args_on_stack,
 597                             const BasicType *sig_bt,
 598                             const VMRegPair *regs,
 599                             Label& skip_fixup) {
 600   // Before we get into the guts of the C2I adapter, see if we should be here
 601   // at all.  We've come from compiled code and are attempting to jump to the
 602   // interpreter, which means the caller made a static call to get here
 603   // (vcalls always get a compiled target if there is one).  Check for a
 604   // compiled target.  If there is one, we need to patch the caller's call.
 605   patch_callers_callsite(masm);
 606 
 607   __ bind(skip_fixup);
 608 
 609 #ifdef COMPILER2
 610   // C2 may leave the stack dirty if not in SSE2+ mode
 611   if (UseSSE >= 2) {
 612     __ verify_FPU(0, "c2i transition should have clean FPU stack");
 613   } else {
 614     __ empty_FPU_stack();
 615   }
 616 #endif /* COMPILER2 */
 617 
 618   // Since all args are passed on the stack, total_args_passed * interpreter_
 619   // stack_element_size  is the
 620   // space we need.
 621   int extraspace = total_args_passed * Interpreter::stackElementSize;
 622 
 623   // Get return address
 624   __ pop(rax);
 625 
 626   // set senderSP value
 627   __ movptr(rsi, rsp);
 628 
 629   __ subptr(rsp, extraspace);
 630 
 631   // Now write the args into the outgoing interpreter space
 632   for (int i = 0; i < total_args_passed; i++) {
 633     if (sig_bt[i] == T_VOID) {
 634       assert(i > 0 && (sig_bt[i-1] == T_LONG || sig_bt[i-1] == T_DOUBLE), "missing half");
 635       continue;
 636     }
 637 
 638     // st_off points to lowest address on stack.
 639     int st_off = ((total_args_passed - 1) - i) * Interpreter::stackElementSize;
 640     int next_off = st_off - Interpreter::stackElementSize;
 641 
 642     // Say 4 args:
 643     // i   st_off
 644     // 0   12 T_LONG
 645     // 1    8 T_VOID
 646     // 2    4 T_OBJECT
 647     // 3    0 T_BOOL
 648     VMReg r_1 = regs[i].first();
 649     VMReg r_2 = regs[i].second();
 650     if (!r_1->is_valid()) {
 651       assert(!r_2->is_valid(), "");
 652       continue;
 653     }
 654 
 655     if (r_1->is_stack()) {
 656       // memory to memory use fpu stack top
 657       int ld_off = r_1->reg2stack() * VMRegImpl::stack_slot_size + extraspace;
 658 
 659       if (!r_2->is_valid()) {
 660         __ movl(rdi, Address(rsp, ld_off));
 661         __ movptr(Address(rsp, st_off), rdi);
 662       } else {
 663 
 664         // ld_off == LSW, ld_off+VMRegImpl::stack_slot_size == MSW
 665         // st_off == MSW, st_off-wordSize == LSW
 666 
 667         __ movptr(rdi, Address(rsp, ld_off));
 668         __ movptr(Address(rsp, next_off), rdi);
 669         __ movptr(rdi, Address(rsp, ld_off + wordSize));
 670         __ movptr(Address(rsp, st_off), rdi);
 671       }
 672     } else if (r_1->is_Register()) {
 673       Register r = r_1->as_Register();
 674       if (!r_2->is_valid()) {
 675         __ movl(Address(rsp, st_off), r);
 676       } else {
 677         // long/double in gpr
 678         ShouldNotReachHere();
 679       }
 680     } else {
 681       assert(r_1->is_XMMRegister(), "");
 682       if (!r_2->is_valid()) {
 683         __ movflt(Address(rsp, st_off), r_1->as_XMMRegister());
 684       } else {
 685         assert(sig_bt[i] == T_DOUBLE || sig_bt[i] == T_LONG, "wrong type");
 686         move_c2i_double(masm, r_1->as_XMMRegister(), st_off);
 687       }
 688     }
 689   }
 690 
 691   // Schedule the branch target address early.
 692   __ movptr(rcx, Address(rbx, in_bytes(Method::interpreter_entry_offset())));
 693   // And repush original return address
 694   __ push(rax);
 695   __ jmp(rcx);
 696 }
 697 
 698 
 699 static void move_i2c_double(MacroAssembler *masm, XMMRegister r, Register saved_sp, int ld_off) {
 700   int next_val_off = ld_off - Interpreter::stackElementSize;
 701   __ movdbl(r, Address(saved_sp, next_val_off));
 702 }
 703 
 704 static void range_check(MacroAssembler* masm, Register pc_reg, Register temp_reg,
 705                         address code_start, address code_end,
 706                         Label& L_ok) {
 707   Label L_fail;
 708   __ lea(temp_reg, ExternalAddress(code_start));
 709   __ cmpptr(pc_reg, temp_reg);
 710   __ jcc(Assembler::belowEqual, L_fail);
 711   __ lea(temp_reg, ExternalAddress(code_end));
 712   __ cmpptr(pc_reg, temp_reg);
 713   __ jcc(Assembler::below, L_ok);
 714   __ bind(L_fail);
 715 }
 716 
 717 void SharedRuntime::gen_i2c_adapter(MacroAssembler *masm,
 718                                     int total_args_passed,
 719                                     int comp_args_on_stack,
 720                                     const BasicType *sig_bt,
 721                                     const VMRegPair *regs) {
 722   // Note: rsi contains the senderSP on entry. We must preserve it since
 723   // we may do a i2c -> c2i transition if we lose a race where compiled
 724   // code goes non-entrant while we get args ready.
 725 
 726   // Adapters can be frameless because they do not require the caller
 727   // to perform additional cleanup work, such as correcting the stack pointer.
 728   // An i2c adapter is frameless because the *caller* frame, which is interpreted,
 729   // routinely repairs its own stack pointer (from interpreter_frame_last_sp),
 730   // even if a callee has modified the stack pointer.
 731   // A c2i adapter is frameless because the *callee* frame, which is interpreted,
 732   // routinely repairs its caller's stack pointer (from sender_sp, which is set
 733   // up via the senderSP register).
 734   // In other words, if *either* the caller or callee is interpreted, we can
 735   // get the stack pointer repaired after a call.
 736   // This is why c2i and i2c adapters cannot be indefinitely composed.
 737   // In particular, if a c2i adapter were to somehow call an i2c adapter,
 738   // both caller and callee would be compiled methods, and neither would
 739   // clean up the stack pointer changes performed by the two adapters.
 740   // If this happens, control eventually transfers back to the compiled
 741   // caller, but with an uncorrected stack, causing delayed havoc.
 742 
 743   // Pick up the return address
 744   __ movptr(rax, Address(rsp, 0));
 745 
 746   if (VerifyAdapterCalls &&
 747       (Interpreter::code() != nullptr || StubRoutines::final_stubs_code() != nullptr)) {
 748     // So, let's test for cascading c2i/i2c adapters right now.
 749     //  assert(Interpreter::contains($return_addr) ||
 750     //         StubRoutines::contains($return_addr),
 751     //         "i2c adapter must return to an interpreter frame");
 752     __ block_comment("verify_i2c { ");
 753     Label L_ok;
 754     if (Interpreter::code() != nullptr) {
 755       range_check(masm, rax, rdi,
 756                   Interpreter::code()->code_start(), Interpreter::code()->code_end(),
 757                   L_ok);
 758     }
 759     if (StubRoutines::initial_stubs_code() != nullptr) {
 760       range_check(masm, rax, rdi,
 761                   StubRoutines::initial_stubs_code()->code_begin(),
 762                   StubRoutines::initial_stubs_code()->code_end(),
 763                   L_ok);
 764     }
 765     if (StubRoutines::final_stubs_code() != nullptr) {
 766       range_check(masm, rax, rdi,
 767                   StubRoutines::final_stubs_code()->code_begin(),
 768                   StubRoutines::final_stubs_code()->code_end(),
 769                   L_ok);
 770     }
 771     const char* msg = "i2c adapter must return to an interpreter frame";
 772     __ block_comment(msg);
 773     __ stop(msg);
 774     __ bind(L_ok);
 775     __ block_comment("} verify_i2ce ");
 776   }
 777 
 778   // Must preserve original SP for loading incoming arguments because
 779   // we need to align the outgoing SP for compiled code.
 780   __ movptr(rdi, rsp);
 781 
 782   // Cut-out for having no stack args.  Since up to 2 int/oop args are passed
 783   // in registers, we will occasionally have no stack args.
 784   int comp_words_on_stack = 0;
 785   if (comp_args_on_stack) {
 786     // Sig words on the stack are greater-than VMRegImpl::stack0.  Those in
 787     // registers are below.  By subtracting stack0, we either get a negative
 788     // number (all values in registers) or the maximum stack slot accessed.
 789     // int comp_args_on_stack = VMRegImpl::reg2stack(max_arg);
 790     // Convert 4-byte stack slots to words.
 791     comp_words_on_stack = align_up(comp_args_on_stack*4, wordSize)>>LogBytesPerWord;
 792     // Round up to miminum stack alignment, in wordSize
 793     comp_words_on_stack = align_up(comp_words_on_stack, 2);
 794     __ subptr(rsp, comp_words_on_stack * wordSize);
 795   }
 796 
 797   // Align the outgoing SP
 798   __ andptr(rsp, -(StackAlignmentInBytes));
 799 
 800   // push the return address on the stack (note that pushing, rather
 801   // than storing it, yields the correct frame alignment for the callee)
 802   __ push(rax);
 803 
 804   // Put saved SP in another register
 805   const Register saved_sp = rax;
 806   __ movptr(saved_sp, rdi);
 807 
 808 
 809   // Will jump to the compiled code just as if compiled code was doing it.
 810   // Pre-load the register-jump target early, to schedule it better.
 811   __ movptr(rdi, Address(rbx, in_bytes(Method::from_compiled_offset())));
 812 
 813   // Now generate the shuffle code.  Pick up all register args and move the
 814   // rest through the floating point stack top.
 815   for (int i = 0; i < total_args_passed; i++) {
 816     if (sig_bt[i] == T_VOID) {
 817       // Longs and doubles are passed in native word order, but misaligned
 818       // in the 32-bit build.
 819       assert(i > 0 && (sig_bt[i-1] == T_LONG || sig_bt[i-1] == T_DOUBLE), "missing half");
 820       continue;
 821     }
 822 
 823     // Pick up 0, 1 or 2 words from SP+offset.
 824 
 825     assert(!regs[i].second()->is_valid() || regs[i].first()->next() == regs[i].second(),
 826             "scrambled load targets?");
 827     // Load in argument order going down.
 828     int ld_off = (total_args_passed - i) * Interpreter::stackElementSize;
 829     // Point to interpreter value (vs. tag)
 830     int next_off = ld_off - Interpreter::stackElementSize;
 831     //
 832     //
 833     //
 834     VMReg r_1 = regs[i].first();
 835     VMReg r_2 = regs[i].second();
 836     if (!r_1->is_valid()) {
 837       assert(!r_2->is_valid(), "");
 838       continue;
 839     }
 840     if (r_1->is_stack()) {
 841       // Convert stack slot to an SP offset (+ wordSize to account for return address )
 842       int st_off = regs[i].first()->reg2stack()*VMRegImpl::stack_slot_size + wordSize;
 843 
 844       // We can use rsi as a temp here because compiled code doesn't need rsi as an input
 845       // and if we end up going thru a c2i because of a miss a reasonable value of rsi
 846       // we be generated.
 847       if (!r_2->is_valid()) {
 848         // __ fld_s(Address(saved_sp, ld_off));
 849         // __ fstp_s(Address(rsp, st_off));
 850         __ movl(rsi, Address(saved_sp, ld_off));
 851         __ movptr(Address(rsp, st_off), rsi);
 852       } else {
 853         // Interpreter local[n] == MSW, local[n+1] == LSW however locals
 854         // are accessed as negative so LSW is at LOW address
 855 
 856         // ld_off is MSW so get LSW
 857         // st_off is LSW (i.e. reg.first())
 858         // __ fld_d(Address(saved_sp, next_off));
 859         // __ fstp_d(Address(rsp, st_off));
 860         //
 861         // We are using two VMRegs. This can be either T_OBJECT, T_ADDRESS, T_LONG, or T_DOUBLE
 862         // the interpreter allocates two slots but only uses one for thr T_LONG or T_DOUBLE case
 863         // So we must adjust where to pick up the data to match the interpreter.
 864         //
 865         // Interpreter local[n] == MSW, local[n+1] == LSW however locals
 866         // are accessed as negative so LSW is at LOW address
 867 
 868         // ld_off is MSW so get LSW
 869         __ movptr(rsi, Address(saved_sp, next_off));
 870         __ movptr(Address(rsp, st_off), rsi);
 871         __ movptr(rsi, Address(saved_sp, ld_off));
 872         __ movptr(Address(rsp, st_off + wordSize), rsi);
 873       }
 874     } else if (r_1->is_Register()) {  // Register argument
 875       Register r = r_1->as_Register();
 876       assert(r != rax, "must be different");
 877       if (r_2->is_valid()) {
 878         //
 879         // We are using two VMRegs. This can be either T_OBJECT, T_ADDRESS, T_LONG, or T_DOUBLE
 880         // the interpreter allocates two slots but only uses one for thr T_LONG or T_DOUBLE case
 881         // So we must adjust where to pick up the data to match the interpreter.
 882 
 883         // this can be a misaligned move
 884         __ movptr(r, Address(saved_sp, next_off));
 885         assert(r_2->as_Register() != rax, "need another temporary register");
 886         // Remember r_1 is low address (and LSB on x86)
 887         // So r_2 gets loaded from high address regardless of the platform
 888         __ movptr(r_2->as_Register(), Address(saved_sp, ld_off));
 889       } else {
 890         __ movl(r, Address(saved_sp, ld_off));
 891       }
 892     } else {
 893       assert(r_1->is_XMMRegister(), "");
 894       if (!r_2->is_valid()) {
 895         __ movflt(r_1->as_XMMRegister(), Address(saved_sp, ld_off));
 896       } else {
 897         move_i2c_double(masm, r_1->as_XMMRegister(), saved_sp, ld_off);
 898       }
 899     }
 900   }
 901 
 902   // 6243940 We might end up in handle_wrong_method if
 903   // the callee is deoptimized as we race thru here. If that
 904   // happens we don't want to take a safepoint because the
 905   // caller frame will look interpreted and arguments are now
 906   // "compiled" so it is much better to make this transition
 907   // invisible to the stack walking code. Unfortunately if
 908   // we try and find the callee by normal means a safepoint
 909   // is possible. So we stash the desired callee in the thread
 910   // and the vm will find there should this case occur.
 911 
 912   __ get_thread(rax);
 913   __ movptr(Address(rax, JavaThread::callee_target_offset()), rbx);
 914 
 915   // move Method* to rax, in case we end up in an c2i adapter.
 916   // the c2i adapters expect Method* in rax, (c2) because c2's
 917   // resolve stubs return the result (the method) in rax,.
 918   // I'd love to fix this.
 919   __ mov(rax, rbx);
 920 
 921   __ jmp(rdi);
 922 }
 923 
 924 // ---------------------------------------------------------------
 925 AdapterHandlerEntry* SharedRuntime::generate_i2c2i_adapters(MacroAssembler *masm,
 926                                                             int total_args_passed,
 927                                                             int comp_args_on_stack,
 928                                                             const BasicType *sig_bt,
 929                                                             const VMRegPair *regs,
 930                                                             AdapterFingerPrint* fingerprint) {
 931   address i2c_entry = __ pc();
 932 
 933   gen_i2c_adapter(masm, total_args_passed, comp_args_on_stack, sig_bt, regs);
 934 
 935   // -------------------------------------------------------------------------
 936   // Generate a C2I adapter.  On entry we know rbx, holds the Method* during calls
 937   // to the interpreter.  The args start out packed in the compiled layout.  They
 938   // need to be unpacked into the interpreter layout.  This will almost always
 939   // require some stack space.  We grow the current (compiled) stack, then repack
 940   // the args.  We  finally end in a jump to the generic interpreter entry point.
 941   // On exit from the interpreter, the interpreter will restore our SP (lest the
 942   // compiled code, which relies solely on SP and not EBP, get sick).
 943 
 944   address c2i_unverified_entry = __ pc();
 945   Label skip_fixup;
 946 
 947   Register data = rax;
 948   Register receiver = rcx;
 949   Register temp = rbx;
 950 
 951   {
 952     __ ic_check(1 /* end_alignment */);
 953     __ movptr(rbx, Address(data, CompiledICData::speculated_method_offset()));
 954     // Method might have been compiled since the call site was patched to
 955     // interpreted if that is the case treat it as a miss so we can get
 956     // the call site corrected.
 957     __ cmpptr(Address(rbx, in_bytes(Method::code_offset())), NULL_WORD);
 958     __ jcc(Assembler::equal, skip_fixup);
 959   }
 960 
 961   address c2i_entry = __ pc();
 962 
 963   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
 964   bs->c2i_entry_barrier(masm);
 965 
 966   gen_c2i_adapter(masm, total_args_passed, comp_args_on_stack, sig_bt, regs, skip_fixup);
 967 
 968   return AdapterHandlerLibrary::new_entry(fingerprint, i2c_entry, c2i_entry, c2i_unverified_entry);
 969 }
 970 
 971 int SharedRuntime::c_calling_convention(const BasicType *sig_bt,
 972                                          VMRegPair *regs,
 973                                          int total_args_passed) {
 974 
 975 // We return the amount of VMRegImpl stack slots we need to reserve for all
 976 // the arguments NOT counting out_preserve_stack_slots.
 977 
 978   uint    stack = 0;        // All arguments on stack
 979 
 980   for( int i = 0; i < total_args_passed; i++) {
 981     // From the type and the argument number (count) compute the location
 982     switch( sig_bt[i] ) {
 983     case T_BOOLEAN:
 984     case T_CHAR:
 985     case T_FLOAT:
 986     case T_BYTE:
 987     case T_SHORT:
 988     case T_INT:
 989     case T_OBJECT:
 990     case T_ARRAY:
 991     case T_ADDRESS:
 992     case T_METADATA:
 993       regs[i].set1(VMRegImpl::stack2reg(stack++));
 994       break;
 995     case T_LONG:
 996     case T_DOUBLE: // The stack numbering is reversed from Java
 997       // Since C arguments do not get reversed, the ordering for
 998       // doubles on the stack must be opposite the Java convention
 999       assert((i + 1) < total_args_passed && sig_bt[i+1] == T_VOID, "missing Half" );
1000       regs[i].set2(VMRegImpl::stack2reg(stack));
1001       stack += 2;
1002       break;
1003     case T_VOID: regs[i].set_bad(); break;
1004     default:
1005       ShouldNotReachHere();
1006       break;
1007     }
1008   }
1009   return stack;
1010 }
1011 
1012 int SharedRuntime::vector_calling_convention(VMRegPair *regs,
1013                                              uint num_bits,
1014                                              uint total_args_passed) {
1015   Unimplemented();
1016   return 0;
1017 }
1018 
1019 // A simple move of integer like type
1020 static void simple_move32(MacroAssembler* masm, VMRegPair src, VMRegPair dst) {
1021   if (src.first()->is_stack()) {
1022     if (dst.first()->is_stack()) {
1023       // stack to stack
1024       // __ ld(FP, reg2offset(src.first()), L5);
1025       // __ st(L5, SP, reg2offset(dst.first()));
1026       __ movl2ptr(rax, Address(rbp, reg2offset_in(src.first())));
1027       __ movptr(Address(rsp, reg2offset_out(dst.first())), rax);
1028     } else {
1029       // stack to reg
1030       __ movl2ptr(dst.first()->as_Register(),  Address(rbp, reg2offset_in(src.first())));
1031     }
1032   } else if (dst.first()->is_stack()) {
1033     // reg to stack
1034     // no need to sign extend on 64bit
1035     __ movptr(Address(rsp, reg2offset_out(dst.first())), src.first()->as_Register());
1036   } else {
1037     if (dst.first() != src.first()) {
1038       __ mov(dst.first()->as_Register(), src.first()->as_Register());
1039     }
1040   }
1041 }
1042 
1043 // An oop arg. Must pass a handle not the oop itself
1044 static void object_move(MacroAssembler* masm,
1045                         OopMap* map,
1046                         int oop_handle_offset,
1047                         int framesize_in_slots,
1048                         VMRegPair src,
1049                         VMRegPair dst,
1050                         bool is_receiver,
1051                         int* receiver_offset) {
1052 
1053   // Because of the calling conventions we know that src can be a
1054   // register or a stack location. dst can only be a stack location.
1055 
1056   assert(dst.first()->is_stack(), "must be stack");
1057   // must pass a handle. First figure out the location we use as a handle
1058 
1059   if (src.first()->is_stack()) {
1060     // Oop is already on the stack as an argument
1061     Register rHandle = rax;
1062     Label nil;
1063     __ xorptr(rHandle, rHandle);
1064     __ cmpptr(Address(rbp, reg2offset_in(src.first())), NULL_WORD);
1065     __ jcc(Assembler::equal, nil);
1066     __ lea(rHandle, Address(rbp, reg2offset_in(src.first())));
1067     __ bind(nil);
1068     __ movptr(Address(rsp, reg2offset_out(dst.first())), rHandle);
1069 
1070     int offset_in_older_frame = src.first()->reg2stack() + SharedRuntime::out_preserve_stack_slots();
1071     map->set_oop(VMRegImpl::stack2reg(offset_in_older_frame + framesize_in_slots));
1072     if (is_receiver) {
1073       *receiver_offset = (offset_in_older_frame + framesize_in_slots) * VMRegImpl::stack_slot_size;
1074     }
1075   } else {
1076     // Oop is in a register we must store it to the space we reserve
1077     // on the stack for oop_handles
1078     const Register rOop = src.first()->as_Register();
1079     const Register rHandle = rax;
1080     int oop_slot = (rOop == rcx ? 0 : 1) * VMRegImpl::slots_per_word + oop_handle_offset;
1081     int offset = oop_slot*VMRegImpl::stack_slot_size;
1082     Label skip;
1083     __ movptr(Address(rsp, offset), rOop);
1084     map->set_oop(VMRegImpl::stack2reg(oop_slot));
1085     __ xorptr(rHandle, rHandle);
1086     __ cmpptr(rOop, NULL_WORD);
1087     __ jcc(Assembler::equal, skip);
1088     __ lea(rHandle, Address(rsp, offset));
1089     __ bind(skip);
1090     // Store the handle parameter
1091     __ movptr(Address(rsp, reg2offset_out(dst.first())), rHandle);
1092     if (is_receiver) {
1093       *receiver_offset = offset;
1094     }
1095   }
1096 }
1097 
1098 // A float arg may have to do float reg int reg conversion
1099 static void float_move(MacroAssembler* masm, VMRegPair src, VMRegPair dst) {
1100   assert(!src.second()->is_valid() && !dst.second()->is_valid(), "bad float_move");
1101 
1102   // Because of the calling convention we know that src is either a stack location
1103   // or an xmm register. dst can only be a stack location.
1104 
1105   assert(dst.first()->is_stack() && ( src.first()->is_stack() || src.first()->is_XMMRegister()), "bad parameters");
1106 
1107   if (src.first()->is_stack()) {
1108     __ movl(rax, Address(rbp, reg2offset_in(src.first())));
1109     __ movptr(Address(rsp, reg2offset_out(dst.first())), rax);
1110   } else {
1111     // reg to stack
1112     __ movflt(Address(rsp, reg2offset_out(dst.first())), src.first()->as_XMMRegister());
1113   }
1114 }
1115 
1116 // A long move
1117 static void long_move(MacroAssembler* masm, VMRegPair src, VMRegPair dst) {
1118 
1119   // The only legal possibility for a long_move VMRegPair is:
1120   // 1: two stack slots (possibly unaligned)
1121   // as neither the java  or C calling convention will use registers
1122   // for longs.
1123 
1124   if (src.first()->is_stack() && dst.first()->is_stack()) {
1125     assert(src.second()->is_stack() && dst.second()->is_stack(), "must be all stack");
1126     __ movptr(rax, Address(rbp, reg2offset_in(src.first())));
1127     __ movptr(rbx, Address(rbp, reg2offset_in(src.second())));
1128     __ movptr(Address(rsp, reg2offset_out(dst.first())), rax);
1129     __ movptr(Address(rsp, reg2offset_out(dst.second())), rbx);
1130   } else {
1131     ShouldNotReachHere();
1132   }
1133 }
1134 
1135 // A double move
1136 static void double_move(MacroAssembler* masm, VMRegPair src, VMRegPair dst) {
1137 
1138   // The only legal possibilities for a double_move VMRegPair are:
1139   // The painful thing here is that like long_move a VMRegPair might be
1140 
1141   // Because of the calling convention we know that src is either
1142   //   1: a single physical register (xmm registers only)
1143   //   2: two stack slots (possibly unaligned)
1144   // dst can only be a pair of stack slots.
1145 
1146   assert(dst.first()->is_stack() && (src.first()->is_XMMRegister() || src.first()->is_stack()), "bad args");
1147 
1148   if (src.first()->is_stack()) {
1149     // source is all stack
1150     __ movptr(rax, Address(rbp, reg2offset_in(src.first())));
1151     __ movptr(rbx, Address(rbp, reg2offset_in(src.second())));
1152     __ movptr(Address(rsp, reg2offset_out(dst.first())), rax);
1153     __ movptr(Address(rsp, reg2offset_out(dst.second())), rbx);
1154   } else {
1155     // reg to stack
1156     // No worries about stack alignment
1157     __ movdbl(Address(rsp, reg2offset_out(dst.first())), src.first()->as_XMMRegister());
1158   }
1159 }
1160 
1161 
1162 void SharedRuntime::save_native_result(MacroAssembler *masm, BasicType ret_type, int frame_slots) {
1163   // We always ignore the frame_slots arg and just use the space just below frame pointer
1164   // which by this time is free to use
1165   switch (ret_type) {
1166   case T_FLOAT:
1167     __ fstp_s(Address(rbp, -wordSize));
1168     break;
1169   case T_DOUBLE:
1170     __ fstp_d(Address(rbp, -2*wordSize));
1171     break;
1172   case T_VOID:  break;
1173   case T_LONG:
1174     __ movptr(Address(rbp, -wordSize), rax);
1175     __ movptr(Address(rbp, -2*wordSize), rdx);
1176     break;
1177   default: {
1178     __ movptr(Address(rbp, -wordSize), rax);
1179     }
1180   }
1181 }
1182 
1183 void SharedRuntime::restore_native_result(MacroAssembler *masm, BasicType ret_type, int frame_slots) {
1184   // We always ignore the frame_slots arg and just use the space just below frame pointer
1185   // which by this time is free to use
1186   switch (ret_type) {
1187   case T_FLOAT:
1188     __ fld_s(Address(rbp, -wordSize));
1189     break;
1190   case T_DOUBLE:
1191     __ fld_d(Address(rbp, -2*wordSize));
1192     break;
1193   case T_LONG:
1194     __ movptr(rax, Address(rbp, -wordSize));
1195     __ movptr(rdx, Address(rbp, -2*wordSize));
1196     break;
1197   case T_VOID:  break;
1198   default: {
1199     __ movptr(rax, Address(rbp, -wordSize));
1200     }
1201   }
1202 }
1203 
1204 static void verify_oop_args(MacroAssembler* masm,
1205                             const methodHandle& method,
1206                             const BasicType* sig_bt,
1207                             const VMRegPair* regs) {
1208   Register temp_reg = rbx;  // not part of any compiled calling seq
1209   if (VerifyOops) {
1210     for (int i = 0; i < method->size_of_parameters(); i++) {
1211       if (is_reference_type(sig_bt[i])) {
1212         VMReg r = regs[i].first();
1213         assert(r->is_valid(), "bad oop arg");
1214         if (r->is_stack()) {
1215           __ movptr(temp_reg, Address(rsp, r->reg2stack() * VMRegImpl::stack_slot_size + wordSize));
1216           __ verify_oop(temp_reg);
1217         } else {
1218           __ verify_oop(r->as_Register());
1219         }
1220       }
1221     }
1222   }
1223 }
1224 
1225 static void gen_special_dispatch(MacroAssembler* masm,
1226                                  const methodHandle& method,
1227                                  const BasicType* sig_bt,
1228                                  const VMRegPair* regs) {
1229   verify_oop_args(masm, method, sig_bt, regs);
1230   vmIntrinsics::ID iid = method->intrinsic_id();
1231 
1232   // Now write the args into the outgoing interpreter space
1233   bool     has_receiver   = false;
1234   Register receiver_reg   = noreg;
1235   int      member_arg_pos = -1;
1236   Register member_reg     = noreg;
1237   int      ref_kind       = MethodHandles::signature_polymorphic_intrinsic_ref_kind(iid);
1238   if (ref_kind != 0) {
1239     member_arg_pos = method->size_of_parameters() - 1;  // trailing MemberName argument
1240     member_reg = rbx;  // known to be free at this point
1241     has_receiver = MethodHandles::ref_kind_has_receiver(ref_kind);
1242   } else if (iid == vmIntrinsics::_invokeBasic) {
1243     has_receiver = true;
1244   } else {
1245     fatal("unexpected intrinsic id %d", vmIntrinsics::as_int(iid));
1246   }
1247 
1248   if (member_reg != noreg) {
1249     // Load the member_arg into register, if necessary.
1250     SharedRuntime::check_member_name_argument_is_last_argument(method, sig_bt, regs);
1251     VMReg r = regs[member_arg_pos].first();
1252     if (r->is_stack()) {
1253       __ movptr(member_reg, Address(rsp, r->reg2stack() * VMRegImpl::stack_slot_size + wordSize));
1254     } else {
1255       // no data motion is needed
1256       member_reg = r->as_Register();
1257     }
1258   }
1259 
1260   if (has_receiver) {
1261     // Make sure the receiver is loaded into a register.
1262     assert(method->size_of_parameters() > 0, "oob");
1263     assert(sig_bt[0] == T_OBJECT, "receiver argument must be an object");
1264     VMReg r = regs[0].first();
1265     assert(r->is_valid(), "bad receiver arg");
1266     if (r->is_stack()) {
1267       // Porting note:  This assumes that compiled calling conventions always
1268       // pass the receiver oop in a register.  If this is not true on some
1269       // platform, pick a temp and load the receiver from stack.
1270       fatal("receiver always in a register");
1271       receiver_reg = rcx;  // known to be free at this point
1272       __ movptr(receiver_reg, Address(rsp, r->reg2stack() * VMRegImpl::stack_slot_size + wordSize));
1273     } else {
1274       // no data motion is needed
1275       receiver_reg = r->as_Register();
1276     }
1277   }
1278 
1279   // Figure out which address we are really jumping to:
1280   MethodHandles::generate_method_handle_dispatch(masm, iid,
1281                                                  receiver_reg, member_reg, /*for_compiler_entry:*/ true);
1282 }
1283 
1284 // ---------------------------------------------------------------------------
1285 // Generate a native wrapper for a given method.  The method takes arguments
1286 // in the Java compiled code convention, marshals them to the native
1287 // convention (handlizes oops, etc), transitions to native, makes the call,
1288 // returns to java state (possibly blocking), unhandlizes any result and
1289 // returns.
1290 //
1291 // Critical native functions are a shorthand for the use of
1292 // GetPrimtiveArrayCritical and disallow the use of any other JNI
1293 // functions.  The wrapper is expected to unpack the arguments before
1294 // passing them to the callee. Critical native functions leave the state _in_Java,
1295 // since they cannot stop for GC.
1296 // Some other parts of JNI setup are skipped like the tear down of the JNI handle
1297 // block and the check for pending exceptions it's impossible for them
1298 // to be thrown.
1299 //
1300 //
1301 nmethod* SharedRuntime::generate_native_wrapper(MacroAssembler* masm,
1302                                                 const methodHandle& method,
1303                                                 int compile_id,
1304                                                 BasicType* in_sig_bt,
1305                                                 VMRegPair* in_regs,
1306                                                 BasicType ret_type) {
1307   if (method->is_method_handle_intrinsic()) {
1308     vmIntrinsics::ID iid = method->intrinsic_id();
1309     intptr_t start = (intptr_t)__ pc();
1310     int vep_offset = ((intptr_t)__ pc()) - start;
1311     gen_special_dispatch(masm,
1312                          method,
1313                          in_sig_bt,
1314                          in_regs);
1315     int frame_complete = ((intptr_t)__ pc()) - start;  // not complete, period
1316     __ flush();
1317     int stack_slots = SharedRuntime::out_preserve_stack_slots();  // no out slots at all, actually
1318     return nmethod::new_native_nmethod(method,
1319                                        compile_id,
1320                                        masm->code(),
1321                                        vep_offset,
1322                                        frame_complete,
1323                                        stack_slots / VMRegImpl::slots_per_word,
1324                                        in_ByteSize(-1),
1325                                        in_ByteSize(-1),
1326                                        (OopMapSet*)nullptr);
1327   }
1328   address native_func = method->native_function();
1329   assert(native_func != nullptr, "must have function");
1330 
1331   // An OopMap for lock (and class if static)
1332   OopMapSet *oop_maps = new OopMapSet();
1333 
1334   // We have received a description of where all the java arg are located
1335   // on entry to the wrapper. We need to convert these args to where
1336   // the jni function will expect them. To figure out where they go
1337   // we convert the java signature to a C signature by inserting
1338   // the hidden arguments as arg[0] and possibly arg[1] (static method)
1339 
1340   const int total_in_args = method->size_of_parameters();
1341   int  total_c_args       = total_in_args + (method->is_static() ? 2 : 1);
1342 
1343   BasicType* out_sig_bt = NEW_RESOURCE_ARRAY(BasicType, total_c_args);
1344   VMRegPair* out_regs   = NEW_RESOURCE_ARRAY(VMRegPair, total_c_args);
1345   BasicType* in_elem_bt = nullptr;
1346 
1347   int argc = 0;
1348   out_sig_bt[argc++] = T_ADDRESS;
1349   if (method->is_static()) {
1350     out_sig_bt[argc++] = T_OBJECT;
1351   }
1352 
1353   for (int i = 0; i < total_in_args ; i++ ) {
1354     out_sig_bt[argc++] = in_sig_bt[i];
1355   }
1356 
1357   // Now figure out where the args must be stored and how much stack space
1358   // they require.
1359   int out_arg_slots;
1360   out_arg_slots = c_calling_convention(out_sig_bt, out_regs, total_c_args);
1361 
1362   // Compute framesize for the wrapper.  We need to handlize all oops in
1363   // registers a max of 2 on x86.
1364 
1365   // Calculate the total number of stack slots we will need.
1366 
1367   // First count the abi requirement plus all of the outgoing args
1368   int stack_slots = SharedRuntime::out_preserve_stack_slots() + out_arg_slots;
1369 
1370   // Now the space for the inbound oop handle area
1371   int total_save_slots = 2 * VMRegImpl::slots_per_word; // 2 arguments passed in registers
1372 
1373   int oop_handle_offset = stack_slots;
1374   stack_slots += total_save_slots;
1375 
1376   // Now any space we need for handlizing a klass if static method
1377 
1378   int klass_slot_offset = 0;
1379   int klass_offset = -1;
1380   int lock_slot_offset = 0;
1381   bool is_static = false;
1382 
1383   if (method->is_static()) {
1384     klass_slot_offset = stack_slots;
1385     stack_slots += VMRegImpl::slots_per_word;
1386     klass_offset = klass_slot_offset * VMRegImpl::stack_slot_size;
1387     is_static = true;
1388   }
1389 
1390   // Plus a lock if needed
1391 
1392   if (method->is_synchronized()) {
1393     lock_slot_offset = stack_slots;
1394     stack_slots += VMRegImpl::slots_per_word;
1395   }
1396 
1397   // Now a place (+2) to save return values or temp during shuffling
1398   // + 2 for return address (which we own) and saved rbp,
1399   stack_slots += 4;
1400 
1401   // Ok The space we have allocated will look like:
1402   //
1403   //
1404   // FP-> |                     |
1405   //      |---------------------|
1406   //      | 2 slots for moves   |
1407   //      |---------------------|
1408   //      | lock box (if sync)  |
1409   //      |---------------------| <- lock_slot_offset  (-lock_slot_rbp_offset)
1410   //      | klass (if static)   |
1411   //      |---------------------| <- klass_slot_offset
1412   //      | oopHandle area      |
1413   //      |---------------------| <- oop_handle_offset (a max of 2 registers)
1414   //      | outbound memory     |
1415   //      | based arguments     |
1416   //      |                     |
1417   //      |---------------------|
1418   //      |                     |
1419   // SP-> | out_preserved_slots |
1420   //
1421   //
1422   // ****************************************************************************
1423   // WARNING - on Windows Java Natives use pascal calling convention and pop the
1424   // arguments off of the stack after the jni call. Before the call we can use
1425   // instructions that are SP relative. After the jni call we switch to FP
1426   // relative instructions instead of re-adjusting the stack on windows.
1427   // ****************************************************************************
1428 
1429 
1430   // Now compute actual number of stack words we need rounding to make
1431   // stack properly aligned.
1432   stack_slots = align_up(stack_slots, StackAlignmentInSlots);
1433 
1434   int stack_size = stack_slots * VMRegImpl::stack_slot_size;
1435 
1436   intptr_t start = (intptr_t)__ pc();
1437 
1438   // First thing make an ic check to see if we should even be here
1439 
1440   // We are free to use all registers as temps without saving them and
1441   // restoring them except rbp. rbp is the only callee save register
1442   // as far as the interpreter and the compiler(s) are concerned.
1443 
1444 
1445   const Register receiver = rcx;
1446   Label exception_pending;
1447 
1448   __ verify_oop(receiver);
1449   // verified entry must be aligned for code patching.
1450   __ ic_check(8 /* end_alignment */);
1451 
1452   int vep_offset = ((intptr_t)__ pc()) - start;
1453 
1454 #ifdef COMPILER1
1455   // For Object.hashCode, System.identityHashCode try to pull hashCode from object header if available.
1456   if ((InlineObjectHash && method->intrinsic_id() == vmIntrinsics::_hashCode) || (method->intrinsic_id() == vmIntrinsics::_identityHashCode)) {
1457     inline_check_hashcode_from_object_header(masm, method, rcx /*obj_reg*/, rax /*result*/);
1458    }
1459 #endif // COMPILER1
1460 
1461   // The instruction at the verified entry point must be 5 bytes or longer
1462   // because it can be patched on the fly by make_non_entrant. The stack bang
1463   // instruction fits that requirement.
1464 
1465   // Generate stack overflow check
1466   __ bang_stack_with_offset((int)StackOverflow::stack_shadow_zone_size());
1467 
1468   // Generate a new frame for the wrapper.
1469   __ enter();
1470   // -2 because return address is already present and so is saved rbp
1471   __ subptr(rsp, stack_size - 2*wordSize);
1472 
1473 
1474   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
1475   bs->nmethod_entry_barrier(masm, nullptr /* slow_path */, nullptr /* continuation */);
1476 
1477   // Frame is now completed as far as size and linkage.
1478   int frame_complete = ((intptr_t)__ pc()) - start;
1479 
1480   if (UseRTMLocking) {
1481     // Abort RTM transaction before calling JNI
1482     // because critical section will be large and will be
1483     // aborted anyway. Also nmethod could be deoptimized.
1484     __ xabort(0);
1485   }
1486 
1487   // Calculate the difference between rsp and rbp,. We need to know it
1488   // after the native call because on windows Java Natives will pop
1489   // the arguments and it is painful to do rsp relative addressing
1490   // in a platform independent way. So after the call we switch to
1491   // rbp, relative addressing.
1492 
1493   int fp_adjustment = stack_size - 2*wordSize;
1494 
1495 #ifdef COMPILER2
1496   // C2 may leave the stack dirty if not in SSE2+ mode
1497   if (UseSSE >= 2) {
1498     __ verify_FPU(0, "c2i transition should have clean FPU stack");
1499   } else {
1500     __ empty_FPU_stack();
1501   }
1502 #endif /* COMPILER2 */
1503 
1504   // Compute the rbp, offset for any slots used after the jni call
1505 
1506   int lock_slot_rbp_offset = (lock_slot_offset*VMRegImpl::stack_slot_size) - fp_adjustment;
1507 
1508   // We use rdi as a thread pointer because it is callee save and
1509   // if we load it once it is usable thru the entire wrapper
1510   const Register thread = rdi;
1511 
1512    // We use rsi as the oop handle for the receiver/klass
1513    // It is callee save so it survives the call to native
1514 
1515    const Register oop_handle_reg = rsi;
1516 
1517    __ get_thread(thread);
1518 
1519   //
1520   // We immediately shuffle the arguments so that any vm call we have to
1521   // make from here on out (sync slow path, jvmti, etc.) we will have
1522   // captured the oops from our caller and have a valid oopMap for
1523   // them.
1524 
1525   // -----------------
1526   // The Grand Shuffle
1527   //
1528   // Natives require 1 or 2 extra arguments over the normal ones: the JNIEnv*
1529   // and, if static, the class mirror instead of a receiver.  This pretty much
1530   // guarantees that register layout will not match (and x86 doesn't use reg
1531   // parms though amd does).  Since the native abi doesn't use register args
1532   // and the java conventions does we don't have to worry about collisions.
1533   // All of our moved are reg->stack or stack->stack.
1534   // We ignore the extra arguments during the shuffle and handle them at the
1535   // last moment. The shuffle is described by the two calling convention
1536   // vectors we have in our possession. We simply walk the java vector to
1537   // get the source locations and the c vector to get the destinations.
1538 
1539   int c_arg = method->is_static() ? 2 : 1;
1540 
1541   // Record rsp-based slot for receiver on stack for non-static methods
1542   int receiver_offset = -1;
1543 
1544   // This is a trick. We double the stack slots so we can claim
1545   // the oops in the caller's frame. Since we are sure to have
1546   // more args than the caller doubling is enough to make
1547   // sure we can capture all the incoming oop args from the
1548   // caller.
1549   //
1550   OopMap* map = new OopMap(stack_slots * 2, 0 /* arg_slots*/);
1551 
1552   // Mark location of rbp,
1553   // map->set_callee_saved(VMRegImpl::stack2reg( stack_slots - 2), stack_slots * 2, 0, rbp->as_VMReg());
1554 
1555   // We know that we only have args in at most two integer registers (rcx, rdx). So rax, rbx
1556   // Are free to temporaries if we have to do  stack to steck moves.
1557   // All inbound args are referenced based on rbp, and all outbound args via rsp.
1558 
1559   for (int i = 0; i < total_in_args ; i++, c_arg++ ) {
1560     switch (in_sig_bt[i]) {
1561       case T_ARRAY:
1562       case T_OBJECT:
1563         object_move(masm, map, oop_handle_offset, stack_slots, in_regs[i], out_regs[c_arg],
1564                     ((i == 0) && (!is_static)),
1565                     &receiver_offset);
1566         break;
1567       case T_VOID:
1568         break;
1569 
1570       case T_FLOAT:
1571         float_move(masm, in_regs[i], out_regs[c_arg]);
1572           break;
1573 
1574       case T_DOUBLE:
1575         assert( i + 1 < total_in_args &&
1576                 in_sig_bt[i + 1] == T_VOID &&
1577                 out_sig_bt[c_arg+1] == T_VOID, "bad arg list");
1578         double_move(masm, in_regs[i], out_regs[c_arg]);
1579         break;
1580 
1581       case T_LONG :
1582         long_move(masm, in_regs[i], out_regs[c_arg]);
1583         break;
1584 
1585       case T_ADDRESS: assert(false, "found T_ADDRESS in java args");
1586 
1587       default:
1588         simple_move32(masm, in_regs[i], out_regs[c_arg]);
1589     }
1590   }
1591 
1592   // Pre-load a static method's oop into rsi.  Used both by locking code and
1593   // the normal JNI call code.
1594   if (method->is_static()) {
1595 
1596     //  load opp into a register
1597     __ movoop(oop_handle_reg, JNIHandles::make_local(method->method_holder()->java_mirror()));
1598 
1599     // Now handlize the static class mirror it's known not-null.
1600     __ movptr(Address(rsp, klass_offset), oop_handle_reg);
1601     map->set_oop(VMRegImpl::stack2reg(klass_slot_offset));
1602 
1603     // Now get the handle
1604     __ lea(oop_handle_reg, Address(rsp, klass_offset));
1605     // store the klass handle as second argument
1606     __ movptr(Address(rsp, wordSize), oop_handle_reg);
1607   }
1608 
1609   // Change state to native (we save the return address in the thread, since it might not
1610   // be pushed on the stack when we do a stack traversal). It is enough that the pc()
1611   // points into the right code segment. It does not have to be the correct return pc.
1612   // We use the same pc/oopMap repeatedly when we call out
1613 
1614   intptr_t the_pc = (intptr_t) __ pc();
1615   oop_maps->add_gc_map(the_pc - start, map);
1616 
1617   __ set_last_Java_frame(thread, rsp, noreg, (address)the_pc, noreg);
1618 
1619 
1620   // We have all of the arguments setup at this point. We must not touch any register
1621   // argument registers at this point (what if we save/restore them there are no oop?
1622 
1623   {
1624     SkipIfEqual skip_if(masm, &DTraceMethodProbes, 0, noreg);
1625     __ mov_metadata(rax, method());
1626     __ call_VM_leaf(
1627          CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
1628          thread, rax);
1629   }
1630 
1631   // RedefineClasses() tracing support for obsolete method entry
1632   if (log_is_enabled(Trace, redefine, class, obsolete)) {
1633     __ mov_metadata(rax, method());
1634     __ call_VM_leaf(
1635          CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
1636          thread, rax);
1637   }
1638 
1639   // These are register definitions we need for locking/unlocking
1640   const Register swap_reg = rax;  // Must use rax, for cmpxchg instruction
1641   const Register obj_reg  = rcx;  // Will contain the oop
1642   const Register lock_reg = rdx;  // Address of compiler lock object (BasicLock)
1643 
1644   Label slow_path_lock;
1645   Label lock_done;
1646 
1647   // Lock a synchronized method
1648   if (method->is_synchronized()) {
1649     Label count_mon;
1650 
1651     const int mark_word_offset = BasicLock::displaced_header_offset_in_bytes();
1652 
1653     // Get the handle (the 2nd argument)
1654     __ movptr(oop_handle_reg, Address(rsp, wordSize));
1655 
1656     // Get address of the box
1657 
1658     __ lea(lock_reg, Address(rbp, lock_slot_rbp_offset));
1659 
1660     // Load the oop from the handle
1661     __ movptr(obj_reg, Address(oop_handle_reg, 0));
1662 
1663     if (LockingMode == LM_MONITOR) {
1664       __ jmp(slow_path_lock);
1665     } else if (LockingMode == LM_LEGACY) {
1666       // Load immediate 1 into swap_reg %rax,
1667       __ movptr(swap_reg, 1);
1668 
1669       // Load (object->mark() | 1) into swap_reg %rax,
1670       __ orptr(swap_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
1671 
1672       // Save (object->mark() | 1) into BasicLock's displaced header
1673       __ movptr(Address(lock_reg, mark_word_offset), swap_reg);
1674 
1675       // src -> dest iff dest == rax, else rax, <- dest
1676       // *obj_reg = lock_reg iff *obj_reg == rax, else rax, = *(obj_reg)
1677       __ lock();
1678       __ cmpxchgptr(lock_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
1679       __ jcc(Assembler::equal, count_mon);
1680 
1681       // Test if the oopMark is an obvious stack pointer, i.e.,
1682       //  1) (mark & 3) == 0, and
1683       //  2) rsp <= mark < mark + os::pagesize()
1684       // These 3 tests can be done by evaluating the following
1685       // expression: ((mark - rsp) & (3 - os::vm_page_size())),
1686       // assuming both stack pointer and pagesize have their
1687       // least significant 2 bits clear.
1688       // NOTE: the oopMark is in swap_reg %rax, as the result of cmpxchg
1689 
1690       __ subptr(swap_reg, rsp);
1691       __ andptr(swap_reg, 3 - (int)os::vm_page_size());
1692 
1693       // Save the test result, for recursive case, the result is zero
1694       __ movptr(Address(lock_reg, mark_word_offset), swap_reg);
1695       __ jcc(Assembler::notEqual, slow_path_lock);
1696     } else {
1697       assert(LockingMode == LM_LIGHTWEIGHT, "must be");
1698       __ lightweight_lock(lock_reg, obj_reg, swap_reg, thread, lock_reg, slow_path_lock);
1699     }
1700     __ bind(count_mon);
1701     __ inc_held_monitor_count();
1702 
1703     // Slow path will re-enter here
1704     __ bind(lock_done);
1705   }
1706 
1707 
1708   // Finally just about ready to make the JNI call
1709 
1710   // get JNIEnv* which is first argument to native
1711   __ lea(rdx, Address(thread, in_bytes(JavaThread::jni_environment_offset())));
1712   __ movptr(Address(rsp, 0), rdx);
1713 
1714   // Now set thread in native
1715   __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_native);
1716 
1717   __ call(RuntimeAddress(native_func));
1718 
1719   // Verify or restore cpu control state after JNI call
1720   __ restore_cpu_control_state_after_jni(noreg);
1721 
1722   // WARNING - on Windows Java Natives use pascal calling convention and pop the
1723   // arguments off of the stack. We could just re-adjust the stack pointer here
1724   // and continue to do SP relative addressing but we instead switch to FP
1725   // relative addressing.
1726 
1727   // Unpack native results.
1728   switch (ret_type) {
1729   case T_BOOLEAN: __ c2bool(rax);            break;
1730   case T_CHAR   : __ andptr(rax, 0xFFFF);    break;
1731   case T_BYTE   : __ sign_extend_byte (rax); break;
1732   case T_SHORT  : __ sign_extend_short(rax); break;
1733   case T_INT    : /* nothing to do */        break;
1734   case T_DOUBLE :
1735   case T_FLOAT  :
1736     // Result is in st0 we'll save as needed
1737     break;
1738   case T_ARRAY:                 // Really a handle
1739   case T_OBJECT:                // Really a handle
1740       break; // can't de-handlize until after safepoint check
1741   case T_VOID: break;
1742   case T_LONG: break;
1743   default       : ShouldNotReachHere();
1744   }
1745 
1746   Label after_transition;
1747 
1748   // Switch thread to "native transition" state before reading the synchronization state.
1749   // This additional state is necessary because reading and testing the synchronization
1750   // state is not atomic w.r.t. GC, as this scenario demonstrates:
1751   //     Java thread A, in _thread_in_native state, loads _not_synchronized and is preempted.
1752   //     VM thread changes sync state to synchronizing and suspends threads for GC.
1753   //     Thread A is resumed to finish this native method, but doesn't block here since it
1754   //     didn't see any synchronization is progress, and escapes.
1755   __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_native_trans);
1756 
1757   // Force this write out before the read below
1758   if (!UseSystemMemoryBarrier) {
1759     __ membar(Assembler::Membar_mask_bits(
1760               Assembler::LoadLoad | Assembler::LoadStore |
1761               Assembler::StoreLoad | Assembler::StoreStore));
1762   }
1763 
1764   if (AlwaysRestoreFPU) {
1765     // Make sure the control word is correct.
1766     __ fldcw(ExternalAddress(StubRoutines::x86::addr_fpu_cntrl_wrd_std()));
1767   }
1768 
1769   // check for safepoint operation in progress and/or pending suspend requests
1770   { Label Continue, slow_path;
1771 
1772     __ safepoint_poll(slow_path, thread, true /* at_return */, false /* in_nmethod */);
1773 
1774     __ cmpl(Address(thread, JavaThread::suspend_flags_offset()), 0);
1775     __ jcc(Assembler::equal, Continue);
1776     __ bind(slow_path);
1777 
1778     // Don't use call_VM as it will see a possible pending exception and forward it
1779     // and never return here preventing us from clearing _last_native_pc down below.
1780     // Also can't use call_VM_leaf either as it will check to see if rsi & rdi are
1781     // preserved and correspond to the bcp/locals pointers. So we do a runtime call
1782     // by hand.
1783     //
1784     __ vzeroupper();
1785 
1786     save_native_result(masm, ret_type, stack_slots);
1787     __ push(thread);
1788     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address,
1789                                               JavaThread::check_special_condition_for_native_trans)));
1790     __ increment(rsp, wordSize);
1791     // Restore any method result value
1792     restore_native_result(masm, ret_type, stack_slots);
1793     __ bind(Continue);
1794   }
1795 
1796   // change thread state
1797   __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_Java);
1798   __ bind(after_transition);
1799 
1800   Label reguard;
1801   Label reguard_done;
1802   __ cmpl(Address(thread, JavaThread::stack_guard_state_offset()), StackOverflow::stack_guard_yellow_reserved_disabled);
1803   __ jcc(Assembler::equal, reguard);
1804 
1805   // slow path reguard  re-enters here
1806   __ bind(reguard_done);
1807 
1808   // Handle possible exception (will unlock if necessary)
1809 
1810   // native result if any is live
1811 
1812   // Unlock
1813   Label slow_path_unlock;
1814   Label unlock_done;
1815   if (method->is_synchronized()) {
1816 
1817     Label fast_done;
1818 
1819     // Get locked oop from the handle we passed to jni
1820     __ movptr(obj_reg, Address(oop_handle_reg, 0));
1821 
1822     if (LockingMode == LM_LEGACY) {
1823       Label not_recur;
1824       // Simple recursive lock?
1825       __ cmpptr(Address(rbp, lock_slot_rbp_offset), NULL_WORD);
1826       __ jcc(Assembler::notEqual, not_recur);
1827       __ dec_held_monitor_count();
1828       __ jmpb(fast_done);
1829       __ bind(not_recur);
1830     }
1831 
1832     // Must save rax, if it is live now because cmpxchg must use it
1833     if (ret_type != T_FLOAT && ret_type != T_DOUBLE && ret_type != T_VOID) {
1834       save_native_result(masm, ret_type, stack_slots);
1835     }
1836 
1837     if (LockingMode == LM_MONITOR) {
1838       __ jmp(slow_path_unlock);
1839     } else if (LockingMode == LM_LEGACY) {
1840       //  get old displaced header
1841       __ movptr(rbx, Address(rbp, lock_slot_rbp_offset));
1842 
1843       // get address of the stack lock
1844       __ lea(rax, Address(rbp, lock_slot_rbp_offset));
1845 
1846       // Atomic swap old header if oop still contains the stack lock
1847       // src -> dest iff dest == rax, else rax, <- dest
1848       // *obj_reg = rbx, iff *obj_reg == rax, else rax, = *(obj_reg)
1849       __ lock();
1850       __ cmpxchgptr(rbx, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
1851       __ jcc(Assembler::notEqual, slow_path_unlock);
1852       __ dec_held_monitor_count();
1853     } else {
1854       assert(LockingMode == LM_LIGHTWEIGHT, "must be");
1855       __ lightweight_unlock(obj_reg, swap_reg, thread, lock_reg, slow_path_unlock);
1856       __ dec_held_monitor_count();
1857     }
1858 
1859     // slow path re-enters here
1860     __ bind(unlock_done);
1861     if (ret_type != T_FLOAT && ret_type != T_DOUBLE && ret_type != T_VOID) {
1862       restore_native_result(masm, ret_type, stack_slots);
1863     }
1864 
1865     __ bind(fast_done);
1866   }
1867 
1868   {
1869     SkipIfEqual skip_if(masm, &DTraceMethodProbes, 0, noreg);
1870     // Tell dtrace about this method exit
1871     save_native_result(masm, ret_type, stack_slots);
1872     __ mov_metadata(rax, method());
1873     __ call_VM_leaf(
1874          CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
1875          thread, rax);
1876     restore_native_result(masm, ret_type, stack_slots);
1877   }
1878 
1879   // We can finally stop using that last_Java_frame we setup ages ago
1880 
1881   __ reset_last_Java_frame(thread, false);
1882 
1883   // Unbox oop result, e.g. JNIHandles::resolve value.
1884   if (is_reference_type(ret_type)) {
1885     __ resolve_jobject(rax /* value */,
1886                        thread /* thread */,
1887                        rcx /* tmp */);
1888   }
1889 
1890   if (CheckJNICalls) {
1891     // clear_pending_jni_exception_check
1892     __ movptr(Address(thread, JavaThread::pending_jni_exception_check_fn_offset()), NULL_WORD);
1893   }
1894 
1895   // reset handle block
1896   __ movptr(rcx, Address(thread, JavaThread::active_handles_offset()));
1897   __ movl(Address(rcx, JNIHandleBlock::top_offset()), NULL_WORD);
1898 
1899   // Any exception pending?
1900   __ cmpptr(Address(thread, in_bytes(Thread::pending_exception_offset())), NULL_WORD);
1901   __ jcc(Assembler::notEqual, exception_pending);
1902 
1903   // no exception, we're almost done
1904 
1905   // check that only result value is on FPU stack
1906   __ verify_FPU(ret_type == T_FLOAT || ret_type == T_DOUBLE ? 1 : 0, "native_wrapper normal exit");
1907 
1908   // Fixup floating pointer results so that result looks like a return from a compiled method
1909   if (ret_type == T_FLOAT) {
1910     if (UseSSE >= 1) {
1911       // Pop st0 and store as float and reload into xmm register
1912       __ fstp_s(Address(rbp, -4));
1913       __ movflt(xmm0, Address(rbp, -4));
1914     }
1915   } else if (ret_type == T_DOUBLE) {
1916     if (UseSSE >= 2) {
1917       // Pop st0 and store as double and reload into xmm register
1918       __ fstp_d(Address(rbp, -8));
1919       __ movdbl(xmm0, Address(rbp, -8));
1920     }
1921   }
1922 
1923   // Return
1924 
1925   __ leave();
1926   __ ret(0);
1927 
1928   // Unexpected paths are out of line and go here
1929 
1930   // Slow path locking & unlocking
1931   if (method->is_synchronized()) {
1932 
1933     // BEGIN Slow path lock
1934 
1935     __ bind(slow_path_lock);
1936 
1937     if (LockingMode == LM_LIGHTWEIGHT) {
1938       // Reload the lock addr. Clobbered by lightweight_lock.
1939       __ lea(lock_reg, Address(rbp, lock_slot_rbp_offset));
1940     }
1941 
1942     // has last_Java_frame setup. No exceptions so do vanilla call not call_VM
1943     // args are (oop obj, BasicLock* lock, JavaThread* thread)
1944     __ push(thread);
1945     __ push(lock_reg);
1946     __ push(obj_reg);
1947     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::complete_monitor_locking_C)));
1948     __ addptr(rsp, 3*wordSize);
1949 
1950 #ifdef ASSERT
1951     { Label L;
1952     __ cmpptr(Address(thread, in_bytes(Thread::pending_exception_offset())), NULL_WORD);
1953     __ jcc(Assembler::equal, L);
1954     __ stop("no pending exception allowed on exit from monitorenter");
1955     __ bind(L);
1956     }
1957 #endif
1958     __ jmp(lock_done);
1959 
1960     // END Slow path lock
1961 
1962     // BEGIN Slow path unlock
1963     __ bind(slow_path_unlock);
1964     __ vzeroupper();
1965     // Slow path unlock
1966 
1967     if (ret_type == T_FLOAT || ret_type == T_DOUBLE ) {
1968       save_native_result(masm, ret_type, stack_slots);
1969     }
1970     // Save pending exception around call to VM (which contains an EXCEPTION_MARK)
1971 
1972     __ pushptr(Address(thread, in_bytes(Thread::pending_exception_offset())));
1973     __ movptr(Address(thread, in_bytes(Thread::pending_exception_offset())), NULL_WORD);
1974 
1975 
1976     // should be a peal
1977     // +wordSize because of the push above
1978     // args are (oop obj, BasicLock* lock, JavaThread* thread)
1979     __ push(thread);
1980     __ lea(rax, Address(rbp, lock_slot_rbp_offset));
1981     __ push(rax);
1982 
1983     __ push(obj_reg);
1984     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::complete_monitor_unlocking_C)));
1985     __ addptr(rsp, 3*wordSize);
1986 #ifdef ASSERT
1987     {
1988       Label L;
1989       __ cmpptr(Address(thread, in_bytes(Thread::pending_exception_offset())), NULL_WORD);
1990       __ jcc(Assembler::equal, L);
1991       __ stop("no pending exception allowed on exit complete_monitor_unlocking_C");
1992       __ bind(L);
1993     }
1994 #endif /* ASSERT */
1995 
1996     __ popptr(Address(thread, in_bytes(Thread::pending_exception_offset())));
1997 
1998     if (ret_type == T_FLOAT || ret_type == T_DOUBLE ) {
1999       restore_native_result(masm, ret_type, stack_slots);
2000     }
2001     __ jmp(unlock_done);
2002     // END Slow path unlock
2003 
2004   }
2005 
2006   // SLOW PATH Reguard the stack if needed
2007 
2008   __ bind(reguard);
2009   __ vzeroupper();
2010   save_native_result(masm, ret_type, stack_slots);
2011   {
2012     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
2013   }
2014   restore_native_result(masm, ret_type, stack_slots);
2015   __ jmp(reguard_done);
2016 
2017 
2018   // BEGIN EXCEPTION PROCESSING
2019 
2020   // Forward  the exception
2021   __ bind(exception_pending);
2022 
2023   // remove possible return value from FPU register stack
2024   __ empty_FPU_stack();
2025 
2026   // pop our frame
2027   __ leave();
2028   // and forward the exception
2029   __ jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
2030 
2031   __ flush();
2032 
2033   nmethod *nm = nmethod::new_native_nmethod(method,
2034                                             compile_id,
2035                                             masm->code(),
2036                                             vep_offset,
2037                                             frame_complete,
2038                                             stack_slots / VMRegImpl::slots_per_word,
2039                                             (is_static ? in_ByteSize(klass_offset) : in_ByteSize(receiver_offset)),
2040                                             in_ByteSize(lock_slot_offset*VMRegImpl::stack_slot_size),
2041                                             oop_maps);
2042 
2043   return nm;
2044 
2045 }
2046 
2047 // this function returns the adjust size (in number of words) to a c2i adapter
2048 // activation for use during deoptimization
2049 int Deoptimization::last_frame_adjust(int callee_parameters, int callee_locals ) {
2050   return (callee_locals - callee_parameters) * Interpreter::stackElementWords;
2051 }
2052 
2053 
2054 // Number of stack slots between incoming argument block and the start of
2055 // a new frame.  The PROLOG must add this many slots to the stack.  The
2056 // EPILOG must remove this many slots.  Intel needs one slot for
2057 // return address and one for rbp, (must save rbp)
2058 uint SharedRuntime::in_preserve_stack_slots() {
2059   return 2+VerifyStackAtCalls;
2060 }
2061 
2062 uint SharedRuntime::out_preserve_stack_slots() {
2063   return 0;
2064 }
2065 
2066 //------------------------------generate_deopt_blob----------------------------
2067 void SharedRuntime::generate_deopt_blob() {
2068   // allocate space for the code
2069   ResourceMark rm;
2070   // setup code generation tools
2071   // note: the buffer code size must account for StackShadowPages=50
2072   CodeBuffer   buffer("deopt_blob", 1536, 1024);
2073   MacroAssembler* masm = new MacroAssembler(&buffer);
2074   int frame_size_in_words;
2075   OopMap* map = nullptr;
2076   // Account for the extra args we place on the stack
2077   // by the time we call fetch_unroll_info
2078   const int additional_words = 2; // deopt kind, thread
2079 
2080   OopMapSet *oop_maps = new OopMapSet();
2081 
2082   // -------------
2083   // This code enters when returning to a de-optimized nmethod.  A return
2084   // address has been pushed on the stack, and return values are in
2085   // registers.
2086   // If we are doing a normal deopt then we were called from the patched
2087   // nmethod from the point we returned to the nmethod. So the return
2088   // address on the stack is wrong by NativeCall::instruction_size
2089   // We will adjust the value to it looks like we have the original return
2090   // address on the stack (like when we eagerly deoptimized).
2091   // In the case of an exception pending with deoptimized then we enter
2092   // with a return address on the stack that points after the call we patched
2093   // into the exception handler. We have the following register state:
2094   //    rax,: exception
2095   //    rbx,: exception handler
2096   //    rdx: throwing pc
2097   // So in this case we simply jam rdx into the useless return address and
2098   // the stack looks just like we want.
2099   //
2100   // At this point we need to de-opt.  We save the argument return
2101   // registers.  We call the first C routine, fetch_unroll_info().  This
2102   // routine captures the return values and returns a structure which
2103   // describes the current frame size and the sizes of all replacement frames.
2104   // The current frame is compiled code and may contain many inlined
2105   // functions, each with their own JVM state.  We pop the current frame, then
2106   // push all the new frames.  Then we call the C routine unpack_frames() to
2107   // populate these frames.  Finally unpack_frames() returns us the new target
2108   // address.  Notice that callee-save registers are BLOWN here; they have
2109   // already been captured in the vframeArray at the time the return PC was
2110   // patched.
2111   address start = __ pc();
2112   Label cont;
2113 
2114   // Prolog for non exception case!
2115 
2116   // Save everything in sight.
2117 
2118   map = RegisterSaver::save_live_registers(masm, additional_words, &frame_size_in_words, false);
2119   // Normal deoptimization
2120   __ push(Deoptimization::Unpack_deopt);
2121   __ jmp(cont);
2122 
2123   int reexecute_offset = __ pc() - start;
2124 
2125   // Reexecute case
2126   // return address is the pc describes what bci to do re-execute at
2127 
2128   // No need to update map as each call to save_live_registers will produce identical oopmap
2129   (void) RegisterSaver::save_live_registers(masm, additional_words, &frame_size_in_words, false);
2130 
2131   __ push(Deoptimization::Unpack_reexecute);
2132   __ jmp(cont);
2133 
2134   int exception_offset = __ pc() - start;
2135 
2136   // Prolog for exception case
2137 
2138   // all registers are dead at this entry point, except for rax, and
2139   // rdx which contain the exception oop and exception pc
2140   // respectively.  Set them in TLS and fall thru to the
2141   // unpack_with_exception_in_tls entry point.
2142 
2143   __ get_thread(rdi);
2144   __ movptr(Address(rdi, JavaThread::exception_pc_offset()), rdx);
2145   __ movptr(Address(rdi, JavaThread::exception_oop_offset()), rax);
2146 
2147   int exception_in_tls_offset = __ pc() - start;
2148 
2149   // new implementation because exception oop is now passed in JavaThread
2150 
2151   // Prolog for exception case
2152   // All registers must be preserved because they might be used by LinearScan
2153   // Exceptiop oop and throwing PC are passed in JavaThread
2154   // tos: stack at point of call to method that threw the exception (i.e. only
2155   // args are on the stack, no return address)
2156 
2157   // make room on stack for the return address
2158   // It will be patched later with the throwing pc. The correct value is not
2159   // available now because loading it from memory would destroy registers.
2160   __ push(0);
2161 
2162   // Save everything in sight.
2163 
2164   // No need to update map as each call to save_live_registers will produce identical oopmap
2165   (void) RegisterSaver::save_live_registers(masm, additional_words, &frame_size_in_words, false);
2166 
2167   // Now it is safe to overwrite any register
2168 
2169   // store the correct deoptimization type
2170   __ push(Deoptimization::Unpack_exception);
2171 
2172   // load throwing pc from JavaThread and patch it as the return address
2173   // of the current frame. Then clear the field in JavaThread
2174   __ get_thread(rdi);
2175   __ movptr(rdx, Address(rdi, JavaThread::exception_pc_offset()));
2176   __ movptr(Address(rbp, wordSize), rdx);
2177   __ movptr(Address(rdi, JavaThread::exception_pc_offset()), NULL_WORD);
2178 
2179 #ifdef ASSERT
2180   // verify that there is really an exception oop in JavaThread
2181   __ movptr(rax, Address(rdi, JavaThread::exception_oop_offset()));
2182   __ verify_oop(rax);
2183 
2184   // verify that there is no pending exception
2185   Label no_pending_exception;
2186   __ movptr(rax, Address(rdi, Thread::pending_exception_offset()));
2187   __ testptr(rax, rax);
2188   __ jcc(Assembler::zero, no_pending_exception);
2189   __ stop("must not have pending exception here");
2190   __ bind(no_pending_exception);
2191 #endif
2192 
2193   __ bind(cont);
2194 
2195   // Compiled code leaves the floating point stack dirty, empty it.
2196   __ empty_FPU_stack();
2197 
2198 
2199   // Call C code.  Need thread and this frame, but NOT official VM entry
2200   // crud.  We cannot block on this call, no GC can happen.
2201   __ get_thread(rcx);
2202   __ push(rcx);
2203   // fetch_unroll_info needs to call last_java_frame()
2204   __ set_last_Java_frame(rcx, noreg, noreg, nullptr, noreg);
2205 
2206   __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, Deoptimization::fetch_unroll_info)));
2207 
2208   // Need to have an oopmap that tells fetch_unroll_info where to
2209   // find any register it might need.
2210 
2211   oop_maps->add_gc_map( __ pc()-start, map);
2212 
2213   // Discard args to fetch_unroll_info
2214   __ pop(rcx);
2215   __ pop(rcx);
2216 
2217   __ get_thread(rcx);
2218   __ reset_last_Java_frame(rcx, false);
2219 
2220   // Load UnrollBlock into EDI
2221   __ mov(rdi, rax);
2222 
2223   // Move the unpack kind to a safe place in the UnrollBlock because
2224   // we are very short of registers
2225 
2226   Address unpack_kind(rdi, Deoptimization::UnrollBlock::unpack_kind_offset());
2227   // retrieve the deopt kind from the UnrollBlock.
2228   __ movl(rax, unpack_kind);
2229 
2230    Label noException;
2231   __ cmpl(rax, Deoptimization::Unpack_exception);   // Was exception pending?
2232   __ jcc(Assembler::notEqual, noException);
2233   __ movptr(rax, Address(rcx, JavaThread::exception_oop_offset()));
2234   __ movptr(rdx, Address(rcx, JavaThread::exception_pc_offset()));
2235   __ movptr(Address(rcx, JavaThread::exception_oop_offset()), NULL_WORD);
2236   __ movptr(Address(rcx, JavaThread::exception_pc_offset()), NULL_WORD);
2237 
2238   __ verify_oop(rax);
2239 
2240   // Overwrite the result registers with the exception results.
2241   __ movptr(Address(rsp, RegisterSaver::raxOffset()*wordSize), rax);
2242   __ movptr(Address(rsp, RegisterSaver::rdxOffset()*wordSize), rdx);
2243 
2244   __ bind(noException);
2245 
2246   // Stack is back to only having register save data on the stack.
2247   // Now restore the result registers. Everything else is either dead or captured
2248   // in the vframeArray.
2249 
2250   RegisterSaver::restore_result_registers(masm);
2251 
2252   // Non standard control word may be leaked out through a safepoint blob, and we can
2253   // deopt at a poll point with the non standard control word. However, we should make
2254   // sure the control word is correct after restore_result_registers.
2255   __ fldcw(ExternalAddress(StubRoutines::x86::addr_fpu_cntrl_wrd_std()));
2256 
2257   // All of the register save area has been popped of the stack. Only the
2258   // return address remains.
2259 
2260   // Pop all the frames we must move/replace.
2261   //
2262   // Frame picture (youngest to oldest)
2263   // 1: self-frame (no frame link)
2264   // 2: deopting frame  (no frame link)
2265   // 3: caller of deopting frame (could be compiled/interpreted).
2266   //
2267   // Note: by leaving the return address of self-frame on the stack
2268   // and using the size of frame 2 to adjust the stack
2269   // when we are done the return to frame 3 will still be on the stack.
2270 
2271   // Pop deoptimized frame
2272   __ addptr(rsp, Address(rdi,Deoptimization::UnrollBlock::size_of_deoptimized_frame_offset()));
2273 
2274   // sp should be pointing at the return address to the caller (3)
2275 
2276   // Pick up the initial fp we should save
2277   // restore rbp before stack bang because if stack overflow is thrown it needs to be pushed (and preserved)
2278   __ movptr(rbp, Address(rdi, Deoptimization::UnrollBlock::initial_info_offset()));
2279 
2280 #ifdef ASSERT
2281   // Compilers generate code that bang the stack by as much as the
2282   // interpreter would need. So this stack banging should never
2283   // trigger a fault. Verify that it does not on non product builds.
2284   __ movl(rbx, Address(rdi ,Deoptimization::UnrollBlock::total_frame_sizes_offset()));
2285   __ bang_stack_size(rbx, rcx);
2286 #endif
2287 
2288   // Load array of frame pcs into ECX
2289   __ movptr(rcx,Address(rdi,Deoptimization::UnrollBlock::frame_pcs_offset()));
2290 
2291   __ pop(rsi); // trash the old pc
2292 
2293   // Load array of frame sizes into ESI
2294   __ movptr(rsi,Address(rdi,Deoptimization::UnrollBlock::frame_sizes_offset()));
2295 
2296   Address counter(rdi, Deoptimization::UnrollBlock::counter_temp_offset());
2297 
2298   __ movl(rbx, Address(rdi, Deoptimization::UnrollBlock::number_of_frames_offset()));
2299   __ movl(counter, rbx);
2300 
2301   // Now adjust the caller's stack to make up for the extra locals
2302   // but record the original sp so that we can save it in the skeletal interpreter
2303   // frame and the stack walking of interpreter_sender will get the unextended sp
2304   // value and not the "real" sp value.
2305 
2306   Address sp_temp(rdi, Deoptimization::UnrollBlock::sender_sp_temp_offset());
2307   __ movptr(sp_temp, rsp);
2308   __ movl2ptr(rbx, Address(rdi, Deoptimization::UnrollBlock::caller_adjustment_offset()));
2309   __ subptr(rsp, rbx);
2310 
2311   // Push interpreter frames in a loop
2312   Label loop;
2313   __ bind(loop);
2314   __ movptr(rbx, Address(rsi, 0));      // Load frame size
2315   __ subptr(rbx, 2*wordSize);           // we'll push pc and rbp, by hand
2316   __ pushptr(Address(rcx, 0));          // save return address
2317   __ enter();                           // save old & set new rbp,
2318   __ subptr(rsp, rbx);                  // Prolog!
2319   __ movptr(rbx, sp_temp);              // sender's sp
2320   // This value is corrected by layout_activation_impl
2321   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
2322   __ movptr(Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize), rbx); // Make it walkable
2323   __ movptr(sp_temp, rsp);              // pass to next frame
2324   __ addptr(rsi, wordSize);             // Bump array pointer (sizes)
2325   __ addptr(rcx, wordSize);             // Bump array pointer (pcs)
2326   __ decrementl(counter);             // decrement counter
2327   __ jcc(Assembler::notZero, loop);
2328   __ pushptr(Address(rcx, 0));          // save final return address
2329 
2330   // Re-push self-frame
2331   __ enter();                           // save old & set new rbp,
2332 
2333   //  Return address and rbp, are in place
2334   // We'll push additional args later. Just allocate a full sized
2335   // register save area
2336   __ subptr(rsp, (frame_size_in_words-additional_words - 2) * wordSize);
2337 
2338   // Restore frame locals after moving the frame
2339   __ movptr(Address(rsp, RegisterSaver::raxOffset()*wordSize), rax);
2340   __ movptr(Address(rsp, RegisterSaver::rdxOffset()*wordSize), rdx);
2341   __ fstp_d(Address(rsp, RegisterSaver::fpResultOffset()*wordSize));   // Pop float stack and store in local
2342   if( UseSSE>=2 ) __ movdbl(Address(rsp, RegisterSaver::xmm0Offset()*wordSize), xmm0);
2343   if( UseSSE==1 ) __ movflt(Address(rsp, RegisterSaver::xmm0Offset()*wordSize), xmm0);
2344 
2345   // Set up the args to unpack_frame
2346 
2347   __ pushl(unpack_kind);                     // get the unpack_kind value
2348   __ get_thread(rcx);
2349   __ push(rcx);
2350 
2351   // set last_Java_sp, last_Java_fp
2352   __ set_last_Java_frame(rcx, noreg, rbp, nullptr, noreg);
2353 
2354   // Call C code.  Need thread but NOT official VM entry
2355   // crud.  We cannot block on this call, no GC can happen.  Call should
2356   // restore return values to their stack-slots with the new SP.
2357   __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, Deoptimization::unpack_frames)));
2358   // Set an oopmap for the call site
2359   oop_maps->add_gc_map( __ pc()-start, new OopMap( frame_size_in_words, 0 ));
2360 
2361   // rax, contains the return result type
2362   __ push(rax);
2363 
2364   __ get_thread(rcx);
2365   __ reset_last_Java_frame(rcx, false);
2366 
2367   // Collect return values
2368   __ movptr(rax,Address(rsp, (RegisterSaver::raxOffset() + additional_words + 1)*wordSize));
2369   __ movptr(rdx,Address(rsp, (RegisterSaver::rdxOffset() + additional_words + 1)*wordSize));
2370 
2371   // Clear floating point stack before returning to interpreter
2372   __ empty_FPU_stack();
2373 
2374   // Check if we should push the float or double return value.
2375   Label results_done, yes_double_value;
2376   __ cmpl(Address(rsp, 0), T_DOUBLE);
2377   __ jcc (Assembler::zero, yes_double_value);
2378   __ cmpl(Address(rsp, 0), T_FLOAT);
2379   __ jcc (Assembler::notZero, results_done);
2380 
2381   // return float value as expected by interpreter
2382   if( UseSSE>=1 ) __ movflt(xmm0, Address(rsp, (RegisterSaver::xmm0Offset() + additional_words + 1)*wordSize));
2383   else            __ fld_d(Address(rsp, (RegisterSaver::fpResultOffset() + additional_words + 1)*wordSize));
2384   __ jmp(results_done);
2385 
2386   // return double value as expected by interpreter
2387   __ bind(yes_double_value);
2388   if( UseSSE>=2 ) __ movdbl(xmm0, Address(rsp, (RegisterSaver::xmm0Offset() + additional_words + 1)*wordSize));
2389   else            __ fld_d(Address(rsp, (RegisterSaver::fpResultOffset() + additional_words + 1)*wordSize));
2390 
2391   __ bind(results_done);
2392 
2393   // Pop self-frame.
2394   __ leave();                              // Epilog!
2395 
2396   // Jump to interpreter
2397   __ ret(0);
2398 
2399   // -------------
2400   // make sure all code is generated
2401   masm->flush();
2402 
2403   _deopt_blob = DeoptimizationBlob::create( &buffer, oop_maps, 0, exception_offset, reexecute_offset, frame_size_in_words);
2404   _deopt_blob->set_unpack_with_exception_in_tls_offset(exception_in_tls_offset);
2405 }
2406 
2407 
2408 #ifdef COMPILER2
2409 //------------------------------generate_uncommon_trap_blob--------------------
2410 void SharedRuntime::generate_uncommon_trap_blob() {
2411   // allocate space for the code
2412   ResourceMark rm;
2413   // setup code generation tools
2414   CodeBuffer   buffer("uncommon_trap_blob", 512, 512);
2415   MacroAssembler* masm = new MacroAssembler(&buffer);
2416 
2417   enum frame_layout {
2418     arg0_off,      // thread                     sp + 0 // Arg location for
2419     arg1_off,      // unloaded_class_index       sp + 1 // calling C
2420     arg2_off,      // exec_mode                  sp + 2
2421     // The frame sender code expects that rbp will be in the "natural" place and
2422     // will override any oopMap setting for it. We must therefore force the layout
2423     // so that it agrees with the frame sender code.
2424     rbp_off,       // callee saved register      sp + 3
2425     return_off,    // slot for return address    sp + 4
2426     framesize
2427   };
2428 
2429   address start = __ pc();
2430 
2431   if (UseRTMLocking) {
2432     // Abort RTM transaction before possible nmethod deoptimization.
2433     __ xabort(0);
2434   }
2435 
2436   // Push self-frame.
2437   __ subptr(rsp, return_off*wordSize);     // Epilog!
2438 
2439   // rbp, is an implicitly saved callee saved register (i.e. the calling
2440   // convention will save restore it in prolog/epilog) Other than that
2441   // there are no callee save registers no that adapter frames are gone.
2442   __ movptr(Address(rsp, rbp_off*wordSize), rbp);
2443 
2444   // Clear the floating point exception stack
2445   __ empty_FPU_stack();
2446 
2447   // set last_Java_sp
2448   __ get_thread(rdx);
2449   __ set_last_Java_frame(rdx, noreg, noreg, nullptr, noreg);
2450 
2451   // Call C code.  Need thread but NOT official VM entry
2452   // crud.  We cannot block on this call, no GC can happen.  Call should
2453   // capture callee-saved registers as well as return values.
2454   __ movptr(Address(rsp, arg0_off*wordSize), rdx);
2455   // argument already in ECX
2456   __ movl(Address(rsp, arg1_off*wordSize),rcx);
2457   __ movl(Address(rsp, arg2_off*wordSize), Deoptimization::Unpack_uncommon_trap);
2458   __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, Deoptimization::uncommon_trap)));
2459 
2460   // Set an oopmap for the call site
2461   OopMapSet *oop_maps = new OopMapSet();
2462   OopMap* map =  new OopMap( framesize, 0 );
2463   // No oopMap for rbp, it is known implicitly
2464 
2465   oop_maps->add_gc_map( __ pc()-start, map);
2466 
2467   __ get_thread(rcx);
2468 
2469   __ reset_last_Java_frame(rcx, false);
2470 
2471   // Load UnrollBlock into EDI
2472   __ movptr(rdi, rax);
2473 
2474 #ifdef ASSERT
2475   { Label L;
2476     __ cmpptr(Address(rdi, Deoptimization::UnrollBlock::unpack_kind_offset()),
2477             (int32_t)Deoptimization::Unpack_uncommon_trap);
2478     __ jcc(Assembler::equal, L);
2479     __ stop("SharedRuntime::generate_uncommon_trap_blob: expected Unpack_uncommon_trap");
2480     __ bind(L);
2481   }
2482 #endif
2483 
2484   // Pop all the frames we must move/replace.
2485   //
2486   // Frame picture (youngest to oldest)
2487   // 1: self-frame (no frame link)
2488   // 2: deopting frame  (no frame link)
2489   // 3: caller of deopting frame (could be compiled/interpreted).
2490 
2491   // Pop self-frame.  We have no frame, and must rely only on EAX and ESP.
2492   __ addptr(rsp,(framesize-1)*wordSize);     // Epilog!
2493 
2494   // Pop deoptimized frame
2495   __ movl2ptr(rcx, Address(rdi,Deoptimization::UnrollBlock::size_of_deoptimized_frame_offset()));
2496   __ addptr(rsp, rcx);
2497 
2498   // sp should be pointing at the return address to the caller (3)
2499 
2500   // Pick up the initial fp we should save
2501   // restore rbp before stack bang because if stack overflow is thrown it needs to be pushed (and preserved)
2502   __ movptr(rbp, Address(rdi, Deoptimization::UnrollBlock::initial_info_offset()));
2503 
2504 #ifdef ASSERT
2505   // Compilers generate code that bang the stack by as much as the
2506   // interpreter would need. So this stack banging should never
2507   // trigger a fault. Verify that it does not on non product builds.
2508   __ movl(rbx, Address(rdi ,Deoptimization::UnrollBlock::total_frame_sizes_offset()));
2509   __ bang_stack_size(rbx, rcx);
2510 #endif
2511 
2512   // Load array of frame pcs into ECX
2513   __ movl(rcx,Address(rdi,Deoptimization::UnrollBlock::frame_pcs_offset()));
2514 
2515   __ pop(rsi); // trash the pc
2516 
2517   // Load array of frame sizes into ESI
2518   __ movptr(rsi,Address(rdi,Deoptimization::UnrollBlock::frame_sizes_offset()));
2519 
2520   Address counter(rdi, Deoptimization::UnrollBlock::counter_temp_offset());
2521 
2522   __ movl(rbx, Address(rdi, Deoptimization::UnrollBlock::number_of_frames_offset()));
2523   __ movl(counter, rbx);
2524 
2525   // Now adjust the caller's stack to make up for the extra locals
2526   // but record the original sp so that we can save it in the skeletal interpreter
2527   // frame and the stack walking of interpreter_sender will get the unextended sp
2528   // value and not the "real" sp value.
2529 
2530   Address sp_temp(rdi, Deoptimization::UnrollBlock::sender_sp_temp_offset());
2531   __ movptr(sp_temp, rsp);
2532   __ movl(rbx, Address(rdi, Deoptimization::UnrollBlock::caller_adjustment_offset()));
2533   __ subptr(rsp, rbx);
2534 
2535   // Push interpreter frames in a loop
2536   Label loop;
2537   __ bind(loop);
2538   __ movptr(rbx, Address(rsi, 0));      // Load frame size
2539   __ subptr(rbx, 2*wordSize);           // we'll push pc and rbp, by hand
2540   __ pushptr(Address(rcx, 0));          // save return address
2541   __ enter();                           // save old & set new rbp,
2542   __ subptr(rsp, rbx);                  // Prolog!
2543   __ movptr(rbx, sp_temp);              // sender's sp
2544   // This value is corrected by layout_activation_impl
2545   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD );
2546   __ movptr(Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize), rbx); // Make it walkable
2547   __ movptr(sp_temp, rsp);              // pass to next frame
2548   __ addptr(rsi, wordSize);             // Bump array pointer (sizes)
2549   __ addptr(rcx, wordSize);             // Bump array pointer (pcs)
2550   __ decrementl(counter);             // decrement counter
2551   __ jcc(Assembler::notZero, loop);
2552   __ pushptr(Address(rcx, 0));            // save final return address
2553 
2554   // Re-push self-frame
2555   __ enter();                           // save old & set new rbp,
2556   __ subptr(rsp, (framesize-2) * wordSize);   // Prolog!
2557 
2558 
2559   // set last_Java_sp, last_Java_fp
2560   __ get_thread(rdi);
2561   __ set_last_Java_frame(rdi, noreg, rbp, nullptr, noreg);
2562 
2563   // Call C code.  Need thread but NOT official VM entry
2564   // crud.  We cannot block on this call, no GC can happen.  Call should
2565   // restore return values to their stack-slots with the new SP.
2566   __ movptr(Address(rsp,arg0_off*wordSize),rdi);
2567   __ movl(Address(rsp,arg1_off*wordSize), Deoptimization::Unpack_uncommon_trap);
2568   __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, Deoptimization::unpack_frames)));
2569   // Set an oopmap for the call site
2570   oop_maps->add_gc_map( __ pc()-start, new OopMap( framesize, 0 ) );
2571 
2572   __ get_thread(rdi);
2573   __ reset_last_Java_frame(rdi, true);
2574 
2575   // Pop self-frame.
2576   __ leave();     // Epilog!
2577 
2578   // Jump to interpreter
2579   __ ret(0);
2580 
2581   // -------------
2582   // make sure all code is generated
2583   masm->flush();
2584 
2585    _uncommon_trap_blob = UncommonTrapBlob::create(&buffer, oop_maps, framesize);
2586 }
2587 #endif // COMPILER2
2588 
2589 //------------------------------generate_handler_blob------
2590 //
2591 // Generate a special Compile2Runtime blob that saves all registers,
2592 // setup oopmap, and calls safepoint code to stop the compiled code for
2593 // a safepoint.
2594 //
2595 SafepointBlob* SharedRuntime::generate_handler_blob(address call_ptr, int poll_type) {
2596 
2597   // Account for thread arg in our frame
2598   const int additional_words = 1;
2599   int frame_size_in_words;
2600 
2601   assert (StubRoutines::forward_exception_entry() != nullptr, "must be generated before");
2602 
2603   ResourceMark rm;
2604   OopMapSet *oop_maps = new OopMapSet();
2605   OopMap* map;
2606 
2607   // allocate space for the code
2608   // setup code generation tools
2609   CodeBuffer   buffer("handler_blob", 2048, 1024);
2610   MacroAssembler* masm = new MacroAssembler(&buffer);
2611 
2612   const Register java_thread = rdi; // callee-saved for VC++
2613   address start   = __ pc();
2614   address call_pc = nullptr;
2615   bool cause_return = (poll_type == POLL_AT_RETURN);
2616   bool save_vectors = (poll_type == POLL_AT_VECTOR_LOOP);
2617 
2618   if (UseRTMLocking) {
2619     // Abort RTM transaction before calling runtime
2620     // because critical section will be large and will be
2621     // aborted anyway. Also nmethod could be deoptimized.
2622     __ xabort(0);
2623   }
2624 
2625   // If cause_return is true we are at a poll_return and there is
2626   // the return address on the stack to the caller on the nmethod
2627   // that is safepoint. We can leave this return on the stack and
2628   // effectively complete the return and safepoint in the caller.
2629   // Otherwise we push space for a return address that the safepoint
2630   // handler will install later to make the stack walking sensible.
2631   if (!cause_return)
2632     __ push(rbx);  // Make room for return address (or push it again)
2633 
2634   map = RegisterSaver::save_live_registers(masm, additional_words, &frame_size_in_words, false, save_vectors);
2635 
2636   // The following is basically a call_VM. However, we need the precise
2637   // address of the call in order to generate an oopmap. Hence, we do all the
2638   // work ourselves.
2639 
2640   // Push thread argument and setup last_Java_sp
2641   __ get_thread(java_thread);
2642   __ push(java_thread);
2643   __ set_last_Java_frame(java_thread, noreg, noreg, nullptr, noreg);
2644 
2645   // if this was not a poll_return then we need to correct the return address now.
2646   if (!cause_return) {
2647     // Get the return pc saved by the signal handler and stash it in its appropriate place on the stack.
2648     // Additionally, rbx is a callee saved register and we can look at it later to determine
2649     // if someone changed the return address for us!
2650     __ movptr(rbx, Address(java_thread, JavaThread::saved_exception_pc_offset()));
2651     __ movptr(Address(rbp, wordSize), rbx);
2652   }
2653 
2654   // do the call
2655   __ call(RuntimeAddress(call_ptr));
2656 
2657   // Set an oopmap for the call site.  This oopmap will map all
2658   // oop-registers and debug-info registers as callee-saved.  This
2659   // will allow deoptimization at this safepoint to find all possible
2660   // debug-info recordings, as well as let GC find all oops.
2661 
2662   oop_maps->add_gc_map( __ pc() - start, map);
2663 
2664   // Discard arg
2665   __ pop(rcx);
2666 
2667   Label noException;
2668 
2669   // Clear last_Java_sp again
2670   __ get_thread(java_thread);
2671   __ reset_last_Java_frame(java_thread, false);
2672 
2673   __ cmpptr(Address(java_thread, Thread::pending_exception_offset()), NULL_WORD);
2674   __ jcc(Assembler::equal, noException);
2675 
2676   // Exception pending
2677   RegisterSaver::restore_live_registers(masm, save_vectors);
2678 
2679   __ jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
2680 
2681   __ bind(noException);
2682 
2683   Label no_adjust, bail, not_special;
2684   if (!cause_return) {
2685     // If our stashed return pc was modified by the runtime we avoid touching it
2686     __ cmpptr(rbx, Address(rbp, wordSize));
2687     __ jccb(Assembler::notEqual, no_adjust);
2688 
2689     // Skip over the poll instruction.
2690     // See NativeInstruction::is_safepoint_poll()
2691     // Possible encodings:
2692     //      85 00       test   %eax,(%rax)
2693     //      85 01       test   %eax,(%rcx)
2694     //      85 02       test   %eax,(%rdx)
2695     //      85 03       test   %eax,(%rbx)
2696     //      85 06       test   %eax,(%rsi)
2697     //      85 07       test   %eax,(%rdi)
2698     //
2699     //      85 04 24    test   %eax,(%rsp)
2700     //      85 45 00    test   %eax,0x0(%rbp)
2701 
2702 #ifdef ASSERT
2703     __ movptr(rax, rbx); // remember where 0x85 should be, for verification below
2704 #endif
2705     // rsp/rbp base encoding takes 3 bytes with the following register values:
2706     // rsp 0x04
2707     // rbp 0x05
2708     __ movzbl(rcx, Address(rbx, 1));
2709     __ andptr(rcx, 0x07); // looking for 0x04 .. 0x05
2710     __ subptr(rcx, 4);    // looking for 0x00 .. 0x01
2711     __ cmpptr(rcx, 1);
2712     __ jcc(Assembler::above, not_special);
2713     __ addptr(rbx, 1);
2714     __ bind(not_special);
2715 #ifdef ASSERT
2716     // Verify the correct encoding of the poll we're about to skip.
2717     __ cmpb(Address(rax, 0), NativeTstRegMem::instruction_code_memXregl);
2718     __ jcc(Assembler::notEqual, bail);
2719     // Mask out the modrm bits
2720     __ testb(Address(rax, 1), NativeTstRegMem::modrm_mask);
2721     // rax encodes to 0, so if the bits are nonzero it's incorrect
2722     __ jcc(Assembler::notZero, bail);
2723 #endif
2724     // Adjust return pc forward to step over the safepoint poll instruction
2725     __ addptr(rbx, 2);
2726     __ movptr(Address(rbp, wordSize), rbx);
2727   }
2728 
2729   __ bind(no_adjust);
2730   // Normal exit, register restoring and exit
2731   RegisterSaver::restore_live_registers(masm, save_vectors);
2732 
2733   __ ret(0);
2734 
2735 #ifdef ASSERT
2736   __ bind(bail);
2737   __ stop("Attempting to adjust pc to skip safepoint poll but the return point is not what we expected");
2738 #endif
2739 
2740   // make sure all code is generated
2741   masm->flush();
2742 
2743   // Fill-out other meta info
2744   return SafepointBlob::create(&buffer, oop_maps, frame_size_in_words);
2745 }
2746 
2747 //
2748 // generate_resolve_blob - call resolution (static/virtual/opt-virtual/ic-miss
2749 //
2750 // Generate a stub that calls into vm to find out the proper destination
2751 // of a java call. All the argument registers are live at this point
2752 // but since this is generic code we don't know what they are and the caller
2753 // must do any gc of the args.
2754 //
2755 RuntimeStub* SharedRuntime::generate_resolve_blob(address destination, const char* name) {
2756   assert (StubRoutines::forward_exception_entry() != nullptr, "must be generated before");
2757 
2758   // allocate space for the code
2759   ResourceMark rm;
2760 
2761   CodeBuffer buffer(name, 1000, 512);
2762   MacroAssembler* masm                = new MacroAssembler(&buffer);
2763 
2764   int frame_size_words;
2765   enum frame_layout {
2766                 thread_off,
2767                 extra_words };
2768 
2769   OopMapSet *oop_maps = new OopMapSet();
2770   OopMap* map = nullptr;
2771 
2772   int start = __ offset();
2773 
2774   map = RegisterSaver::save_live_registers(masm, extra_words, &frame_size_words);
2775 
2776   int frame_complete = __ offset();
2777 
2778   const Register thread = rdi;
2779   __ get_thread(rdi);
2780 
2781   __ push(thread);
2782   __ set_last_Java_frame(thread, noreg, rbp, nullptr, noreg);
2783 
2784   __ call(RuntimeAddress(destination));
2785 
2786 
2787   // Set an oopmap for the call site.
2788   // We need this not only for callee-saved registers, but also for volatile
2789   // registers that the compiler might be keeping live across a safepoint.
2790 
2791   oop_maps->add_gc_map( __ offset() - start, map);
2792 
2793   // rax, contains the address we are going to jump to assuming no exception got installed
2794 
2795   __ addptr(rsp, wordSize);
2796 
2797   // clear last_Java_sp
2798   __ reset_last_Java_frame(thread, true);
2799   // check for pending exceptions
2800   Label pending;
2801   __ cmpptr(Address(thread, Thread::pending_exception_offset()), NULL_WORD);
2802   __ jcc(Assembler::notEqual, pending);
2803 
2804   // get the returned Method*
2805   __ get_vm_result_2(rbx, thread);
2806   __ movptr(Address(rsp, RegisterSaver::rbx_offset() * wordSize), rbx);
2807 
2808   __ movptr(Address(rsp, RegisterSaver::rax_offset() * wordSize), rax);
2809 
2810   RegisterSaver::restore_live_registers(masm);
2811 
2812   // We are back to the original state on entry and ready to go.
2813 
2814   __ jmp(rax);
2815 
2816   // Pending exception after the safepoint
2817 
2818   __ bind(pending);
2819 
2820   RegisterSaver::restore_live_registers(masm);
2821 
2822   // exception pending => remove activation and forward to exception handler
2823 
2824   __ get_thread(thread);
2825   __ movptr(Address(thread, JavaThread::vm_result_offset()), NULL_WORD);
2826   __ movptr(rax, Address(thread, Thread::pending_exception_offset()));
2827   __ jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
2828 
2829   // -------------
2830   // make sure all code is generated
2831   masm->flush();
2832 
2833   // return the  blob
2834   // frame_size_words or bytes??
2835   return RuntimeStub::new_runtime_stub(name, &buffer, frame_complete, frame_size_words, oop_maps, true);
2836 }