1 /*
   2  * Copyright (c) 2003, 2013, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 #include "precompiled.hpp"
  26 #include "asm/macroAssembler.hpp"
  27 #include "asm/macroAssembler.inline.hpp"
  28 #include "code/debugInfoRec.hpp"
  29 #include "code/icBuffer.hpp"
  30 #include "code/vtableStubs.hpp"
  31 #include "interpreter/interpreter.hpp"
  32 #include "oops/compiledICHolder.hpp"
  33 #include "prims/jvmtiRedefineClassesTrace.hpp"
  34 #include "runtime/sharedRuntime.hpp"
  35 #include "runtime/vframeArray.hpp"
  36 #include "vmreg_x86.inline.hpp"
  37 #ifdef COMPILER1
  38 #include "c1/c1_Runtime1.hpp"
  39 #endif
  40 #ifdef COMPILER2
  41 #include "opto/runtime.hpp"
  42 #endif
  43 
  44 #define __ masm->
  45 
  46 const int StackAlignmentInSlots = StackAlignmentInBytes / VMRegImpl::stack_slot_size;
  47 
  48 class SimpleRuntimeFrame {
  49 
  50   public:
  51 
  52   // Most of the runtime stubs have this simple frame layout.
  53   // This class exists to make the layout shared in one place.
  54   // Offsets are for compiler stack slots, which are jints.
  55   enum layout {
  56     // The frame sender code expects that rbp will be in the "natural" place and
  57     // will override any oopMap setting for it. We must therefore force the layout
  58     // so that it agrees with the frame sender code.
  59     rbp_off = frame::arg_reg_save_area_bytes/BytesPerInt,
  60     rbp_off2,
  61     return_off, return_off2,
  62     framesize
  63   };
  64 };
  65 
  66 class RegisterSaver {
  67   // Capture info about frame layout.  Layout offsets are in jint
  68   // units because compiler frame slots are jints.
  69 #define DEF_XMM_OFFS(regnum) xmm ## regnum ## _off = xmm_off + (regnum)*16/BytesPerInt, xmm ## regnum ## H_off
  70   enum layout {
  71     fpu_state_off = frame::arg_reg_save_area_bytes/BytesPerInt, // fxsave save area
  72     xmm_off       = fpu_state_off + 160/BytesPerInt,            // offset in fxsave save area
  73     DEF_XMM_OFFS(0),
  74     DEF_XMM_OFFS(1),
  75     DEF_XMM_OFFS(2),
  76     DEF_XMM_OFFS(3),
  77     DEF_XMM_OFFS(4),
  78     DEF_XMM_OFFS(5),
  79     DEF_XMM_OFFS(6),
  80     DEF_XMM_OFFS(7),
  81     DEF_XMM_OFFS(8),
  82     DEF_XMM_OFFS(9),
  83     DEF_XMM_OFFS(10),
  84     DEF_XMM_OFFS(11),
  85     DEF_XMM_OFFS(12),
  86     DEF_XMM_OFFS(13),
  87     DEF_XMM_OFFS(14),
  88     DEF_XMM_OFFS(15),
  89     fpu_state_end = fpu_state_off + ((FPUStateSizeInWords-1)*wordSize / BytesPerInt),
  90     fpu_stateH_end,
  91     r15_off, r15H_off,
  92     r14_off, r14H_off,
  93     r13_off, r13H_off,
  94     r12_off, r12H_off,
  95     r11_off, r11H_off,
  96     r10_off, r10H_off,
  97     r9_off,  r9H_off,
  98     r8_off,  r8H_off,
  99     rdi_off, rdiH_off,
 100     rsi_off, rsiH_off,
 101     ignore_off, ignoreH_off,  // extra copy of rbp
 102     rsp_off, rspH_off,
 103     rbx_off, rbxH_off,
 104     rdx_off, rdxH_off,
 105     rcx_off, rcxH_off,
 106     rax_off, raxH_off,
 107     // 16-byte stack alignment fill word: see MacroAssembler::push/pop_IU_state
 108     align_off, alignH_off,
 109     flags_off, flagsH_off,
 110     // The frame sender code expects that rbp will be in the "natural" place and
 111     // will override any oopMap setting for it. We must therefore force the layout
 112     // so that it agrees with the frame sender code.
 113     rbp_off, rbpH_off,        // copy of rbp we will restore
 114     return_off, returnH_off,  // slot for return address
 115     reg_save_size             // size in compiler stack slots
 116   };
 117 
 118  public:
 119   static OopMap* save_live_registers(MacroAssembler* masm, int additional_frame_words, int* total_frame_words, bool save_vectors = false);
 120   static void restore_live_registers(MacroAssembler* masm, bool restore_vectors = false);
 121 
 122   // Offsets into the register save area
 123   // Used by deoptimization when it is managing result register
 124   // values on its own
 125 
 126   static int rax_offset_in_bytes(void)    { return BytesPerInt * rax_off; }
 127   static int rdx_offset_in_bytes(void)    { return BytesPerInt * rdx_off; }
 128   static int rbx_offset_in_bytes(void)    { return BytesPerInt * rbx_off; }
 129   static int xmm0_offset_in_bytes(void)   { return BytesPerInt * xmm0_off; }
 130   static int return_offset_in_bytes(void) { return BytesPerInt * return_off; }
 131 
 132   // During deoptimization only the result registers need to be restored,
 133   // all the other values have already been extracted.
 134   static void restore_result_registers(MacroAssembler* masm);
 135 };
 136 
 137 OopMap* RegisterSaver::save_live_registers(MacroAssembler* masm, int additional_frame_words, int* total_frame_words, bool save_vectors) {
 138   int vect_words = 0;
 139 #ifdef COMPILER2
 140   if (save_vectors) {
 141     assert(UseAVX > 0, "256bit vectors are supported only with AVX");
 142     assert(MaxVectorSize == 32, "only 256bit vectors are supported now");
 143     // Save upper half of YMM registes
 144     vect_words = 16 * 16 / wordSize;
 145     additional_frame_words += vect_words;
 146   }
 147 #else
 148   assert(!save_vectors, "vectors are generated only by C2");
 149 #endif
 150 
 151   // Always make the frame size 16-byte aligned
 152   int frame_size_in_bytes = round_to(additional_frame_words*wordSize +
 153                                      reg_save_size*BytesPerInt, 16);
 154   // OopMap frame size is in compiler stack slots (jint's) not bytes or words
 155   int frame_size_in_slots = frame_size_in_bytes / BytesPerInt;
 156   // The caller will allocate additional_frame_words
 157   int additional_frame_slots = additional_frame_words*wordSize / BytesPerInt;
 158   // CodeBlob frame size is in words.
 159   int frame_size_in_words = frame_size_in_bytes / wordSize;
 160   *total_frame_words = frame_size_in_words;
 161 
 162   // Save registers, fpu state, and flags.
 163   // We assume caller has already pushed the return address onto the
 164   // stack, so rsp is 8-byte aligned here.
 165   // We push rpb twice in this sequence because we want the real rbp
 166   // to be under the return like a normal enter.
 167 
 168   __ enter();          // rsp becomes 16-byte aligned here
 169   __ push_CPU_state(); // Push a multiple of 16 bytes
 170 
 171   if (vect_words > 0) {
 172     assert(vect_words*wordSize == 256, "");
 173     __ subptr(rsp, 256); // Save upper half of YMM registes
 174     __ vextractf128h(Address(rsp,  0),xmm0);
 175     __ vextractf128h(Address(rsp, 16),xmm1);
 176     __ vextractf128h(Address(rsp, 32),xmm2);
 177     __ vextractf128h(Address(rsp, 48),xmm3);
 178     __ vextractf128h(Address(rsp, 64),xmm4);
 179     __ vextractf128h(Address(rsp, 80),xmm5);
 180     __ vextractf128h(Address(rsp, 96),xmm6);
 181     __ vextractf128h(Address(rsp,112),xmm7);
 182     __ vextractf128h(Address(rsp,128),xmm8);
 183     __ vextractf128h(Address(rsp,144),xmm9);
 184     __ vextractf128h(Address(rsp,160),xmm10);
 185     __ vextractf128h(Address(rsp,176),xmm11);
 186     __ vextractf128h(Address(rsp,192),xmm12);
 187     __ vextractf128h(Address(rsp,208),xmm13);
 188     __ vextractf128h(Address(rsp,224),xmm14);
 189     __ vextractf128h(Address(rsp,240),xmm15);
 190   }
 191   if (frame::arg_reg_save_area_bytes != 0) {
 192     // Allocate argument register save area
 193     __ subptr(rsp, frame::arg_reg_save_area_bytes);
 194   }
 195 
 196   // Set an oopmap for the call site.  This oopmap will map all
 197   // oop-registers and debug-info registers as callee-saved.  This
 198   // will allow deoptimization at this safepoint to find all possible
 199   // debug-info recordings, as well as let GC find all oops.
 200 
 201   OopMapSet *oop_maps = new OopMapSet();
 202   OopMap* map = new OopMap(frame_size_in_slots, 0);
 203 
 204 #define STACK_OFFSET(x) VMRegImpl::stack2reg((x) + additional_frame_slots)
 205 
 206   map->set_callee_saved(STACK_OFFSET( rax_off ), rax->as_VMReg());
 207   map->set_callee_saved(STACK_OFFSET( rcx_off ), rcx->as_VMReg());
 208   map->set_callee_saved(STACK_OFFSET( rdx_off ), rdx->as_VMReg());
 209   map->set_callee_saved(STACK_OFFSET( rbx_off ), rbx->as_VMReg());
 210   // rbp location is known implicitly by the frame sender code, needs no oopmap
 211   // and the location where rbp was saved by is ignored
 212   map->set_callee_saved(STACK_OFFSET( rsi_off ), rsi->as_VMReg());
 213   map->set_callee_saved(STACK_OFFSET( rdi_off ), rdi->as_VMReg());
 214   map->set_callee_saved(STACK_OFFSET( r8_off  ), r8->as_VMReg());
 215   map->set_callee_saved(STACK_OFFSET( r9_off  ), r9->as_VMReg());
 216   map->set_callee_saved(STACK_OFFSET( r10_off ), r10->as_VMReg());
 217   map->set_callee_saved(STACK_OFFSET( r11_off ), r11->as_VMReg());
 218   map->set_callee_saved(STACK_OFFSET( r12_off ), r12->as_VMReg());
 219   map->set_callee_saved(STACK_OFFSET( r13_off ), r13->as_VMReg());
 220   map->set_callee_saved(STACK_OFFSET( r14_off ), r14->as_VMReg());
 221   map->set_callee_saved(STACK_OFFSET( r15_off ), r15->as_VMReg());
 222   map->set_callee_saved(STACK_OFFSET(xmm0_off ), xmm0->as_VMReg());
 223   map->set_callee_saved(STACK_OFFSET(xmm1_off ), xmm1->as_VMReg());
 224   map->set_callee_saved(STACK_OFFSET(xmm2_off ), xmm2->as_VMReg());
 225   map->set_callee_saved(STACK_OFFSET(xmm3_off ), xmm3->as_VMReg());
 226   map->set_callee_saved(STACK_OFFSET(xmm4_off ), xmm4->as_VMReg());
 227   map->set_callee_saved(STACK_OFFSET(xmm5_off ), xmm5->as_VMReg());
 228   map->set_callee_saved(STACK_OFFSET(xmm6_off ), xmm6->as_VMReg());
 229   map->set_callee_saved(STACK_OFFSET(xmm7_off ), xmm7->as_VMReg());
 230   map->set_callee_saved(STACK_OFFSET(xmm8_off ), xmm8->as_VMReg());
 231   map->set_callee_saved(STACK_OFFSET(xmm9_off ), xmm9->as_VMReg());
 232   map->set_callee_saved(STACK_OFFSET(xmm10_off), xmm10->as_VMReg());
 233   map->set_callee_saved(STACK_OFFSET(xmm11_off), xmm11->as_VMReg());
 234   map->set_callee_saved(STACK_OFFSET(xmm12_off), xmm12->as_VMReg());
 235   map->set_callee_saved(STACK_OFFSET(xmm13_off), xmm13->as_VMReg());
 236   map->set_callee_saved(STACK_OFFSET(xmm14_off), xmm14->as_VMReg());
 237   map->set_callee_saved(STACK_OFFSET(xmm15_off), xmm15->as_VMReg());
 238 
 239   // %%% These should all be a waste but we'll keep things as they were for now
 240   if (true) {
 241     map->set_callee_saved(STACK_OFFSET( raxH_off ), rax->as_VMReg()->next());
 242     map->set_callee_saved(STACK_OFFSET( rcxH_off ), rcx->as_VMReg()->next());
 243     map->set_callee_saved(STACK_OFFSET( rdxH_off ), rdx->as_VMReg()->next());
 244     map->set_callee_saved(STACK_OFFSET( rbxH_off ), rbx->as_VMReg()->next());
 245     // rbp location is known implicitly by the frame sender code, needs no oopmap
 246     map->set_callee_saved(STACK_OFFSET( rsiH_off ), rsi->as_VMReg()->next());
 247     map->set_callee_saved(STACK_OFFSET( rdiH_off ), rdi->as_VMReg()->next());
 248     map->set_callee_saved(STACK_OFFSET( r8H_off  ), r8->as_VMReg()->next());
 249     map->set_callee_saved(STACK_OFFSET( r9H_off  ), r9->as_VMReg()->next());
 250     map->set_callee_saved(STACK_OFFSET( r10H_off ), r10->as_VMReg()->next());
 251     map->set_callee_saved(STACK_OFFSET( r11H_off ), r11->as_VMReg()->next());
 252     map->set_callee_saved(STACK_OFFSET( r12H_off ), r12->as_VMReg()->next());
 253     map->set_callee_saved(STACK_OFFSET( r13H_off ), r13->as_VMReg()->next());
 254     map->set_callee_saved(STACK_OFFSET( r14H_off ), r14->as_VMReg()->next());
 255     map->set_callee_saved(STACK_OFFSET( r15H_off ), r15->as_VMReg()->next());
 256     map->set_callee_saved(STACK_OFFSET(xmm0H_off ), xmm0->as_VMReg()->next());
 257     map->set_callee_saved(STACK_OFFSET(xmm1H_off ), xmm1->as_VMReg()->next());
 258     map->set_callee_saved(STACK_OFFSET(xmm2H_off ), xmm2->as_VMReg()->next());
 259     map->set_callee_saved(STACK_OFFSET(xmm3H_off ), xmm3->as_VMReg()->next());
 260     map->set_callee_saved(STACK_OFFSET(xmm4H_off ), xmm4->as_VMReg()->next());
 261     map->set_callee_saved(STACK_OFFSET(xmm5H_off ), xmm5->as_VMReg()->next());
 262     map->set_callee_saved(STACK_OFFSET(xmm6H_off ), xmm6->as_VMReg()->next());
 263     map->set_callee_saved(STACK_OFFSET(xmm7H_off ), xmm7->as_VMReg()->next());
 264     map->set_callee_saved(STACK_OFFSET(xmm8H_off ), xmm8->as_VMReg()->next());
 265     map->set_callee_saved(STACK_OFFSET(xmm9H_off ), xmm9->as_VMReg()->next());
 266     map->set_callee_saved(STACK_OFFSET(xmm10H_off), xmm10->as_VMReg()->next());
 267     map->set_callee_saved(STACK_OFFSET(xmm11H_off), xmm11->as_VMReg()->next());
 268     map->set_callee_saved(STACK_OFFSET(xmm12H_off), xmm12->as_VMReg()->next());
 269     map->set_callee_saved(STACK_OFFSET(xmm13H_off), xmm13->as_VMReg()->next());
 270     map->set_callee_saved(STACK_OFFSET(xmm14H_off), xmm14->as_VMReg()->next());
 271     map->set_callee_saved(STACK_OFFSET(xmm15H_off), xmm15->as_VMReg()->next());
 272   }
 273 
 274   return map;
 275 }
 276 
 277 void RegisterSaver::restore_live_registers(MacroAssembler* masm, bool restore_vectors) {
 278   if (frame::arg_reg_save_area_bytes != 0) {
 279     // Pop arg register save area
 280     __ addptr(rsp, frame::arg_reg_save_area_bytes);
 281   }
 282 #ifdef COMPILER2
 283   if (restore_vectors) {
 284     // Restore upper half of YMM registes.
 285     assert(UseAVX > 0, "256bit vectors are supported only with AVX");
 286     assert(MaxVectorSize == 32, "only 256bit vectors are supported now");
 287     __ vinsertf128h(xmm0, Address(rsp,  0));
 288     __ vinsertf128h(xmm1, Address(rsp, 16));
 289     __ vinsertf128h(xmm2, Address(rsp, 32));
 290     __ vinsertf128h(xmm3, Address(rsp, 48));
 291     __ vinsertf128h(xmm4, Address(rsp, 64));
 292     __ vinsertf128h(xmm5, Address(rsp, 80));
 293     __ vinsertf128h(xmm6, Address(rsp, 96));
 294     __ vinsertf128h(xmm7, Address(rsp,112));
 295     __ vinsertf128h(xmm8, Address(rsp,128));
 296     __ vinsertf128h(xmm9, Address(rsp,144));
 297     __ vinsertf128h(xmm10, Address(rsp,160));
 298     __ vinsertf128h(xmm11, Address(rsp,176));
 299     __ vinsertf128h(xmm12, Address(rsp,192));
 300     __ vinsertf128h(xmm13, Address(rsp,208));
 301     __ vinsertf128h(xmm14, Address(rsp,224));
 302     __ vinsertf128h(xmm15, Address(rsp,240));
 303     __ addptr(rsp, 256);
 304   }
 305 #else
 306   assert(!restore_vectors, "vectors are generated only by C2");
 307 #endif
 308   // Recover CPU state
 309   __ pop_CPU_state();
 310   // Get the rbp described implicitly by the calling convention (no oopMap)
 311   __ pop(rbp);
 312 }
 313 
 314 void RegisterSaver::restore_result_registers(MacroAssembler* masm) {
 315 
 316   // Just restore result register. Only used by deoptimization. By
 317   // now any callee save register that needs to be restored to a c2
 318   // caller of the deoptee has been extracted into the vframeArray
 319   // and will be stuffed into the c2i adapter we create for later
 320   // restoration so only result registers need to be restored here.
 321 
 322   // Restore fp result register
 323   __ movdbl(xmm0, Address(rsp, xmm0_offset_in_bytes()));
 324   // Restore integer result register
 325   __ movptr(rax, Address(rsp, rax_offset_in_bytes()));
 326   __ movptr(rdx, Address(rsp, rdx_offset_in_bytes()));
 327 
 328   // Pop all of the register save are off the stack except the return address
 329   __ addptr(rsp, return_offset_in_bytes());
 330 }
 331 
 332 // Is vector's size (in bytes) bigger than a size saved by default?
 333 // 16 bytes XMM registers are saved by default using fxsave/fxrstor instructions.
 334 bool SharedRuntime::is_wide_vector(int size) {
 335   return size > 16;
 336 }
 337 
 338 // The java_calling_convention describes stack locations as ideal slots on
 339 // a frame with no abi restrictions. Since we must observe abi restrictions
 340 // (like the placement of the register window) the slots must be biased by
 341 // the following value.
 342 static int reg2offset_in(VMReg r) {
 343   // Account for saved rbp and return address
 344   // This should really be in_preserve_stack_slots
 345   return (r->reg2stack() + 4) * VMRegImpl::stack_slot_size;
 346 }
 347 
 348 static int reg2offset_out(VMReg r) {
 349   return (r->reg2stack() + SharedRuntime::out_preserve_stack_slots()) * VMRegImpl::stack_slot_size;
 350 }
 351 
 352 // ---------------------------------------------------------------------------
 353 // Read the array of BasicTypes from a signature, and compute where the
 354 // arguments should go.  Values in the VMRegPair regs array refer to 4-byte
 355 // quantities.  Values less than VMRegImpl::stack0 are registers, those above
 356 // refer to 4-byte stack slots.  All stack slots are based off of the stack pointer
 357 // as framesizes are fixed.
 358 // VMRegImpl::stack0 refers to the first slot 0(sp).
 359 // and VMRegImpl::stack0+1 refers to the memory word 4-byes higher.  Register
 360 // up to RegisterImpl::number_of_registers) are the 64-bit
 361 // integer registers.
 362 
 363 // Note: the INPUTS in sig_bt are in units of Java argument words, which are
 364 // either 32-bit or 64-bit depending on the build.  The OUTPUTS are in 32-bit
 365 // units regardless of build. Of course for i486 there is no 64 bit build
 366 
 367 // The Java calling convention is a "shifted" version of the C ABI.
 368 // By skipping the first C ABI register we can call non-static jni methods
 369 // with small numbers of arguments without having to shuffle the arguments
 370 // at all. Since we control the java ABI we ought to at least get some
 371 // advantage out of it.
 372 
 373 int SharedRuntime::java_calling_convention(const BasicType *sig_bt,
 374                                            VMRegPair *regs,
 375                                            int total_args_passed,
 376                                            int is_outgoing) {
 377 
 378   // Create the mapping between argument positions and
 379   // registers.
 380   static const Register INT_ArgReg[Argument::n_int_register_parameters_j] = {
 381     j_rarg0, j_rarg1, j_rarg2, j_rarg3, j_rarg4, j_rarg5
 382   };
 383   static const XMMRegister FP_ArgReg[Argument::n_float_register_parameters_j] = {
 384     j_farg0, j_farg1, j_farg2, j_farg3,
 385     j_farg4, j_farg5, j_farg6, j_farg7
 386   };
 387 
 388 
 389   uint int_args = 0;
 390   uint fp_args = 0;
 391   uint stk_args = 0; // inc by 2 each time
 392 
 393   for (int i = 0; i < total_args_passed; i++) {
 394     switch (sig_bt[i]) {
 395     case T_BOOLEAN:
 396     case T_CHAR:
 397     case T_BYTE:
 398     case T_SHORT:
 399     case T_INT:
 400       if (int_args < Argument::n_int_register_parameters_j) {
 401         regs[i].set1(INT_ArgReg[int_args++]->as_VMReg());
 402       } else {
 403         regs[i].set1(VMRegImpl::stack2reg(stk_args));
 404         stk_args += 2;
 405       }
 406       break;
 407     case T_VOID:
 408       // halves of T_LONG or T_DOUBLE
 409       assert(i != 0 && (sig_bt[i - 1] == T_LONG || sig_bt[i - 1] == T_DOUBLE), "expecting half");
 410       regs[i].set_bad();
 411       break;
 412     case T_LONG:
 413       assert(sig_bt[i + 1] == T_VOID, "expecting half");
 414       // fall through
 415     case T_OBJECT:
 416     case T_ARRAY:
 417     case T_ADDRESS:
 418       if (int_args < Argument::n_int_register_parameters_j) {
 419         regs[i].set2(INT_ArgReg[int_args++]->as_VMReg());
 420       } else {
 421         regs[i].set2(VMRegImpl::stack2reg(stk_args));
 422         stk_args += 2;
 423       }
 424       break;
 425     case T_FLOAT:
 426       if (fp_args < Argument::n_float_register_parameters_j) {
 427         regs[i].set1(FP_ArgReg[fp_args++]->as_VMReg());
 428       } else {
 429         regs[i].set1(VMRegImpl::stack2reg(stk_args));
 430         stk_args += 2;
 431       }
 432       break;
 433     case T_DOUBLE:
 434       assert(sig_bt[i + 1] == T_VOID, "expecting half");
 435       if (fp_args < Argument::n_float_register_parameters_j) {
 436         regs[i].set2(FP_ArgReg[fp_args++]->as_VMReg());
 437       } else {
 438         regs[i].set2(VMRegImpl::stack2reg(stk_args));
 439         stk_args += 2;
 440       }
 441       break;
 442     default:
 443       ShouldNotReachHere();
 444       break;
 445     }
 446   }
 447 
 448   return round_to(stk_args, 2);
 449 }
 450 
 451 // Patch the callers callsite with entry to compiled code if it exists.
 452 static void patch_callers_callsite(MacroAssembler *masm) {
 453   Label L;
 454   __ cmpptr(Address(rbx, in_bytes(Method::code_offset())), (int32_t)NULL_WORD);
 455   __ jcc(Assembler::equal, L);
 456 
 457   // Save the current stack pointer
 458   __ mov(r13, rsp);
 459   // Schedule the branch target address early.
 460   // Call into the VM to patch the caller, then jump to compiled callee
 461   // rax isn't live so capture return address while we easily can
 462   __ movptr(rax, Address(rsp, 0));
 463 
 464   // align stack so push_CPU_state doesn't fault
 465   __ andptr(rsp, -(StackAlignmentInBytes));
 466   __ push_CPU_state();
 467 
 468   // VM needs caller's callsite
 469   // VM needs target method
 470   // This needs to be a long call since we will relocate this adapter to
 471   // the codeBuffer and it may not reach
 472 
 473   // Allocate argument register save area
 474   if (frame::arg_reg_save_area_bytes != 0) {
 475     __ subptr(rsp, frame::arg_reg_save_area_bytes);
 476   }
 477   __ mov(c_rarg0, rbx);
 478   __ mov(c_rarg1, rax);
 479   __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::fixup_callers_callsite)));
 480 
 481   // De-allocate argument register save area
 482   if (frame::arg_reg_save_area_bytes != 0) {
 483     __ addptr(rsp, frame::arg_reg_save_area_bytes);
 484   }
 485 
 486   __ pop_CPU_state();
 487   // restore sp
 488   __ mov(rsp, r13);
 489   __ bind(L);
 490 }
 491 
 492 
 493 static void gen_c2i_adapter(MacroAssembler *masm,
 494                             int total_args_passed,
 495                             int comp_args_on_stack,
 496                             const BasicType *sig_bt,
 497                             const VMRegPair *regs,
 498                             Label& skip_fixup) {
 499   // Before we get into the guts of the C2I adapter, see if we should be here
 500   // at all.  We've come from compiled code and are attempting to jump to the
 501   // interpreter, which means the caller made a static call to get here
 502   // (vcalls always get a compiled target if there is one).  Check for a
 503   // compiled target.  If there is one, we need to patch the caller's call.
 504   patch_callers_callsite(masm);
 505 
 506   __ bind(skip_fixup);
 507 
 508   // Since all args are passed on the stack, total_args_passed *
 509   // Interpreter::stackElementSize is the space we need. Plus 1 because
 510   // we also account for the return address location since
 511   // we store it first rather than hold it in rax across all the shuffling
 512 
 513   int extraspace = (total_args_passed * Interpreter::stackElementSize) + wordSize;
 514 
 515   // stack is aligned, keep it that way
 516   extraspace = round_to(extraspace, 2*wordSize);
 517 
 518   // Get return address
 519   __ pop(rax);
 520 
 521   // set senderSP value
 522   __ mov(r13, rsp);
 523 
 524   __ subptr(rsp, extraspace);
 525 
 526   // Store the return address in the expected location
 527   __ movptr(Address(rsp, 0), rax);
 528 
 529   // Now write the args into the outgoing interpreter space
 530   for (int i = 0; i < total_args_passed; i++) {
 531     if (sig_bt[i] == T_VOID) {
 532       assert(i > 0 && (sig_bt[i-1] == T_LONG || sig_bt[i-1] == T_DOUBLE), "missing half");
 533       continue;
 534     }
 535 
 536     // offset to start parameters
 537     int st_off   = (total_args_passed - i) * Interpreter::stackElementSize;
 538     int next_off = st_off - Interpreter::stackElementSize;
 539 
 540     // Say 4 args:
 541     // i   st_off
 542     // 0   32 T_LONG
 543     // 1   24 T_VOID
 544     // 2   16 T_OBJECT
 545     // 3    8 T_BOOL
 546     // -    0 return address
 547     //
 548     // However to make thing extra confusing. Because we can fit a long/double in
 549     // a single slot on a 64 bt vm and it would be silly to break them up, the interpreter
 550     // leaves one slot empty and only stores to a single slot. In this case the
 551     // slot that is occupied is the T_VOID slot. See I said it was confusing.
 552 
 553     VMReg r_1 = regs[i].first();
 554     VMReg r_2 = regs[i].second();
 555     if (!r_1->is_valid()) {
 556       assert(!r_2->is_valid(), "");
 557       continue;
 558     }
 559     if (r_1->is_stack()) {
 560       // memory to memory use rax
 561       int ld_off = r_1->reg2stack() * VMRegImpl::stack_slot_size + extraspace;
 562       if (!r_2->is_valid()) {
 563         // sign extend??
 564         __ movl(rax, Address(rsp, ld_off));
 565         __ movptr(Address(rsp, st_off), rax);
 566 
 567       } else {
 568 
 569         __ movq(rax, Address(rsp, ld_off));
 570 
 571         // Two VMREgs|OptoRegs can be T_OBJECT, T_ADDRESS, T_DOUBLE, T_LONG
 572         // T_DOUBLE and T_LONG use two slots in the interpreter
 573         if ( sig_bt[i] == T_LONG || sig_bt[i] == T_DOUBLE) {
 574           // ld_off == LSW, ld_off+wordSize == MSW
 575           // st_off == MSW, next_off == LSW
 576           __ movq(Address(rsp, next_off), rax);
 577 #ifdef ASSERT
 578           // Overwrite the unused slot with known junk
 579           __ mov64(rax, CONST64(0xdeadffffdeadaaaa));
 580           __ movptr(Address(rsp, st_off), rax);
 581 #endif /* ASSERT */
 582         } else {
 583           __ movq(Address(rsp, st_off), rax);
 584         }
 585       }
 586     } else if (r_1->is_Register()) {
 587       Register r = r_1->as_Register();
 588       if (!r_2->is_valid()) {
 589         // must be only an int (or less ) so move only 32bits to slot
 590         // why not sign extend??
 591         __ movl(Address(rsp, st_off), r);
 592       } else {
 593         // Two VMREgs|OptoRegs can be T_OBJECT, T_ADDRESS, T_DOUBLE, T_LONG
 594         // T_DOUBLE and T_LONG use two slots in the interpreter
 595         if ( sig_bt[i] == T_LONG || sig_bt[i] == T_DOUBLE) {
 596           // long/double in gpr
 597 #ifdef ASSERT
 598           // Overwrite the unused slot with known junk
 599           __ mov64(rax, CONST64(0xdeadffffdeadaaab));
 600           __ movptr(Address(rsp, st_off), rax);
 601 #endif /* ASSERT */
 602           __ movq(Address(rsp, next_off), r);
 603         } else {
 604           __ movptr(Address(rsp, st_off), r);
 605         }
 606       }
 607     } else {
 608       assert(r_1->is_XMMRegister(), "");
 609       if (!r_2->is_valid()) {
 610         // only a float use just part of the slot
 611         __ movflt(Address(rsp, st_off), r_1->as_XMMRegister());
 612       } else {
 613 #ifdef ASSERT
 614         // Overwrite the unused slot with known junk
 615         __ mov64(rax, CONST64(0xdeadffffdeadaaac));
 616         __ movptr(Address(rsp, st_off), rax);
 617 #endif /* ASSERT */
 618         __ movdbl(Address(rsp, next_off), r_1->as_XMMRegister());
 619       }
 620     }
 621   }
 622 
 623   // Schedule the branch target address early.
 624   __ movptr(rcx, Address(rbx, in_bytes(Method::interpreter_entry_offset())));
 625   __ jmp(rcx);
 626 }
 627 
 628 static void range_check(MacroAssembler* masm, Register pc_reg, Register temp_reg,
 629                         address code_start, address code_end,
 630                         Label& L_ok) {
 631   Label L_fail;
 632   __ lea(temp_reg, ExternalAddress(code_start));
 633   __ cmpptr(pc_reg, temp_reg);
 634   __ jcc(Assembler::belowEqual, L_fail);
 635   __ lea(temp_reg, ExternalAddress(code_end));
 636   __ cmpptr(pc_reg, temp_reg);
 637   __ jcc(Assembler::below, L_ok);
 638   __ bind(L_fail);
 639 }
 640 
 641 static void gen_i2c_adapter(MacroAssembler *masm,
 642                             int total_args_passed,
 643                             int comp_args_on_stack,
 644                             const BasicType *sig_bt,
 645                             const VMRegPair *regs) {
 646 
 647   // Note: r13 contains the senderSP on entry. We must preserve it since
 648   // we may do a i2c -> c2i transition if we lose a race where compiled
 649   // code goes non-entrant while we get args ready.
 650   // In addition we use r13 to locate all the interpreter args as
 651   // we must align the stack to 16 bytes on an i2c entry else we
 652   // lose alignment we expect in all compiled code and register
 653   // save code can segv when fxsave instructions find improperly
 654   // aligned stack pointer.
 655 
 656   // Adapters can be frameless because they do not require the caller
 657   // to perform additional cleanup work, such as correcting the stack pointer.
 658   // An i2c adapter is frameless because the *caller* frame, which is interpreted,
 659   // routinely repairs its own stack pointer (from interpreter_frame_last_sp),
 660   // even if a callee has modified the stack pointer.
 661   // A c2i adapter is frameless because the *callee* frame, which is interpreted,
 662   // routinely repairs its caller's stack pointer (from sender_sp, which is set
 663   // up via the senderSP register).
 664   // In other words, if *either* the caller or callee is interpreted, we can
 665   // get the stack pointer repaired after a call.
 666   // This is why c2i and i2c adapters cannot be indefinitely composed.
 667   // In particular, if a c2i adapter were to somehow call an i2c adapter,
 668   // both caller and callee would be compiled methods, and neither would
 669   // clean up the stack pointer changes performed by the two adapters.
 670   // If this happens, control eventually transfers back to the compiled
 671   // caller, but with an uncorrected stack, causing delayed havoc.
 672 
 673   // Pick up the return address
 674   __ movptr(rax, Address(rsp, 0));
 675 
 676   if (VerifyAdapterCalls &&
 677       (Interpreter::code() != NULL || StubRoutines::code1() != NULL)) {
 678     // So, let's test for cascading c2i/i2c adapters right now.
 679     //  assert(Interpreter::contains($return_addr) ||
 680     //         StubRoutines::contains($return_addr),
 681     //         "i2c adapter must return to an interpreter frame");
 682     __ block_comment("verify_i2c { ");
 683     Label L_ok;
 684     if (Interpreter::code() != NULL)
 685       range_check(masm, rax, r11,
 686                   Interpreter::code()->code_start(), Interpreter::code()->code_end(),
 687                   L_ok);
 688     if (StubRoutines::code1() != NULL)
 689       range_check(masm, rax, r11,
 690                   StubRoutines::code1()->code_begin(), StubRoutines::code1()->code_end(),
 691                   L_ok);
 692     if (StubRoutines::code2() != NULL)
 693       range_check(masm, rax, r11,
 694                   StubRoutines::code2()->code_begin(), StubRoutines::code2()->code_end(),
 695                   L_ok);
 696     const char* msg = "i2c adapter must return to an interpreter frame";
 697     __ block_comment(msg);
 698     __ stop(msg);
 699     __ bind(L_ok);
 700     __ block_comment("} verify_i2ce ");
 701   }
 702 
 703   // Must preserve original SP for loading incoming arguments because
 704   // we need to align the outgoing SP for compiled code.
 705   __ movptr(r11, rsp);
 706 
 707   // Cut-out for having no stack args.  Since up to 2 int/oop args are passed
 708   // in registers, we will occasionally have no stack args.
 709   int comp_words_on_stack = 0;
 710   if (comp_args_on_stack) {
 711     // Sig words on the stack are greater-than VMRegImpl::stack0.  Those in
 712     // registers are below.  By subtracting stack0, we either get a negative
 713     // number (all values in registers) or the maximum stack slot accessed.
 714 
 715     // Convert 4-byte c2 stack slots to words.
 716     comp_words_on_stack = round_to(comp_args_on_stack*VMRegImpl::stack_slot_size, wordSize)>>LogBytesPerWord;
 717     // Round up to miminum stack alignment, in wordSize
 718     comp_words_on_stack = round_to(comp_words_on_stack, 2);
 719     __ subptr(rsp, comp_words_on_stack * wordSize);
 720   }
 721 
 722 
 723   // Ensure compiled code always sees stack at proper alignment
 724   __ andptr(rsp, -16);
 725 
 726   // push the return address and misalign the stack that youngest frame always sees
 727   // as far as the placement of the call instruction
 728   __ push(rax);
 729 
 730   // Put saved SP in another register
 731   const Register saved_sp = rax;
 732   __ movptr(saved_sp, r11);
 733 
 734   // Will jump to the compiled code just as if compiled code was doing it.
 735   // Pre-load the register-jump target early, to schedule it better.
 736   __ movptr(r11, Address(rbx, in_bytes(Method::from_compiled_offset())));
 737 
 738   // Now generate the shuffle code.  Pick up all register args and move the
 739   // rest through the floating point stack top.
 740   for (int i = 0; i < total_args_passed; i++) {
 741     if (sig_bt[i] == T_VOID) {
 742       // Longs and doubles are passed in native word order, but misaligned
 743       // in the 32-bit build.
 744       assert(i > 0 && (sig_bt[i-1] == T_LONG || sig_bt[i-1] == T_DOUBLE), "missing half");
 745       continue;
 746     }
 747 
 748     // Pick up 0, 1 or 2 words from SP+offset.
 749 
 750     assert(!regs[i].second()->is_valid() || regs[i].first()->next() == regs[i].second(),
 751             "scrambled load targets?");
 752     // Load in argument order going down.
 753     int ld_off = (total_args_passed - i)*Interpreter::stackElementSize;
 754     // Point to interpreter value (vs. tag)
 755     int next_off = ld_off - Interpreter::stackElementSize;
 756     //
 757     //
 758     //
 759     VMReg r_1 = regs[i].first();
 760     VMReg r_2 = regs[i].second();
 761     if (!r_1->is_valid()) {
 762       assert(!r_2->is_valid(), "");
 763       continue;
 764     }
 765     if (r_1->is_stack()) {
 766       // Convert stack slot to an SP offset (+ wordSize to account for return address )
 767       int st_off = regs[i].first()->reg2stack()*VMRegImpl::stack_slot_size + wordSize;
 768 
 769       // We can use r13 as a temp here because compiled code doesn't need r13 as an input
 770       // and if we end up going thru a c2i because of a miss a reasonable value of r13
 771       // will be generated.
 772       if (!r_2->is_valid()) {
 773         // sign extend???
 774         __ movl(r13, Address(saved_sp, ld_off));
 775         __ movptr(Address(rsp, st_off), r13);
 776       } else {
 777         //
 778         // We are using two optoregs. This can be either T_OBJECT, T_ADDRESS, T_LONG, or T_DOUBLE
 779         // the interpreter allocates two slots but only uses one for thr T_LONG or T_DOUBLE case
 780         // So we must adjust where to pick up the data to match the interpreter.
 781         //
 782         // Interpreter local[n] == MSW, local[n+1] == LSW however locals
 783         // are accessed as negative so LSW is at LOW address
 784 
 785         // ld_off is MSW so get LSW
 786         const int offset = (sig_bt[i]==T_LONG||sig_bt[i]==T_DOUBLE)?
 787                            next_off : ld_off;
 788         __ movq(r13, Address(saved_sp, offset));
 789         // st_off is LSW (i.e. reg.first())
 790         __ movq(Address(rsp, st_off), r13);
 791       }
 792     } else if (r_1->is_Register()) {  // Register argument
 793       Register r = r_1->as_Register();
 794       assert(r != rax, "must be different");
 795       if (r_2->is_valid()) {
 796         //
 797         // We are using two VMRegs. This can be either T_OBJECT, T_ADDRESS, T_LONG, or T_DOUBLE
 798         // the interpreter allocates two slots but only uses one for thr T_LONG or T_DOUBLE case
 799         // So we must adjust where to pick up the data to match the interpreter.
 800 
 801         const int offset = (sig_bt[i]==T_LONG||sig_bt[i]==T_DOUBLE)?
 802                            next_off : ld_off;
 803 
 804         // this can be a misaligned move
 805         __ movq(r, Address(saved_sp, offset));
 806       } else {
 807         // sign extend and use a full word?
 808         __ movl(r, Address(saved_sp, ld_off));
 809       }
 810     } else {
 811       if (!r_2->is_valid()) {
 812         __ movflt(r_1->as_XMMRegister(), Address(saved_sp, ld_off));
 813       } else {
 814         __ movdbl(r_1->as_XMMRegister(), Address(saved_sp, next_off));
 815       }
 816     }
 817   }
 818 
 819   // 6243940 We might end up in handle_wrong_method if
 820   // the callee is deoptimized as we race thru here. If that
 821   // happens we don't want to take a safepoint because the
 822   // caller frame will look interpreted and arguments are now
 823   // "compiled" so it is much better to make this transition
 824   // invisible to the stack walking code. Unfortunately if
 825   // we try and find the callee by normal means a safepoint
 826   // is possible. So we stash the desired callee in the thread
 827   // and the vm will find there should this case occur.
 828 
 829   __ movptr(Address(r15_thread, JavaThread::callee_target_offset()), rbx);
 830 
 831   // put Method* where a c2i would expect should we end up there
 832   // only needed becaus eof c2 resolve stubs return Method* as a result in
 833   // rax
 834   __ mov(rax, rbx);
 835   __ jmp(r11);
 836 }
 837 
 838 // ---------------------------------------------------------------
 839 AdapterHandlerEntry* SharedRuntime::generate_i2c2i_adapters(MacroAssembler *masm,
 840                                                             int total_args_passed,
 841                                                             int comp_args_on_stack,
 842                                                             const BasicType *sig_bt,
 843                                                             const VMRegPair *regs,
 844                                                             AdapterFingerPrint* fingerprint) {
 845   address i2c_entry = __ pc();
 846 
 847   gen_i2c_adapter(masm, total_args_passed, comp_args_on_stack, sig_bt, regs);
 848 
 849   // -------------------------------------------------------------------------
 850   // Generate a C2I adapter.  On entry we know rbx holds the Method* during calls
 851   // to the interpreter.  The args start out packed in the compiled layout.  They
 852   // need to be unpacked into the interpreter layout.  This will almost always
 853   // require some stack space.  We grow the current (compiled) stack, then repack
 854   // the args.  We  finally end in a jump to the generic interpreter entry point.
 855   // On exit from the interpreter, the interpreter will restore our SP (lest the
 856   // compiled code, which relys solely on SP and not RBP, get sick).
 857 
 858   address c2i_unverified_entry = __ pc();
 859   Label skip_fixup;
 860   Label ok;
 861 
 862   Register holder = rax;
 863   Register receiver = j_rarg0;
 864   Register temp = rbx;
 865 
 866   {
 867     __ load_klass(temp, receiver);
 868     __ cmpptr(temp, Address(holder, CompiledICHolder::holder_klass_offset()));
 869     __ movptr(rbx, Address(holder, CompiledICHolder::holder_method_offset()));
 870     __ jcc(Assembler::equal, ok);
 871     __ jump(RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
 872 
 873     __ bind(ok);
 874     // Method might have been compiled since the call site was patched to
 875     // interpreted if that is the case treat it as a miss so we can get
 876     // the call site corrected.
 877     __ cmpptr(Address(rbx, in_bytes(Method::code_offset())), (int32_t)NULL_WORD);
 878     __ jcc(Assembler::equal, skip_fixup);
 879     __ jump(RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
 880   }
 881 
 882   address c2i_entry = __ pc();
 883 
 884   gen_c2i_adapter(masm, total_args_passed, comp_args_on_stack, sig_bt, regs, skip_fixup);
 885 
 886   __ flush();
 887   return AdapterHandlerLibrary::new_entry(fingerprint, i2c_entry, c2i_entry, c2i_unverified_entry);
 888 }
 889 
 890 int SharedRuntime::c_calling_convention(const BasicType *sig_bt,
 891                                          VMRegPair *regs,
 892                                          VMRegPair *regs2,
 893                                          int total_args_passed) {
 894   assert(regs2 == NULL, "not needed on x86");
 895 // We return the amount of VMRegImpl stack slots we need to reserve for all
 896 // the arguments NOT counting out_preserve_stack_slots.
 897 
 898 // NOTE: These arrays will have to change when c1 is ported
 899 #ifdef _WIN64
 900     static const Register INT_ArgReg[Argument::n_int_register_parameters_c] = {
 901       c_rarg0, c_rarg1, c_rarg2, c_rarg3
 902     };
 903     static const XMMRegister FP_ArgReg[Argument::n_float_register_parameters_c] = {
 904       c_farg0, c_farg1, c_farg2, c_farg3
 905     };
 906 #else
 907     static const Register INT_ArgReg[Argument::n_int_register_parameters_c] = {
 908       c_rarg0, c_rarg1, c_rarg2, c_rarg3, c_rarg4, c_rarg5
 909     };
 910     static const XMMRegister FP_ArgReg[Argument::n_float_register_parameters_c] = {
 911       c_farg0, c_farg1, c_farg2, c_farg3,
 912       c_farg4, c_farg5, c_farg6, c_farg7
 913     };
 914 #endif // _WIN64
 915 
 916 
 917     uint int_args = 0;
 918     uint fp_args = 0;
 919     uint stk_args = 0; // inc by 2 each time
 920 
 921     for (int i = 0; i < total_args_passed; i++) {
 922       switch (sig_bt[i]) {
 923       case T_BOOLEAN:
 924       case T_CHAR:
 925       case T_BYTE:
 926       case T_SHORT:
 927       case T_INT:
 928         if (int_args < Argument::n_int_register_parameters_c) {
 929           regs[i].set1(INT_ArgReg[int_args++]->as_VMReg());
 930 #ifdef _WIN64
 931           fp_args++;
 932           // Allocate slots for callee to stuff register args the stack.
 933           stk_args += 2;
 934 #endif
 935         } else {
 936           regs[i].set1(VMRegImpl::stack2reg(stk_args));
 937           stk_args += 2;
 938         }
 939         break;
 940       case T_LONG:
 941         assert(sig_bt[i + 1] == T_VOID, "expecting half");
 942         // fall through
 943       case T_OBJECT:
 944       case T_ARRAY:
 945       case T_ADDRESS:
 946       case T_METADATA:
 947         if (int_args < Argument::n_int_register_parameters_c) {
 948           regs[i].set2(INT_ArgReg[int_args++]->as_VMReg());
 949 #ifdef _WIN64
 950           fp_args++;
 951           stk_args += 2;
 952 #endif
 953         } else {
 954           regs[i].set2(VMRegImpl::stack2reg(stk_args));
 955           stk_args += 2;
 956         }
 957         break;
 958       case T_FLOAT:
 959         if (fp_args < Argument::n_float_register_parameters_c) {
 960           regs[i].set1(FP_ArgReg[fp_args++]->as_VMReg());
 961 #ifdef _WIN64
 962           int_args++;
 963           // Allocate slots for callee to stuff register args the stack.
 964           stk_args += 2;
 965 #endif
 966         } else {
 967           regs[i].set1(VMRegImpl::stack2reg(stk_args));
 968           stk_args += 2;
 969         }
 970         break;
 971       case T_DOUBLE:
 972         assert(sig_bt[i + 1] == T_VOID, "expecting half");
 973         if (fp_args < Argument::n_float_register_parameters_c) {
 974           regs[i].set2(FP_ArgReg[fp_args++]->as_VMReg());
 975 #ifdef _WIN64
 976           int_args++;
 977           // Allocate slots for callee to stuff register args the stack.
 978           stk_args += 2;
 979 #endif
 980         } else {
 981           regs[i].set2(VMRegImpl::stack2reg(stk_args));
 982           stk_args += 2;
 983         }
 984         break;
 985       case T_VOID: // Halves of longs and doubles
 986         assert(i != 0 && (sig_bt[i - 1] == T_LONG || sig_bt[i - 1] == T_DOUBLE), "expecting half");
 987         regs[i].set_bad();
 988         break;
 989       default:
 990         ShouldNotReachHere();
 991         break;
 992       }
 993     }
 994 #ifdef _WIN64
 995   // windows abi requires that we always allocate enough stack space
 996   // for 4 64bit registers to be stored down.
 997   if (stk_args < 8) {
 998     stk_args = 8;
 999   }
1000 #endif // _WIN64
1001 
1002   return stk_args;
1003 }
1004 
1005 // On 64 bit we will store integer like items to the stack as
1006 // 64 bits items (sparc abi) even though java would only store
1007 // 32bits for a parameter. On 32bit it will simply be 32 bits
1008 // So this routine will do 32->32 on 32bit and 32->64 on 64bit
1009 static void move32_64(MacroAssembler* masm, VMRegPair src, VMRegPair dst) {
1010   if (src.first()->is_stack()) {
1011     if (dst.first()->is_stack()) {
1012       // stack to stack
1013       __ movslq(rax, Address(rbp, reg2offset_in(src.first())));
1014       __ movq(Address(rsp, reg2offset_out(dst.first())), rax);
1015     } else {
1016       // stack to reg
1017       __ movslq(dst.first()->as_Register(), Address(rbp, reg2offset_in(src.first())));
1018     }
1019   } else if (dst.first()->is_stack()) {
1020     // reg to stack
1021     // Do we really have to sign extend???
1022     // __ movslq(src.first()->as_Register(), src.first()->as_Register());
1023     __ movq(Address(rsp, reg2offset_out(dst.first())), src.first()->as_Register());
1024   } else {
1025     // Do we really have to sign extend???
1026     // __ movslq(dst.first()->as_Register(), src.first()->as_Register());
1027     if (dst.first() != src.first()) {
1028       __ movq(dst.first()->as_Register(), src.first()->as_Register());
1029     }
1030   }
1031 }
1032 
1033 static void move_ptr(MacroAssembler* masm, VMRegPair src, VMRegPair dst) {
1034   if (src.first()->is_stack()) {
1035     if (dst.first()->is_stack()) {
1036       // stack to stack
1037       __ movq(rax, Address(rbp, reg2offset_in(src.first())));
1038       __ movq(Address(rsp, reg2offset_out(dst.first())), rax);
1039     } else {
1040       // stack to reg
1041       __ movq(dst.first()->as_Register(), Address(rbp, reg2offset_in(src.first())));
1042     }
1043   } else if (dst.first()->is_stack()) {
1044     // reg to stack
1045     __ movq(Address(rsp, reg2offset_out(dst.first())), src.first()->as_Register());
1046   } else {
1047     if (dst.first() != src.first()) {
1048       __ movq(dst.first()->as_Register(), src.first()->as_Register());
1049     }
1050   }
1051 }
1052 
1053 // An oop arg. Must pass a handle not the oop itself
1054 static void object_move(MacroAssembler* masm,
1055                         OopMap* map,
1056                         int oop_handle_offset,
1057                         int framesize_in_slots,
1058                         VMRegPair src,
1059                         VMRegPair dst,
1060                         bool is_receiver,
1061                         int* receiver_offset) {
1062 
1063   // must pass a handle. First figure out the location we use as a handle
1064 
1065   Register rHandle = dst.first()->is_stack() ? rax : dst.first()->as_Register();
1066 
1067   // See if oop is NULL if it is we need no handle
1068 
1069   if (src.first()->is_stack()) {
1070 
1071     // Oop is already on the stack as an argument
1072     int offset_in_older_frame = src.first()->reg2stack() + SharedRuntime::out_preserve_stack_slots();
1073     map->set_oop(VMRegImpl::stack2reg(offset_in_older_frame + framesize_in_slots));
1074     if (is_receiver) {
1075       *receiver_offset = (offset_in_older_frame + framesize_in_slots) * VMRegImpl::stack_slot_size;
1076     }
1077 
1078     __ cmpptr(Address(rbp, reg2offset_in(src.first())), (int32_t)NULL_WORD);
1079     __ lea(rHandle, Address(rbp, reg2offset_in(src.first())));
1080     // conditionally move a NULL
1081     __ cmovptr(Assembler::equal, rHandle, Address(rbp, reg2offset_in(src.first())));
1082   } else {
1083 
1084     // Oop is in an a register we must store it to the space we reserve
1085     // on the stack for oop_handles and pass a handle if oop is non-NULL
1086 
1087     const Register rOop = src.first()->as_Register();
1088     int oop_slot;
1089     if (rOop == j_rarg0)
1090       oop_slot = 0;
1091     else if (rOop == j_rarg1)
1092       oop_slot = 1;
1093     else if (rOop == j_rarg2)
1094       oop_slot = 2;
1095     else if (rOop == j_rarg3)
1096       oop_slot = 3;
1097     else if (rOop == j_rarg4)
1098       oop_slot = 4;
1099     else {
1100       assert(rOop == j_rarg5, "wrong register");
1101       oop_slot = 5;
1102     }
1103 
1104     oop_slot = oop_slot * VMRegImpl::slots_per_word + oop_handle_offset;
1105     int offset = oop_slot*VMRegImpl::stack_slot_size;
1106 
1107     map->set_oop(VMRegImpl::stack2reg(oop_slot));
1108     // Store oop in handle area, may be NULL
1109     __ movptr(Address(rsp, offset), rOop);
1110     if (is_receiver) {
1111       *receiver_offset = offset;
1112     }
1113 
1114     __ cmpptr(rOop, (int32_t)NULL_WORD);
1115     __ lea(rHandle, Address(rsp, offset));
1116     // conditionally move a NULL from the handle area where it was just stored
1117     __ cmovptr(Assembler::equal, rHandle, Address(rsp, offset));
1118   }
1119 
1120   // If arg is on the stack then place it otherwise it is already in correct reg.
1121   if (dst.first()->is_stack()) {
1122     __ movptr(Address(rsp, reg2offset_out(dst.first())), rHandle);
1123   }
1124 }
1125 
1126 // A float arg may have to do float reg int reg conversion
1127 static void float_move(MacroAssembler* masm, VMRegPair src, VMRegPair dst) {
1128   assert(!src.second()->is_valid() && !dst.second()->is_valid(), "bad float_move");
1129 
1130   // The calling conventions assures us that each VMregpair is either
1131   // all really one physical register or adjacent stack slots.
1132   // This greatly simplifies the cases here compared to sparc.
1133 
1134   if (src.first()->is_stack()) {
1135     if (dst.first()->is_stack()) {
1136       __ movl(rax, Address(rbp, reg2offset_in(src.first())));
1137       __ movptr(Address(rsp, reg2offset_out(dst.first())), rax);
1138     } else {
1139       // stack to reg
1140       assert(dst.first()->is_XMMRegister(), "only expect xmm registers as parameters");
1141       __ movflt(dst.first()->as_XMMRegister(), Address(rbp, reg2offset_in(src.first())));
1142     }
1143   } else if (dst.first()->is_stack()) {
1144     // reg to stack
1145     assert(src.first()->is_XMMRegister(), "only expect xmm registers as parameters");
1146     __ movflt(Address(rsp, reg2offset_out(dst.first())), src.first()->as_XMMRegister());
1147   } else {
1148     // reg to reg
1149     // In theory these overlap but the ordering is such that this is likely a nop
1150     if ( src.first() != dst.first()) {
1151       __ movdbl(dst.first()->as_XMMRegister(),  src.first()->as_XMMRegister());
1152     }
1153   }
1154 }
1155 
1156 // A long move
1157 static void long_move(MacroAssembler* masm, VMRegPair src, VMRegPair dst) {
1158 
1159   // The calling conventions assures us that each VMregpair is either
1160   // all really one physical register or adjacent stack slots.
1161   // This greatly simplifies the cases here compared to sparc.
1162 
1163   if (src.is_single_phys_reg() ) {
1164     if (dst.is_single_phys_reg()) {
1165       if (dst.first() != src.first()) {
1166         __ mov(dst.first()->as_Register(), src.first()->as_Register());
1167       }
1168     } else {
1169       assert(dst.is_single_reg(), "not a stack pair");
1170       __ movq(Address(rsp, reg2offset_out(dst.first())), src.first()->as_Register());
1171     }
1172   } else if (dst.is_single_phys_reg()) {
1173     assert(src.is_single_reg(),  "not a stack pair");
1174     __ movq(dst.first()->as_Register(), Address(rbp, reg2offset_out(src.first())));
1175   } else {
1176     assert(src.is_single_reg() && dst.is_single_reg(), "not stack pairs");
1177     __ movq(rax, Address(rbp, reg2offset_in(src.first())));
1178     __ movq(Address(rsp, reg2offset_out(dst.first())), rax);
1179   }
1180 }
1181 
1182 // A double move
1183 static void double_move(MacroAssembler* masm, VMRegPair src, VMRegPair dst) {
1184 
1185   // The calling conventions assures us that each VMregpair is either
1186   // all really one physical register or adjacent stack slots.
1187   // This greatly simplifies the cases here compared to sparc.
1188 
1189   if (src.is_single_phys_reg() ) {
1190     if (dst.is_single_phys_reg()) {
1191       // In theory these overlap but the ordering is such that this is likely a nop
1192       if ( src.first() != dst.first()) {
1193         __ movdbl(dst.first()->as_XMMRegister(), src.first()->as_XMMRegister());
1194       }
1195     } else {
1196       assert(dst.is_single_reg(), "not a stack pair");
1197       __ movdbl(Address(rsp, reg2offset_out(dst.first())), src.first()->as_XMMRegister());
1198     }
1199   } else if (dst.is_single_phys_reg()) {
1200     assert(src.is_single_reg(),  "not a stack pair");
1201     __ movdbl(dst.first()->as_XMMRegister(), Address(rbp, reg2offset_out(src.first())));
1202   } else {
1203     assert(src.is_single_reg() && dst.is_single_reg(), "not stack pairs");
1204     __ movq(rax, Address(rbp, reg2offset_in(src.first())));
1205     __ movq(Address(rsp, reg2offset_out(dst.first())), rax);
1206   }
1207 }
1208 
1209 
1210 void SharedRuntime::save_native_result(MacroAssembler *masm, BasicType ret_type, int frame_slots) {
1211   // We always ignore the frame_slots arg and just use the space just below frame pointer
1212   // which by this time is free to use
1213   switch (ret_type) {
1214   case T_FLOAT:
1215     __ movflt(Address(rbp, -wordSize), xmm0);
1216     break;
1217   case T_DOUBLE:
1218     __ movdbl(Address(rbp, -wordSize), xmm0);
1219     break;
1220   case T_VOID:  break;
1221   default: {
1222     __ movptr(Address(rbp, -wordSize), rax);
1223     }
1224   }
1225 }
1226 
1227 void SharedRuntime::restore_native_result(MacroAssembler *masm, BasicType ret_type, int frame_slots) {
1228   // We always ignore the frame_slots arg and just use the space just below frame pointer
1229   // which by this time is free to use
1230   switch (ret_type) {
1231   case T_FLOAT:
1232     __ movflt(xmm0, Address(rbp, -wordSize));
1233     break;
1234   case T_DOUBLE:
1235     __ movdbl(xmm0, Address(rbp, -wordSize));
1236     break;
1237   case T_VOID:  break;
1238   default: {
1239     __ movptr(rax, Address(rbp, -wordSize));
1240     }
1241   }
1242 }
1243 
1244 static void save_args(MacroAssembler *masm, int arg_count, int first_arg, VMRegPair *args) {
1245     for ( int i = first_arg ; i < arg_count ; i++ ) {
1246       if (args[i].first()->is_Register()) {
1247         __ push(args[i].first()->as_Register());
1248       } else if (args[i].first()->is_XMMRegister()) {
1249         __ subptr(rsp, 2*wordSize);
1250         __ movdbl(Address(rsp, 0), args[i].first()->as_XMMRegister());
1251       }
1252     }
1253 }
1254 
1255 static void restore_args(MacroAssembler *masm, int arg_count, int first_arg, VMRegPair *args) {
1256     for ( int i = arg_count - 1 ; i >= first_arg ; i-- ) {
1257       if (args[i].first()->is_Register()) {
1258         __ pop(args[i].first()->as_Register());
1259       } else if (args[i].first()->is_XMMRegister()) {
1260         __ movdbl(args[i].first()->as_XMMRegister(), Address(rsp, 0));
1261         __ addptr(rsp, 2*wordSize);
1262       }
1263     }
1264 }
1265 
1266 
1267 static void save_or_restore_arguments(MacroAssembler* masm,
1268                                       const int stack_slots,
1269                                       const int total_in_args,
1270                                       const int arg_save_area,
1271                                       OopMap* map,
1272                                       VMRegPair* in_regs,
1273                                       BasicType* in_sig_bt) {
1274   // if map is non-NULL then the code should store the values,
1275   // otherwise it should load them.
1276   int slot = arg_save_area;
1277   // Save down double word first
1278   for ( int i = 0; i < total_in_args; i++) {
1279     if (in_regs[i].first()->is_XMMRegister() && in_sig_bt[i] == T_DOUBLE) {
1280       int offset = slot * VMRegImpl::stack_slot_size;
1281       slot += VMRegImpl::slots_per_word;
1282       assert(slot <= stack_slots, "overflow");
1283       if (map != NULL) {
1284         __ movdbl(Address(rsp, offset), in_regs[i].first()->as_XMMRegister());
1285       } else {
1286         __ movdbl(in_regs[i].first()->as_XMMRegister(), Address(rsp, offset));
1287       }
1288     }
1289     if (in_regs[i].first()->is_Register() &&
1290         (in_sig_bt[i] == T_LONG || in_sig_bt[i] == T_ARRAY)) {
1291       int offset = slot * VMRegImpl::stack_slot_size;
1292       if (map != NULL) {
1293         __ movq(Address(rsp, offset), in_regs[i].first()->as_Register());
1294         if (in_sig_bt[i] == T_ARRAY) {
1295           map->set_oop(VMRegImpl::stack2reg(slot));;
1296         }
1297       } else {
1298         __ movq(in_regs[i].first()->as_Register(), Address(rsp, offset));
1299       }
1300       slot += VMRegImpl::slots_per_word;
1301     }
1302   }
1303   // Save or restore single word registers
1304   for ( int i = 0; i < total_in_args; i++) {
1305     if (in_regs[i].first()->is_Register()) {
1306       int offset = slot * VMRegImpl::stack_slot_size;
1307       slot++;
1308       assert(slot <= stack_slots, "overflow");
1309 
1310       // Value is in an input register pass we must flush it to the stack
1311       const Register reg = in_regs[i].first()->as_Register();
1312       switch (in_sig_bt[i]) {
1313         case T_BOOLEAN:
1314         case T_CHAR:
1315         case T_BYTE:
1316         case T_SHORT:
1317         case T_INT:
1318           if (map != NULL) {
1319             __ movl(Address(rsp, offset), reg);
1320           } else {
1321             __ movl(reg, Address(rsp, offset));
1322           }
1323           break;
1324         case T_ARRAY:
1325         case T_LONG:
1326           // handled above
1327           break;
1328         case T_OBJECT:
1329         default: ShouldNotReachHere();
1330       }
1331     } else if (in_regs[i].first()->is_XMMRegister()) {
1332       if (in_sig_bt[i] == T_FLOAT) {
1333         int offset = slot * VMRegImpl::stack_slot_size;
1334         slot++;
1335         assert(slot <= stack_slots, "overflow");
1336         if (map != NULL) {
1337           __ movflt(Address(rsp, offset), in_regs[i].first()->as_XMMRegister());
1338         } else {
1339           __ movflt(in_regs[i].first()->as_XMMRegister(), Address(rsp, offset));
1340         }
1341       }
1342     } else if (in_regs[i].first()->is_stack()) {
1343       if (in_sig_bt[i] == T_ARRAY && map != NULL) {
1344         int offset_in_older_frame = in_regs[i].first()->reg2stack() + SharedRuntime::out_preserve_stack_slots();
1345         map->set_oop(VMRegImpl::stack2reg(offset_in_older_frame + stack_slots));
1346       }
1347     }
1348   }
1349 }
1350 
1351 
1352 // Check GC_locker::needs_gc and enter the runtime if it's true.  This
1353 // keeps a new JNI critical region from starting until a GC has been
1354 // forced.  Save down any oops in registers and describe them in an
1355 // OopMap.
1356 static void check_needs_gc_for_critical_native(MacroAssembler* masm,
1357                                                int stack_slots,
1358                                                int total_c_args,
1359                                                int total_in_args,
1360                                                int arg_save_area,
1361                                                OopMapSet* oop_maps,
1362                                                VMRegPair* in_regs,
1363                                                BasicType* in_sig_bt) {
1364   __ block_comment("check GC_locker::needs_gc");
1365   Label cont;
1366   __ cmp8(ExternalAddress((address)GC_locker::needs_gc_address()), false);
1367   __ jcc(Assembler::equal, cont);
1368 
1369   // Save down any incoming oops and call into the runtime to halt for a GC
1370 
1371   OopMap* map = new OopMap(stack_slots * 2, 0 /* arg_slots*/);
1372   save_or_restore_arguments(masm, stack_slots, total_in_args,
1373                             arg_save_area, map, in_regs, in_sig_bt);
1374 
1375   address the_pc = __ pc();
1376   oop_maps->add_gc_map( __ offset(), map);
1377   __ set_last_Java_frame(rsp, noreg, the_pc);
1378 
1379   __ block_comment("block_for_jni_critical");
1380   __ movptr(c_rarg0, r15_thread);
1381   __ mov(r12, rsp); // remember sp
1382   __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1383   __ andptr(rsp, -16); // align stack as required by ABI
1384   __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::block_for_jni_critical)));
1385   __ mov(rsp, r12); // restore sp
1386   __ reinit_heapbase();
1387 
1388   __ reset_last_Java_frame(false, true);
1389 
1390   save_or_restore_arguments(masm, stack_slots, total_in_args,
1391                             arg_save_area, NULL, in_regs, in_sig_bt);
1392 
1393   __ bind(cont);
1394 #ifdef ASSERT
1395   if (StressCriticalJNINatives) {
1396     // Stress register saving
1397     OopMap* map = new OopMap(stack_slots * 2, 0 /* arg_slots*/);
1398     save_or_restore_arguments(masm, stack_slots, total_in_args,
1399                               arg_save_area, map, in_regs, in_sig_bt);
1400     // Destroy argument registers
1401     for (int i = 0; i < total_in_args - 1; i++) {
1402       if (in_regs[i].first()->is_Register()) {
1403         const Register reg = in_regs[i].first()->as_Register();
1404         __ xorptr(reg, reg);
1405       } else if (in_regs[i].first()->is_XMMRegister()) {
1406         __ xorpd(in_regs[i].first()->as_XMMRegister(), in_regs[i].first()->as_XMMRegister());
1407       } else if (in_regs[i].first()->is_FloatRegister()) {
1408         ShouldNotReachHere();
1409       } else if (in_regs[i].first()->is_stack()) {
1410         // Nothing to do
1411       } else {
1412         ShouldNotReachHere();
1413       }
1414       if (in_sig_bt[i] == T_LONG || in_sig_bt[i] == T_DOUBLE) {
1415         i++;
1416       }
1417     }
1418 
1419     save_or_restore_arguments(masm, stack_slots, total_in_args,
1420                               arg_save_area, NULL, in_regs, in_sig_bt);
1421   }
1422 #endif
1423 }
1424 
1425 // Unpack an array argument into a pointer to the body and the length
1426 // if the array is non-null, otherwise pass 0 for both.
1427 static void unpack_array_argument(MacroAssembler* masm, VMRegPair reg, BasicType in_elem_type, VMRegPair body_arg, VMRegPair length_arg) {
1428   Register tmp_reg = rax;
1429   assert(!body_arg.first()->is_Register() || body_arg.first()->as_Register() != tmp_reg,
1430          "possible collision");
1431   assert(!length_arg.first()->is_Register() || length_arg.first()->as_Register() != tmp_reg,
1432          "possible collision");
1433 
1434   __ block_comment("unpack_array_argument {");
1435 
1436   // Pass the length, ptr pair
1437   Label is_null, done;
1438   VMRegPair tmp;
1439   tmp.set_ptr(tmp_reg->as_VMReg());
1440   if (reg.first()->is_stack()) {
1441     // Load the arg up from the stack
1442     move_ptr(masm, reg, tmp);
1443     reg = tmp;
1444   }
1445   __ testptr(reg.first()->as_Register(), reg.first()->as_Register());
1446   __ jccb(Assembler::equal, is_null);
1447   __ lea(tmp_reg, Address(reg.first()->as_Register(), arrayOopDesc::base_offset_in_bytes(in_elem_type)));
1448   move_ptr(masm, tmp, body_arg);
1449   // load the length relative to the body.
1450   __ movl(tmp_reg, Address(tmp_reg, arrayOopDesc::length_offset_in_bytes() -
1451                            arrayOopDesc::base_offset_in_bytes(in_elem_type)));
1452   move32_64(masm, tmp, length_arg);
1453   __ jmpb(done);
1454   __ bind(is_null);
1455   // Pass zeros
1456   __ xorptr(tmp_reg, tmp_reg);
1457   move_ptr(masm, tmp, body_arg);
1458   move32_64(masm, tmp, length_arg);
1459   __ bind(done);
1460 
1461   __ block_comment("} unpack_array_argument");
1462 }
1463 
1464 
1465 // Different signatures may require very different orders for the move
1466 // to avoid clobbering other arguments.  There's no simple way to
1467 // order them safely.  Compute a safe order for issuing stores and
1468 // break any cycles in those stores.  This code is fairly general but
1469 // it's not necessary on the other platforms so we keep it in the
1470 // platform dependent code instead of moving it into a shared file.
1471 // (See bugs 7013347 & 7145024.)
1472 // Note that this code is specific to LP64.
1473 class ComputeMoveOrder: public StackObj {
1474   class MoveOperation: public ResourceObj {
1475     friend class ComputeMoveOrder;
1476    private:
1477     VMRegPair        _src;
1478     VMRegPair        _dst;
1479     int              _src_index;
1480     int              _dst_index;
1481     bool             _processed;
1482     MoveOperation*  _next;
1483     MoveOperation*  _prev;
1484 
1485     static int get_id(VMRegPair r) {
1486       return r.first()->value();
1487     }
1488 
1489    public:
1490     MoveOperation(int src_index, VMRegPair src, int dst_index, VMRegPair dst):
1491       _src(src)
1492     , _src_index(src_index)
1493     , _dst(dst)
1494     , _dst_index(dst_index)
1495     , _next(NULL)
1496     , _prev(NULL)
1497     , _processed(false) {
1498     }
1499 
1500     VMRegPair src() const              { return _src; }
1501     int src_id() const                 { return get_id(src()); }
1502     int src_index() const              { return _src_index; }
1503     VMRegPair dst() const              { return _dst; }
1504     void set_dst(int i, VMRegPair dst) { _dst_index = i, _dst = dst; }
1505     int dst_index() const              { return _dst_index; }
1506     int dst_id() const                 { return get_id(dst()); }
1507     MoveOperation* next() const       { return _next; }
1508     MoveOperation* prev() const       { return _prev; }
1509     void set_processed()               { _processed = true; }
1510     bool is_processed() const          { return _processed; }
1511 
1512     // insert
1513     void break_cycle(VMRegPair temp_register) {
1514       // create a new store following the last store
1515       // to move from the temp_register to the original
1516       MoveOperation* new_store = new MoveOperation(-1, temp_register, dst_index(), dst());
1517 
1518       // break the cycle of links and insert new_store at the end
1519       // break the reverse link.
1520       MoveOperation* p = prev();
1521       assert(p->next() == this, "must be");
1522       _prev = NULL;
1523       p->_next = new_store;
1524       new_store->_prev = p;
1525 
1526       // change the original store to save it's value in the temp.
1527       set_dst(-1, temp_register);
1528     }
1529 
1530     void link(GrowableArray<MoveOperation*>& killer) {
1531       // link this store in front the store that it depends on
1532       MoveOperation* n = killer.at_grow(src_id(), NULL);
1533       if (n != NULL) {
1534         assert(_next == NULL && n->_prev == NULL, "shouldn't have been set yet");
1535         _next = n;
1536         n->_prev = this;
1537       }
1538     }
1539   };
1540 
1541  private:
1542   GrowableArray<MoveOperation*> edges;
1543 
1544  public:
1545   ComputeMoveOrder(int total_in_args, VMRegPair* in_regs, int total_c_args, VMRegPair* out_regs,
1546                     BasicType* in_sig_bt, GrowableArray<int>& arg_order, VMRegPair tmp_vmreg) {
1547     // Move operations where the dest is the stack can all be
1548     // scheduled first since they can't interfere with the other moves.
1549     for (int i = total_in_args - 1, c_arg = total_c_args - 1; i >= 0; i--, c_arg--) {
1550       if (in_sig_bt[i] == T_ARRAY) {
1551         c_arg--;
1552         if (out_regs[c_arg].first()->is_stack() &&
1553             out_regs[c_arg + 1].first()->is_stack()) {
1554           arg_order.push(i);
1555           arg_order.push(c_arg);
1556         } else {
1557           if (out_regs[c_arg].first()->is_stack() ||
1558               in_regs[i].first() == out_regs[c_arg].first()) {
1559             add_edge(i, in_regs[i].first(), c_arg, out_regs[c_arg + 1]);
1560           } else {
1561             add_edge(i, in_regs[i].first(), c_arg, out_regs[c_arg]);
1562           }
1563         }
1564       } else if (in_sig_bt[i] == T_VOID) {
1565         arg_order.push(i);
1566         arg_order.push(c_arg);
1567       } else {
1568         if (out_regs[c_arg].first()->is_stack() ||
1569             in_regs[i].first() == out_regs[c_arg].first()) {
1570           arg_order.push(i);
1571           arg_order.push(c_arg);
1572         } else {
1573           add_edge(i, in_regs[i].first(), c_arg, out_regs[c_arg]);
1574         }
1575       }
1576     }
1577     // Break any cycles in the register moves and emit the in the
1578     // proper order.
1579     GrowableArray<MoveOperation*>* stores = get_store_order(tmp_vmreg);
1580     for (int i = 0; i < stores->length(); i++) {
1581       arg_order.push(stores->at(i)->src_index());
1582       arg_order.push(stores->at(i)->dst_index());
1583     }
1584  }
1585 
1586   // Collected all the move operations
1587   void add_edge(int src_index, VMRegPair src, int dst_index, VMRegPair dst) {
1588     if (src.first() == dst.first()) return;
1589     edges.append(new MoveOperation(src_index, src, dst_index, dst));
1590   }
1591 
1592   // Walk the edges breaking cycles between moves.  The result list
1593   // can be walked in order to produce the proper set of loads
1594   GrowableArray<MoveOperation*>* get_store_order(VMRegPair temp_register) {
1595     // Record which moves kill which values
1596     GrowableArray<MoveOperation*> killer;
1597     for (int i = 0; i < edges.length(); i++) {
1598       MoveOperation* s = edges.at(i);
1599       assert(killer.at_grow(s->dst_id(), NULL) == NULL, "only one killer");
1600       killer.at_put_grow(s->dst_id(), s, NULL);
1601     }
1602     assert(killer.at_grow(MoveOperation::get_id(temp_register), NULL) == NULL,
1603            "make sure temp isn't in the registers that are killed");
1604 
1605     // create links between loads and stores
1606     for (int i = 0; i < edges.length(); i++) {
1607       edges.at(i)->link(killer);
1608     }
1609 
1610     // at this point, all the move operations are chained together
1611     // in a doubly linked list.  Processing it backwards finds
1612     // the beginning of the chain, forwards finds the end.  If there's
1613     // a cycle it can be broken at any point,  so pick an edge and walk
1614     // backward until the list ends or we end where we started.
1615     GrowableArray<MoveOperation*>* stores = new GrowableArray<MoveOperation*>();
1616     for (int e = 0; e < edges.length(); e++) {
1617       MoveOperation* s = edges.at(e);
1618       if (!s->is_processed()) {
1619         MoveOperation* start = s;
1620         // search for the beginning of the chain or cycle
1621         while (start->prev() != NULL && start->prev() != s) {
1622           start = start->prev();
1623         }
1624         if (start->prev() == s) {
1625           start->break_cycle(temp_register);
1626         }
1627         // walk the chain forward inserting to store list
1628         while (start != NULL) {
1629           stores->append(start);
1630           start->set_processed();
1631           start = start->next();
1632         }
1633       }
1634     }
1635     return stores;
1636   }
1637 };
1638 
1639 static void verify_oop_args(MacroAssembler* masm,
1640                             methodHandle method,
1641                             const BasicType* sig_bt,
1642                             const VMRegPair* regs) {
1643   Register temp_reg = rbx;  // not part of any compiled calling seq
1644   if (VerifyOops) {
1645     for (int i = 0; i < method->size_of_parameters(); i++) {
1646       if (sig_bt[i] == T_OBJECT ||
1647           sig_bt[i] == T_ARRAY) {
1648         VMReg r = regs[i].first();
1649         assert(r->is_valid(), "bad oop arg");
1650         if (r->is_stack()) {
1651           __ movptr(temp_reg, Address(rsp, r->reg2stack() * VMRegImpl::stack_slot_size + wordSize));
1652           __ verify_oop(temp_reg);
1653         } else {
1654           __ verify_oop(r->as_Register());
1655         }
1656       }
1657     }
1658   }
1659 }
1660 
1661 static void gen_special_dispatch(MacroAssembler* masm,
1662                                  methodHandle method,
1663                                  const BasicType* sig_bt,
1664                                  const VMRegPair* regs) {
1665   verify_oop_args(masm, method, sig_bt, regs);
1666   vmIntrinsics::ID iid = method->intrinsic_id();
1667 
1668   // Now write the args into the outgoing interpreter space
1669   bool     has_receiver   = false;
1670   Register receiver_reg   = noreg;
1671   int      member_arg_pos = -1;
1672   Register member_reg     = noreg;
1673   int      ref_kind       = MethodHandles::signature_polymorphic_intrinsic_ref_kind(iid);
1674   if (ref_kind != 0) {
1675     member_arg_pos = method->size_of_parameters() - 1;  // trailing MemberName argument
1676     member_reg = rbx;  // known to be free at this point
1677     has_receiver = MethodHandles::ref_kind_has_receiver(ref_kind);
1678   } else if (iid == vmIntrinsics::_invokeBasic) {
1679     has_receiver = true;
1680   } else {
1681     fatal(err_msg_res("unexpected intrinsic id %d", iid));
1682   }
1683 
1684   if (member_reg != noreg) {
1685     // Load the member_arg into register, if necessary.
1686     SharedRuntime::check_member_name_argument_is_last_argument(method, sig_bt, regs);
1687     VMReg r = regs[member_arg_pos].first();
1688     if (r->is_stack()) {
1689       __ movptr(member_reg, Address(rsp, r->reg2stack() * VMRegImpl::stack_slot_size + wordSize));
1690     } else {
1691       // no data motion is needed
1692       member_reg = r->as_Register();
1693     }
1694   }
1695 
1696   if (has_receiver) {
1697     // Make sure the receiver is loaded into a register.
1698     assert(method->size_of_parameters() > 0, "oob");
1699     assert(sig_bt[0] == T_OBJECT, "receiver argument must be an object");
1700     VMReg r = regs[0].first();
1701     assert(r->is_valid(), "bad receiver arg");
1702     if (r->is_stack()) {
1703       // Porting note:  This assumes that compiled calling conventions always
1704       // pass the receiver oop in a register.  If this is not true on some
1705       // platform, pick a temp and load the receiver from stack.
1706       fatal("receiver always in a register");
1707       receiver_reg = j_rarg0;  // known to be free at this point
1708       __ movptr(receiver_reg, Address(rsp, r->reg2stack() * VMRegImpl::stack_slot_size + wordSize));
1709     } else {
1710       // no data motion is needed
1711       receiver_reg = r->as_Register();
1712     }
1713   }
1714 
1715   // Figure out which address we are really jumping to:
1716   MethodHandles::generate_method_handle_dispatch(masm, iid,
1717                                                  receiver_reg, member_reg, /*for_compiler_entry:*/ true);
1718 }
1719 
1720 // ---------------------------------------------------------------------------
1721 // Generate a native wrapper for a given method.  The method takes arguments
1722 // in the Java compiled code convention, marshals them to the native
1723 // convention (handlizes oops, etc), transitions to native, makes the call,
1724 // returns to java state (possibly blocking), unhandlizes any result and
1725 // returns.
1726 //
1727 // Critical native functions are a shorthand for the use of
1728 // GetPrimtiveArrayCritical and disallow the use of any other JNI
1729 // functions.  The wrapper is expected to unpack the arguments before
1730 // passing them to the callee and perform checks before and after the
1731 // native call to ensure that they GC_locker
1732 // lock_critical/unlock_critical semantics are followed.  Some other
1733 // parts of JNI setup are skipped like the tear down of the JNI handle
1734 // block and the check for pending exceptions it's impossible for them
1735 // to be thrown.
1736 //
1737 // They are roughly structured like this:
1738 //    if (GC_locker::needs_gc())
1739 //      SharedRuntime::block_for_jni_critical();
1740 //    tranistion to thread_in_native
1741 //    unpack arrray arguments and call native entry point
1742 //    check for safepoint in progress
1743 //    check if any thread suspend flags are set
1744 //      call into JVM and possible unlock the JNI critical
1745 //      if a GC was suppressed while in the critical native.
1746 //    transition back to thread_in_Java
1747 //    return to caller
1748 //
1749 nmethod* SharedRuntime::generate_native_wrapper(MacroAssembler* masm,
1750                                                 methodHandle method,
1751                                                 int compile_id,
1752                                                 BasicType* in_sig_bt,
1753                                                 VMRegPair* in_regs,
1754                                                 BasicType ret_type) {
1755   if (method->is_method_handle_intrinsic()) {
1756     vmIntrinsics::ID iid = method->intrinsic_id();
1757     intptr_t start = (intptr_t)__ pc();
1758     int vep_offset = ((intptr_t)__ pc()) - start;
1759     gen_special_dispatch(masm,
1760                          method,
1761                          in_sig_bt,
1762                          in_regs);
1763     int frame_complete = ((intptr_t)__ pc()) - start;  // not complete, period
1764     __ flush();
1765     int stack_slots = SharedRuntime::out_preserve_stack_slots();  // no out slots at all, actually
1766     return nmethod::new_native_nmethod(method,
1767                                        compile_id,
1768                                        masm->code(),
1769                                        vep_offset,
1770                                        frame_complete,
1771                                        stack_slots / VMRegImpl::slots_per_word,
1772                                        in_ByteSize(-1),
1773                                        in_ByteSize(-1),
1774                                        (OopMapSet*)NULL);
1775   }
1776   bool is_critical_native = true;
1777   address native_func = method->critical_native_function();
1778   if (native_func == NULL) {
1779     native_func = method->native_function();
1780     is_critical_native = false;
1781   }
1782   assert(native_func != NULL, "must have function");
1783 
1784   // An OopMap for lock (and class if static)
1785   OopMapSet *oop_maps = new OopMapSet();
1786   intptr_t start = (intptr_t)__ pc();
1787 
1788   // We have received a description of where all the java arg are located
1789   // on entry to the wrapper. We need to convert these args to where
1790   // the jni function will expect them. To figure out where they go
1791   // we convert the java signature to a C signature by inserting
1792   // the hidden arguments as arg[0] and possibly arg[1] (static method)
1793 
1794   const int total_in_args = method->size_of_parameters();
1795   int total_c_args = total_in_args;
1796   if (!is_critical_native) {
1797     total_c_args += 1;
1798     if (method->is_static()) {
1799       total_c_args++;
1800     }
1801   } else {
1802     for (int i = 0; i < total_in_args; i++) {
1803       if (in_sig_bt[i] == T_ARRAY) {
1804         total_c_args++;
1805       }
1806     }
1807   }
1808 
1809   BasicType* out_sig_bt = NEW_RESOURCE_ARRAY(BasicType, total_c_args);
1810   VMRegPair* out_regs   = NEW_RESOURCE_ARRAY(VMRegPair, total_c_args);
1811   BasicType* in_elem_bt = NULL;
1812 
1813   int argc = 0;
1814   if (!is_critical_native) {
1815     out_sig_bt[argc++] = T_ADDRESS;
1816     if (method->is_static()) {
1817       out_sig_bt[argc++] = T_OBJECT;
1818     }
1819 
1820     for (int i = 0; i < total_in_args ; i++ ) {
1821       out_sig_bt[argc++] = in_sig_bt[i];
1822     }
1823   } else {
1824     Thread* THREAD = Thread::current();
1825     in_elem_bt = NEW_RESOURCE_ARRAY(BasicType, total_in_args);
1826     SignatureStream ss(method->signature());
1827     for (int i = 0; i < total_in_args ; i++ ) {
1828       if (in_sig_bt[i] == T_ARRAY) {
1829         // Arrays are passed as int, elem* pair
1830         out_sig_bt[argc++] = T_INT;
1831         out_sig_bt[argc++] = T_ADDRESS;
1832         Symbol* atype = ss.as_symbol(CHECK_NULL);
1833         const char* at = atype->as_C_string();
1834         if (strlen(at) == 2) {
1835           assert(at[0] == '[', "must be");
1836           switch (at[1]) {
1837             case 'B': in_elem_bt[i]  = T_BYTE; break;
1838             case 'C': in_elem_bt[i]  = T_CHAR; break;
1839             case 'D': in_elem_bt[i]  = T_DOUBLE; break;
1840             case 'F': in_elem_bt[i]  = T_FLOAT; break;
1841             case 'I': in_elem_bt[i]  = T_INT; break;
1842             case 'J': in_elem_bt[i]  = T_LONG; break;
1843             case 'S': in_elem_bt[i]  = T_SHORT; break;
1844             case 'Z': in_elem_bt[i]  = T_BOOLEAN; break;
1845             default: ShouldNotReachHere();
1846           }
1847         }
1848       } else {
1849         out_sig_bt[argc++] = in_sig_bt[i];
1850         in_elem_bt[i] = T_VOID;
1851       }
1852       if (in_sig_bt[i] != T_VOID) {
1853         assert(in_sig_bt[i] == ss.type(), "must match");
1854         ss.next();
1855       }
1856     }
1857   }
1858 
1859   // Now figure out where the args must be stored and how much stack space
1860   // they require.
1861   int out_arg_slots;
1862   out_arg_slots = c_calling_convention(out_sig_bt, out_regs, NULL, total_c_args);
1863 
1864   // Compute framesize for the wrapper.  We need to handlize all oops in
1865   // incoming registers
1866 
1867   // Calculate the total number of stack slots we will need.
1868 
1869   // First count the abi requirement plus all of the outgoing args
1870   int stack_slots = SharedRuntime::out_preserve_stack_slots() + out_arg_slots;
1871 
1872   // Now the space for the inbound oop handle area
1873   int total_save_slots = 6 * VMRegImpl::slots_per_word;  // 6 arguments passed in registers
1874   if (is_critical_native) {
1875     // Critical natives may have to call out so they need a save area
1876     // for register arguments.
1877     int double_slots = 0;
1878     int single_slots = 0;
1879     for ( int i = 0; i < total_in_args; i++) {
1880       if (in_regs[i].first()->is_Register()) {
1881         const Register reg = in_regs[i].first()->as_Register();
1882         switch (in_sig_bt[i]) {
1883           case T_BOOLEAN:
1884           case T_BYTE:
1885           case T_SHORT:
1886           case T_CHAR:
1887           case T_INT:  single_slots++; break;
1888           case T_ARRAY:  // specific to LP64 (7145024)
1889           case T_LONG: double_slots++; break;
1890           default:  ShouldNotReachHere();
1891         }
1892       } else if (in_regs[i].first()->is_XMMRegister()) {
1893         switch (in_sig_bt[i]) {
1894           case T_FLOAT:  single_slots++; break;
1895           case T_DOUBLE: double_slots++; break;
1896           default:  ShouldNotReachHere();
1897         }
1898       } else if (in_regs[i].first()->is_FloatRegister()) {
1899         ShouldNotReachHere();
1900       }
1901     }
1902     total_save_slots = double_slots * 2 + single_slots;
1903     // align the save area
1904     if (double_slots != 0) {
1905       stack_slots = round_to(stack_slots, 2);
1906     }
1907   }
1908 
1909   int oop_handle_offset = stack_slots;
1910   stack_slots += total_save_slots;
1911 
1912   // Now any space we need for handlizing a klass if static method
1913 
1914   int klass_slot_offset = 0;
1915   int klass_offset = -1;
1916   int lock_slot_offset = 0;
1917   bool is_static = false;
1918 
1919   if (method->is_static()) {
1920     klass_slot_offset = stack_slots;
1921     stack_slots += VMRegImpl::slots_per_word;
1922     klass_offset = klass_slot_offset * VMRegImpl::stack_slot_size;
1923     is_static = true;
1924   }
1925 
1926   // Plus a lock if needed
1927 
1928   if (method->is_synchronized()) {
1929     lock_slot_offset = stack_slots;
1930     stack_slots += VMRegImpl::slots_per_word;
1931   }
1932 
1933   // Now a place (+2) to save return values or temp during shuffling
1934   // + 4 for return address (which we own) and saved rbp
1935   stack_slots += 6;
1936 
1937   // Ok The space we have allocated will look like:
1938   //
1939   //
1940   // FP-> |                     |
1941   //      |---------------------|
1942   //      | 2 slots for moves   |
1943   //      |---------------------|
1944   //      | lock box (if sync)  |
1945   //      |---------------------| <- lock_slot_offset
1946   //      | klass (if static)   |
1947   //      |---------------------| <- klass_slot_offset
1948   //      | oopHandle area      |
1949   //      |---------------------| <- oop_handle_offset (6 java arg registers)
1950   //      | outbound memory     |
1951   //      | based arguments     |
1952   //      |                     |
1953   //      |---------------------|
1954   //      |                     |
1955   // SP-> | out_preserved_slots |
1956   //
1957   //
1958 
1959 
1960   // Now compute actual number of stack words we need rounding to make
1961   // stack properly aligned.
1962   stack_slots = round_to(stack_slots, StackAlignmentInSlots);
1963 
1964   int stack_size = stack_slots * VMRegImpl::stack_slot_size;
1965 
1966   // First thing make an ic check to see if we should even be here
1967 
1968   // We are free to use all registers as temps without saving them and
1969   // restoring them except rbp. rbp is the only callee save register
1970   // as far as the interpreter and the compiler(s) are concerned.
1971 
1972 
1973   const Register ic_reg = rax;
1974   const Register receiver = j_rarg0;
1975 
1976   Label hit;
1977   Label exception_pending;
1978 
1979   assert_different_registers(ic_reg, receiver, rscratch1);
1980   __ verify_oop(receiver);
1981   __ load_klass(rscratch1, receiver);
1982   __ cmpq(ic_reg, rscratch1);
1983   __ jcc(Assembler::equal, hit);
1984 
1985   __ jump(RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
1986 
1987   // Verified entry point must be aligned
1988   __ align(8);
1989 
1990   __ bind(hit);
1991 
1992   int vep_offset = ((intptr_t)__ pc()) - start;
1993 
1994   // The instruction at the verified entry point must be 5 bytes or longer
1995   // because it can be patched on the fly by make_non_entrant. The stack bang
1996   // instruction fits that requirement.
1997 
1998   // Generate stack overflow check
1999 
2000   if (UseStackBanging) {
2001     __ bang_stack_with_offset(StackShadowPages*os::vm_page_size());
2002   } else {
2003     // need a 5 byte instruction to allow MT safe patching to non-entrant
2004     __ fat_nop();
2005   }
2006 
2007   // Generate a new frame for the wrapper.
2008   __ enter();
2009   // -2 because return address is already present and so is saved rbp
2010   __ subptr(rsp, stack_size - 2*wordSize);
2011 
2012   // Frame is now completed as far as size and linkage.
2013   int frame_complete = ((intptr_t)__ pc()) - start;
2014 
2015     if (UseRTMLocking) {
2016       // Abort RTM transaction before calling JNI
2017       // because critical section will be large and will be
2018       // aborted anyway. Also nmethod could be deoptimized.
2019       __ xabort(0);
2020     }
2021 
2022 #ifdef ASSERT
2023     {
2024       Label L;
2025       __ mov(rax, rsp);
2026       __ andptr(rax, -16); // must be 16 byte boundary (see amd64 ABI)
2027       __ cmpptr(rax, rsp);
2028       __ jcc(Assembler::equal, L);
2029       __ stop("improperly aligned stack");
2030       __ bind(L);
2031     }
2032 #endif /* ASSERT */
2033 
2034 
2035   // We use r14 as the oop handle for the receiver/klass
2036   // It is callee save so it survives the call to native
2037 
2038   const Register oop_handle_reg = r14;
2039 
2040   if (is_critical_native) {
2041     check_needs_gc_for_critical_native(masm, stack_slots, total_c_args, total_in_args,
2042                                        oop_handle_offset, oop_maps, in_regs, in_sig_bt);
2043   }
2044 
2045   //
2046   // We immediately shuffle the arguments so that any vm call we have to
2047   // make from here on out (sync slow path, jvmti, etc.) we will have
2048   // captured the oops from our caller and have a valid oopMap for
2049   // them.
2050 
2051   // -----------------
2052   // The Grand Shuffle
2053 
2054   // The Java calling convention is either equal (linux) or denser (win64) than the
2055   // c calling convention. However the because of the jni_env argument the c calling
2056   // convention always has at least one more (and two for static) arguments than Java.
2057   // Therefore if we move the args from java -> c backwards then we will never have
2058   // a register->register conflict and we don't have to build a dependency graph
2059   // and figure out how to break any cycles.
2060   //
2061 
2062   // Record esp-based slot for receiver on stack for non-static methods
2063   int receiver_offset = -1;
2064 
2065   // This is a trick. We double the stack slots so we can claim
2066   // the oops in the caller's frame. Since we are sure to have
2067   // more args than the caller doubling is enough to make
2068   // sure we can capture all the incoming oop args from the
2069   // caller.
2070   //
2071   OopMap* map = new OopMap(stack_slots * 2, 0 /* arg_slots*/);
2072 
2073   // Mark location of rbp (someday)
2074   // map->set_callee_saved(VMRegImpl::stack2reg( stack_slots - 2), stack_slots * 2, 0, vmreg(rbp));
2075 
2076   // Use eax, ebx as temporaries during any memory-memory moves we have to do
2077   // All inbound args are referenced based on rbp and all outbound args via rsp.
2078 
2079 
2080 #ifdef ASSERT
2081   bool reg_destroyed[RegisterImpl::number_of_registers];
2082   bool freg_destroyed[XMMRegisterImpl::number_of_registers];
2083   for ( int r = 0 ; r < RegisterImpl::number_of_registers ; r++ ) {
2084     reg_destroyed[r] = false;
2085   }
2086   for ( int f = 0 ; f < XMMRegisterImpl::number_of_registers ; f++ ) {
2087     freg_destroyed[f] = false;
2088   }
2089 
2090 #endif /* ASSERT */
2091 
2092   // This may iterate in two different directions depending on the
2093   // kind of native it is.  The reason is that for regular JNI natives
2094   // the incoming and outgoing registers are offset upwards and for
2095   // critical natives they are offset down.
2096   GrowableArray<int> arg_order(2 * total_in_args);
2097   VMRegPair tmp_vmreg;
2098   tmp_vmreg.set1(rbx->as_VMReg());
2099 
2100   if (!is_critical_native) {
2101     for (int i = total_in_args - 1, c_arg = total_c_args - 1; i >= 0; i--, c_arg--) {
2102       arg_order.push(i);
2103       arg_order.push(c_arg);
2104     }
2105   } else {
2106     // Compute a valid move order, using tmp_vmreg to break any cycles
2107     ComputeMoveOrder cmo(total_in_args, in_regs, total_c_args, out_regs, in_sig_bt, arg_order, tmp_vmreg);
2108   }
2109 
2110   int temploc = -1;
2111   for (int ai = 0; ai < arg_order.length(); ai += 2) {
2112     int i = arg_order.at(ai);
2113     int c_arg = arg_order.at(ai + 1);
2114     __ block_comment(err_msg("move %d -> %d", i, c_arg));
2115     if (c_arg == -1) {
2116       assert(is_critical_native, "should only be required for critical natives");
2117       // This arg needs to be moved to a temporary
2118       __ mov(tmp_vmreg.first()->as_Register(), in_regs[i].first()->as_Register());
2119       in_regs[i] = tmp_vmreg;
2120       temploc = i;
2121       continue;
2122     } else if (i == -1) {
2123       assert(is_critical_native, "should only be required for critical natives");
2124       // Read from the temporary location
2125       assert(temploc != -1, "must be valid");
2126       i = temploc;
2127       temploc = -1;
2128     }
2129 #ifdef ASSERT
2130     if (in_regs[i].first()->is_Register()) {
2131       assert(!reg_destroyed[in_regs[i].first()->as_Register()->encoding()], "destroyed reg!");
2132     } else if (in_regs[i].first()->is_XMMRegister()) {
2133       assert(!freg_destroyed[in_regs[i].first()->as_XMMRegister()->encoding()], "destroyed reg!");
2134     }
2135     if (out_regs[c_arg].first()->is_Register()) {
2136       reg_destroyed[out_regs[c_arg].first()->as_Register()->encoding()] = true;
2137     } else if (out_regs[c_arg].first()->is_XMMRegister()) {
2138       freg_destroyed[out_regs[c_arg].first()->as_XMMRegister()->encoding()] = true;
2139     }
2140 #endif /* ASSERT */
2141     switch (in_sig_bt[i]) {
2142       case T_ARRAY:
2143         if (is_critical_native) {
2144           unpack_array_argument(masm, in_regs[i], in_elem_bt[i], out_regs[c_arg + 1], out_regs[c_arg]);
2145           c_arg++;
2146 #ifdef ASSERT
2147           if (out_regs[c_arg].first()->is_Register()) {
2148             reg_destroyed[out_regs[c_arg].first()->as_Register()->encoding()] = true;
2149           } else if (out_regs[c_arg].first()->is_XMMRegister()) {
2150             freg_destroyed[out_regs[c_arg].first()->as_XMMRegister()->encoding()] = true;
2151           }
2152 #endif
2153           break;
2154         }
2155       case T_OBJECT:
2156         assert(!is_critical_native, "no oop arguments");
2157         object_move(masm, map, oop_handle_offset, stack_slots, in_regs[i], out_regs[c_arg],
2158                     ((i == 0) && (!is_static)),
2159                     &receiver_offset);
2160         break;
2161       case T_VOID:
2162         break;
2163 
2164       case T_FLOAT:
2165         float_move(masm, in_regs[i], out_regs[c_arg]);
2166           break;
2167 
2168       case T_DOUBLE:
2169         assert( i + 1 < total_in_args &&
2170                 in_sig_bt[i + 1] == T_VOID &&
2171                 out_sig_bt[c_arg+1] == T_VOID, "bad arg list");
2172         double_move(masm, in_regs[i], out_regs[c_arg]);
2173         break;
2174 
2175       case T_LONG :
2176         long_move(masm, in_regs[i], out_regs[c_arg]);
2177         break;
2178 
2179       case T_ADDRESS: assert(false, "found T_ADDRESS in java args");
2180 
2181       default:
2182         move32_64(masm, in_regs[i], out_regs[c_arg]);
2183     }
2184   }
2185 
2186   int c_arg;
2187 
2188   // Pre-load a static method's oop into r14.  Used both by locking code and
2189   // the normal JNI call code.
2190   if (!is_critical_native) {
2191     // point c_arg at the first arg that is already loaded in case we
2192     // need to spill before we call out
2193     c_arg = total_c_args - total_in_args;
2194 
2195     if (method->is_static()) {
2196 
2197       //  load oop into a register
2198       __ movoop(oop_handle_reg, JNIHandles::make_local(method->method_holder()->java_mirror()));
2199 
2200       // Now handlize the static class mirror it's known not-null.
2201       __ movptr(Address(rsp, klass_offset), oop_handle_reg);
2202       map->set_oop(VMRegImpl::stack2reg(klass_slot_offset));
2203 
2204       // Now get the handle
2205       __ lea(oop_handle_reg, Address(rsp, klass_offset));
2206       // store the klass handle as second argument
2207       __ movptr(c_rarg1, oop_handle_reg);
2208       // and protect the arg if we must spill
2209       c_arg--;
2210     }
2211   } else {
2212     // For JNI critical methods we need to save all registers in save_args.
2213     c_arg = 0;
2214   }
2215 
2216   // Change state to native (we save the return address in the thread, since it might not
2217   // be pushed on the stack when we do a a stack traversal). It is enough that the pc()
2218   // points into the right code segment. It does not have to be the correct return pc.
2219   // We use the same pc/oopMap repeatedly when we call out
2220 
2221   intptr_t the_pc = (intptr_t) __ pc();
2222   oop_maps->add_gc_map(the_pc - start, map);
2223 
2224   __ set_last_Java_frame(rsp, noreg, (address)the_pc);
2225 
2226 
2227   // We have all of the arguments setup at this point. We must not touch any register
2228   // argument registers at this point (what if we save/restore them there are no oop?
2229 
2230   {
2231     SkipIfEqual skip(masm, &DTraceMethodProbes, false);
2232     // protect the args we've loaded
2233     save_args(masm, total_c_args, c_arg, out_regs);
2234     __ mov_metadata(c_rarg1, method());
2235     __ call_VM_leaf(
2236       CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
2237       r15_thread, c_rarg1);
2238     restore_args(masm, total_c_args, c_arg, out_regs);
2239   }
2240 
2241   // RedefineClasses() tracing support for obsolete method entry
2242   if (RC_TRACE_IN_RANGE(0x00001000, 0x00002000)) {
2243     // protect the args we've loaded
2244     save_args(masm, total_c_args, c_arg, out_regs);
2245     __ mov_metadata(c_rarg1, method());
2246     __ call_VM_leaf(
2247       CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
2248       r15_thread, c_rarg1);
2249     restore_args(masm, total_c_args, c_arg, out_regs);
2250   }
2251 
2252   // Lock a synchronized method
2253 
2254   // Register definitions used by locking and unlocking
2255 
2256   const Register swap_reg = rax;  // Must use rax for cmpxchg instruction
2257   const Register obj_reg  = rbx;  // Will contain the oop
2258   const Register lock_reg = r13;  // Address of compiler lock object (BasicLock)
2259   const Register old_hdr  = r13;  // value of old header at unlock time
2260 
2261   Label slow_path_lock;
2262   Label lock_done;
2263 
2264   if (method->is_synchronized()) {
2265     assert(!is_critical_native, "unhandled");
2266 
2267 
2268     const int mark_word_offset = BasicLock::displaced_header_offset_in_bytes();
2269 
2270     // Get the handle (the 2nd argument)
2271     __ mov(oop_handle_reg, c_rarg1);
2272 
2273     // Get address of the box
2274 
2275     __ lea(lock_reg, Address(rsp, lock_slot_offset * VMRegImpl::stack_slot_size));
2276 
2277     // Load the oop from the handle
2278     __ movptr(obj_reg, Address(oop_handle_reg, 0));
2279 
2280     if (UseBiasedLocking) {
2281       __ biased_locking_enter(lock_reg, obj_reg, swap_reg, rscratch1, false, lock_done, &slow_path_lock);
2282     }
2283 
2284     // Load immediate 1 into swap_reg %rax
2285     __ movl(swap_reg, 1);
2286 
2287     // Load (object->mark() | 1) into swap_reg %rax
2288     __ orptr(swap_reg, Address(obj_reg, 0));
2289 
2290     // Save (object->mark() | 1) into BasicLock's displaced header
2291     __ movptr(Address(lock_reg, mark_word_offset), swap_reg);
2292 
2293     if (os::is_MP()) {
2294       __ lock();
2295     }
2296 
2297     // src -> dest iff dest == rax else rax <- dest
2298     __ cmpxchgptr(lock_reg, Address(obj_reg, 0));
2299     __ jcc(Assembler::equal, lock_done);
2300 
2301     // Hmm should this move to the slow path code area???
2302 
2303     // Test if the oopMark is an obvious stack pointer, i.e.,
2304     //  1) (mark & 3) == 0, and
2305     //  2) rsp <= mark < mark + os::pagesize()
2306     // These 3 tests can be done by evaluating the following
2307     // expression: ((mark - rsp) & (3 - os::vm_page_size())),
2308     // assuming both stack pointer and pagesize have their
2309     // least significant 2 bits clear.
2310     // NOTE: the oopMark is in swap_reg %rax as the result of cmpxchg
2311 
2312     __ subptr(swap_reg, rsp);
2313     __ andptr(swap_reg, 3 - os::vm_page_size());
2314 
2315     // Save the test result, for recursive case, the result is zero
2316     __ movptr(Address(lock_reg, mark_word_offset), swap_reg);
2317     __ jcc(Assembler::notEqual, slow_path_lock);
2318 
2319     // Slow path will re-enter here
2320 
2321     __ bind(lock_done);
2322   }
2323 
2324 
2325   // Finally just about ready to make the JNI call
2326 
2327 
2328   // get JNIEnv* which is first argument to native
2329   if (!is_critical_native) {
2330     __ lea(c_rarg0, Address(r15_thread, in_bytes(JavaThread::jni_environment_offset())));
2331   }
2332 
2333   // Now set thread in native
2334   __ movl(Address(r15_thread, JavaThread::thread_state_offset()), _thread_in_native);
2335 
2336   __ call(RuntimeAddress(native_func));
2337 
2338   // Verify or restore cpu control state after JNI call
2339   __ restore_cpu_control_state_after_jni();
2340 
2341   // Unpack native results.
2342   switch (ret_type) {
2343   case T_BOOLEAN: __ c2bool(rax);            break;
2344   case T_CHAR   : __ movzwl(rax, rax);      break;
2345   case T_BYTE   : __ sign_extend_byte (rax); break;
2346   case T_SHORT  : __ sign_extend_short(rax); break;
2347   case T_INT    : /* nothing to do */        break;
2348   case T_DOUBLE :
2349   case T_FLOAT  :
2350     // Result is in xmm0 we'll save as needed
2351     break;
2352   case T_ARRAY:                 // Really a handle
2353   case T_OBJECT:                // Really a handle
2354       break; // can't de-handlize until after safepoint check
2355   case T_VOID: break;
2356   case T_LONG: break;
2357   default       : ShouldNotReachHere();
2358   }
2359 
2360   // Switch thread to "native transition" state before reading the synchronization state.
2361   // This additional state is necessary because reading and testing the synchronization
2362   // state is not atomic w.r.t. GC, as this scenario demonstrates:
2363   //     Java thread A, in _thread_in_native state, loads _not_synchronized and is preempted.
2364   //     VM thread changes sync state to synchronizing and suspends threads for GC.
2365   //     Thread A is resumed to finish this native method, but doesn't block here since it
2366   //     didn't see any synchronization is progress, and escapes.
2367   __ movl(Address(r15_thread, JavaThread::thread_state_offset()), _thread_in_native_trans);
2368 
2369   if(os::is_MP()) {
2370     if (UseMembar) {
2371       // Force this write out before the read below
2372       __ membar(Assembler::Membar_mask_bits(
2373            Assembler::LoadLoad | Assembler::LoadStore |
2374            Assembler::StoreLoad | Assembler::StoreStore));
2375     } else {
2376       // Write serialization page so VM thread can do a pseudo remote membar.
2377       // We use the current thread pointer to calculate a thread specific
2378       // offset to write to within the page. This minimizes bus traffic
2379       // due to cache line collision.
2380       __ serialize_memory(r15_thread, rcx);
2381     }
2382   }
2383 
2384   Label after_transition;
2385 
2386   // check for safepoint operation in progress and/or pending suspend requests
2387   {
2388     Label Continue;
2389 
2390     __ cmp32(ExternalAddress((address)SafepointSynchronize::address_of_state()),
2391              SafepointSynchronize::_not_synchronized);
2392 
2393     Label L;
2394     __ jcc(Assembler::notEqual, L);
2395     __ cmpl(Address(r15_thread, JavaThread::suspend_flags_offset()), 0);
2396     __ jcc(Assembler::equal, Continue);
2397     __ bind(L);
2398 
2399     // Don't use call_VM as it will see a possible pending exception and forward it
2400     // and never return here preventing us from clearing _last_native_pc down below.
2401     // Also can't use call_VM_leaf either as it will check to see if rsi & rdi are
2402     // preserved and correspond to the bcp/locals pointers. So we do a runtime call
2403     // by hand.
2404     //
2405     save_native_result(masm, ret_type, stack_slots);
2406     __ mov(c_rarg0, r15_thread);
2407     __ mov(r12, rsp); // remember sp
2408     __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
2409     __ andptr(rsp, -16); // align stack as required by ABI
2410     if (!is_critical_native) {
2411       __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans)));
2412     } else {
2413       __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans_and_transition)));
2414     }
2415     __ mov(rsp, r12); // restore sp
2416     __ reinit_heapbase();
2417     // Restore any method result value
2418     restore_native_result(masm, ret_type, stack_slots);
2419 
2420     if (is_critical_native) {
2421       // The call above performed the transition to thread_in_Java so
2422       // skip the transition logic below.
2423       __ jmpb(after_transition);
2424     }
2425 
2426     __ bind(Continue);
2427   }
2428 
2429   // change thread state
2430   __ movl(Address(r15_thread, JavaThread::thread_state_offset()), _thread_in_Java);
2431   __ bind(after_transition);
2432 
2433   Label reguard;
2434   Label reguard_done;
2435   __ cmpl(Address(r15_thread, JavaThread::stack_guard_state_offset()), JavaThread::stack_guard_yellow_disabled);
2436   __ jcc(Assembler::equal, reguard);
2437   __ bind(reguard_done);
2438 
2439   // native result if any is live
2440 
2441   // Unlock
2442   Label unlock_done;
2443   Label slow_path_unlock;
2444   if (method->is_synchronized()) {
2445 
2446     // Get locked oop from the handle we passed to jni
2447     __ movptr(obj_reg, Address(oop_handle_reg, 0));
2448 
2449     Label done;
2450 
2451     if (UseBiasedLocking) {
2452       __ biased_locking_exit(obj_reg, old_hdr, done);
2453     }
2454 
2455     // Simple recursive lock?
2456 
2457     __ cmpptr(Address(rsp, lock_slot_offset * VMRegImpl::stack_slot_size), (int32_t)NULL_WORD);
2458     __ jcc(Assembler::equal, done);
2459 
2460     // Must save rax if if it is live now because cmpxchg must use it
2461     if (ret_type != T_FLOAT && ret_type != T_DOUBLE && ret_type != T_VOID) {
2462       save_native_result(masm, ret_type, stack_slots);
2463     }
2464 
2465 
2466     // get address of the stack lock
2467     __ lea(rax, Address(rsp, lock_slot_offset * VMRegImpl::stack_slot_size));
2468     //  get old displaced header
2469     __ movptr(old_hdr, Address(rax, 0));
2470 
2471     // Atomic swap old header if oop still contains the stack lock
2472     if (os::is_MP()) {
2473       __ lock();
2474     }
2475     __ cmpxchgptr(old_hdr, Address(obj_reg, 0));
2476     __ jcc(Assembler::notEqual, slow_path_unlock);
2477 
2478     // slow path re-enters here
2479     __ bind(unlock_done);
2480     if (ret_type != T_FLOAT && ret_type != T_DOUBLE && ret_type != T_VOID) {
2481       restore_native_result(masm, ret_type, stack_slots);
2482     }
2483 
2484     __ bind(done);
2485 
2486   }
2487   {
2488     SkipIfEqual skip(masm, &DTraceMethodProbes, false);
2489     save_native_result(masm, ret_type, stack_slots);
2490     __ mov_metadata(c_rarg1, method());
2491     __ call_VM_leaf(
2492          CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
2493          r15_thread, c_rarg1);
2494     restore_native_result(masm, ret_type, stack_slots);
2495   }
2496 
2497   __ reset_last_Java_frame(false, true);
2498 
2499   // Unpack oop result
2500   if (ret_type == T_OBJECT || ret_type == T_ARRAY) {
2501       Label L;
2502       __ testptr(rax, rax);
2503       __ jcc(Assembler::zero, L);
2504       __ movptr(rax, Address(rax, 0));
2505       __ bind(L);
2506       __ verify_oop(rax);
2507   }
2508 
2509   if (!is_critical_native) {
2510     // reset handle block
2511     __ movptr(rcx, Address(r15_thread, JavaThread::active_handles_offset()));
2512     __ movl(Address(rcx, JNIHandleBlock::top_offset_in_bytes()), (int32_t)NULL_WORD);
2513   }
2514 
2515   // pop our frame
2516 
2517   __ leave();
2518 
2519   if (!is_critical_native) {
2520     // Any exception pending?
2521     __ cmpptr(Address(r15_thread, in_bytes(Thread::pending_exception_offset())), (int32_t)NULL_WORD);
2522     __ jcc(Assembler::notEqual, exception_pending);
2523   }
2524 
2525   // Return
2526 
2527   __ ret(0);
2528 
2529   // Unexpected paths are out of line and go here
2530 
2531   if (!is_critical_native) {
2532     // forward the exception
2533     __ bind(exception_pending);
2534 
2535     // and forward the exception
2536     __ jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
2537   }
2538 
2539   // Slow path locking & unlocking
2540   if (method->is_synchronized()) {
2541 
2542     // BEGIN Slow path lock
2543     __ bind(slow_path_lock);
2544 
2545     // has last_Java_frame setup. No exceptions so do vanilla call not call_VM
2546     // args are (oop obj, BasicLock* lock, JavaThread* thread)
2547 
2548     // protect the args we've loaded
2549     save_args(masm, total_c_args, c_arg, out_regs);
2550 
2551     __ mov(c_rarg0, obj_reg);
2552     __ mov(c_rarg1, lock_reg);
2553     __ mov(c_rarg2, r15_thread);
2554 
2555     // Not a leaf but we have last_Java_frame setup as we want
2556     __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::complete_monitor_locking_C), 3);
2557     restore_args(masm, total_c_args, c_arg, out_regs);
2558 
2559 #ifdef ASSERT
2560     { Label L;
2561     __ cmpptr(Address(r15_thread, in_bytes(Thread::pending_exception_offset())), (int32_t)NULL_WORD);
2562     __ jcc(Assembler::equal, L);
2563     __ stop("no pending exception allowed on exit from monitorenter");
2564     __ bind(L);
2565     }
2566 #endif
2567     __ jmp(lock_done);
2568 
2569     // END Slow path lock
2570 
2571     // BEGIN Slow path unlock
2572     __ bind(slow_path_unlock);
2573 
2574     // If we haven't already saved the native result we must save it now as xmm registers
2575     // are still exposed.
2576 
2577     if (ret_type == T_FLOAT || ret_type == T_DOUBLE ) {
2578       save_native_result(masm, ret_type, stack_slots);
2579     }
2580 
2581     __ lea(c_rarg1, Address(rsp, lock_slot_offset * VMRegImpl::stack_slot_size));
2582 
2583     __ mov(c_rarg0, obj_reg);
2584     __ mov(r12, rsp); // remember sp
2585     __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
2586     __ andptr(rsp, -16); // align stack as required by ABI
2587 
2588     // Save pending exception around call to VM (which contains an EXCEPTION_MARK)
2589     // NOTE that obj_reg == rbx currently
2590     __ movptr(rbx, Address(r15_thread, in_bytes(Thread::pending_exception_offset())));
2591     __ movptr(Address(r15_thread, in_bytes(Thread::pending_exception_offset())), (int32_t)NULL_WORD);
2592 
2593     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::complete_monitor_unlocking_C)));
2594     __ mov(rsp, r12); // restore sp
2595     __ reinit_heapbase();
2596 #ifdef ASSERT
2597     {
2598       Label L;
2599       __ cmpptr(Address(r15_thread, in_bytes(Thread::pending_exception_offset())), (int)NULL_WORD);
2600       __ jcc(Assembler::equal, L);
2601       __ stop("no pending exception allowed on exit complete_monitor_unlocking_C");
2602       __ bind(L);
2603     }
2604 #endif /* ASSERT */
2605 
2606     __ movptr(Address(r15_thread, in_bytes(Thread::pending_exception_offset())), rbx);
2607 
2608     if (ret_type == T_FLOAT || ret_type == T_DOUBLE ) {
2609       restore_native_result(masm, ret_type, stack_slots);
2610     }
2611     __ jmp(unlock_done);
2612 
2613     // END Slow path unlock
2614 
2615   } // synchronized
2616 
2617   // SLOW PATH Reguard the stack if needed
2618 
2619   __ bind(reguard);
2620   save_native_result(masm, ret_type, stack_slots);
2621   __ mov(r12, rsp); // remember sp
2622   __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
2623   __ andptr(rsp, -16); // align stack as required by ABI
2624   __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
2625   __ mov(rsp, r12); // restore sp
2626   __ reinit_heapbase();
2627   restore_native_result(masm, ret_type, stack_slots);
2628   // and continue
2629   __ jmp(reguard_done);
2630 
2631 
2632 
2633   __ flush();
2634 
2635   nmethod *nm = nmethod::new_native_nmethod(method,
2636                                             compile_id,
2637                                             masm->code(),
2638                                             vep_offset,
2639                                             frame_complete,
2640                                             stack_slots / VMRegImpl::slots_per_word,
2641                                             (is_static ? in_ByteSize(klass_offset) : in_ByteSize(receiver_offset)),
2642                                             in_ByteSize(lock_slot_offset*VMRegImpl::stack_slot_size),
2643                                             oop_maps);
2644 
2645   if (is_critical_native) {
2646     nm->set_lazy_critical_native(true);
2647   }
2648 
2649   return nm;
2650 
2651 }
2652 
2653 #ifdef HAVE_DTRACE_H
2654 // ---------------------------------------------------------------------------
2655 // Generate a dtrace nmethod for a given signature.  The method takes arguments
2656 // in the Java compiled code convention, marshals them to the native
2657 // abi and then leaves nops at the position you would expect to call a native
2658 // function. When the probe is enabled the nops are replaced with a trap
2659 // instruction that dtrace inserts and the trace will cause a notification
2660 // to dtrace.
2661 //
2662 // The probes are only able to take primitive types and java/lang/String as
2663 // arguments.  No other java types are allowed. Strings are converted to utf8
2664 // strings so that from dtrace point of view java strings are converted to C
2665 // strings. There is an arbitrary fixed limit on the total space that a method
2666 // can use for converting the strings. (256 chars per string in the signature).
2667 // So any java string larger then this is truncated.
2668 
2669 static int  fp_offset[ConcreteRegisterImpl::number_of_registers] = { 0 };
2670 static bool offsets_initialized = false;
2671 
2672 
2673 nmethod *SharedRuntime::generate_dtrace_nmethod(MacroAssembler *masm,
2674                                                 methodHandle method) {
2675 
2676 
2677   // generate_dtrace_nmethod is guarded by a mutex so we are sure to
2678   // be single threaded in this method.
2679   assert(AdapterHandlerLibrary_lock->owned_by_self(), "must be");
2680 
2681   if (!offsets_initialized) {
2682     fp_offset[c_rarg0->as_VMReg()->value()] = -1 * wordSize;
2683     fp_offset[c_rarg1->as_VMReg()->value()] = -2 * wordSize;
2684     fp_offset[c_rarg2->as_VMReg()->value()] = -3 * wordSize;
2685     fp_offset[c_rarg3->as_VMReg()->value()] = -4 * wordSize;
2686     fp_offset[c_rarg4->as_VMReg()->value()] = -5 * wordSize;
2687     fp_offset[c_rarg5->as_VMReg()->value()] = -6 * wordSize;
2688 
2689     fp_offset[c_farg0->as_VMReg()->value()] = -7 * wordSize;
2690     fp_offset[c_farg1->as_VMReg()->value()] = -8 * wordSize;
2691     fp_offset[c_farg2->as_VMReg()->value()] = -9 * wordSize;
2692     fp_offset[c_farg3->as_VMReg()->value()] = -10 * wordSize;
2693     fp_offset[c_farg4->as_VMReg()->value()] = -11 * wordSize;
2694     fp_offset[c_farg5->as_VMReg()->value()] = -12 * wordSize;
2695     fp_offset[c_farg6->as_VMReg()->value()] = -13 * wordSize;
2696     fp_offset[c_farg7->as_VMReg()->value()] = -14 * wordSize;
2697 
2698     offsets_initialized = true;
2699   }
2700   // Fill in the signature array, for the calling-convention call.
2701   int total_args_passed = method->size_of_parameters();
2702 
2703   BasicType* in_sig_bt  = NEW_RESOURCE_ARRAY(BasicType, total_args_passed);
2704   VMRegPair  *in_regs   = NEW_RESOURCE_ARRAY(VMRegPair, total_args_passed);
2705 
2706   // The signature we are going to use for the trap that dtrace will see
2707   // java/lang/String is converted. We drop "this" and any other object
2708   // is converted to NULL.  (A one-slot java/lang/Long object reference
2709   // is converted to a two-slot long, which is why we double the allocation).
2710   BasicType* out_sig_bt = NEW_RESOURCE_ARRAY(BasicType, total_args_passed * 2);
2711   VMRegPair* out_regs   = NEW_RESOURCE_ARRAY(VMRegPair, total_args_passed * 2);
2712 
2713   int i=0;
2714   int total_strings = 0;
2715   int first_arg_to_pass = 0;
2716   int total_c_args = 0;
2717 
2718   // Skip the receiver as dtrace doesn't want to see it
2719   if( !method->is_static() ) {
2720     in_sig_bt[i++] = T_OBJECT;
2721     first_arg_to_pass = 1;
2722   }
2723 
2724   // We need to convert the java args to where a native (non-jni) function
2725   // would expect them. To figure out where they go we convert the java
2726   // signature to a C signature.
2727 
2728   SignatureStream ss(method->signature());
2729   for ( ; !ss.at_return_type(); ss.next()) {
2730     BasicType bt = ss.type();
2731     in_sig_bt[i++] = bt;  // Collect remaining bits of signature
2732     out_sig_bt[total_c_args++] = bt;
2733     if( bt == T_OBJECT) {
2734       Symbol* s = ss.as_symbol_or_null();   // symbol is created
2735       if (s == vmSymbols::java_lang_String()) {
2736         total_strings++;
2737         out_sig_bt[total_c_args-1] = T_ADDRESS;
2738       } else if (s == vmSymbols::java_lang_Boolean() ||
2739                  s == vmSymbols::java_lang_Character() ||
2740                  s == vmSymbols::java_lang_Byte() ||
2741                  s == vmSymbols::java_lang_Short() ||
2742                  s == vmSymbols::java_lang_Integer() ||
2743                  s == vmSymbols::java_lang_Float()) {
2744         out_sig_bt[total_c_args-1] = T_INT;
2745       } else if (s == vmSymbols::java_lang_Long() ||
2746                  s == vmSymbols::java_lang_Double()) {
2747         out_sig_bt[total_c_args-1] = T_LONG;
2748         out_sig_bt[total_c_args++] = T_VOID;
2749       }
2750     } else if ( bt == T_LONG || bt == T_DOUBLE ) {
2751       in_sig_bt[i++] = T_VOID;   // Longs & doubles take 2 Java slots
2752       // We convert double to long
2753       out_sig_bt[total_c_args-1] = T_LONG;
2754       out_sig_bt[total_c_args++] = T_VOID;
2755     } else if ( bt == T_FLOAT) {
2756       // We convert float to int
2757       out_sig_bt[total_c_args-1] = T_INT;
2758     }
2759   }
2760 
2761   assert(i==total_args_passed, "validly parsed signature");
2762 
2763   // Now get the compiled-Java layout as input arguments
2764   int comp_args_on_stack;
2765   comp_args_on_stack = SharedRuntime::java_calling_convention(
2766       in_sig_bt, in_regs, total_args_passed, false);
2767 
2768   // Now figure out where the args must be stored and how much stack space
2769   // they require (neglecting out_preserve_stack_slots but space for storing
2770   // the 1st six register arguments). It's weird see int_stk_helper.
2771 
2772   int out_arg_slots;
2773   out_arg_slots = c_calling_convention(out_sig_bt, out_regs, NULL, total_c_args);
2774 
2775   // Calculate the total number of stack slots we will need.
2776 
2777   // First count the abi requirement plus all of the outgoing args
2778   int stack_slots = SharedRuntime::out_preserve_stack_slots() + out_arg_slots;
2779 
2780   // Now space for the string(s) we must convert
2781   int* string_locs   = NEW_RESOURCE_ARRAY(int, total_strings + 1);
2782   for (i = 0; i < total_strings ; i++) {
2783     string_locs[i] = stack_slots;
2784     stack_slots += max_dtrace_string_size / VMRegImpl::stack_slot_size;
2785   }
2786 
2787   // Plus the temps we might need to juggle register args
2788   // regs take two slots each
2789   stack_slots += (Argument::n_int_register_parameters_c +
2790                   Argument::n_float_register_parameters_c) * 2;
2791 
2792 
2793   // + 4 for return address (which we own) and saved rbp,
2794 
2795   stack_slots += 4;
2796 
2797   // Ok The space we have allocated will look like:
2798   //
2799   //
2800   // FP-> |                     |
2801   //      |---------------------|
2802   //      | string[n]           |
2803   //      |---------------------| <- string_locs[n]
2804   //      | string[n-1]         |
2805   //      |---------------------| <- string_locs[n-1]
2806   //      | ...                 |
2807   //      | ...                 |
2808   //      |---------------------| <- string_locs[1]
2809   //      | string[0]           |
2810   //      |---------------------| <- string_locs[0]
2811   //      | outbound memory     |
2812   //      | based arguments     |
2813   //      |                     |
2814   //      |---------------------|
2815   //      |                     |
2816   // SP-> | out_preserved_slots |
2817   //
2818   //
2819 
2820   // Now compute actual number of stack words we need rounding to make
2821   // stack properly aligned.
2822   stack_slots = round_to(stack_slots, 4 * VMRegImpl::slots_per_word);
2823 
2824   int stack_size = stack_slots * VMRegImpl::stack_slot_size;
2825 
2826   intptr_t start = (intptr_t)__ pc();
2827 
2828   // First thing make an ic check to see if we should even be here
2829 
2830   // We are free to use all registers as temps without saving them and
2831   // restoring them except rbp. rbp, is the only callee save register
2832   // as far as the interpreter and the compiler(s) are concerned.
2833 
2834   const Register ic_reg = rax;
2835   const Register receiver = rcx;
2836   Label hit;
2837   Label exception_pending;
2838 
2839 
2840   __ verify_oop(receiver);
2841   __ cmpl(ic_reg, Address(receiver, oopDesc::klass_offset_in_bytes()));
2842   __ jcc(Assembler::equal, hit);
2843 
2844   __ jump(RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
2845 
2846   // verified entry must be aligned for code patching.
2847   // and the first 5 bytes must be in the same cache line
2848   // if we align at 8 then we will be sure 5 bytes are in the same line
2849   __ align(8);
2850 
2851   __ bind(hit);
2852 
2853   int vep_offset = ((intptr_t)__ pc()) - start;
2854 
2855 
2856   // The instruction at the verified entry point must be 5 bytes or longer
2857   // because it can be patched on the fly by make_non_entrant. The stack bang
2858   // instruction fits that requirement.
2859 
2860   // Generate stack overflow check
2861 
2862   if (UseStackBanging) {
2863     if (stack_size <= StackShadowPages*os::vm_page_size()) {
2864       __ bang_stack_with_offset(StackShadowPages*os::vm_page_size());
2865     } else {
2866       __ movl(rax, stack_size);
2867       __ bang_stack_size(rax, rbx);
2868     }
2869   } else {
2870     // need a 5 byte instruction to allow MT safe patching to non-entrant
2871     __ fat_nop();
2872   }
2873 
2874   assert(((uintptr_t)__ pc() - start - vep_offset) >= 5,
2875          "valid size for make_non_entrant");
2876 
2877   // Generate a new frame for the wrapper.
2878   __ enter();
2879 
2880   // -4 because return address is already present and so is saved rbp,
2881   if (stack_size - 2*wordSize != 0) {
2882     __ subq(rsp, stack_size - 2*wordSize);
2883   }
2884 
2885   // Frame is now completed as far a size and linkage.
2886 
2887   int frame_complete = ((intptr_t)__ pc()) - start;
2888 
2889   int c_arg, j_arg;
2890 
2891   // State of input register args
2892 
2893   bool  live[ConcreteRegisterImpl::number_of_registers];
2894 
2895   live[j_rarg0->as_VMReg()->value()] = false;
2896   live[j_rarg1->as_VMReg()->value()] = false;
2897   live[j_rarg2->as_VMReg()->value()] = false;
2898   live[j_rarg3->as_VMReg()->value()] = false;
2899   live[j_rarg4->as_VMReg()->value()] = false;
2900   live[j_rarg5->as_VMReg()->value()] = false;
2901 
2902   live[j_farg0->as_VMReg()->value()] = false;
2903   live[j_farg1->as_VMReg()->value()] = false;
2904   live[j_farg2->as_VMReg()->value()] = false;
2905   live[j_farg3->as_VMReg()->value()] = false;
2906   live[j_farg4->as_VMReg()->value()] = false;
2907   live[j_farg5->as_VMReg()->value()] = false;
2908   live[j_farg6->as_VMReg()->value()] = false;
2909   live[j_farg7->as_VMReg()->value()] = false;
2910 
2911 
2912   bool rax_is_zero = false;
2913 
2914   // All args (except strings) destined for the stack are moved first
2915   for (j_arg = first_arg_to_pass, c_arg = 0 ;
2916        j_arg < total_args_passed ; j_arg++, c_arg++ ) {
2917     VMRegPair src = in_regs[j_arg];
2918     VMRegPair dst = out_regs[c_arg];
2919 
2920     // Get the real reg value or a dummy (rsp)
2921 
2922     int src_reg = src.first()->is_reg() ?
2923                   src.first()->value() :
2924                   rsp->as_VMReg()->value();
2925 
2926     bool useless =  in_sig_bt[j_arg] == T_ARRAY ||
2927                     (in_sig_bt[j_arg] == T_OBJECT &&
2928                      out_sig_bt[c_arg] != T_INT &&
2929                      out_sig_bt[c_arg] != T_ADDRESS &&
2930                      out_sig_bt[c_arg] != T_LONG);
2931 
2932     live[src_reg] = !useless;
2933 
2934     if (dst.first()->is_stack()) {
2935 
2936       // Even though a string arg in a register is still live after this loop
2937       // after the string conversion loop (next) it will be dead so we take
2938       // advantage of that now for simpler code to manage live.
2939 
2940       live[src_reg] = false;
2941       switch (in_sig_bt[j_arg]) {
2942 
2943         case T_ARRAY:
2944         case T_OBJECT:
2945           {
2946             Address stack_dst(rsp, reg2offset_out(dst.first()));
2947 
2948             if (out_sig_bt[c_arg] == T_INT || out_sig_bt[c_arg] == T_LONG) {
2949               // need to unbox a one-word value
2950               Register in_reg = rax;
2951               if ( src.first()->is_reg() ) {
2952                 in_reg = src.first()->as_Register();
2953               } else {
2954                 __ movq(rax, Address(rbp, reg2offset_in(src.first())));
2955                 rax_is_zero = false;
2956               }
2957               Label skipUnbox;
2958               __ movptr(Address(rsp, reg2offset_out(dst.first())),
2959                         (int32_t)NULL_WORD);
2960               __ testq(in_reg, in_reg);
2961               __ jcc(Assembler::zero, skipUnbox);
2962 
2963               BasicType bt = out_sig_bt[c_arg];
2964               int box_offset = java_lang_boxing_object::value_offset_in_bytes(bt);
2965               Address src1(in_reg, box_offset);
2966               if ( bt == T_LONG ) {
2967                 __ movq(in_reg,  src1);
2968                 __ movq(stack_dst, in_reg);
2969                 assert(out_sig_bt[c_arg+1] == T_VOID, "must be");
2970                 ++c_arg; // skip over T_VOID to keep the loop indices in sync
2971               } else {
2972                 __ movl(in_reg,  src1);
2973                 __ movl(stack_dst, in_reg);
2974               }
2975 
2976               __ bind(skipUnbox);
2977             } else if (out_sig_bt[c_arg] != T_ADDRESS) {
2978               // Convert the arg to NULL
2979               if (!rax_is_zero) {
2980                 __ xorq(rax, rax);
2981                 rax_is_zero = true;
2982               }
2983               __ movq(stack_dst, rax);
2984             }
2985           }
2986           break;
2987 
2988         case T_VOID:
2989           break;
2990 
2991         case T_FLOAT:
2992           // This does the right thing since we know it is destined for the
2993           // stack
2994           float_move(masm, src, dst);
2995           break;
2996 
2997         case T_DOUBLE:
2998           // This does the right thing since we know it is destined for the
2999           // stack
3000           double_move(masm, src, dst);
3001           break;
3002 
3003         case T_LONG :
3004           long_move(masm, src, dst);
3005           break;
3006 
3007         case T_ADDRESS: assert(false, "found T_ADDRESS in java args");
3008 
3009         default:
3010           move32_64(masm, src, dst);
3011       }
3012     }
3013 
3014   }
3015 
3016   // If we have any strings we must store any register based arg to the stack
3017   // This includes any still live xmm registers too.
3018 
3019   int sid = 0;
3020 
3021   if (total_strings > 0 ) {
3022     for (j_arg = first_arg_to_pass, c_arg = 0 ;
3023          j_arg < total_args_passed ; j_arg++, c_arg++ ) {
3024       VMRegPair src = in_regs[j_arg];
3025       VMRegPair dst = out_regs[c_arg];
3026 
3027       if (src.first()->is_reg()) {
3028         Address src_tmp(rbp, fp_offset[src.first()->value()]);
3029 
3030         // string oops were left untouched by the previous loop even if the
3031         // eventual (converted) arg is destined for the stack so park them
3032         // away now (except for first)
3033 
3034         if (out_sig_bt[c_arg] == T_ADDRESS) {
3035           Address utf8_addr = Address(
3036               rsp, string_locs[sid++] * VMRegImpl::stack_slot_size);
3037           if (sid != 1) {
3038             // The first string arg won't be killed until after the utf8
3039             // conversion
3040             __ movq(utf8_addr, src.first()->as_Register());
3041           }
3042         } else if (dst.first()->is_reg()) {
3043           if (in_sig_bt[j_arg] == T_FLOAT || in_sig_bt[j_arg] == T_DOUBLE) {
3044 
3045             // Convert the xmm register to an int and store it in the reserved
3046             // location for the eventual c register arg
3047             XMMRegister f = src.first()->as_XMMRegister();
3048             if (in_sig_bt[j_arg] == T_FLOAT) {
3049               __ movflt(src_tmp, f);
3050             } else {
3051               __ movdbl(src_tmp, f);
3052             }
3053           } else {
3054             // If the arg is an oop type we don't support don't bother to store
3055             // it remember string was handled above.
3056             bool useless =  in_sig_bt[j_arg] == T_ARRAY ||
3057                             (in_sig_bt[j_arg] == T_OBJECT &&
3058                              out_sig_bt[c_arg] != T_INT &&
3059                              out_sig_bt[c_arg] != T_LONG);
3060 
3061             if (!useless) {
3062               __ movq(src_tmp, src.first()->as_Register());
3063             }
3064           }
3065         }
3066       }
3067       if (in_sig_bt[j_arg] == T_OBJECT && out_sig_bt[c_arg] == T_LONG) {
3068         assert(out_sig_bt[c_arg+1] == T_VOID, "must be");
3069         ++c_arg; // skip over T_VOID to keep the loop indices in sync
3070       }
3071     }
3072 
3073     // Now that the volatile registers are safe, convert all the strings
3074     sid = 0;
3075 
3076     for (j_arg = first_arg_to_pass, c_arg = 0 ;
3077          j_arg < total_args_passed ; j_arg++, c_arg++ ) {
3078       if (out_sig_bt[c_arg] == T_ADDRESS) {
3079         // It's a string
3080         Address utf8_addr = Address(
3081             rsp, string_locs[sid++] * VMRegImpl::stack_slot_size);
3082         // The first string we find might still be in the original java arg
3083         // register
3084 
3085         VMReg src = in_regs[j_arg].first();
3086 
3087         // We will need to eventually save the final argument to the trap
3088         // in the von-volatile location dedicated to src. This is the offset
3089         // from fp we will use.
3090         int src_off = src->is_reg() ?
3091             fp_offset[src->value()] : reg2offset_in(src);
3092 
3093         // This is where the argument will eventually reside
3094         VMRegPair dst = out_regs[c_arg];
3095 
3096         if (src->is_reg()) {
3097           if (sid == 1) {
3098             __ movq(c_rarg0, src->as_Register());
3099           } else {
3100             __ movq(c_rarg0, utf8_addr);
3101           }
3102         } else {
3103           // arg is still in the original location
3104           __ movq(c_rarg0, Address(rbp, reg2offset_in(src)));
3105         }
3106         Label done, convert;
3107 
3108         // see if the oop is NULL
3109         __ testq(c_rarg0, c_rarg0);
3110         __ jcc(Assembler::notEqual, convert);
3111 
3112         if (dst.first()->is_reg()) {
3113           // Save the ptr to utf string in the origina src loc or the tmp
3114           // dedicated to it
3115           __ movq(Address(rbp, src_off), c_rarg0);
3116         } else {
3117           __ movq(Address(rsp, reg2offset_out(dst.first())), c_rarg0);
3118         }
3119         __ jmp(done);
3120 
3121         __ bind(convert);
3122 
3123         __ lea(c_rarg1, utf8_addr);
3124         if (dst.first()->is_reg()) {
3125           __ movq(Address(rbp, src_off), c_rarg1);
3126         } else {
3127           __ movq(Address(rsp, reg2offset_out(dst.first())), c_rarg1);
3128         }
3129         // And do the conversion
3130         __ call(RuntimeAddress(
3131                 CAST_FROM_FN_PTR(address, SharedRuntime::get_utf)));
3132 
3133         __ bind(done);
3134       }
3135       if (in_sig_bt[j_arg] == T_OBJECT && out_sig_bt[c_arg] == T_LONG) {
3136         assert(out_sig_bt[c_arg+1] == T_VOID, "must be");
3137         ++c_arg; // skip over T_VOID to keep the loop indices in sync
3138       }
3139     }
3140     // The get_utf call killed all the c_arg registers
3141     live[c_rarg0->as_VMReg()->value()] = false;
3142     live[c_rarg1->as_VMReg()->value()] = false;
3143     live[c_rarg2->as_VMReg()->value()] = false;
3144     live[c_rarg3->as_VMReg()->value()] = false;
3145     live[c_rarg4->as_VMReg()->value()] = false;
3146     live[c_rarg5->as_VMReg()->value()] = false;
3147 
3148     live[c_farg0->as_VMReg()->value()] = false;
3149     live[c_farg1->as_VMReg()->value()] = false;
3150     live[c_farg2->as_VMReg()->value()] = false;
3151     live[c_farg3->as_VMReg()->value()] = false;
3152     live[c_farg4->as_VMReg()->value()] = false;
3153     live[c_farg5->as_VMReg()->value()] = false;
3154     live[c_farg6->as_VMReg()->value()] = false;
3155     live[c_farg7->as_VMReg()->value()] = false;
3156   }
3157 
3158   // Now we can finally move the register args to their desired locations
3159 
3160   rax_is_zero = false;
3161 
3162   for (j_arg = first_arg_to_pass, c_arg = 0 ;
3163        j_arg < total_args_passed ; j_arg++, c_arg++ ) {
3164 
3165     VMRegPair src = in_regs[j_arg];
3166     VMRegPair dst = out_regs[c_arg];
3167 
3168     // Only need to look for args destined for the interger registers (since we
3169     // convert float/double args to look like int/long outbound)
3170     if (dst.first()->is_reg()) {
3171       Register r =  dst.first()->as_Register();
3172 
3173       // Check if the java arg is unsupported and thereofre useless
3174       bool useless =  in_sig_bt[j_arg] == T_ARRAY ||
3175                       (in_sig_bt[j_arg] == T_OBJECT &&
3176                        out_sig_bt[c_arg] != T_INT &&
3177                        out_sig_bt[c_arg] != T_ADDRESS &&
3178                        out_sig_bt[c_arg] != T_LONG);
3179 
3180 
3181       // If we're going to kill an existing arg save it first
3182       if (live[dst.first()->value()]) {
3183         // you can't kill yourself
3184         if (src.first() != dst.first()) {
3185           __ movq(Address(rbp, fp_offset[dst.first()->value()]), r);
3186         }
3187       }
3188       if (src.first()->is_reg()) {
3189         if (live[src.first()->value()] ) {
3190           if (in_sig_bt[j_arg] == T_FLOAT) {
3191             __ movdl(r, src.first()->as_XMMRegister());
3192           } else if (in_sig_bt[j_arg] == T_DOUBLE) {
3193             __ movdq(r, src.first()->as_XMMRegister());
3194           } else if (r != src.first()->as_Register()) {
3195             if (!useless) {
3196               __ movq(r, src.first()->as_Register());
3197             }
3198           }
3199         } else {
3200           // If the arg is an oop type we don't support don't bother to store
3201           // it
3202           if (!useless) {
3203             if (in_sig_bt[j_arg] == T_DOUBLE ||
3204                 in_sig_bt[j_arg] == T_LONG  ||
3205                 in_sig_bt[j_arg] == T_OBJECT ) {
3206               __ movq(r, Address(rbp, fp_offset[src.first()->value()]));
3207             } else {
3208               __ movl(r, Address(rbp, fp_offset[src.first()->value()]));
3209             }
3210           }
3211         }
3212         live[src.first()->value()] = false;
3213       } else if (!useless) {
3214         // full sized move even for int should be ok
3215         __ movq(r, Address(rbp, reg2offset_in(src.first())));
3216       }
3217 
3218       // At this point r has the original java arg in the final location
3219       // (assuming it wasn't useless). If the java arg was an oop
3220       // we have a bit more to do
3221 
3222       if (in_sig_bt[j_arg] == T_ARRAY || in_sig_bt[j_arg] == T_OBJECT ) {
3223         if (out_sig_bt[c_arg] == T_INT || out_sig_bt[c_arg] == T_LONG) {
3224           // need to unbox a one-word value
3225           Label skip;
3226           __ testq(r, r);
3227           __ jcc(Assembler::equal, skip);
3228           BasicType bt = out_sig_bt[c_arg];
3229           int box_offset = java_lang_boxing_object::value_offset_in_bytes(bt);
3230           Address src1(r, box_offset);
3231           if ( bt == T_LONG ) {
3232             __ movq(r, src1);
3233           } else {
3234             __ movl(r, src1);
3235           }
3236           __ bind(skip);
3237 
3238         } else if (out_sig_bt[c_arg] != T_ADDRESS) {
3239           // Convert the arg to NULL
3240           __ xorq(r, r);
3241         }
3242       }
3243 
3244       // dst can longer be holding an input value
3245       live[dst.first()->value()] = false;
3246     }
3247     if (in_sig_bt[j_arg] == T_OBJECT && out_sig_bt[c_arg] == T_LONG) {
3248       assert(out_sig_bt[c_arg+1] == T_VOID, "must be");
3249       ++c_arg; // skip over T_VOID to keep the loop indices in sync
3250     }
3251   }
3252 
3253 
3254   // Ok now we are done. Need to place the nop that dtrace wants in order to
3255   // patch in the trap
3256   int patch_offset = ((intptr_t)__ pc()) - start;
3257 
3258   __ nop();
3259 
3260 
3261   // Return
3262 
3263   __ leave();
3264   __ ret(0);
3265 
3266   __ flush();
3267 
3268   nmethod *nm = nmethod::new_dtrace_nmethod(
3269       method, masm->code(), vep_offset, patch_offset, frame_complete,
3270       stack_slots / VMRegImpl::slots_per_word);
3271   return nm;
3272 
3273 }
3274 
3275 #endif // HAVE_DTRACE_H
3276 
3277 // this function returns the adjust size (in number of words) to a c2i adapter
3278 // activation for use during deoptimization
3279 int Deoptimization::last_frame_adjust(int callee_parameters, int callee_locals ) {
3280   return (callee_locals - callee_parameters) * Interpreter::stackElementWords;
3281 }
3282 
3283 
3284 uint SharedRuntime::out_preserve_stack_slots() {
3285   return 0;
3286 }
3287 
3288 //------------------------------generate_deopt_blob----------------------------
3289 void SharedRuntime::generate_deopt_blob() {
3290   // Allocate space for the code
3291   ResourceMark rm;
3292   // Setup code generation tools
3293   CodeBuffer buffer("deopt_blob", 2048, 1024);
3294   MacroAssembler* masm = new MacroAssembler(&buffer);
3295   int frame_size_in_words;
3296   OopMap* map = NULL;
3297   OopMapSet *oop_maps = new OopMapSet();
3298 
3299   // -------------
3300   // This code enters when returning to a de-optimized nmethod.  A return
3301   // address has been pushed on the the stack, and return values are in
3302   // registers.
3303   // If we are doing a normal deopt then we were called from the patched
3304   // nmethod from the point we returned to the nmethod. So the return
3305   // address on the stack is wrong by NativeCall::instruction_size
3306   // We will adjust the value so it looks like we have the original return
3307   // address on the stack (like when we eagerly deoptimized).
3308   // In the case of an exception pending when deoptimizing, we enter
3309   // with a return address on the stack that points after the call we patched
3310   // into the exception handler. We have the following register state from,
3311   // e.g., the forward exception stub (see stubGenerator_x86_64.cpp).
3312   //    rax: exception oop
3313   //    rbx: exception handler
3314   //    rdx: throwing pc
3315   // So in this case we simply jam rdx into the useless return address and
3316   // the stack looks just like we want.
3317   //
3318   // At this point we need to de-opt.  We save the argument return
3319   // registers.  We call the first C routine, fetch_unroll_info().  This
3320   // routine captures the return values and returns a structure which
3321   // describes the current frame size and the sizes of all replacement frames.
3322   // The current frame is compiled code and may contain many inlined
3323   // functions, each with their own JVM state.  We pop the current frame, then
3324   // push all the new frames.  Then we call the C routine unpack_frames() to
3325   // populate these frames.  Finally unpack_frames() returns us the new target
3326   // address.  Notice that callee-save registers are BLOWN here; they have
3327   // already been captured in the vframeArray at the time the return PC was
3328   // patched.
3329   address start = __ pc();
3330   Label cont;
3331 
3332   // Prolog for non exception case!
3333 
3334   // Save everything in sight.
3335   map = RegisterSaver::save_live_registers(masm, 0, &frame_size_in_words);
3336 
3337   // Normal deoptimization.  Save exec mode for unpack_frames.
3338   __ movl(r14, Deoptimization::Unpack_deopt); // callee-saved
3339   __ jmp(cont);
3340 
3341   int reexecute_offset = __ pc() - start;
3342 
3343   // Reexecute case
3344   // return address is the pc describes what bci to do re-execute at
3345 
3346   // No need to update map as each call to save_live_registers will produce identical oopmap
3347   (void) RegisterSaver::save_live_registers(masm, 0, &frame_size_in_words);
3348 
3349   __ movl(r14, Deoptimization::Unpack_reexecute); // callee-saved
3350   __ jmp(cont);
3351 
3352   int exception_offset = __ pc() - start;
3353 
3354   // Prolog for exception case
3355 
3356   // all registers are dead at this entry point, except for rax, and
3357   // rdx which contain the exception oop and exception pc
3358   // respectively.  Set them in TLS and fall thru to the
3359   // unpack_with_exception_in_tls entry point.
3360 
3361   __ movptr(Address(r15_thread, JavaThread::exception_pc_offset()), rdx);
3362   __ movptr(Address(r15_thread, JavaThread::exception_oop_offset()), rax);
3363 
3364   int exception_in_tls_offset = __ pc() - start;
3365 
3366   // new implementation because exception oop is now passed in JavaThread
3367 
3368   // Prolog for exception case
3369   // All registers must be preserved because they might be used by LinearScan
3370   // Exceptiop oop and throwing PC are passed in JavaThread
3371   // tos: stack at point of call to method that threw the exception (i.e. only
3372   // args are on the stack, no return address)
3373 
3374   // make room on stack for the return address
3375   // It will be patched later with the throwing pc. The correct value is not
3376   // available now because loading it from memory would destroy registers.
3377   __ push(0);
3378 
3379   // Save everything in sight.
3380   map = RegisterSaver::save_live_registers(masm, 0, &frame_size_in_words);
3381 
3382   // Now it is safe to overwrite any register
3383 
3384   // Deopt during an exception.  Save exec mode for unpack_frames.
3385   __ movl(r14, Deoptimization::Unpack_exception); // callee-saved
3386 
3387   // load throwing pc from JavaThread and patch it as the return address
3388   // of the current frame. Then clear the field in JavaThread
3389 
3390   __ movptr(rdx, Address(r15_thread, JavaThread::exception_pc_offset()));
3391   __ movptr(Address(rbp, wordSize), rdx);
3392   __ movptr(Address(r15_thread, JavaThread::exception_pc_offset()), (int32_t)NULL_WORD);
3393 
3394 #ifdef ASSERT
3395   // verify that there is really an exception oop in JavaThread
3396   __ movptr(rax, Address(r15_thread, JavaThread::exception_oop_offset()));
3397   __ verify_oop(rax);
3398 
3399   // verify that there is no pending exception
3400   Label no_pending_exception;
3401   __ movptr(rax, Address(r15_thread, Thread::pending_exception_offset()));
3402   __ testptr(rax, rax);
3403   __ jcc(Assembler::zero, no_pending_exception);
3404   __ stop("must not have pending exception here");
3405   __ bind(no_pending_exception);
3406 #endif
3407 
3408   __ bind(cont);
3409 
3410   // Call C code.  Need thread and this frame, but NOT official VM entry
3411   // crud.  We cannot block on this call, no GC can happen.
3412   //
3413   // UnrollBlock* fetch_unroll_info(JavaThread* thread)
3414 
3415   // fetch_unroll_info needs to call last_java_frame().
3416 
3417   __ set_last_Java_frame(noreg, noreg, NULL);
3418 #ifdef ASSERT
3419   { Label L;
3420     __ cmpptr(Address(r15_thread,
3421                     JavaThread::last_Java_fp_offset()),
3422             (int32_t)0);
3423     __ jcc(Assembler::equal, L);
3424     __ stop("SharedRuntime::generate_deopt_blob: last_Java_fp not cleared");
3425     __ bind(L);
3426   }
3427 #endif // ASSERT
3428   __ mov(c_rarg0, r15_thread);
3429   __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, Deoptimization::fetch_unroll_info)));
3430 
3431   // Need to have an oopmap that tells fetch_unroll_info where to
3432   // find any register it might need.
3433   oop_maps->add_gc_map(__ pc() - start, map);
3434 
3435   __ reset_last_Java_frame(false, false);
3436 
3437   // Load UnrollBlock* into rdi
3438   __ mov(rdi, rax);
3439 
3440    Label noException;
3441   __ cmpl(r14, Deoptimization::Unpack_exception);   // Was exception pending?
3442   __ jcc(Assembler::notEqual, noException);
3443   __ movptr(rax, Address(r15_thread, JavaThread::exception_oop_offset()));
3444   // QQQ this is useless it was NULL above
3445   __ movptr(rdx, Address(r15_thread, JavaThread::exception_pc_offset()));
3446   __ movptr(Address(r15_thread, JavaThread::exception_oop_offset()), (int32_t)NULL_WORD);
3447   __ movptr(Address(r15_thread, JavaThread::exception_pc_offset()), (int32_t)NULL_WORD);
3448 
3449   __ verify_oop(rax);
3450 
3451   // Overwrite the result registers with the exception results.
3452   __ movptr(Address(rsp, RegisterSaver::rax_offset_in_bytes()), rax);
3453   // I think this is useless
3454   __ movptr(Address(rsp, RegisterSaver::rdx_offset_in_bytes()), rdx);
3455 
3456   __ bind(noException);
3457 
3458   // Only register save data is on the stack.
3459   // Now restore the result registers.  Everything else is either dead
3460   // or captured in the vframeArray.
3461   RegisterSaver::restore_result_registers(masm);
3462 
3463   // All of the register save area has been popped of the stack. Only the
3464   // return address remains.
3465 
3466   // Pop all the frames we must move/replace.
3467   //
3468   // Frame picture (youngest to oldest)
3469   // 1: self-frame (no frame link)
3470   // 2: deopting frame  (no frame link)
3471   // 3: caller of deopting frame (could be compiled/interpreted).
3472   //
3473   // Note: by leaving the return address of self-frame on the stack
3474   // and using the size of frame 2 to adjust the stack
3475   // when we are done the return to frame 3 will still be on the stack.
3476 
3477   // Pop deoptimized frame
3478   __ movl(rcx, Address(rdi, Deoptimization::UnrollBlock::size_of_deoptimized_frame_offset_in_bytes()));
3479   __ addptr(rsp, rcx);
3480 
3481   // rsp should be pointing at the return address to the caller (3)
3482 
3483   // Pick up the initial fp we should save
3484   // restore rbp before stack bang because if stack overflow is thrown it needs to be pushed (and preserved)
3485   __ movptr(rbp, Address(rdi, Deoptimization::UnrollBlock::initial_info_offset_in_bytes()));
3486 
3487   // Stack bang to make sure there's enough room for these interpreter frames.
3488   if (UseStackBanging) {
3489     __ movl(rbx, Address(rdi, Deoptimization::UnrollBlock::total_frame_sizes_offset_in_bytes()));
3490     __ bang_stack_size(rbx, rcx);
3491   }
3492 
3493   // Load address of array of frame pcs into rcx
3494   __ movptr(rcx, Address(rdi, Deoptimization::UnrollBlock::frame_pcs_offset_in_bytes()));
3495 
3496   // Trash the old pc
3497   __ addptr(rsp, wordSize);
3498 
3499   // Load address of array of frame sizes into rsi
3500   __ movptr(rsi, Address(rdi, Deoptimization::UnrollBlock::frame_sizes_offset_in_bytes()));
3501 
3502   // Load counter into rdx
3503   __ movl(rdx, Address(rdi, Deoptimization::UnrollBlock::number_of_frames_offset_in_bytes()));
3504 
3505   // Now adjust the caller's stack to make up for the extra locals
3506   // but record the original sp so that we can save it in the skeletal interpreter
3507   // frame and the stack walking of interpreter_sender will get the unextended sp
3508   // value and not the "real" sp value.
3509 
3510   const Register sender_sp = r8;
3511 
3512   __ mov(sender_sp, rsp);
3513   __ movl(rbx, Address(rdi,
3514                        Deoptimization::UnrollBlock::
3515                        caller_adjustment_offset_in_bytes()));
3516   __ subptr(rsp, rbx);
3517 
3518   // Push interpreter frames in a loop
3519   Label loop;
3520   __ bind(loop);
3521   __ movptr(rbx, Address(rsi, 0));      // Load frame size
3522 #ifdef CC_INTERP
3523   __ subptr(rbx, 4*wordSize);           // we'll push pc and ebp by hand and
3524 #ifdef ASSERT
3525   __ push(0xDEADDEAD);                  // Make a recognizable pattern
3526   __ push(0xDEADDEAD);
3527 #else /* ASSERT */
3528   __ subptr(rsp, 2*wordSize);           // skip the "static long no_param"
3529 #endif /* ASSERT */
3530 #else
3531   __ subptr(rbx, 2*wordSize);           // We'll push pc and ebp by hand
3532 #endif // CC_INTERP
3533   __ pushptr(Address(rcx, 0));          // Save return address
3534   __ enter();                           // Save old & set new ebp
3535   __ subptr(rsp, rbx);                  // Prolog
3536 #ifdef CC_INTERP
3537   __ movptr(Address(rbp,
3538                   -(sizeof(BytecodeInterpreter)) + in_bytes(byte_offset_of(BytecodeInterpreter, _sender_sp))),
3539             sender_sp); // Make it walkable
3540 #else /* CC_INTERP */
3541   // This value is corrected by layout_activation_impl
3542   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD );
3543   __ movptr(Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize), sender_sp); // Make it walkable
3544 #endif /* CC_INTERP */
3545   __ mov(sender_sp, rsp);               // Pass sender_sp to next frame
3546   __ addptr(rsi, wordSize);             // Bump array pointer (sizes)
3547   __ addptr(rcx, wordSize);             // Bump array pointer (pcs)
3548   __ decrementl(rdx);                   // Decrement counter
3549   __ jcc(Assembler::notZero, loop);
3550   __ pushptr(Address(rcx, 0));          // Save final return address
3551 
3552   // Re-push self-frame
3553   __ enter();                           // Save old & set new ebp
3554 
3555   // Allocate a full sized register save area.
3556   // Return address and rbp are in place, so we allocate two less words.
3557   __ subptr(rsp, (frame_size_in_words - 2) * wordSize);
3558 
3559   // Restore frame locals after moving the frame
3560   __ movdbl(Address(rsp, RegisterSaver::xmm0_offset_in_bytes()), xmm0);
3561   __ movptr(Address(rsp, RegisterSaver::rax_offset_in_bytes()), rax);
3562 
3563   // Call C code.  Need thread but NOT official VM entry
3564   // crud.  We cannot block on this call, no GC can happen.  Call should
3565   // restore return values to their stack-slots with the new SP.
3566   //
3567   // void Deoptimization::unpack_frames(JavaThread* thread, int exec_mode)
3568 
3569   // Use rbp because the frames look interpreted now
3570   // Save "the_pc" since it cannot easily be retrieved using the last_java_SP after we aligned SP.
3571   // Don't need the precise return PC here, just precise enough to point into this code blob.
3572   address the_pc = __ pc();
3573   __ set_last_Java_frame(noreg, rbp, the_pc);
3574 
3575   __ andptr(rsp, -(StackAlignmentInBytes));  // Fix stack alignment as required by ABI
3576   __ mov(c_rarg0, r15_thread);
3577   __ movl(c_rarg1, r14); // second arg: exec_mode
3578   __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, Deoptimization::unpack_frames)));
3579   // Revert SP alignment after call since we're going to do some SP relative addressing below
3580   __ movptr(rsp, Address(r15_thread, JavaThread::last_Java_sp_offset()));
3581 
3582   // Set an oopmap for the call site
3583   // Use the same PC we used for the last java frame
3584   oop_maps->add_gc_map(the_pc - start,
3585                        new OopMap( frame_size_in_words, 0 ));
3586 
3587   // Clear fp AND pc
3588   __ reset_last_Java_frame(true, true);
3589 
3590   // Collect return values
3591   __ movdbl(xmm0, Address(rsp, RegisterSaver::xmm0_offset_in_bytes()));
3592   __ movptr(rax, Address(rsp, RegisterSaver::rax_offset_in_bytes()));
3593   // I think this is useless (throwing pc?)
3594   __ movptr(rdx, Address(rsp, RegisterSaver::rdx_offset_in_bytes()));
3595 
3596   // Pop self-frame.
3597   __ leave();                           // Epilog
3598 
3599   // Jump to interpreter
3600   __ ret(0);
3601 
3602   // Make sure all code is generated
3603   masm->flush();
3604 
3605   _deopt_blob = DeoptimizationBlob::create(&buffer, oop_maps, 0, exception_offset, reexecute_offset, frame_size_in_words);
3606   _deopt_blob->set_unpack_with_exception_in_tls_offset(exception_in_tls_offset);
3607 }
3608 
3609 #ifdef COMPILER2
3610 //------------------------------generate_uncommon_trap_blob--------------------
3611 void SharedRuntime::generate_uncommon_trap_blob() {
3612   // Allocate space for the code
3613   ResourceMark rm;
3614   // Setup code generation tools
3615   CodeBuffer buffer("uncommon_trap_blob", 2048, 1024);
3616   MacroAssembler* masm = new MacroAssembler(&buffer);
3617 
3618   assert(SimpleRuntimeFrame::framesize % 4 == 0, "sp not 16-byte aligned");
3619 
3620   address start = __ pc();
3621 
3622   if (UseRTMLocking) {
3623     // Abort RTM transaction before possible nmethod deoptimization.
3624     __ xabort(0);
3625   }
3626 
3627   // Push self-frame.  We get here with a return address on the
3628   // stack, so rsp is 8-byte aligned until we allocate our frame.
3629   __ subptr(rsp, SimpleRuntimeFrame::return_off << LogBytesPerInt); // Epilog!
3630 
3631   // No callee saved registers. rbp is assumed implicitly saved
3632   __ movptr(Address(rsp, SimpleRuntimeFrame::rbp_off << LogBytesPerInt), rbp);
3633 
3634   // compiler left unloaded_class_index in j_rarg0 move to where the
3635   // runtime expects it.
3636   __ movl(c_rarg1, j_rarg0);
3637 
3638   __ set_last_Java_frame(noreg, noreg, NULL);
3639 
3640   // Call C code.  Need thread but NOT official VM entry
3641   // crud.  We cannot block on this call, no GC can happen.  Call should
3642   // capture callee-saved registers as well as return values.
3643   // Thread is in rdi already.
3644   //
3645   // UnrollBlock* uncommon_trap(JavaThread* thread, jint unloaded_class_index);
3646 
3647   __ mov(c_rarg0, r15_thread);
3648   __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, Deoptimization::uncommon_trap)));
3649 
3650   // Set an oopmap for the call site
3651   OopMapSet* oop_maps = new OopMapSet();
3652   OopMap* map = new OopMap(SimpleRuntimeFrame::framesize, 0);
3653 
3654   // location of rbp is known implicitly by the frame sender code
3655 
3656   oop_maps->add_gc_map(__ pc() - start, map);
3657 
3658   __ reset_last_Java_frame(false, false);
3659 
3660   // Load UnrollBlock* into rdi
3661   __ mov(rdi, rax);
3662 
3663   // Pop all the frames we must move/replace.
3664   //
3665   // Frame picture (youngest to oldest)
3666   // 1: self-frame (no frame link)
3667   // 2: deopting frame  (no frame link)
3668   // 3: caller of deopting frame (could be compiled/interpreted).
3669 
3670   // Pop self-frame.  We have no frame, and must rely only on rax and rsp.
3671   __ addptr(rsp, (SimpleRuntimeFrame::framesize - 2) << LogBytesPerInt); // Epilog!
3672 
3673   // Pop deoptimized frame (int)
3674   __ movl(rcx, Address(rdi,
3675                        Deoptimization::UnrollBlock::
3676                        size_of_deoptimized_frame_offset_in_bytes()));
3677   __ addptr(rsp, rcx);
3678 
3679   // rsp should be pointing at the return address to the caller (3)
3680 
3681   // Pick up the initial fp we should save
3682   // restore rbp before stack bang because if stack overflow is thrown it needs to be pushed (and preserved)
3683   __ movptr(rbp, Address(rdi, Deoptimization::UnrollBlock::initial_info_offset_in_bytes()));
3684 
3685   // Stack bang to make sure there's enough room for these interpreter frames.
3686   if (UseStackBanging) {
3687     __ movl(rbx, Address(rdi ,Deoptimization::UnrollBlock::total_frame_sizes_offset_in_bytes()));
3688     __ bang_stack_size(rbx, rcx);
3689   }
3690 
3691   // Load address of array of frame pcs into rcx (address*)
3692   __ movptr(rcx, Address(rdi, Deoptimization::UnrollBlock::frame_pcs_offset_in_bytes()));
3693 
3694   // Trash the return pc
3695   __ addptr(rsp, wordSize);
3696 
3697   // Load address of array of frame sizes into rsi (intptr_t*)
3698   __ movptr(rsi, Address(rdi, Deoptimization::UnrollBlock:: frame_sizes_offset_in_bytes()));
3699 
3700   // Counter
3701   __ movl(rdx, Address(rdi, Deoptimization::UnrollBlock:: number_of_frames_offset_in_bytes())); // (int)
3702 
3703   // Now adjust the caller's stack to make up for the extra locals but
3704   // record the original sp so that we can save it in the skeletal
3705   // interpreter frame and the stack walking of interpreter_sender
3706   // will get the unextended sp value and not the "real" sp value.
3707 
3708   const Register sender_sp = r8;
3709 
3710   __ mov(sender_sp, rsp);
3711   __ movl(rbx, Address(rdi, Deoptimization::UnrollBlock:: caller_adjustment_offset_in_bytes())); // (int)
3712   __ subptr(rsp, rbx);
3713 
3714   // Push interpreter frames in a loop
3715   Label loop;
3716   __ bind(loop);
3717   __ movptr(rbx, Address(rsi, 0)); // Load frame size
3718   __ subptr(rbx, 2 * wordSize);    // We'll push pc and rbp by hand
3719   __ pushptr(Address(rcx, 0));     // Save return address
3720   __ enter();                      // Save old & set new rbp
3721   __ subptr(rsp, rbx);             // Prolog
3722 #ifdef CC_INTERP
3723   __ movptr(Address(rbp,
3724                   -(sizeof(BytecodeInterpreter)) + in_bytes(byte_offset_of(BytecodeInterpreter, _sender_sp))),
3725             sender_sp); // Make it walkable
3726 #else // CC_INTERP
3727   __ movptr(Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize),
3728             sender_sp);            // Make it walkable
3729   // This value is corrected by layout_activation_impl
3730   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD );
3731 #endif // CC_INTERP
3732   __ mov(sender_sp, rsp);          // Pass sender_sp to next frame
3733   __ addptr(rsi, wordSize);        // Bump array pointer (sizes)
3734   __ addptr(rcx, wordSize);        // Bump array pointer (pcs)
3735   __ decrementl(rdx);              // Decrement counter
3736   __ jcc(Assembler::notZero, loop);
3737   __ pushptr(Address(rcx, 0));     // Save final return address
3738 
3739   // Re-push self-frame
3740   __ enter();                 // Save old & set new rbp
3741   __ subptr(rsp, (SimpleRuntimeFrame::framesize - 4) << LogBytesPerInt);
3742                               // Prolog
3743 
3744   // Use rbp because the frames look interpreted now
3745   // Save "the_pc" since it cannot easily be retrieved using the last_java_SP after we aligned SP.
3746   // Don't need the precise return PC here, just precise enough to point into this code blob.
3747   address the_pc = __ pc();
3748   __ set_last_Java_frame(noreg, rbp, the_pc);
3749 
3750   // Call C code.  Need thread but NOT official VM entry
3751   // crud.  We cannot block on this call, no GC can happen.  Call should
3752   // restore return values to their stack-slots with the new SP.
3753   // Thread is in rdi already.
3754   //
3755   // BasicType unpack_frames(JavaThread* thread, int exec_mode);
3756 
3757   __ andptr(rsp, -(StackAlignmentInBytes)); // Align SP as required by ABI
3758   __ mov(c_rarg0, r15_thread);
3759   __ movl(c_rarg1, Deoptimization::Unpack_uncommon_trap);
3760   __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, Deoptimization::unpack_frames)));
3761 
3762   // Set an oopmap for the call site
3763   // Use the same PC we used for the last java frame
3764   oop_maps->add_gc_map(the_pc - start, new OopMap(SimpleRuntimeFrame::framesize, 0));
3765 
3766   // Clear fp AND pc
3767   __ reset_last_Java_frame(true, true);
3768 
3769   // Pop self-frame.
3770   __ leave();                 // Epilog
3771 
3772   // Jump to interpreter
3773   __ ret(0);
3774 
3775   // Make sure all code is generated
3776   masm->flush();
3777 
3778   _uncommon_trap_blob =  UncommonTrapBlob::create(&buffer, oop_maps,
3779                                                  SimpleRuntimeFrame::framesize >> 1);
3780 }
3781 #endif // COMPILER2
3782 
3783 
3784 //------------------------------generate_handler_blob------
3785 //
3786 // Generate a special Compile2Runtime blob that saves all registers,
3787 // and setup oopmap.
3788 //
3789 SafepointBlob* SharedRuntime::generate_handler_blob(address call_ptr, int poll_type) {
3790   assert(StubRoutines::forward_exception_entry() != NULL,
3791          "must be generated before");
3792 
3793   ResourceMark rm;
3794   OopMapSet *oop_maps = new OopMapSet();
3795   OopMap* map;
3796 
3797   // Allocate space for the code.  Setup code generation tools.
3798   CodeBuffer buffer("handler_blob", 2048, 1024);
3799   MacroAssembler* masm = new MacroAssembler(&buffer);
3800 
3801   address start   = __ pc();
3802   address call_pc = NULL;
3803   int frame_size_in_words;
3804   bool cause_return = (poll_type == POLL_AT_RETURN);
3805   bool save_vectors = (poll_type == POLL_AT_VECTOR_LOOP);
3806 
3807   if (UseRTMLocking) {
3808     // Abort RTM transaction before calling runtime
3809     // because critical section will be large and will be
3810     // aborted anyway. Also nmethod could be deoptimized.
3811     __ xabort(0);
3812   }
3813 
3814   // Make room for return address (or push it again)
3815   if (!cause_return) {
3816     __ push(rbx);
3817   }
3818 
3819   // Save registers, fpu state, and flags
3820   map = RegisterSaver::save_live_registers(masm, 0, &frame_size_in_words, save_vectors);
3821 
3822   // The following is basically a call_VM.  However, we need the precise
3823   // address of the call in order to generate an oopmap. Hence, we do all the
3824   // work outselves.
3825 
3826   __ set_last_Java_frame(noreg, noreg, NULL);
3827 
3828   // The return address must always be correct so that frame constructor never
3829   // sees an invalid pc.
3830 
3831   if (!cause_return) {
3832     // overwrite the dummy value we pushed on entry
3833     __ movptr(c_rarg0, Address(r15_thread, JavaThread::saved_exception_pc_offset()));
3834     __ movptr(Address(rbp, wordSize), c_rarg0);
3835   }
3836 
3837   // Do the call
3838   __ mov(c_rarg0, r15_thread);
3839   __ call(RuntimeAddress(call_ptr));
3840 
3841   // Set an oopmap for the call site.  This oopmap will map all
3842   // oop-registers and debug-info registers as callee-saved.  This
3843   // will allow deoptimization at this safepoint to find all possible
3844   // debug-info recordings, as well as let GC find all oops.
3845 
3846   oop_maps->add_gc_map( __ pc() - start, map);
3847 
3848   Label noException;
3849 
3850   __ reset_last_Java_frame(false, false);
3851 
3852   __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
3853   __ jcc(Assembler::equal, noException);
3854 
3855   // Exception pending
3856 
3857   RegisterSaver::restore_live_registers(masm, save_vectors);
3858 
3859   __ jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
3860 
3861   // No exception case
3862   __ bind(noException);
3863 
3864   // Normal exit, restore registers and exit.
3865   RegisterSaver::restore_live_registers(masm, save_vectors);
3866 
3867   __ ret(0);
3868 
3869   // Make sure all code is generated
3870   masm->flush();
3871 
3872   // Fill-out other meta info
3873   return SafepointBlob::create(&buffer, oop_maps, frame_size_in_words);
3874 }
3875 
3876 //
3877 // generate_resolve_blob - call resolution (static/virtual/opt-virtual/ic-miss
3878 //
3879 // Generate a stub that calls into vm to find out the proper destination
3880 // of a java call. All the argument registers are live at this point
3881 // but since this is generic code we don't know what they are and the caller
3882 // must do any gc of the args.
3883 //
3884 RuntimeStub* SharedRuntime::generate_resolve_blob(address destination, const char* name) {
3885   assert (StubRoutines::forward_exception_entry() != NULL, "must be generated before");
3886 
3887   // allocate space for the code
3888   ResourceMark rm;
3889 
3890   CodeBuffer buffer(name, 1000, 512);
3891   MacroAssembler* masm                = new MacroAssembler(&buffer);
3892 
3893   int frame_size_in_words;
3894 
3895   OopMapSet *oop_maps = new OopMapSet();
3896   OopMap* map = NULL;
3897 
3898   int start = __ offset();
3899 
3900   map = RegisterSaver::save_live_registers(masm, 0, &frame_size_in_words);
3901 
3902   int frame_complete = __ offset();
3903 
3904   __ set_last_Java_frame(noreg, noreg, NULL);
3905 
3906   __ mov(c_rarg0, r15_thread);
3907 
3908   __ call(RuntimeAddress(destination));
3909 
3910 
3911   // Set an oopmap for the call site.
3912   // We need this not only for callee-saved registers, but also for volatile
3913   // registers that the compiler might be keeping live across a safepoint.
3914 
3915   oop_maps->add_gc_map( __ offset() - start, map);
3916 
3917   // rax contains the address we are going to jump to assuming no exception got installed
3918 
3919   // clear last_Java_sp
3920   __ reset_last_Java_frame(false, false);
3921   // check for pending exceptions
3922   Label pending;
3923   __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
3924   __ jcc(Assembler::notEqual, pending);
3925 
3926   // get the returned Method*
3927   __ get_vm_result_2(rbx, r15_thread);
3928   __ movptr(Address(rsp, RegisterSaver::rbx_offset_in_bytes()), rbx);
3929 
3930   __ movptr(Address(rsp, RegisterSaver::rax_offset_in_bytes()), rax);
3931 
3932   RegisterSaver::restore_live_registers(masm);
3933 
3934   // We are back the the original state on entry and ready to go.
3935 
3936   __ jmp(rax);
3937 
3938   // Pending exception after the safepoint
3939 
3940   __ bind(pending);
3941 
3942   RegisterSaver::restore_live_registers(masm);
3943 
3944   // exception pending => remove activation and forward to exception handler
3945 
3946   __ movptr(Address(r15_thread, JavaThread::vm_result_offset()), (int)NULL_WORD);
3947 
3948   __ movptr(rax, Address(r15_thread, Thread::pending_exception_offset()));
3949   __ jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
3950 
3951   // -------------
3952   // make sure all code is generated
3953   masm->flush();
3954 
3955   // return the  blob
3956   // frame_size_words or bytes??
3957   return RuntimeStub::new_runtime_stub(name, &buffer, frame_complete, frame_size_in_words, oop_maps, true);
3958 }
3959 
3960 
3961 #ifdef COMPILER2
3962 // This is here instead of runtime_x86_64.cpp because it uses SimpleRuntimeFrame
3963 //
3964 //------------------------------generate_exception_blob---------------------------
3965 // creates exception blob at the end
3966 // Using exception blob, this code is jumped from a compiled method.
3967 // (see emit_exception_handler in x86_64.ad file)
3968 //
3969 // Given an exception pc at a call we call into the runtime for the
3970 // handler in this method. This handler might merely restore state
3971 // (i.e. callee save registers) unwind the frame and jump to the
3972 // exception handler for the nmethod if there is no Java level handler
3973 // for the nmethod.
3974 //
3975 // This code is entered with a jmp.
3976 //
3977 // Arguments:
3978 //   rax: exception oop
3979 //   rdx: exception pc
3980 //
3981 // Results:
3982 //   rax: exception oop
3983 //   rdx: exception pc in caller or ???
3984 //   destination: exception handler of caller
3985 //
3986 // Note: the exception pc MUST be at a call (precise debug information)
3987 //       Registers rax, rdx, rcx, rsi, rdi, r8-r11 are not callee saved.
3988 //
3989 
3990 void OptoRuntime::generate_exception_blob() {
3991   assert(!OptoRuntime::is_callee_saved_register(RDX_num), "");
3992   assert(!OptoRuntime::is_callee_saved_register(RAX_num), "");
3993   assert(!OptoRuntime::is_callee_saved_register(RCX_num), "");
3994 
3995   assert(SimpleRuntimeFrame::framesize % 4 == 0, "sp not 16-byte aligned");
3996 
3997   // Allocate space for the code
3998   ResourceMark rm;
3999   // Setup code generation tools
4000   CodeBuffer buffer("exception_blob", 2048, 1024);
4001   MacroAssembler* masm = new MacroAssembler(&buffer);
4002 
4003 
4004   address start = __ pc();
4005 
4006   // Exception pc is 'return address' for stack walker
4007   __ push(rdx);
4008   __ subptr(rsp, SimpleRuntimeFrame::return_off << LogBytesPerInt); // Prolog
4009 
4010   // Save callee-saved registers.  See x86_64.ad.
4011 
4012   // rbp is an implicitly saved callee saved register (i.e. the calling
4013   // convention will save restore it in prolog/epilog) Other than that
4014   // there are no callee save registers now that adapter frames are gone.
4015 
4016   __ movptr(Address(rsp, SimpleRuntimeFrame::rbp_off << LogBytesPerInt), rbp);
4017 
4018   // Store exception in Thread object. We cannot pass any arguments to the
4019   // handle_exception call, since we do not want to make any assumption
4020   // about the size of the frame where the exception happened in.
4021   // c_rarg0 is either rdi (Linux) or rcx (Windows).
4022   __ movptr(Address(r15_thread, JavaThread::exception_oop_offset()),rax);
4023   __ movptr(Address(r15_thread, JavaThread::exception_pc_offset()), rdx);
4024 
4025   // This call does all the hard work.  It checks if an exception handler
4026   // exists in the method.
4027   // If so, it returns the handler address.
4028   // If not, it prepares for stack-unwinding, restoring the callee-save
4029   // registers of the frame being removed.
4030   //
4031   // address OptoRuntime::handle_exception_C(JavaThread* thread)
4032 
4033   // At a method handle call, the stack may not be properly aligned
4034   // when returning with an exception.
4035   address the_pc = __ pc();
4036   __ set_last_Java_frame(noreg, noreg, the_pc);
4037   __ mov(c_rarg0, r15_thread);
4038   __ andptr(rsp, -(StackAlignmentInBytes));    // Align stack
4039   __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, OptoRuntime::handle_exception_C)));
4040 
4041   // Set an oopmap for the call site.  This oopmap will only be used if we
4042   // are unwinding the stack.  Hence, all locations will be dead.
4043   // Callee-saved registers will be the same as the frame above (i.e.,
4044   // handle_exception_stub), since they were restored when we got the
4045   // exception.
4046 
4047   OopMapSet* oop_maps = new OopMapSet();
4048 
4049   oop_maps->add_gc_map(the_pc - start, new OopMap(SimpleRuntimeFrame::framesize, 0));
4050 
4051   __ reset_last_Java_frame(false, true);
4052 
4053   // Restore callee-saved registers
4054 
4055   // rbp is an implicitly saved callee saved register (i.e. the calling
4056   // convention will save restore it in prolog/epilog) Other than that
4057   // there are no callee save registers no that adapter frames are gone.
4058 
4059   __ movptr(rbp, Address(rsp, SimpleRuntimeFrame::rbp_off << LogBytesPerInt));
4060 
4061   __ addptr(rsp, SimpleRuntimeFrame::return_off << LogBytesPerInt); // Epilog
4062   __ pop(rdx);                  // No need for exception pc anymore
4063 
4064   // rax: exception handler
4065 
4066   // Restore SP from BP if the exception PC is a MethodHandle call site.
4067   __ cmpl(Address(r15_thread, JavaThread::is_method_handle_return_offset()), 0);
4068   __ cmovptr(Assembler::notEqual, rsp, rbp_mh_SP_save);
4069 
4070   // We have a handler in rax (could be deopt blob).
4071   __ mov(r8, rax);
4072 
4073   // Get the exception oop
4074   __ movptr(rax, Address(r15_thread, JavaThread::exception_oop_offset()));
4075   // Get the exception pc in case we are deoptimized
4076   __ movptr(rdx, Address(r15_thread, JavaThread::exception_pc_offset()));
4077 #ifdef ASSERT
4078   __ movptr(Address(r15_thread, JavaThread::exception_handler_pc_offset()), (int)NULL_WORD);
4079   __ movptr(Address(r15_thread, JavaThread::exception_pc_offset()), (int)NULL_WORD);
4080 #endif
4081   // Clear the exception oop so GC no longer processes it as a root.
4082   __ movptr(Address(r15_thread, JavaThread::exception_oop_offset()), (int)NULL_WORD);
4083 
4084   // rax: exception oop
4085   // r8:  exception handler
4086   // rdx: exception pc
4087   // Jump to handler
4088 
4089   __ jmp(r8);
4090 
4091   // Make sure all code is generated
4092   masm->flush();
4093 
4094   // Set exception blob
4095   _exception_blob =  ExceptionBlob::create(&buffer, oop_maps, SimpleRuntimeFrame::framesize >> 1);
4096 }
4097 #endif // COMPILER2