1 /*
   2  * Copyright 2003-2009 Sun Microsystems, Inc.  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 Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  20  * CA 95054 USA or visit www.sun.com if you need additional information or
  21  * have any questions.
  22  *
  23  */
  24 
  25 #include "incls/_precompiled.incl"
  26 #include "incls/_sharedRuntime_x86_64.cpp.incl"
  27 
  28 DeoptimizationBlob *SharedRuntime::_deopt_blob;
  29 #ifdef COMPILER2
  30 UncommonTrapBlob   *SharedRuntime::_uncommon_trap_blob;
  31 ExceptionBlob      *OptoRuntime::_exception_blob;
  32 #endif // COMPILER2
  33 
  34 SafepointBlob      *SharedRuntime::_polling_page_safepoint_handler_blob;
  35 SafepointBlob      *SharedRuntime::_polling_page_return_handler_blob;
  36 RuntimeStub*       SharedRuntime::_wrong_method_blob;
  37 RuntimeStub*       SharedRuntime::_ic_miss_blob;
  38 RuntimeStub*       SharedRuntime::_resolve_opt_virtual_call_blob;
  39 RuntimeStub*       SharedRuntime::_resolve_virtual_call_blob;
  40 RuntimeStub*       SharedRuntime::_resolve_static_call_blob;
  41 
  42 const int StackAlignmentInSlots = StackAlignmentInBytes / VMRegImpl::stack_slot_size;
  43 
  44 #define __ masm->
  45 
  46 class SimpleRuntimeFrame {
  47 
  48   public:
  49 
  50   // Most of the runtime stubs have this simple frame layout.
  51   // This class exists to make the layout shared in one place.
  52   // Offsets are for compiler stack slots, which are jints.
  53   enum layout {
  54     // The frame sender code expects that rbp will be in the "natural" place and
  55     // will override any oopMap setting for it. We must therefore force the layout
  56     // so that it agrees with the frame sender code.
  57     rbp_off = frame::arg_reg_save_area_bytes/BytesPerInt,
  58     rbp_off2,
  59     return_off, return_off2,
  60     framesize
  61   };
  62 };
  63 
  64 class RegisterSaver {
  65   // Capture info about frame layout.  Layout offsets are in jint
  66   // units because compiler frame slots are jints.
  67 #define DEF_XMM_OFFS(regnum) xmm ## regnum ## _off = xmm_off + (regnum)*16/BytesPerInt, xmm ## regnum ## H_off
  68   enum layout {
  69     fpu_state_off = frame::arg_reg_save_area_bytes/BytesPerInt, // fxsave save area
  70     xmm_off       = fpu_state_off + 160/BytesPerInt,            // offset in fxsave save area
  71     DEF_XMM_OFFS(0),
  72     DEF_XMM_OFFS(1),
  73     DEF_XMM_OFFS(2),
  74     DEF_XMM_OFFS(3),
  75     DEF_XMM_OFFS(4),
  76     DEF_XMM_OFFS(5),
  77     DEF_XMM_OFFS(6),
  78     DEF_XMM_OFFS(7),
  79     DEF_XMM_OFFS(8),
  80     DEF_XMM_OFFS(9),
  81     DEF_XMM_OFFS(10),
  82     DEF_XMM_OFFS(11),
  83     DEF_XMM_OFFS(12),
  84     DEF_XMM_OFFS(13),
  85     DEF_XMM_OFFS(14),
  86     DEF_XMM_OFFS(15),
  87     fpu_state_end = fpu_state_off + ((FPUStateSizeInWords-1)*wordSize / BytesPerInt),
  88     fpu_stateH_end,
  89     r15_off, r15H_off,
  90     r14_off, r14H_off,
  91     r13_off, r13H_off,
  92     r12_off, r12H_off,
  93     r11_off, r11H_off,
  94     r10_off, r10H_off,
  95     r9_off,  r9H_off,
  96     r8_off,  r8H_off,
  97     rdi_off, rdiH_off,
  98     rsi_off, rsiH_off,
  99     ignore_off, ignoreH_off,  // extra copy of rbp
 100     rsp_off, rspH_off,
 101     rbx_off, rbxH_off,
 102     rdx_off, rdxH_off,
 103     rcx_off, rcxH_off,
 104     rax_off, raxH_off,
 105     // 16-byte stack alignment fill word: see MacroAssembler::push/pop_IU_state
 106     align_off, alignH_off,
 107     flags_off, flagsH_off,
 108     // The frame sender code expects that rbp will be in the "natural" place and
 109     // will override any oopMap setting for it. We must therefore force the layout
 110     // so that it agrees with the frame sender code.
 111     rbp_off, rbpH_off,        // copy of rbp we will restore
 112     return_off, returnH_off,  // slot for return address
 113     reg_save_size             // size in compiler stack slots
 114   };
 115 
 116  public:
 117   static OopMap* save_live_registers(MacroAssembler* masm, int additional_frame_words, int* total_frame_words);
 118   static void restore_live_registers(MacroAssembler* masm);
 119 
 120   // Offsets into the register save area
 121   // Used by deoptimization when it is managing result register
 122   // values on its own
 123 
 124   static int rax_offset_in_bytes(void)    { return BytesPerInt * rax_off; }
 125   static int rdx_offset_in_bytes(void)    { return BytesPerInt * rdx_off; }
 126   static int rbx_offset_in_bytes(void)    { return BytesPerInt * rbx_off; }
 127   static int xmm0_offset_in_bytes(void)   { return BytesPerInt * xmm0_off; }
 128   static int return_offset_in_bytes(void) { return BytesPerInt * return_off; }
 129 
 130   // During deoptimization only the result registers need to be restored,
 131   // all the other values have already been extracted.
 132   static void restore_result_registers(MacroAssembler* masm);
 133 };
 134 
 135 OopMap* RegisterSaver::save_live_registers(MacroAssembler* masm, int additional_frame_words, int* total_frame_words) {
 136 
 137   // Always make the frame size 16-byte aligned
 138   int frame_size_in_bytes = round_to(additional_frame_words*wordSize +
 139                                      reg_save_size*BytesPerInt, 16);
 140   // OopMap frame size is in compiler stack slots (jint's) not bytes or words
 141   int frame_size_in_slots = frame_size_in_bytes / BytesPerInt;
 142   // The caller will allocate additional_frame_words
 143   int additional_frame_slots = additional_frame_words*wordSize / BytesPerInt;
 144   // CodeBlob frame size is in words.
 145   int frame_size_in_words = frame_size_in_bytes / wordSize;
 146   *total_frame_words = frame_size_in_words;
 147 
 148   // Save registers, fpu state, and flags.
 149   // We assume caller has already pushed the return address onto the
 150   // stack, so rsp is 8-byte aligned here.
 151   // We push rpb twice in this sequence because we want the real rbp
 152   // to be under the return like a normal enter.
 153 
 154   __ enter();          // rsp becomes 16-byte aligned here
 155   __ push_CPU_state(); // Push a multiple of 16 bytes
 156   if (frame::arg_reg_save_area_bytes != 0) {
 157     // Allocate argument register save area
 158     __ subptr(rsp, frame::arg_reg_save_area_bytes);
 159   }
 160 
 161   // Set an oopmap for the call site.  This oopmap will map all
 162   // oop-registers and debug-info registers as callee-saved.  This
 163   // will allow deoptimization at this safepoint to find all possible
 164   // debug-info recordings, as well as let GC find all oops.
 165 
 166   OopMapSet *oop_maps = new OopMapSet();
 167   OopMap* map = new OopMap(frame_size_in_slots, 0);
 168   map->set_callee_saved(VMRegImpl::stack2reg( rax_off  + additional_frame_slots), rax->as_VMReg());
 169   map->set_callee_saved(VMRegImpl::stack2reg( rcx_off  + additional_frame_slots), rcx->as_VMReg());
 170   map->set_callee_saved(VMRegImpl::stack2reg( rdx_off  + additional_frame_slots), rdx->as_VMReg());
 171   map->set_callee_saved(VMRegImpl::stack2reg( rbx_off  + additional_frame_slots), rbx->as_VMReg());
 172   // rbp location is known implicitly by the frame sender code, needs no oopmap
 173   // and the location where rbp was saved by is ignored
 174   map->set_callee_saved(VMRegImpl::stack2reg( rsi_off  + additional_frame_slots), rsi->as_VMReg());
 175   map->set_callee_saved(VMRegImpl::stack2reg( rdi_off  + additional_frame_slots), rdi->as_VMReg());
 176   map->set_callee_saved(VMRegImpl::stack2reg( r8_off   + additional_frame_slots), r8->as_VMReg());
 177   map->set_callee_saved(VMRegImpl::stack2reg( r9_off   + additional_frame_slots), r9->as_VMReg());
 178   map->set_callee_saved(VMRegImpl::stack2reg( r10_off  + additional_frame_slots), r10->as_VMReg());
 179   map->set_callee_saved(VMRegImpl::stack2reg( r11_off  + additional_frame_slots), r11->as_VMReg());
 180   map->set_callee_saved(VMRegImpl::stack2reg( r12_off  + additional_frame_slots), r12->as_VMReg());
 181   map->set_callee_saved(VMRegImpl::stack2reg( r13_off  + additional_frame_slots), r13->as_VMReg());
 182   map->set_callee_saved(VMRegImpl::stack2reg( r14_off  + additional_frame_slots), r14->as_VMReg());
 183   map->set_callee_saved(VMRegImpl::stack2reg( r15_off  + additional_frame_slots), r15->as_VMReg());
 184   map->set_callee_saved(VMRegImpl::stack2reg(xmm0_off  + additional_frame_slots), xmm0->as_VMReg());
 185   map->set_callee_saved(VMRegImpl::stack2reg(xmm1_off  + additional_frame_slots), xmm1->as_VMReg());
 186   map->set_callee_saved(VMRegImpl::stack2reg(xmm2_off  + additional_frame_slots), xmm2->as_VMReg());
 187   map->set_callee_saved(VMRegImpl::stack2reg(xmm3_off  + additional_frame_slots), xmm3->as_VMReg());
 188   map->set_callee_saved(VMRegImpl::stack2reg(xmm4_off  + additional_frame_slots), xmm4->as_VMReg());
 189   map->set_callee_saved(VMRegImpl::stack2reg(xmm5_off  + additional_frame_slots), xmm5->as_VMReg());
 190   map->set_callee_saved(VMRegImpl::stack2reg(xmm6_off  + additional_frame_slots), xmm6->as_VMReg());
 191   map->set_callee_saved(VMRegImpl::stack2reg(xmm7_off  + additional_frame_slots), xmm7->as_VMReg());
 192   map->set_callee_saved(VMRegImpl::stack2reg(xmm8_off  + additional_frame_slots), xmm8->as_VMReg());
 193   map->set_callee_saved(VMRegImpl::stack2reg(xmm9_off  + additional_frame_slots), xmm9->as_VMReg());
 194   map->set_callee_saved(VMRegImpl::stack2reg(xmm10_off + additional_frame_slots), xmm10->as_VMReg());
 195   map->set_callee_saved(VMRegImpl::stack2reg(xmm11_off + additional_frame_slots), xmm11->as_VMReg());
 196   map->set_callee_saved(VMRegImpl::stack2reg(xmm12_off + additional_frame_slots), xmm12->as_VMReg());
 197   map->set_callee_saved(VMRegImpl::stack2reg(xmm13_off + additional_frame_slots), xmm13->as_VMReg());
 198   map->set_callee_saved(VMRegImpl::stack2reg(xmm14_off + additional_frame_slots), xmm14->as_VMReg());
 199   map->set_callee_saved(VMRegImpl::stack2reg(xmm15_off + additional_frame_slots), xmm15->as_VMReg());
 200 
 201   // %%% These should all be a waste but we'll keep things as they were for now
 202   if (true) {
 203     map->set_callee_saved(VMRegImpl::stack2reg( raxH_off  + additional_frame_slots),
 204                           rax->as_VMReg()->next());
 205     map->set_callee_saved(VMRegImpl::stack2reg( rcxH_off  + additional_frame_slots),
 206                           rcx->as_VMReg()->next());
 207     map->set_callee_saved(VMRegImpl::stack2reg( rdxH_off  + additional_frame_slots),
 208                           rdx->as_VMReg()->next());
 209     map->set_callee_saved(VMRegImpl::stack2reg( rbxH_off  + additional_frame_slots),
 210                           rbx->as_VMReg()->next());
 211     // rbp location is known implicitly by the frame sender code, needs no oopmap
 212     map->set_callee_saved(VMRegImpl::stack2reg( rsiH_off  + additional_frame_slots),
 213                           rsi->as_VMReg()->next());
 214     map->set_callee_saved(VMRegImpl::stack2reg( rdiH_off  + additional_frame_slots),
 215                           rdi->as_VMReg()->next());
 216     map->set_callee_saved(VMRegImpl::stack2reg( r8H_off   + additional_frame_slots),
 217                           r8->as_VMReg()->next());
 218     map->set_callee_saved(VMRegImpl::stack2reg( r9H_off   + additional_frame_slots),
 219                           r9->as_VMReg()->next());
 220     map->set_callee_saved(VMRegImpl::stack2reg( r10H_off  + additional_frame_slots),
 221                           r10->as_VMReg()->next());
 222     map->set_callee_saved(VMRegImpl::stack2reg( r11H_off  + additional_frame_slots),
 223                           r11->as_VMReg()->next());
 224     map->set_callee_saved(VMRegImpl::stack2reg( r12H_off  + additional_frame_slots),
 225                           r12->as_VMReg()->next());
 226     map->set_callee_saved(VMRegImpl::stack2reg( r13H_off  + additional_frame_slots),
 227                           r13->as_VMReg()->next());
 228     map->set_callee_saved(VMRegImpl::stack2reg( r14H_off  + additional_frame_slots),
 229                           r14->as_VMReg()->next());
 230     map->set_callee_saved(VMRegImpl::stack2reg( r15H_off  + additional_frame_slots),
 231                           r15->as_VMReg()->next());
 232     map->set_callee_saved(VMRegImpl::stack2reg(xmm0H_off  + additional_frame_slots),
 233                           xmm0->as_VMReg()->next());
 234     map->set_callee_saved(VMRegImpl::stack2reg(xmm1H_off  + additional_frame_slots),
 235                           xmm1->as_VMReg()->next());
 236     map->set_callee_saved(VMRegImpl::stack2reg(xmm2H_off  + additional_frame_slots),
 237                           xmm2->as_VMReg()->next());
 238     map->set_callee_saved(VMRegImpl::stack2reg(xmm3H_off  + additional_frame_slots),
 239                           xmm3->as_VMReg()->next());
 240     map->set_callee_saved(VMRegImpl::stack2reg(xmm4H_off  + additional_frame_slots),
 241                           xmm4->as_VMReg()->next());
 242     map->set_callee_saved(VMRegImpl::stack2reg(xmm5H_off  + additional_frame_slots),
 243                           xmm5->as_VMReg()->next());
 244     map->set_callee_saved(VMRegImpl::stack2reg(xmm6H_off  + additional_frame_slots),
 245                           xmm6->as_VMReg()->next());
 246     map->set_callee_saved(VMRegImpl::stack2reg(xmm7H_off  + additional_frame_slots),
 247                           xmm7->as_VMReg()->next());
 248     map->set_callee_saved(VMRegImpl::stack2reg(xmm8H_off  + additional_frame_slots),
 249                           xmm8->as_VMReg()->next());
 250     map->set_callee_saved(VMRegImpl::stack2reg(xmm9H_off  + additional_frame_slots),
 251                           xmm9->as_VMReg()->next());
 252     map->set_callee_saved(VMRegImpl::stack2reg(xmm10H_off + additional_frame_slots),
 253                           xmm10->as_VMReg()->next());
 254     map->set_callee_saved(VMRegImpl::stack2reg(xmm11H_off + additional_frame_slots),
 255                           xmm11->as_VMReg()->next());
 256     map->set_callee_saved(VMRegImpl::stack2reg(xmm12H_off + additional_frame_slots),
 257                           xmm12->as_VMReg()->next());
 258     map->set_callee_saved(VMRegImpl::stack2reg(xmm13H_off + additional_frame_slots),
 259                           xmm13->as_VMReg()->next());
 260     map->set_callee_saved(VMRegImpl::stack2reg(xmm14H_off + additional_frame_slots),
 261                           xmm14->as_VMReg()->next());
 262     map->set_callee_saved(VMRegImpl::stack2reg(xmm15H_off + additional_frame_slots),
 263                           xmm15->as_VMReg()->next());
 264   }
 265 
 266   return map;
 267 }
 268 
 269 void RegisterSaver::restore_live_registers(MacroAssembler* masm) {
 270   if (frame::arg_reg_save_area_bytes != 0) {
 271     // Pop arg register save area
 272     __ addptr(rsp, frame::arg_reg_save_area_bytes);
 273   }
 274   // Recover CPU state
 275   __ pop_CPU_state();
 276   // Get the rbp described implicitly by the calling convention (no oopMap)
 277   __ pop(rbp);
 278 }
 279 
 280 void RegisterSaver::restore_result_registers(MacroAssembler* masm) {
 281 
 282   // Just restore result register. Only used by deoptimization. By
 283   // now any callee save register that needs to be restored to a c2
 284   // caller of the deoptee has been extracted into the vframeArray
 285   // and will be stuffed into the c2i adapter we create for later
 286   // restoration so only result registers need to be restored here.
 287 
 288   // Restore fp result register
 289   __ movdbl(xmm0, Address(rsp, xmm0_offset_in_bytes()));
 290   // Restore integer result register
 291   __ movptr(rax, Address(rsp, rax_offset_in_bytes()));
 292   __ movptr(rdx, Address(rsp, rdx_offset_in_bytes()));
 293 
 294   // Pop all of the register save are off the stack except the return address
 295   __ addptr(rsp, return_offset_in_bytes());
 296 }
 297 
 298 // The java_calling_convention describes stack locations as ideal slots on
 299 // a frame with no abi restrictions. Since we must observe abi restrictions
 300 // (like the placement of the register window) the slots must be biased by
 301 // the following value.
 302 static int reg2offset_in(VMReg r) {
 303   // Account for saved rbp and return address
 304   // This should really be in_preserve_stack_slots
 305   return (r->reg2stack() + 4) * VMRegImpl::stack_slot_size;
 306 }
 307 
 308 static int reg2offset_out(VMReg r) {
 309   return (r->reg2stack() + SharedRuntime::out_preserve_stack_slots()) * VMRegImpl::stack_slot_size;
 310 }
 311 
 312 // ---------------------------------------------------------------------------
 313 // Read the array of BasicTypes from a signature, and compute where the
 314 // arguments should go.  Values in the VMRegPair regs array refer to 4-byte
 315 // quantities.  Values less than VMRegImpl::stack0 are registers, those above
 316 // refer to 4-byte stack slots.  All stack slots are based off of the stack pointer
 317 // as framesizes are fixed.
 318 // VMRegImpl::stack0 refers to the first slot 0(sp).
 319 // and VMRegImpl::stack0+1 refers to the memory word 4-byes higher.  Register
 320 // up to RegisterImpl::number_of_registers) are the 64-bit
 321 // integer registers.
 322 
 323 // Note: the INPUTS in sig_bt are in units of Java argument words, which are
 324 // either 32-bit or 64-bit depending on the build.  The OUTPUTS are in 32-bit
 325 // units regardless of build. Of course for i486 there is no 64 bit build
 326 
 327 // The Java calling convention is a "shifted" version of the C ABI.
 328 // By skipping the first C ABI register we can call non-static jni methods
 329 // with small numbers of arguments without having to shuffle the arguments
 330 // at all. Since we control the java ABI we ought to at least get some
 331 // advantage out of it.
 332 
 333 int SharedRuntime::java_calling_convention(const BasicType *sig_bt,
 334                                            VMRegPair *regs,
 335                                            int total_args_passed,
 336                                            int is_outgoing) {
 337 
 338   // Create the mapping between argument positions and
 339   // registers.
 340   static const Register INT_ArgReg[Argument::n_int_register_parameters_j] = {
 341     j_rarg0, j_rarg1, j_rarg2, j_rarg3, j_rarg4, j_rarg5
 342   };
 343   static const XMMRegister FP_ArgReg[Argument::n_float_register_parameters_j] = {
 344     j_farg0, j_farg1, j_farg2, j_farg3,
 345     j_farg4, j_farg5, j_farg6, j_farg7
 346   };
 347 
 348 
 349   uint int_args = 0;
 350   uint fp_args = 0;
 351   uint stk_args = 0; // inc by 2 each time
 352 
 353   for (int i = 0; i < total_args_passed; i++) {
 354     switch (sig_bt[i]) {
 355     case T_BOOLEAN:
 356     case T_CHAR:
 357     case T_BYTE:
 358     case T_SHORT:
 359     case T_INT:
 360       if (int_args < Argument::n_int_register_parameters_j) {
 361         regs[i].set1(INT_ArgReg[int_args++]->as_VMReg());
 362       } else {
 363         regs[i].set1(VMRegImpl::stack2reg(stk_args));
 364         stk_args += 2;
 365       }
 366       break;
 367     case T_VOID:
 368       // halves of T_LONG or T_DOUBLE
 369       assert(i != 0 && (sig_bt[i - 1] == T_LONG || sig_bt[i - 1] == T_DOUBLE), "expecting half");
 370       regs[i].set_bad();
 371       break;
 372     case T_LONG:
 373       assert(sig_bt[i + 1] == T_VOID, "expecting half");
 374       // fall through
 375     case T_OBJECT:
 376     case T_ARRAY:
 377     case T_ADDRESS:
 378       if (int_args < Argument::n_int_register_parameters_j) {
 379         regs[i].set2(INT_ArgReg[int_args++]->as_VMReg());
 380       } else {
 381         regs[i].set2(VMRegImpl::stack2reg(stk_args));
 382         stk_args += 2;
 383       }
 384       break;
 385     case T_FLOAT:
 386       if (fp_args < Argument::n_float_register_parameters_j) {
 387         regs[i].set1(FP_ArgReg[fp_args++]->as_VMReg());
 388       } else {
 389         regs[i].set1(VMRegImpl::stack2reg(stk_args));
 390         stk_args += 2;
 391       }
 392       break;
 393     case T_DOUBLE:
 394       assert(sig_bt[i + 1] == T_VOID, "expecting half");
 395       if (fp_args < Argument::n_float_register_parameters_j) {
 396         regs[i].set2(FP_ArgReg[fp_args++]->as_VMReg());
 397       } else {
 398         regs[i].set2(VMRegImpl::stack2reg(stk_args));
 399         stk_args += 2;
 400       }
 401       break;
 402     default:
 403       ShouldNotReachHere();
 404       break;
 405     }
 406   }
 407 
 408   return round_to(stk_args, 2);
 409 }
 410 
 411 // Patch the callers callsite with entry to compiled code if it exists.
 412 static void patch_callers_callsite(MacroAssembler *masm) {
 413   Label L;
 414   __ verify_oop(rbx);
 415   __ cmpptr(Address(rbx, in_bytes(methodOopDesc::code_offset())), (int32_t)NULL_WORD);
 416   __ jcc(Assembler::equal, L);
 417 
 418   // Save the current stack pointer
 419   __ mov(r13, rsp);
 420   // Schedule the branch target address early.
 421   // Call into the VM to patch the caller, then jump to compiled callee
 422   // rax isn't live so capture return address while we easily can
 423   __ movptr(rax, Address(rsp, 0));
 424 
 425   // align stack so push_CPU_state doesn't fault
 426   __ andptr(rsp, -(StackAlignmentInBytes));
 427   __ push_CPU_state();
 428 
 429 
 430   __ verify_oop(rbx);
 431   // VM needs caller's callsite
 432   // VM needs target method
 433   // This needs to be a long call since we will relocate this adapter to
 434   // the codeBuffer and it may not reach
 435 
 436   // Allocate argument register save area
 437   if (frame::arg_reg_save_area_bytes != 0) {
 438     __ subptr(rsp, frame::arg_reg_save_area_bytes);
 439   }
 440   __ mov(c_rarg0, rbx);
 441   __ mov(c_rarg1, rax);
 442   __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::fixup_callers_callsite)));
 443 
 444   // De-allocate argument register save area
 445   if (frame::arg_reg_save_area_bytes != 0) {
 446     __ addptr(rsp, frame::arg_reg_save_area_bytes);
 447   }
 448 
 449   __ pop_CPU_state();
 450   // restore sp
 451   __ mov(rsp, r13);
 452   __ bind(L);
 453 }
 454 
 455 // Helper function to put tags in interpreter stack.
 456 static void  tag_stack(MacroAssembler *masm, const BasicType sig, int st_off) {
 457   if (TaggedStackInterpreter) {
 458     int tag_offset = st_off + Interpreter::expr_tag_offset_in_bytes(0);
 459     if (sig == T_OBJECT || sig == T_ARRAY) {
 460       __ movptr(Address(rsp, tag_offset), (int32_t) frame::TagReference);
 461     } else if (sig == T_LONG || sig == T_DOUBLE) {
 462       int next_tag_offset = st_off + Interpreter::expr_tag_offset_in_bytes(1);
 463       __ movptr(Address(rsp, next_tag_offset), (int32_t) frame::TagValue);
 464       __ movptr(Address(rsp, tag_offset), (int32_t) frame::TagValue);
 465     } else {
 466       __ movptr(Address(rsp, tag_offset), (int32_t) frame::TagValue);
 467     }
 468   }
 469 }
 470 
 471 
 472 static void gen_c2i_adapter(MacroAssembler *masm,
 473                             int total_args_passed,
 474                             int comp_args_on_stack,
 475                             const BasicType *sig_bt,
 476                             const VMRegPair *regs,
 477                             Label& skip_fixup) {
 478   // Before we get into the guts of the C2I adapter, see if we should be here
 479   // at all.  We've come from compiled code and are attempting to jump to the
 480   // interpreter, which means the caller made a static call to get here
 481   // (vcalls always get a compiled target if there is one).  Check for a
 482   // compiled target.  If there is one, we need to patch the caller's call.
 483   patch_callers_callsite(masm);
 484 
 485   __ bind(skip_fixup);
 486 
 487   // Since all args are passed on the stack, total_args_passed *
 488   // Interpreter::stackElementSize is the space we need. Plus 1 because
 489   // we also account for the return address location since
 490   // we store it first rather than hold it in rax across all the shuffling
 491 
 492   int extraspace = (total_args_passed * Interpreter::stackElementSize()) + wordSize;
 493 
 494   // stack is aligned, keep it that way
 495   extraspace = round_to(extraspace, 2*wordSize);
 496 
 497   // Get return address
 498   __ pop(rax);
 499 
 500   // set senderSP value
 501   __ mov(r13, rsp);
 502 
 503   __ subptr(rsp, extraspace);
 504 
 505   // Store the return address in the expected location
 506   __ movptr(Address(rsp, 0), rax);
 507 
 508   // Now write the args into the outgoing interpreter space
 509   for (int i = 0; i < total_args_passed; i++) {
 510     if (sig_bt[i] == T_VOID) {
 511       assert(i > 0 && (sig_bt[i-1] == T_LONG || sig_bt[i-1] == T_DOUBLE), "missing half");
 512       continue;
 513     }
 514 
 515     // offset to start parameters
 516     int st_off   = (total_args_passed - i) * Interpreter::stackElementSize() +
 517                    Interpreter::value_offset_in_bytes();
 518     int next_off = st_off - Interpreter::stackElementSize();
 519 
 520     // Say 4 args:
 521     // i   st_off
 522     // 0   32 T_LONG
 523     // 1   24 T_VOID
 524     // 2   16 T_OBJECT
 525     // 3    8 T_BOOL
 526     // -    0 return address
 527     //
 528     // However to make thing extra confusing. Because we can fit a long/double in
 529     // a single slot on a 64 bt vm and it would be silly to break them up, the interpreter
 530     // leaves one slot empty and only stores to a single slot. In this case the
 531     // slot that is occupied is the T_VOID slot. See I said it was confusing.
 532 
 533     VMReg r_1 = regs[i].first();
 534     VMReg r_2 = regs[i].second();
 535     if (!r_1->is_valid()) {
 536       assert(!r_2->is_valid(), "");
 537       continue;
 538     }
 539     if (r_1->is_stack()) {
 540       // memory to memory use rax
 541       int ld_off = r_1->reg2stack() * VMRegImpl::stack_slot_size + extraspace;
 542       if (!r_2->is_valid()) {
 543         // sign extend??
 544         __ movl(rax, Address(rsp, ld_off));
 545         __ movptr(Address(rsp, st_off), rax);
 546         tag_stack(masm, sig_bt[i], st_off);
 547 
 548       } else {
 549 
 550         __ movq(rax, Address(rsp, ld_off));
 551 
 552         // Two VMREgs|OptoRegs can be T_OBJECT, T_ADDRESS, T_DOUBLE, T_LONG
 553         // T_DOUBLE and T_LONG use two slots in the interpreter
 554         if ( sig_bt[i] == T_LONG || sig_bt[i] == T_DOUBLE) {
 555           // ld_off == LSW, ld_off+wordSize == MSW
 556           // st_off == MSW, next_off == LSW
 557           __ movq(Address(rsp, next_off), rax);
 558 #ifdef ASSERT
 559           // Overwrite the unused slot with known junk
 560           __ mov64(rax, CONST64(0xdeadffffdeadaaaa));
 561           __ movptr(Address(rsp, st_off), rax);
 562 #endif /* ASSERT */
 563           tag_stack(masm, sig_bt[i], next_off);
 564         } else {
 565           __ movq(Address(rsp, st_off), rax);
 566           tag_stack(masm, sig_bt[i], st_off);
 567         }
 568       }
 569     } else if (r_1->is_Register()) {
 570       Register r = r_1->as_Register();
 571       if (!r_2->is_valid()) {
 572         // must be only an int (or less ) so move only 32bits to slot
 573         // why not sign extend??
 574         __ movl(Address(rsp, st_off), r);
 575         tag_stack(masm, sig_bt[i], st_off);
 576       } else {
 577         // Two VMREgs|OptoRegs can be T_OBJECT, T_ADDRESS, T_DOUBLE, T_LONG
 578         // T_DOUBLE and T_LONG use two slots in the interpreter
 579         if ( sig_bt[i] == T_LONG || sig_bt[i] == T_DOUBLE) {
 580           // long/double in gpr
 581 #ifdef ASSERT
 582           // Overwrite the unused slot with known junk
 583           __ mov64(rax, CONST64(0xdeadffffdeadaaab));
 584           __ movptr(Address(rsp, st_off), rax);
 585 #endif /* ASSERT */
 586           __ movq(Address(rsp, next_off), r);
 587           tag_stack(masm, sig_bt[i], next_off);
 588         } else {
 589           __ movptr(Address(rsp, st_off), r);
 590           tag_stack(masm, sig_bt[i], st_off);
 591         }
 592       }
 593     } else {
 594       assert(r_1->is_XMMRegister(), "");
 595       if (!r_2->is_valid()) {
 596         // only a float use just part of the slot
 597         __ movflt(Address(rsp, st_off), r_1->as_XMMRegister());
 598         tag_stack(masm, sig_bt[i], st_off);
 599       } else {
 600 #ifdef ASSERT
 601         // Overwrite the unused slot with known junk
 602         __ mov64(rax, CONST64(0xdeadffffdeadaaac));
 603         __ movptr(Address(rsp, st_off), rax);
 604 #endif /* ASSERT */
 605         __ movdbl(Address(rsp, next_off), r_1->as_XMMRegister());
 606         tag_stack(masm, sig_bt[i], next_off);
 607       }
 608     }
 609   }
 610 
 611   // Schedule the branch target address early.
 612   __ movptr(rcx, Address(rbx, in_bytes(methodOopDesc::interpreter_entry_offset())));
 613   __ jmp(rcx);
 614 }
 615 
 616 static void gen_i2c_adapter(MacroAssembler *masm,
 617                             int total_args_passed,
 618                             int comp_args_on_stack,
 619                             const BasicType *sig_bt,
 620                             const VMRegPair *regs) {
 621 
 622   //
 623   // We will only enter here from an interpreted frame and never from after
 624   // passing thru a c2i. Azul allowed this but we do not. If we lose the
 625   // race and use a c2i we will remain interpreted for the race loser(s).
 626   // This removes all sorts of headaches on the x86 side and also eliminates
 627   // the possibility of having c2i -> i2c -> c2i -> ... endless transitions.
 628 
 629 
 630   // Note: r13 contains the senderSP on entry. We must preserve it since
 631   // we may do a i2c -> c2i transition if we lose a race where compiled
 632   // code goes non-entrant while we get args ready.
 633   // In addition we use r13 to locate all the interpreter args as
 634   // we must align the stack to 16 bytes on an i2c entry else we
 635   // lose alignment we expect in all compiled code and register
 636   // save code can segv when fxsave instructions find improperly
 637   // aligned stack pointer.
 638 
 639   __ movptr(rax, Address(rsp, 0));
 640 
 641   // Cut-out for having no stack args.  Since up to 2 int/oop args are passed
 642   // in registers, we will occasionally have no stack args.
 643   int comp_words_on_stack = 0;
 644   if (comp_args_on_stack) {
 645     // Sig words on the stack are greater-than VMRegImpl::stack0.  Those in
 646     // registers are below.  By subtracting stack0, we either get a negative
 647     // number (all values in registers) or the maximum stack slot accessed.
 648 
 649     // Convert 4-byte c2 stack slots to words.
 650     comp_words_on_stack = round_to(comp_args_on_stack*VMRegImpl::stack_slot_size, wordSize)>>LogBytesPerWord;
 651     // Round up to miminum stack alignment, in wordSize
 652     comp_words_on_stack = round_to(comp_words_on_stack, 2);
 653     __ subptr(rsp, comp_words_on_stack * wordSize);
 654   }
 655 
 656 
 657   // Ensure compiled code always sees stack at proper alignment
 658   __ andptr(rsp, -16);
 659 
 660   // push the return address and misalign the stack that youngest frame always sees
 661   // as far as the placement of the call instruction
 662   __ push(rax);
 663 
 664   // Will jump to the compiled code just as if compiled code was doing it.
 665   // Pre-load the register-jump target early, to schedule it better.
 666   __ movptr(r11, Address(rbx, in_bytes(methodOopDesc::from_compiled_offset())));
 667 
 668   // Now generate the shuffle code.  Pick up all register args and move the
 669   // rest through the floating point stack top.
 670   for (int i = 0; i < total_args_passed; i++) {
 671     if (sig_bt[i] == T_VOID) {
 672       // Longs and doubles are passed in native word order, but misaligned
 673       // in the 32-bit build.
 674       assert(i > 0 && (sig_bt[i-1] == T_LONG || sig_bt[i-1] == T_DOUBLE), "missing half");
 675       continue;
 676     }
 677 
 678     // Pick up 0, 1 or 2 words from SP+offset.
 679 
 680     assert(!regs[i].second()->is_valid() || regs[i].first()->next() == regs[i].second(),
 681             "scrambled load targets?");
 682     // Load in argument order going down.
 683     // int ld_off = (total_args_passed + comp_words_on_stack -i)*wordSize;
 684     // base ld_off on r13 (sender_sp) as the stack alignment makes offsets from rsp
 685     // unpredictable
 686     int ld_off = ((total_args_passed - 1) - i)*Interpreter::stackElementSize();
 687 
 688     // Point to interpreter value (vs. tag)
 689     int next_off = ld_off - Interpreter::stackElementSize();
 690     //
 691     //
 692     //
 693     VMReg r_1 = regs[i].first();
 694     VMReg r_2 = regs[i].second();
 695     if (!r_1->is_valid()) {
 696       assert(!r_2->is_valid(), "");
 697       continue;
 698     }
 699     if (r_1->is_stack()) {
 700       // Convert stack slot to an SP offset (+ wordSize to account for return address )
 701       int st_off = regs[i].first()->reg2stack()*VMRegImpl::stack_slot_size + wordSize;
 702       if (!r_2->is_valid()) {
 703         // sign extend???
 704         __ movl(rax, Address(r13, ld_off));
 705         __ movptr(Address(rsp, st_off), rax);
 706       } else {
 707         //
 708         // We are using two optoregs. This can be either T_OBJECT, T_ADDRESS, T_LONG, or T_DOUBLE
 709         // the interpreter allocates two slots but only uses one for thr T_LONG or T_DOUBLE case
 710         // So we must adjust where to pick up the data to match the interpreter.
 711         //
 712         // Interpreter local[n] == MSW, local[n+1] == LSW however locals
 713         // are accessed as negative so LSW is at LOW address
 714 
 715         // ld_off is MSW so get LSW
 716         const int offset = (sig_bt[i]==T_LONG||sig_bt[i]==T_DOUBLE)?
 717                            next_off : ld_off;
 718         __ movq(rax, Address(r13, offset));
 719         // st_off is LSW (i.e. reg.first())
 720         __ movq(Address(rsp, st_off), rax);
 721       }
 722     } else if (r_1->is_Register()) {  // Register argument
 723       Register r = r_1->as_Register();
 724       assert(r != rax, "must be different");
 725       if (r_2->is_valid()) {
 726         //
 727         // We are using two VMRegs. This can be either T_OBJECT, T_ADDRESS, T_LONG, or T_DOUBLE
 728         // the interpreter allocates two slots but only uses one for thr T_LONG or T_DOUBLE case
 729         // So we must adjust where to pick up the data to match the interpreter.
 730 
 731         const int offset = (sig_bt[i]==T_LONG||sig_bt[i]==T_DOUBLE)?
 732                            next_off : ld_off;
 733 
 734         // this can be a misaligned move
 735         __ movq(r, Address(r13, offset));
 736       } else {
 737         // sign extend and use a full word?
 738         __ movl(r, Address(r13, ld_off));
 739       }
 740     } else {
 741       if (!r_2->is_valid()) {
 742         __ movflt(r_1->as_XMMRegister(), Address(r13, ld_off));
 743       } else {
 744         __ movdbl(r_1->as_XMMRegister(), Address(r13, next_off));
 745       }
 746     }
 747   }
 748 
 749   // 6243940 We might end up in handle_wrong_method if
 750   // the callee is deoptimized as we race thru here. If that
 751   // happens we don't want to take a safepoint because the
 752   // caller frame will look interpreted and arguments are now
 753   // "compiled" so it is much better to make this transition
 754   // invisible to the stack walking code. Unfortunately if
 755   // we try and find the callee by normal means a safepoint
 756   // is possible. So we stash the desired callee in the thread
 757   // and the vm will find there should this case occur.
 758 
 759   __ movptr(Address(r15_thread, JavaThread::callee_target_offset()), rbx);
 760 
 761   // put methodOop where a c2i would expect should we end up there
 762   // only needed becaus eof c2 resolve stubs return methodOop as a result in
 763   // rax
 764   __ mov(rax, rbx);
 765   __ jmp(r11);
 766 }
 767 
 768 // ---------------------------------------------------------------
 769 AdapterHandlerEntry* SharedRuntime::generate_i2c2i_adapters(MacroAssembler *masm,
 770                                                             int total_args_passed,
 771                                                             int comp_args_on_stack,
 772                                                             const BasicType *sig_bt,
 773                                                             const VMRegPair *regs) {
 774   address i2c_entry = __ pc();
 775 
 776   gen_i2c_adapter(masm, total_args_passed, comp_args_on_stack, sig_bt, regs);
 777 
 778   // -------------------------------------------------------------------------
 779   // Generate a C2I adapter.  On entry we know rbx holds the methodOop during calls
 780   // to the interpreter.  The args start out packed in the compiled layout.  They
 781   // need to be unpacked into the interpreter layout.  This will almost always
 782   // require some stack space.  We grow the current (compiled) stack, then repack
 783   // the args.  We  finally end in a jump to the generic interpreter entry point.
 784   // On exit from the interpreter, the interpreter will restore our SP (lest the
 785   // compiled code, which relys solely on SP and not RBP, get sick).
 786 
 787   address c2i_unverified_entry = __ pc();
 788   Label skip_fixup;
 789   Label ok;
 790 
 791   Register holder = rax;
 792   Register receiver = j_rarg0;
 793   Register temp = rbx;
 794 
 795   {
 796     __ verify_oop(holder);
 797     __ load_klass(temp, receiver);
 798     __ verify_oop(temp);
 799 
 800     __ cmpptr(temp, Address(holder, compiledICHolderOopDesc::holder_klass_offset()));
 801     __ movptr(rbx, Address(holder, compiledICHolderOopDesc::holder_method_offset()));
 802     __ jcc(Assembler::equal, ok);
 803     __ jump(RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
 804 
 805     __ bind(ok);
 806     // Method might have been compiled since the call site was patched to
 807     // interpreted if that is the case treat it as a miss so we can get
 808     // the call site corrected.
 809     __ cmpptr(Address(rbx, in_bytes(methodOopDesc::code_offset())), (int32_t)NULL_WORD);
 810     __ jcc(Assembler::equal, skip_fixup);
 811     __ jump(RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
 812   }
 813 
 814   address c2i_entry = __ pc();
 815 
 816   gen_c2i_adapter(masm, total_args_passed, comp_args_on_stack, sig_bt, regs, skip_fixup);
 817 
 818   __ flush();
 819   return new AdapterHandlerEntry(i2c_entry, c2i_entry, c2i_unverified_entry);
 820 }
 821 
 822 int SharedRuntime::c_calling_convention(const BasicType *sig_bt,
 823                                          VMRegPair *regs,
 824                                          int total_args_passed) {
 825 // We return the amount of VMRegImpl stack slots we need to reserve for all
 826 // the arguments NOT counting out_preserve_stack_slots.
 827 
 828 // NOTE: These arrays will have to change when c1 is ported
 829 #ifdef _WIN64
 830     static const Register INT_ArgReg[Argument::n_int_register_parameters_c] = {
 831       c_rarg0, c_rarg1, c_rarg2, c_rarg3
 832     };
 833     static const XMMRegister FP_ArgReg[Argument::n_float_register_parameters_c] = {
 834       c_farg0, c_farg1, c_farg2, c_farg3
 835     };
 836 #else
 837     static const Register INT_ArgReg[Argument::n_int_register_parameters_c] = {
 838       c_rarg0, c_rarg1, c_rarg2, c_rarg3, c_rarg4, c_rarg5
 839     };
 840     static const XMMRegister FP_ArgReg[Argument::n_float_register_parameters_c] = {
 841       c_farg0, c_farg1, c_farg2, c_farg3,
 842       c_farg4, c_farg5, c_farg6, c_farg7
 843     };
 844 #endif // _WIN64
 845 
 846 
 847     uint int_args = 0;
 848     uint fp_args = 0;
 849     uint stk_args = 0; // inc by 2 each time
 850 
 851     for (int i = 0; i < total_args_passed; i++) {
 852       switch (sig_bt[i]) {
 853       case T_BOOLEAN:
 854       case T_CHAR:
 855       case T_BYTE:
 856       case T_SHORT:
 857       case T_INT:
 858         if (int_args < Argument::n_int_register_parameters_c) {
 859           regs[i].set1(INT_ArgReg[int_args++]->as_VMReg());
 860 #ifdef _WIN64
 861           fp_args++;
 862           // Allocate slots for callee to stuff register args the stack.
 863           stk_args += 2;
 864 #endif
 865         } else {
 866           regs[i].set1(VMRegImpl::stack2reg(stk_args));
 867           stk_args += 2;
 868         }
 869         break;
 870       case T_LONG:
 871         assert(sig_bt[i + 1] == T_VOID, "expecting half");
 872         // fall through
 873       case T_OBJECT:
 874       case T_ARRAY:
 875       case T_ADDRESS:
 876         if (int_args < Argument::n_int_register_parameters_c) {
 877           regs[i].set2(INT_ArgReg[int_args++]->as_VMReg());
 878 #ifdef _WIN64
 879           fp_args++;
 880           stk_args += 2;
 881 #endif
 882         } else {
 883           regs[i].set2(VMRegImpl::stack2reg(stk_args));
 884           stk_args += 2;
 885         }
 886         break;
 887       case T_FLOAT:
 888         if (fp_args < Argument::n_float_register_parameters_c) {
 889           regs[i].set1(FP_ArgReg[fp_args++]->as_VMReg());
 890 #ifdef _WIN64
 891           int_args++;
 892           // Allocate slots for callee to stuff register args the stack.
 893           stk_args += 2;
 894 #endif
 895         } else {
 896           regs[i].set1(VMRegImpl::stack2reg(stk_args));
 897           stk_args += 2;
 898         }
 899         break;
 900       case T_DOUBLE:
 901         assert(sig_bt[i + 1] == T_VOID, "expecting half");
 902         if (fp_args < Argument::n_float_register_parameters_c) {
 903           regs[i].set2(FP_ArgReg[fp_args++]->as_VMReg());
 904 #ifdef _WIN64
 905           int_args++;
 906           // Allocate slots for callee to stuff register args the stack.
 907           stk_args += 2;
 908 #endif
 909         } else {
 910           regs[i].set2(VMRegImpl::stack2reg(stk_args));
 911           stk_args += 2;
 912         }
 913         break;
 914       case T_VOID: // Halves of longs and doubles
 915         assert(i != 0 && (sig_bt[i - 1] == T_LONG || sig_bt[i - 1] == T_DOUBLE), "expecting half");
 916         regs[i].set_bad();
 917         break;
 918       default:
 919         ShouldNotReachHere();
 920         break;
 921       }
 922     }
 923 #ifdef _WIN64
 924   // windows abi requires that we always allocate enough stack space
 925   // for 4 64bit registers to be stored down.
 926   if (stk_args < 8) {
 927     stk_args = 8;
 928   }
 929 #endif // _WIN64
 930 
 931   return stk_args;
 932 }
 933 
 934 // On 64 bit we will store integer like items to the stack as
 935 // 64 bits items (sparc abi) even though java would only store
 936 // 32bits for a parameter. On 32bit it will simply be 32 bits
 937 // So this routine will do 32->32 on 32bit and 32->64 on 64bit
 938 static void move32_64(MacroAssembler* masm, VMRegPair src, VMRegPair dst) {
 939   if (src.first()->is_stack()) {
 940     if (dst.first()->is_stack()) {
 941       // stack to stack
 942       __ movslq(rax, Address(rbp, reg2offset_in(src.first())));
 943       __ movq(Address(rsp, reg2offset_out(dst.first())), rax);
 944     } else {
 945       // stack to reg
 946       __ movslq(dst.first()->as_Register(), Address(rbp, reg2offset_in(src.first())));
 947     }
 948   } else if (dst.first()->is_stack()) {
 949     // reg to stack
 950     // Do we really have to sign extend???
 951     // __ movslq(src.first()->as_Register(), src.first()->as_Register());
 952     __ movq(Address(rsp, reg2offset_out(dst.first())), src.first()->as_Register());
 953   } else {
 954     // Do we really have to sign extend???
 955     // __ movslq(dst.first()->as_Register(), src.first()->as_Register());
 956     if (dst.first() != src.first()) {
 957       __ movq(dst.first()->as_Register(), src.first()->as_Register());
 958     }
 959   }
 960 }
 961 
 962 
 963 // An oop arg. Must pass a handle not the oop itself
 964 static void object_move(MacroAssembler* masm,
 965                         OopMap* map,
 966                         int oop_handle_offset,
 967                         int framesize_in_slots,
 968                         VMRegPair src,
 969                         VMRegPair dst,
 970                         bool is_receiver,
 971                         int* receiver_offset) {
 972 
 973   // must pass a handle. First figure out the location we use as a handle
 974 
 975   Register rHandle = dst.first()->is_stack() ? rax : dst.first()->as_Register();
 976 
 977   // See if oop is NULL if it is we need no handle
 978 
 979   if (src.first()->is_stack()) {
 980 
 981     // Oop is already on the stack as an argument
 982     int offset_in_older_frame = src.first()->reg2stack() + SharedRuntime::out_preserve_stack_slots();
 983     map->set_oop(VMRegImpl::stack2reg(offset_in_older_frame + framesize_in_slots));
 984     if (is_receiver) {
 985       *receiver_offset = (offset_in_older_frame + framesize_in_slots) * VMRegImpl::stack_slot_size;
 986     }
 987 
 988     __ cmpptr(Address(rbp, reg2offset_in(src.first())), (int32_t)NULL_WORD);
 989     __ lea(rHandle, Address(rbp, reg2offset_in(src.first())));
 990     // conditionally move a NULL
 991     __ cmovptr(Assembler::equal, rHandle, Address(rbp, reg2offset_in(src.first())));
 992   } else {
 993 
 994     // Oop is in an a register we must store it to the space we reserve
 995     // on the stack for oop_handles and pass a handle if oop is non-NULL
 996 
 997     const Register rOop = src.first()->as_Register();
 998     int oop_slot;
 999     if (rOop == j_rarg0)
1000       oop_slot = 0;
1001     else if (rOop == j_rarg1)
1002       oop_slot = 1;
1003     else if (rOop == j_rarg2)
1004       oop_slot = 2;
1005     else if (rOop == j_rarg3)
1006       oop_slot = 3;
1007     else if (rOop == j_rarg4)
1008       oop_slot = 4;
1009     else {
1010       assert(rOop == j_rarg5, "wrong register");
1011       oop_slot = 5;
1012     }
1013 
1014     oop_slot = oop_slot * VMRegImpl::slots_per_word + oop_handle_offset;
1015     int offset = oop_slot*VMRegImpl::stack_slot_size;
1016 
1017     map->set_oop(VMRegImpl::stack2reg(oop_slot));
1018     // Store oop in handle area, may be NULL
1019     __ movptr(Address(rsp, offset), rOop);
1020     if (is_receiver) {
1021       *receiver_offset = offset;
1022     }
1023 
1024     __ cmpptr(rOop, (int32_t)NULL_WORD);
1025     __ lea(rHandle, Address(rsp, offset));
1026     // conditionally move a NULL from the handle area where it was just stored
1027     __ cmovptr(Assembler::equal, rHandle, Address(rsp, offset));
1028   }
1029 
1030   // If arg is on the stack then place it otherwise it is already in correct reg.
1031   if (dst.first()->is_stack()) {
1032     __ movptr(Address(rsp, reg2offset_out(dst.first())), rHandle);
1033   }
1034 }
1035 
1036 // A float arg may have to do float reg int reg conversion
1037 static void float_move(MacroAssembler* masm, VMRegPair src, VMRegPair dst) {
1038   assert(!src.second()->is_valid() && !dst.second()->is_valid(), "bad float_move");
1039 
1040   // The calling conventions assures us that each VMregpair is either
1041   // all really one physical register or adjacent stack slots.
1042   // This greatly simplifies the cases here compared to sparc.
1043 
1044   if (src.first()->is_stack()) {
1045     if (dst.first()->is_stack()) {
1046       __ movl(rax, Address(rbp, reg2offset_in(src.first())));
1047       __ movptr(Address(rsp, reg2offset_out(dst.first())), rax);
1048     } else {
1049       // stack to reg
1050       assert(dst.first()->is_XMMRegister(), "only expect xmm registers as parameters");
1051       __ movflt(dst.first()->as_XMMRegister(), Address(rbp, reg2offset_in(src.first())));
1052     }
1053   } else if (dst.first()->is_stack()) {
1054     // reg to stack
1055     assert(src.first()->is_XMMRegister(), "only expect xmm registers as parameters");
1056     __ movflt(Address(rsp, reg2offset_out(dst.first())), src.first()->as_XMMRegister());
1057   } else {
1058     // reg to reg
1059     // In theory these overlap but the ordering is such that this is likely a nop
1060     if ( src.first() != dst.first()) {
1061       __ movdbl(dst.first()->as_XMMRegister(),  src.first()->as_XMMRegister());
1062     }
1063   }
1064 }
1065 
1066 // A long move
1067 static void long_move(MacroAssembler* masm, VMRegPair src, VMRegPair dst) {
1068 
1069   // The calling conventions assures us that each VMregpair is either
1070   // all really one physical register or adjacent stack slots.
1071   // This greatly simplifies the cases here compared to sparc.
1072 
1073   if (src.is_single_phys_reg() ) {
1074     if (dst.is_single_phys_reg()) {
1075       if (dst.first() != src.first()) {
1076         __ mov(dst.first()->as_Register(), src.first()->as_Register());
1077       }
1078     } else {
1079       assert(dst.is_single_reg(), "not a stack pair");
1080       __ movq(Address(rsp, reg2offset_out(dst.first())), src.first()->as_Register());
1081     }
1082   } else if (dst.is_single_phys_reg()) {
1083     assert(src.is_single_reg(),  "not a stack pair");
1084     __ movq(dst.first()->as_Register(), Address(rbp, reg2offset_out(src.first())));
1085   } else {
1086     assert(src.is_single_reg() && dst.is_single_reg(), "not stack pairs");
1087     __ movq(rax, Address(rbp, reg2offset_in(src.first())));
1088     __ movq(Address(rsp, reg2offset_out(dst.first())), rax);
1089   }
1090 }
1091 
1092 // A double move
1093 static void double_move(MacroAssembler* masm, VMRegPair src, VMRegPair dst) {
1094 
1095   // The calling conventions assures us that each VMregpair is either
1096   // all really one physical register or adjacent stack slots.
1097   // This greatly simplifies the cases here compared to sparc.
1098 
1099   if (src.is_single_phys_reg() ) {
1100     if (dst.is_single_phys_reg()) {
1101       // In theory these overlap but the ordering is such that this is likely a nop
1102       if ( src.first() != dst.first()) {
1103         __ movdbl(dst.first()->as_XMMRegister(), src.first()->as_XMMRegister());
1104       }
1105     } else {
1106       assert(dst.is_single_reg(), "not a stack pair");
1107       __ movdbl(Address(rsp, reg2offset_out(dst.first())), src.first()->as_XMMRegister());
1108     }
1109   } else if (dst.is_single_phys_reg()) {
1110     assert(src.is_single_reg(),  "not a stack pair");
1111     __ movdbl(dst.first()->as_XMMRegister(), Address(rbp, reg2offset_out(src.first())));
1112   } else {
1113     assert(src.is_single_reg() && dst.is_single_reg(), "not stack pairs");
1114     __ movq(rax, Address(rbp, reg2offset_in(src.first())));
1115     __ movq(Address(rsp, reg2offset_out(dst.first())), rax);
1116   }
1117 }
1118 
1119 
1120 void SharedRuntime::save_native_result(MacroAssembler *masm, BasicType ret_type, int frame_slots) {
1121   // We always ignore the frame_slots arg and just use the space just below frame pointer
1122   // which by this time is free to use
1123   switch (ret_type) {
1124   case T_FLOAT:
1125     __ movflt(Address(rbp, -wordSize), xmm0);
1126     break;
1127   case T_DOUBLE:
1128     __ movdbl(Address(rbp, -wordSize), xmm0);
1129     break;
1130   case T_VOID:  break;
1131   default: {
1132     __ movptr(Address(rbp, -wordSize), rax);
1133     }
1134   }
1135 }
1136 
1137 void SharedRuntime::restore_native_result(MacroAssembler *masm, BasicType ret_type, int frame_slots) {
1138   // We always ignore the frame_slots arg and just use the space just below frame pointer
1139   // which by this time is free to use
1140   switch (ret_type) {
1141   case T_FLOAT:
1142     __ movflt(xmm0, Address(rbp, -wordSize));
1143     break;
1144   case T_DOUBLE:
1145     __ movdbl(xmm0, Address(rbp, -wordSize));
1146     break;
1147   case T_VOID:  break;
1148   default: {
1149     __ movptr(rax, Address(rbp, -wordSize));
1150     }
1151   }
1152 }
1153 
1154 static void save_args(MacroAssembler *masm, int arg_count, int first_arg, VMRegPair *args) {
1155     for ( int i = first_arg ; i < arg_count ; i++ ) {
1156       if (args[i].first()->is_Register()) {
1157         __ push(args[i].first()->as_Register());
1158       } else if (args[i].first()->is_XMMRegister()) {
1159         __ subptr(rsp, 2*wordSize);
1160         __ movdbl(Address(rsp, 0), args[i].first()->as_XMMRegister());
1161       }
1162     }
1163 }
1164 
1165 static void restore_args(MacroAssembler *masm, int arg_count, int first_arg, VMRegPair *args) {
1166     for ( int i = arg_count - 1 ; i >= first_arg ; i-- ) {
1167       if (args[i].first()->is_Register()) {
1168         __ pop(args[i].first()->as_Register());
1169       } else if (args[i].first()->is_XMMRegister()) {
1170         __ movdbl(args[i].first()->as_XMMRegister(), Address(rsp, 0));
1171         __ addptr(rsp, 2*wordSize);
1172       }
1173     }
1174 }
1175 
1176 // ---------------------------------------------------------------------------
1177 // Generate a native wrapper for a given method.  The method takes arguments
1178 // in the Java compiled code convention, marshals them to the native
1179 // convention (handlizes oops, etc), transitions to native, makes the call,
1180 // returns to java state (possibly blocking), unhandlizes any result and
1181 // returns.
1182 nmethod *SharedRuntime::generate_native_wrapper(MacroAssembler *masm,
1183                                                 methodHandle method,
1184                                                 int total_in_args,
1185                                                 int comp_args_on_stack,
1186                                                 BasicType *in_sig_bt,
1187                                                 VMRegPair *in_regs,
1188                                                 BasicType ret_type) {
1189   // Native nmethod wrappers never take possesion of the oop arguments.
1190   // So the caller will gc the arguments. The only thing we need an
1191   // oopMap for is if the call is static
1192   //
1193   // An OopMap for lock (and class if static)
1194   OopMapSet *oop_maps = new OopMapSet();
1195   intptr_t start = (intptr_t)__ pc();
1196 
1197   // We have received a description of where all the java arg are located
1198   // on entry to the wrapper. We need to convert these args to where
1199   // the jni function will expect them. To figure out where they go
1200   // we convert the java signature to a C signature by inserting
1201   // the hidden arguments as arg[0] and possibly arg[1] (static method)
1202 
1203   int total_c_args = total_in_args + 1;
1204   if (method->is_static()) {
1205     total_c_args++;
1206   }
1207 
1208   BasicType* out_sig_bt = NEW_RESOURCE_ARRAY(BasicType, total_c_args);
1209   VMRegPair* out_regs   = NEW_RESOURCE_ARRAY(VMRegPair,   total_c_args);
1210 
1211   int argc = 0;
1212   out_sig_bt[argc++] = T_ADDRESS;
1213   if (method->is_static()) {
1214     out_sig_bt[argc++] = T_OBJECT;
1215   }
1216 
1217   for (int i = 0; i < total_in_args ; i++ ) {
1218     out_sig_bt[argc++] = in_sig_bt[i];
1219   }
1220 
1221   // Now figure out where the args must be stored and how much stack space
1222   // they require.
1223   //
1224   int out_arg_slots;
1225   out_arg_slots = c_calling_convention(out_sig_bt, out_regs, total_c_args);
1226 
1227   // Compute framesize for the wrapper.  We need to handlize all oops in
1228   // incoming registers
1229 
1230   // Calculate the total number of stack slots we will need.
1231 
1232   // First count the abi requirement plus all of the outgoing args
1233   int stack_slots = SharedRuntime::out_preserve_stack_slots() + out_arg_slots;
1234 
1235   // Now the space for the inbound oop handle area
1236 
1237   int oop_handle_offset = stack_slots;
1238   stack_slots += 6*VMRegImpl::slots_per_word;
1239 
1240   // Now any space we need for handlizing a klass if static method
1241 
1242   int oop_temp_slot_offset = 0;
1243   int klass_slot_offset = 0;
1244   int klass_offset = -1;
1245   int lock_slot_offset = 0;
1246   bool is_static = false;
1247 
1248   if (method->is_static()) {
1249     klass_slot_offset = stack_slots;
1250     stack_slots += VMRegImpl::slots_per_word;
1251     klass_offset = klass_slot_offset * VMRegImpl::stack_slot_size;
1252     is_static = true;
1253   }
1254 
1255   // Plus a lock if needed
1256 
1257   if (method->is_synchronized()) {
1258     lock_slot_offset = stack_slots;
1259     stack_slots += VMRegImpl::slots_per_word;
1260   }
1261 
1262   // Now a place (+2) to save return values or temp during shuffling
1263   // + 4 for return address (which we own) and saved rbp
1264   stack_slots += 6;
1265 
1266   // Ok The space we have allocated will look like:
1267   //
1268   //
1269   // FP-> |                     |
1270   //      |---------------------|
1271   //      | 2 slots for moves   |
1272   //      |---------------------|
1273   //      | lock box (if sync)  |
1274   //      |---------------------| <- lock_slot_offset
1275   //      | klass (if static)   |
1276   //      |---------------------| <- klass_slot_offset
1277   //      | oopHandle area      |
1278   //      |---------------------| <- oop_handle_offset (6 java arg registers)
1279   //      | outbound memory     |
1280   //      | based arguments     |
1281   //      |                     |
1282   //      |---------------------|
1283   //      |                     |
1284   // SP-> | out_preserved_slots |
1285   //
1286   //
1287 
1288 
1289   // Now compute actual number of stack words we need rounding to make
1290   // stack properly aligned.
1291   stack_slots = round_to(stack_slots, StackAlignmentInSlots);
1292 
1293   int stack_size = stack_slots * VMRegImpl::stack_slot_size;
1294 
1295 
1296   // First thing make an ic check to see if we should even be here
1297 
1298   // We are free to use all registers as temps without saving them and
1299   // restoring them except rbp. rbp is the only callee save register
1300   // as far as the interpreter and the compiler(s) are concerned.
1301 
1302 
1303   const Register ic_reg = rax;
1304   const Register receiver = j_rarg0;
1305 
1306   Label ok;
1307   Label exception_pending;
1308 
1309   assert_different_registers(ic_reg, receiver, rscratch1);
1310   __ verify_oop(receiver);
1311   __ load_klass(rscratch1, receiver);
1312   __ cmpq(ic_reg, rscratch1);
1313   __ jcc(Assembler::equal, ok);
1314 
1315   __ jump(RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
1316 
1317   __ bind(ok);
1318 
1319   // Verified entry point must be aligned
1320   __ align(8);
1321 
1322   int vep_offset = ((intptr_t)__ pc()) - start;
1323 
1324   // The instruction at the verified entry point must be 5 bytes or longer
1325   // because it can be patched on the fly by make_non_entrant. The stack bang
1326   // instruction fits that requirement.
1327 
1328   // Generate stack overflow check
1329 
1330   if (UseStackBanging) {
1331     __ bang_stack_with_offset(StackShadowPages*os::vm_page_size());
1332   } else {
1333     // need a 5 byte instruction to allow MT safe patching to non-entrant
1334     __ fat_nop();
1335   }
1336 
1337   // Generate a new frame for the wrapper.
1338   __ enter();
1339   // -2 because return address is already present and so is saved rbp
1340   __ subptr(rsp, stack_size - 2*wordSize);
1341 
1342     // Frame is now completed as far as size and linkage.
1343 
1344     int frame_complete = ((intptr_t)__ pc()) - start;
1345 
1346 #ifdef ASSERT
1347     {
1348       Label L;
1349       __ mov(rax, rsp);
1350       __ andptr(rax, -16); // must be 16 byte boundary (see amd64 ABI)
1351       __ cmpptr(rax, rsp);
1352       __ jcc(Assembler::equal, L);
1353       __ stop("improperly aligned stack");
1354       __ bind(L);
1355     }
1356 #endif /* ASSERT */
1357 
1358 
1359   // We use r14 as the oop handle for the receiver/klass
1360   // It is callee save so it survives the call to native
1361 
1362   const Register oop_handle_reg = r14;
1363 
1364 
1365 
1366   //
1367   // We immediately shuffle the arguments so that any vm call we have to
1368   // make from here on out (sync slow path, jvmti, etc.) we will have
1369   // captured the oops from our caller and have a valid oopMap for
1370   // them.
1371 
1372   // -----------------
1373   // The Grand Shuffle
1374 
1375   // The Java calling convention is either equal (linux) or denser (win64) than the
1376   // c calling convention. However the because of the jni_env argument the c calling
1377   // convention always has at least one more (and two for static) arguments than Java.
1378   // Therefore if we move the args from java -> c backwards then we will never have
1379   // a register->register conflict and we don't have to build a dependency graph
1380   // and figure out how to break any cycles.
1381   //
1382 
1383   // Record esp-based slot for receiver on stack for non-static methods
1384   int receiver_offset = -1;
1385 
1386   // This is a trick. We double the stack slots so we can claim
1387   // the oops in the caller's frame. Since we are sure to have
1388   // more args than the caller doubling is enough to make
1389   // sure we can capture all the incoming oop args from the
1390   // caller.
1391   //
1392   OopMap* map = new OopMap(stack_slots * 2, 0 /* arg_slots*/);
1393 
1394   // Mark location of rbp (someday)
1395   // map->set_callee_saved(VMRegImpl::stack2reg( stack_slots - 2), stack_slots * 2, 0, vmreg(rbp));
1396 
1397   // Use eax, ebx as temporaries during any memory-memory moves we have to do
1398   // All inbound args are referenced based on rbp and all outbound args via rsp.
1399 
1400 
1401 #ifdef ASSERT
1402   bool reg_destroyed[RegisterImpl::number_of_registers];
1403   bool freg_destroyed[XMMRegisterImpl::number_of_registers];
1404   for ( int r = 0 ; r < RegisterImpl::number_of_registers ; r++ ) {
1405     reg_destroyed[r] = false;
1406   }
1407   for ( int f = 0 ; f < XMMRegisterImpl::number_of_registers ; f++ ) {
1408     freg_destroyed[f] = false;
1409   }
1410 
1411 #endif /* ASSERT */
1412 
1413 
1414   int c_arg = total_c_args - 1;
1415   for ( int i = total_in_args - 1; i >= 0 ; i--, c_arg-- ) {
1416 #ifdef ASSERT
1417     if (in_regs[i].first()->is_Register()) {
1418       assert(!reg_destroyed[in_regs[i].first()->as_Register()->encoding()], "destroyed reg!");
1419     } else if (in_regs[i].first()->is_XMMRegister()) {
1420       assert(!freg_destroyed[in_regs[i].first()->as_XMMRegister()->encoding()], "destroyed reg!");
1421     }
1422     if (out_regs[c_arg].first()->is_Register()) {
1423       reg_destroyed[out_regs[c_arg].first()->as_Register()->encoding()] = true;
1424     } else if (out_regs[c_arg].first()->is_XMMRegister()) {
1425       freg_destroyed[out_regs[c_arg].first()->as_XMMRegister()->encoding()] = true;
1426     }
1427 #endif /* ASSERT */
1428     switch (in_sig_bt[i]) {
1429       case T_ARRAY:
1430       case T_OBJECT:
1431         object_move(masm, map, oop_handle_offset, stack_slots, in_regs[i], out_regs[c_arg],
1432                     ((i == 0) && (!is_static)),
1433                     &receiver_offset);
1434         break;
1435       case T_VOID:
1436         break;
1437 
1438       case T_FLOAT:
1439         float_move(masm, in_regs[i], out_regs[c_arg]);
1440           break;
1441 
1442       case T_DOUBLE:
1443         assert( i + 1 < total_in_args &&
1444                 in_sig_bt[i + 1] == T_VOID &&
1445                 out_sig_bt[c_arg+1] == T_VOID, "bad arg list");
1446         double_move(masm, in_regs[i], out_regs[c_arg]);
1447         break;
1448 
1449       case T_LONG :
1450         long_move(masm, in_regs[i], out_regs[c_arg]);
1451         break;
1452 
1453       case T_ADDRESS: assert(false, "found T_ADDRESS in java args");
1454 
1455       default:
1456         move32_64(masm, in_regs[i], out_regs[c_arg]);
1457     }
1458   }
1459 
1460   // point c_arg at the first arg that is already loaded in case we
1461   // need to spill before we call out
1462   c_arg++;
1463 
1464   // Pre-load a static method's oop into r14.  Used both by locking code and
1465   // the normal JNI call code.
1466   if (method->is_static()) {
1467 
1468     //  load oop into a register
1469     __ movoop(oop_handle_reg, JNIHandles::make_local(Klass::cast(method->method_holder())->java_mirror()));
1470 
1471     // Now handlize the static class mirror it's known not-null.
1472     __ movptr(Address(rsp, klass_offset), oop_handle_reg);
1473     map->set_oop(VMRegImpl::stack2reg(klass_slot_offset));
1474 
1475     // Now get the handle
1476     __ lea(oop_handle_reg, Address(rsp, klass_offset));
1477     // store the klass handle as second argument
1478     __ movptr(c_rarg1, oop_handle_reg);
1479     // and protect the arg if we must spill
1480     c_arg--;
1481   }
1482 
1483   // Change state to native (we save the return address in the thread, since it might not
1484   // be pushed on the stack when we do a a stack traversal). It is enough that the pc()
1485   // points into the right code segment. It does not have to be the correct return pc.
1486   // We use the same pc/oopMap repeatedly when we call out
1487 
1488   intptr_t the_pc = (intptr_t) __ pc();
1489   oop_maps->add_gc_map(the_pc - start, map);
1490 
1491   __ set_last_Java_frame(rsp, noreg, (address)the_pc);
1492 
1493 
1494   // We have all of the arguments setup at this point. We must not touch any register
1495   // argument registers at this point (what if we save/restore them there are no oop?
1496 
1497   {
1498     SkipIfEqual skip(masm, &DTraceMethodProbes, false);
1499     // protect the args we've loaded
1500     save_args(masm, total_c_args, c_arg, out_regs);
1501     __ movoop(c_rarg1, JNIHandles::make_local(method()));
1502     __ call_VM_leaf(
1503       CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
1504       r15_thread, c_rarg1);
1505     restore_args(masm, total_c_args, c_arg, out_regs);
1506   }
1507 
1508   // RedefineClasses() tracing support for obsolete method entry
1509   if (RC_TRACE_IN_RANGE(0x00001000, 0x00002000)) {
1510     // protect the args we've loaded
1511     save_args(masm, total_c_args, c_arg, out_regs);
1512     __ movoop(c_rarg1, JNIHandles::make_local(method()));
1513     __ call_VM_leaf(
1514       CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
1515       r15_thread, c_rarg1);
1516     restore_args(masm, total_c_args, c_arg, out_regs);
1517   }
1518 
1519   // Lock a synchronized method
1520 
1521   // Register definitions used by locking and unlocking
1522 
1523   const Register swap_reg = rax;  // Must use rax for cmpxchg instruction
1524   const Register obj_reg  = rbx;  // Will contain the oop
1525   const Register lock_reg = r13;  // Address of compiler lock object (BasicLock)
1526   const Register old_hdr  = r13;  // value of old header at unlock time
1527 
1528   Label slow_path_lock;
1529   Label lock_done;
1530 
1531   if (method->is_synchronized()) {
1532 
1533 
1534     const int mark_word_offset = BasicLock::displaced_header_offset_in_bytes();
1535 
1536     // Get the handle (the 2nd argument)
1537     __ mov(oop_handle_reg, c_rarg1);
1538 
1539     // Get address of the box
1540 
1541     __ lea(lock_reg, Address(rsp, lock_slot_offset * VMRegImpl::stack_slot_size));
1542 
1543     // Load the oop from the handle
1544     __ movptr(obj_reg, Address(oop_handle_reg, 0));
1545 
1546     if (UseBiasedLocking) {
1547       __ biased_locking_enter(lock_reg, obj_reg, swap_reg, rscratch1, false, lock_done, &slow_path_lock);
1548     }
1549 
1550     // Load immediate 1 into swap_reg %rax
1551     __ movl(swap_reg, 1);
1552 
1553     // Load (object->mark() | 1) into swap_reg %rax
1554     __ orptr(swap_reg, Address(obj_reg, 0));
1555 
1556     // Save (object->mark() | 1) into BasicLock's displaced header
1557     __ movptr(Address(lock_reg, mark_word_offset), swap_reg);
1558 
1559     if (os::is_MP()) {
1560       __ lock();
1561     }
1562 
1563     // src -> dest iff dest == rax else rax <- dest
1564     __ cmpxchgptr(lock_reg, Address(obj_reg, 0));
1565     __ jcc(Assembler::equal, lock_done);
1566 
1567     // Hmm should this move to the slow path code area???
1568 
1569     // Test if the oopMark is an obvious stack pointer, i.e.,
1570     //  1) (mark & 3) == 0, and
1571     //  2) rsp <= mark < mark + os::pagesize()
1572     // These 3 tests can be done by evaluating the following
1573     // expression: ((mark - rsp) & (3 - os::vm_page_size())),
1574     // assuming both stack pointer and pagesize have their
1575     // least significant 2 bits clear.
1576     // NOTE: the oopMark is in swap_reg %rax as the result of cmpxchg
1577 
1578     __ subptr(swap_reg, rsp);
1579     __ andptr(swap_reg, 3 - os::vm_page_size());
1580 
1581     // Save the test result, for recursive case, the result is zero
1582     __ movptr(Address(lock_reg, mark_word_offset), swap_reg);
1583     __ jcc(Assembler::notEqual, slow_path_lock);
1584 
1585     // Slow path will re-enter here
1586 
1587     __ bind(lock_done);
1588   }
1589 
1590 
1591   // Finally just about ready to make the JNI call
1592 
1593 
1594   // get JNIEnv* which is first argument to native
1595 
1596   __ lea(c_rarg0, Address(r15_thread, in_bytes(JavaThread::jni_environment_offset())));
1597 
1598   // Now set thread in native
1599   __ movl(Address(r15_thread, JavaThread::thread_state_offset()), _thread_in_native);
1600 
1601   __ call(RuntimeAddress(method->native_function()));
1602 
1603     // Either restore the MXCSR register after returning from the JNI Call
1604     // or verify that it wasn't changed.
1605     if (RestoreMXCSROnJNICalls) {
1606       __ ldmxcsr(ExternalAddress(StubRoutines::x86::mxcsr_std()));
1607 
1608     }
1609     else if (CheckJNICalls ) {
1610       __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::x86::verify_mxcsr_entry())));
1611     }
1612 
1613 
1614   // Unpack native results.
1615   switch (ret_type) {
1616   case T_BOOLEAN: __ c2bool(rax);            break;
1617   case T_CHAR   : __ movzwl(rax, rax);      break;
1618   case T_BYTE   : __ sign_extend_byte (rax); break;
1619   case T_SHORT  : __ sign_extend_short(rax); break;
1620   case T_INT    : /* nothing to do */        break;
1621   case T_DOUBLE :
1622   case T_FLOAT  :
1623     // Result is in xmm0 we'll save as needed
1624     break;
1625   case T_ARRAY:                 // Really a handle
1626   case T_OBJECT:                // Really a handle
1627       break; // can't de-handlize until after safepoint check
1628   case T_VOID: break;
1629   case T_LONG: break;
1630   default       : ShouldNotReachHere();
1631   }
1632 
1633   // Switch thread to "native transition" state before reading the synchronization state.
1634   // This additional state is necessary because reading and testing the synchronization
1635   // state is not atomic w.r.t. GC, as this scenario demonstrates:
1636   //     Java thread A, in _thread_in_native state, loads _not_synchronized and is preempted.
1637   //     VM thread changes sync state to synchronizing and suspends threads for GC.
1638   //     Thread A is resumed to finish this native method, but doesn't block here since it
1639   //     didn't see any synchronization is progress, and escapes.
1640   __ movl(Address(r15_thread, JavaThread::thread_state_offset()), _thread_in_native_trans);
1641 
1642   if(os::is_MP()) {
1643     if (UseMembar) {
1644       // Force this write out before the read below
1645       __ membar(Assembler::Membar_mask_bits(
1646            Assembler::LoadLoad | Assembler::LoadStore |
1647            Assembler::StoreLoad | Assembler::StoreStore));
1648     } else {
1649       // Write serialization page so VM thread can do a pseudo remote membar.
1650       // We use the current thread pointer to calculate a thread specific
1651       // offset to write to within the page. This minimizes bus traffic
1652       // due to cache line collision.
1653       __ serialize_memory(r15_thread, rcx);
1654     }
1655   }
1656 
1657 
1658   // check for safepoint operation in progress and/or pending suspend requests
1659   {
1660     Label Continue;
1661 
1662     __ cmp32(ExternalAddress((address)SafepointSynchronize::address_of_state()),
1663              SafepointSynchronize::_not_synchronized);
1664 
1665     Label L;
1666     __ jcc(Assembler::notEqual, L);
1667     __ cmpl(Address(r15_thread, JavaThread::suspend_flags_offset()), 0);
1668     __ jcc(Assembler::equal, Continue);
1669     __ bind(L);
1670 
1671     // Don't use call_VM as it will see a possible pending exception and forward it
1672     // and never return here preventing us from clearing _last_native_pc down below.
1673     // Also can't use call_VM_leaf either as it will check to see if rsi & rdi are
1674     // preserved and correspond to the bcp/locals pointers. So we do a runtime call
1675     // by hand.
1676     //
1677     save_native_result(masm, ret_type, stack_slots);
1678     __ mov(c_rarg0, r15_thread);
1679     __ mov(r12, rsp); // remember sp
1680     __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1681     __ andptr(rsp, -16); // align stack as required by ABI
1682     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans)));
1683     __ mov(rsp, r12); // restore sp
1684     __ reinit_heapbase();
1685     // Restore any method result value
1686     restore_native_result(masm, ret_type, stack_slots);
1687     __ bind(Continue);
1688   }
1689 
1690   // change thread state
1691   __ movl(Address(r15_thread, JavaThread::thread_state_offset()), _thread_in_Java);
1692 
1693   Label reguard;
1694   Label reguard_done;
1695   __ cmpl(Address(r15_thread, JavaThread::stack_guard_state_offset()), JavaThread::stack_guard_yellow_disabled);
1696   __ jcc(Assembler::equal, reguard);
1697   __ bind(reguard_done);
1698 
1699   // native result if any is live
1700 
1701   // Unlock
1702   Label unlock_done;
1703   Label slow_path_unlock;
1704   if (method->is_synchronized()) {
1705 
1706     // Get locked oop from the handle we passed to jni
1707     __ movptr(obj_reg, Address(oop_handle_reg, 0));
1708 
1709     Label done;
1710 
1711     if (UseBiasedLocking) {
1712       __ biased_locking_exit(obj_reg, old_hdr, done);
1713     }
1714 
1715     // Simple recursive lock?
1716 
1717     __ cmpptr(Address(rsp, lock_slot_offset * VMRegImpl::stack_slot_size), (int32_t)NULL_WORD);
1718     __ jcc(Assembler::equal, done);
1719 
1720     // Must save rax if if it is live now because cmpxchg must use it
1721     if (ret_type != T_FLOAT && ret_type != T_DOUBLE && ret_type != T_VOID) {
1722       save_native_result(masm, ret_type, stack_slots);
1723     }
1724 
1725 
1726     // get address of the stack lock
1727     __ lea(rax, Address(rsp, lock_slot_offset * VMRegImpl::stack_slot_size));
1728     //  get old displaced header
1729     __ movptr(old_hdr, Address(rax, 0));
1730 
1731     // Atomic swap old header if oop still contains the stack lock
1732     if (os::is_MP()) {
1733       __ lock();
1734     }
1735     __ cmpxchgptr(old_hdr, Address(obj_reg, 0));
1736     __ jcc(Assembler::notEqual, slow_path_unlock);
1737 
1738     // slow path re-enters here
1739     __ bind(unlock_done);
1740     if (ret_type != T_FLOAT && ret_type != T_DOUBLE && ret_type != T_VOID) {
1741       restore_native_result(masm, ret_type, stack_slots);
1742     }
1743 
1744     __ bind(done);
1745 
1746   }
1747   {
1748     SkipIfEqual skip(masm, &DTraceMethodProbes, false);
1749     save_native_result(masm, ret_type, stack_slots);
1750     __ movoop(c_rarg1, JNIHandles::make_local(method()));
1751     __ call_VM_leaf(
1752          CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
1753          r15_thread, c_rarg1);
1754     restore_native_result(masm, ret_type, stack_slots);
1755   }
1756 
1757   __ reset_last_Java_frame(false, true);
1758 
1759   // Unpack oop result
1760   if (ret_type == T_OBJECT || ret_type == T_ARRAY) {
1761       Label L;
1762       __ testptr(rax, rax);
1763       __ jcc(Assembler::zero, L);
1764       __ movptr(rax, Address(rax, 0));
1765       __ bind(L);
1766       __ verify_oop(rax);
1767   }
1768 
1769   // reset handle block
1770   __ movptr(rcx, Address(r15_thread, JavaThread::active_handles_offset()));
1771   __ movptr(Address(rcx, JNIHandleBlock::top_offset_in_bytes()), (int32_t)NULL_WORD);
1772 
1773   // pop our frame
1774 
1775   __ leave();
1776 
1777   // Any exception pending?
1778   __ cmpptr(Address(r15_thread, in_bytes(Thread::pending_exception_offset())), (int32_t)NULL_WORD);
1779   __ jcc(Assembler::notEqual, exception_pending);
1780 
1781   // Return
1782 
1783   __ ret(0);
1784 
1785   // Unexpected paths are out of line and go here
1786 
1787   // forward the exception
1788   __ bind(exception_pending);
1789 
1790   // and forward the exception
1791   __ jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
1792 
1793 
1794   // Slow path locking & unlocking
1795   if (method->is_synchronized()) {
1796 
1797     // BEGIN Slow path lock
1798     __ bind(slow_path_lock);
1799 
1800     // has last_Java_frame setup. No exceptions so do vanilla call not call_VM
1801     // args are (oop obj, BasicLock* lock, JavaThread* thread)
1802 
1803     // protect the args we've loaded
1804     save_args(masm, total_c_args, c_arg, out_regs);
1805 
1806     __ mov(c_rarg0, obj_reg);
1807     __ mov(c_rarg1, lock_reg);
1808     __ mov(c_rarg2, r15_thread);
1809 
1810     // Not a leaf but we have last_Java_frame setup as we want
1811     __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::complete_monitor_locking_C), 3);
1812     restore_args(masm, total_c_args, c_arg, out_regs);
1813 
1814 #ifdef ASSERT
1815     { Label L;
1816     __ cmpptr(Address(r15_thread, in_bytes(Thread::pending_exception_offset())), (int32_t)NULL_WORD);
1817     __ jcc(Assembler::equal, L);
1818     __ stop("no pending exception allowed on exit from monitorenter");
1819     __ bind(L);
1820     }
1821 #endif
1822     __ jmp(lock_done);
1823 
1824     // END Slow path lock
1825 
1826     // BEGIN Slow path unlock
1827     __ bind(slow_path_unlock);
1828 
1829     // If we haven't already saved the native result we must save it now as xmm registers
1830     // are still exposed.
1831 
1832     if (ret_type == T_FLOAT || ret_type == T_DOUBLE ) {
1833       save_native_result(masm, ret_type, stack_slots);
1834     }
1835 
1836     __ lea(c_rarg1, Address(rsp, lock_slot_offset * VMRegImpl::stack_slot_size));
1837 
1838     __ mov(c_rarg0, obj_reg);
1839     __ mov(r12, rsp); // remember sp
1840     __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1841     __ andptr(rsp, -16); // align stack as required by ABI
1842 
1843     // Save pending exception around call to VM (which contains an EXCEPTION_MARK)
1844     // NOTE that obj_reg == rbx currently
1845     __ movptr(rbx, Address(r15_thread, in_bytes(Thread::pending_exception_offset())));
1846     __ movptr(Address(r15_thread, in_bytes(Thread::pending_exception_offset())), (int32_t)NULL_WORD);
1847 
1848     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::complete_monitor_unlocking_C)));
1849     __ mov(rsp, r12); // restore sp
1850     __ reinit_heapbase();
1851 #ifdef ASSERT
1852     {
1853       Label L;
1854       __ cmpptr(Address(r15_thread, in_bytes(Thread::pending_exception_offset())), (int)NULL_WORD);
1855       __ jcc(Assembler::equal, L);
1856       __ stop("no pending exception allowed on exit complete_monitor_unlocking_C");
1857       __ bind(L);
1858     }
1859 #endif /* ASSERT */
1860 
1861     __ movptr(Address(r15_thread, in_bytes(Thread::pending_exception_offset())), rbx);
1862 
1863     if (ret_type == T_FLOAT || ret_type == T_DOUBLE ) {
1864       restore_native_result(masm, ret_type, stack_slots);
1865     }
1866     __ jmp(unlock_done);
1867 
1868     // END Slow path unlock
1869 
1870   } // synchronized
1871 
1872   // SLOW PATH Reguard the stack if needed
1873 
1874   __ bind(reguard);
1875   save_native_result(masm, ret_type, stack_slots);
1876   __ mov(r12, rsp); // remember sp
1877   __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1878   __ andptr(rsp, -16); // align stack as required by ABI
1879   __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
1880   __ mov(rsp, r12); // restore sp
1881   __ reinit_heapbase();
1882   restore_native_result(masm, ret_type, stack_slots);
1883   // and continue
1884   __ jmp(reguard_done);
1885 
1886 
1887 
1888   __ flush();
1889 
1890   nmethod *nm = nmethod::new_native_nmethod(method,
1891                                             masm->code(),
1892                                             vep_offset,
1893                                             frame_complete,
1894                                             stack_slots / VMRegImpl::slots_per_word,
1895                                             (is_static ? in_ByteSize(klass_offset) : in_ByteSize(receiver_offset)),
1896                                             in_ByteSize(lock_slot_offset*VMRegImpl::stack_slot_size),
1897                                             oop_maps);
1898   return nm;
1899 
1900 }
1901 
1902 #ifdef HAVE_DTRACE_H
1903 // ---------------------------------------------------------------------------
1904 // Generate a dtrace nmethod for a given signature.  The method takes arguments
1905 // in the Java compiled code convention, marshals them to the native
1906 // abi and then leaves nops at the position you would expect to call a native
1907 // function. When the probe is enabled the nops are replaced with a trap
1908 // instruction that dtrace inserts and the trace will cause a notification
1909 // to dtrace.
1910 //
1911 // The probes are only able to take primitive types and java/lang/String as
1912 // arguments.  No other java types are allowed. Strings are converted to utf8
1913 // strings so that from dtrace point of view java strings are converted to C
1914 // strings. There is an arbitrary fixed limit on the total space that a method
1915 // can use for converting the strings. (256 chars per string in the signature).
1916 // So any java string larger then this is truncated.
1917 
1918 static int  fp_offset[ConcreteRegisterImpl::number_of_registers] = { 0 };
1919 static bool offsets_initialized = false;
1920 
1921 
1922 nmethod *SharedRuntime::generate_dtrace_nmethod(MacroAssembler *masm,
1923                                                 methodHandle method) {
1924 
1925 
1926   // generate_dtrace_nmethod is guarded by a mutex so we are sure to
1927   // be single threaded in this method.
1928   assert(AdapterHandlerLibrary_lock->owned_by_self(), "must be");
1929 
1930   if (!offsets_initialized) {
1931     fp_offset[c_rarg0->as_VMReg()->value()] = -1 * wordSize;
1932     fp_offset[c_rarg1->as_VMReg()->value()] = -2 * wordSize;
1933     fp_offset[c_rarg2->as_VMReg()->value()] = -3 * wordSize;
1934     fp_offset[c_rarg3->as_VMReg()->value()] = -4 * wordSize;
1935     fp_offset[c_rarg4->as_VMReg()->value()] = -5 * wordSize;
1936     fp_offset[c_rarg5->as_VMReg()->value()] = -6 * wordSize;
1937 
1938     fp_offset[c_farg0->as_VMReg()->value()] = -7 * wordSize;
1939     fp_offset[c_farg1->as_VMReg()->value()] = -8 * wordSize;
1940     fp_offset[c_farg2->as_VMReg()->value()] = -9 * wordSize;
1941     fp_offset[c_farg3->as_VMReg()->value()] = -10 * wordSize;
1942     fp_offset[c_farg4->as_VMReg()->value()] = -11 * wordSize;
1943     fp_offset[c_farg5->as_VMReg()->value()] = -12 * wordSize;
1944     fp_offset[c_farg6->as_VMReg()->value()] = -13 * wordSize;
1945     fp_offset[c_farg7->as_VMReg()->value()] = -14 * wordSize;
1946 
1947     offsets_initialized = true;
1948   }
1949   // Fill in the signature array, for the calling-convention call.
1950   int total_args_passed = method->size_of_parameters();
1951 
1952   BasicType* in_sig_bt  = NEW_RESOURCE_ARRAY(BasicType, total_args_passed);
1953   VMRegPair  *in_regs   = NEW_RESOURCE_ARRAY(VMRegPair, total_args_passed);
1954 
1955   // The signature we are going to use for the trap that dtrace will see
1956   // java/lang/String is converted. We drop "this" and any other object
1957   // is converted to NULL.  (A one-slot java/lang/Long object reference
1958   // is converted to a two-slot long, which is why we double the allocation).
1959   BasicType* out_sig_bt = NEW_RESOURCE_ARRAY(BasicType, total_args_passed * 2);
1960   VMRegPair* out_regs   = NEW_RESOURCE_ARRAY(VMRegPair, total_args_passed * 2);
1961 
1962   int i=0;
1963   int total_strings = 0;
1964   int first_arg_to_pass = 0;
1965   int total_c_args = 0;
1966 
1967   // Skip the receiver as dtrace doesn't want to see it
1968   if( !method->is_static() ) {
1969     in_sig_bt[i++] = T_OBJECT;
1970     first_arg_to_pass = 1;
1971   }
1972 
1973   // We need to convert the java args to where a native (non-jni) function
1974   // would expect them. To figure out where they go we convert the java
1975   // signature to a C signature.
1976 
1977   SignatureStream ss(method->signature());
1978   for ( ; !ss.at_return_type(); ss.next()) {
1979     BasicType bt = ss.type();
1980     in_sig_bt[i++] = bt;  // Collect remaining bits of signature
1981     out_sig_bt[total_c_args++] = bt;
1982     if( bt == T_OBJECT) {
1983       symbolOop s = ss.as_symbol_or_null();
1984       if (s == vmSymbols::java_lang_String()) {
1985         total_strings++;
1986         out_sig_bt[total_c_args-1] = T_ADDRESS;
1987       } else if (s == vmSymbols::java_lang_Boolean() ||
1988                  s == vmSymbols::java_lang_Character() ||
1989                  s == vmSymbols::java_lang_Byte() ||
1990                  s == vmSymbols::java_lang_Short() ||
1991                  s == vmSymbols::java_lang_Integer() ||
1992                  s == vmSymbols::java_lang_Float()) {
1993         out_sig_bt[total_c_args-1] = T_INT;
1994       } else if (s == vmSymbols::java_lang_Long() ||
1995                  s == vmSymbols::java_lang_Double()) {
1996         out_sig_bt[total_c_args-1] = T_LONG;
1997         out_sig_bt[total_c_args++] = T_VOID;
1998       }
1999     } else if ( bt == T_LONG || bt == T_DOUBLE ) {
2000       in_sig_bt[i++] = T_VOID;   // Longs & doubles take 2 Java slots
2001       // We convert double to long
2002       out_sig_bt[total_c_args-1] = T_LONG;
2003       out_sig_bt[total_c_args++] = T_VOID;
2004     } else if ( bt == T_FLOAT) {
2005       // We convert float to int
2006       out_sig_bt[total_c_args-1] = T_INT;
2007     }
2008   }
2009 
2010   assert(i==total_args_passed, "validly parsed signature");
2011 
2012   // Now get the compiled-Java layout as input arguments
2013   int comp_args_on_stack;
2014   comp_args_on_stack = SharedRuntime::java_calling_convention(
2015       in_sig_bt, in_regs, total_args_passed, false);
2016 
2017   // Now figure out where the args must be stored and how much stack space
2018   // they require (neglecting out_preserve_stack_slots but space for storing
2019   // the 1st six register arguments). It's weird see int_stk_helper.
2020 
2021   int out_arg_slots;
2022   out_arg_slots = c_calling_convention(out_sig_bt, out_regs, total_c_args);
2023 
2024   // Calculate the total number of stack slots we will need.
2025 
2026   // First count the abi requirement plus all of the outgoing args
2027   int stack_slots = SharedRuntime::out_preserve_stack_slots() + out_arg_slots;
2028 
2029   // Now space for the string(s) we must convert
2030   int* string_locs   = NEW_RESOURCE_ARRAY(int, total_strings + 1);
2031   for (i = 0; i < total_strings ; i++) {
2032     string_locs[i] = stack_slots;
2033     stack_slots += max_dtrace_string_size / VMRegImpl::stack_slot_size;
2034   }
2035 
2036   // Plus the temps we might need to juggle register args
2037   // regs take two slots each
2038   stack_slots += (Argument::n_int_register_parameters_c +
2039                   Argument::n_float_register_parameters_c) * 2;
2040 
2041 
2042   // + 4 for return address (which we own) and saved rbp,
2043 
2044   stack_slots += 4;
2045 
2046   // Ok The space we have allocated will look like:
2047   //
2048   //
2049   // FP-> |                     |
2050   //      |---------------------|
2051   //      | string[n]           |
2052   //      |---------------------| <- string_locs[n]
2053   //      | string[n-1]         |
2054   //      |---------------------| <- string_locs[n-1]
2055   //      | ...                 |
2056   //      | ...                 |
2057   //      |---------------------| <- string_locs[1]
2058   //      | string[0]           |
2059   //      |---------------------| <- string_locs[0]
2060   //      | outbound memory     |
2061   //      | based arguments     |
2062   //      |                     |
2063   //      |---------------------|
2064   //      |                     |
2065   // SP-> | out_preserved_slots |
2066   //
2067   //
2068 
2069   // Now compute actual number of stack words we need rounding to make
2070   // stack properly aligned.
2071   stack_slots = round_to(stack_slots, 4 * VMRegImpl::slots_per_word);
2072 
2073   int stack_size = stack_slots * VMRegImpl::stack_slot_size;
2074 
2075   intptr_t start = (intptr_t)__ pc();
2076 
2077   // First thing make an ic check to see if we should even be here
2078 
2079   // We are free to use all registers as temps without saving them and
2080   // restoring them except rbp. rbp, is the only callee save register
2081   // as far as the interpreter and the compiler(s) are concerned.
2082 
2083   const Register ic_reg = rax;
2084   const Register receiver = rcx;
2085   Label hit;
2086   Label exception_pending;
2087 
2088 
2089   __ verify_oop(receiver);
2090   __ cmpl(ic_reg, Address(receiver, oopDesc::klass_offset_in_bytes()));
2091   __ jcc(Assembler::equal, hit);
2092 
2093   __ jump(RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
2094 
2095   // verified entry must be aligned for code patching.
2096   // and the first 5 bytes must be in the same cache line
2097   // if we align at 8 then we will be sure 5 bytes are in the same line
2098   __ align(8);
2099 
2100   __ bind(hit);
2101 
2102   int vep_offset = ((intptr_t)__ pc()) - start;
2103 
2104 
2105   // The instruction at the verified entry point must be 5 bytes or longer
2106   // because it can be patched on the fly by make_non_entrant. The stack bang
2107   // instruction fits that requirement.
2108 
2109   // Generate stack overflow check
2110 
2111   if (UseStackBanging) {
2112     if (stack_size <= StackShadowPages*os::vm_page_size()) {
2113       __ bang_stack_with_offset(StackShadowPages*os::vm_page_size());
2114     } else {
2115       __ movl(rax, stack_size);
2116       __ bang_stack_size(rax, rbx);
2117     }
2118   } else {
2119     // need a 5 byte instruction to allow MT safe patching to non-entrant
2120     __ fat_nop();
2121   }
2122 
2123   assert(((uintptr_t)__ pc() - start - vep_offset) >= 5,
2124          "valid size for make_non_entrant");
2125 
2126   // Generate a new frame for the wrapper.
2127   __ enter();
2128 
2129   // -4 because return address is already present and so is saved rbp,
2130   if (stack_size - 2*wordSize != 0) {
2131     __ subq(rsp, stack_size - 2*wordSize);
2132   }
2133 
2134   // Frame is now completed as far a size and linkage.
2135 
2136   int frame_complete = ((intptr_t)__ pc()) - start;
2137 
2138   int c_arg, j_arg;
2139 
2140   // State of input register args
2141 
2142   bool  live[ConcreteRegisterImpl::number_of_registers];
2143 
2144   live[j_rarg0->as_VMReg()->value()] = false;
2145   live[j_rarg1->as_VMReg()->value()] = false;
2146   live[j_rarg2->as_VMReg()->value()] = false;
2147   live[j_rarg3->as_VMReg()->value()] = false;
2148   live[j_rarg4->as_VMReg()->value()] = false;
2149   live[j_rarg5->as_VMReg()->value()] = false;
2150 
2151   live[j_farg0->as_VMReg()->value()] = false;
2152   live[j_farg1->as_VMReg()->value()] = false;
2153   live[j_farg2->as_VMReg()->value()] = false;
2154   live[j_farg3->as_VMReg()->value()] = false;
2155   live[j_farg4->as_VMReg()->value()] = false;
2156   live[j_farg5->as_VMReg()->value()] = false;
2157   live[j_farg6->as_VMReg()->value()] = false;
2158   live[j_farg7->as_VMReg()->value()] = false;
2159 
2160 
2161   bool rax_is_zero = false;
2162 
2163   // All args (except strings) destined for the stack are moved first
2164   for (j_arg = first_arg_to_pass, c_arg = 0 ;
2165        j_arg < total_args_passed ; j_arg++, c_arg++ ) {
2166     VMRegPair src = in_regs[j_arg];
2167     VMRegPair dst = out_regs[c_arg];
2168 
2169     // Get the real reg value or a dummy (rsp)
2170 
2171     int src_reg = src.first()->is_reg() ?
2172                   src.first()->value() :
2173                   rsp->as_VMReg()->value();
2174 
2175     bool useless =  in_sig_bt[j_arg] == T_ARRAY ||
2176                     (in_sig_bt[j_arg] == T_OBJECT &&
2177                      out_sig_bt[c_arg] != T_INT &&
2178                      out_sig_bt[c_arg] != T_ADDRESS &&
2179                      out_sig_bt[c_arg] != T_LONG);
2180 
2181     live[src_reg] = !useless;
2182 
2183     if (dst.first()->is_stack()) {
2184 
2185       // Even though a string arg in a register is still live after this loop
2186       // after the string conversion loop (next) it will be dead so we take
2187       // advantage of that now for simpler code to manage live.
2188 
2189       live[src_reg] = false;
2190       switch (in_sig_bt[j_arg]) {
2191 
2192         case T_ARRAY:
2193         case T_OBJECT:
2194           {
2195             Address stack_dst(rsp, reg2offset_out(dst.first()));
2196 
2197             if (out_sig_bt[c_arg] == T_INT || out_sig_bt[c_arg] == T_LONG) {
2198               // need to unbox a one-word value
2199               Register in_reg = rax;
2200               if ( src.first()->is_reg() ) {
2201                 in_reg = src.first()->as_Register();
2202               } else {
2203                 __ movq(rax, Address(rbp, reg2offset_in(src.first())));
2204                 rax_is_zero = false;
2205               }
2206               Label skipUnbox;
2207               __ movptr(Address(rsp, reg2offset_out(dst.first())),
2208                         (int32_t)NULL_WORD);
2209               __ testq(in_reg, in_reg);
2210               __ jcc(Assembler::zero, skipUnbox);
2211 
2212               BasicType bt = out_sig_bt[c_arg];
2213               int box_offset = java_lang_boxing_object::value_offset_in_bytes(bt);
2214               Address src1(in_reg, box_offset);
2215               if ( bt == T_LONG ) {
2216                 __ movq(in_reg,  src1);
2217                 __ movq(stack_dst, in_reg);
2218                 assert(out_sig_bt[c_arg+1] == T_VOID, "must be");
2219                 ++c_arg; // skip over T_VOID to keep the loop indices in sync
2220               } else {
2221                 __ movl(in_reg,  src1);
2222                 __ movl(stack_dst, in_reg);
2223               }
2224 
2225               __ bind(skipUnbox);
2226             } else if (out_sig_bt[c_arg] != T_ADDRESS) {
2227               // Convert the arg to NULL
2228               if (!rax_is_zero) {
2229                 __ xorq(rax, rax);
2230                 rax_is_zero = true;
2231               }
2232               __ movq(stack_dst, rax);
2233             }
2234           }
2235           break;
2236 
2237         case T_VOID:
2238           break;
2239 
2240         case T_FLOAT:
2241           // This does the right thing since we know it is destined for the
2242           // stack
2243           float_move(masm, src, dst);
2244           break;
2245 
2246         case T_DOUBLE:
2247           // This does the right thing since we know it is destined for the
2248           // stack
2249           double_move(masm, src, dst);
2250           break;
2251 
2252         case T_LONG :
2253           long_move(masm, src, dst);
2254           break;
2255 
2256         case T_ADDRESS: assert(false, "found T_ADDRESS in java args");
2257 
2258         default:
2259           move32_64(masm, src, dst);
2260       }
2261     }
2262 
2263   }
2264 
2265   // If we have any strings we must store any register based arg to the stack
2266   // This includes any still live xmm registers too.
2267 
2268   int sid = 0;
2269 
2270   if (total_strings > 0 ) {
2271     for (j_arg = first_arg_to_pass, c_arg = 0 ;
2272          j_arg < total_args_passed ; j_arg++, c_arg++ ) {
2273       VMRegPair src = in_regs[j_arg];
2274       VMRegPair dst = out_regs[c_arg];
2275 
2276       if (src.first()->is_reg()) {
2277         Address src_tmp(rbp, fp_offset[src.first()->value()]);
2278 
2279         // string oops were left untouched by the previous loop even if the
2280         // eventual (converted) arg is destined for the stack so park them
2281         // away now (except for first)
2282 
2283         if (out_sig_bt[c_arg] == T_ADDRESS) {
2284           Address utf8_addr = Address(
2285               rsp, string_locs[sid++] * VMRegImpl::stack_slot_size);
2286           if (sid != 1) {
2287             // The first string arg won't be killed until after the utf8
2288             // conversion
2289             __ movq(utf8_addr, src.first()->as_Register());
2290           }
2291         } else if (dst.first()->is_reg()) {
2292           if (in_sig_bt[j_arg] == T_FLOAT || in_sig_bt[j_arg] == T_DOUBLE) {
2293 
2294             // Convert the xmm register to an int and store it in the reserved
2295             // location for the eventual c register arg
2296             XMMRegister f = src.first()->as_XMMRegister();
2297             if (in_sig_bt[j_arg] == T_FLOAT) {
2298               __ movflt(src_tmp, f);
2299             } else {
2300               __ movdbl(src_tmp, f);
2301             }
2302           } else {
2303             // If the arg is an oop type we don't support don't bother to store
2304             // it remember string was handled above.
2305             bool useless =  in_sig_bt[j_arg] == T_ARRAY ||
2306                             (in_sig_bt[j_arg] == T_OBJECT &&
2307                              out_sig_bt[c_arg] != T_INT &&
2308                              out_sig_bt[c_arg] != T_LONG);
2309 
2310             if (!useless) {
2311               __ movq(src_tmp, src.first()->as_Register());
2312             }
2313           }
2314         }
2315       }
2316       if (in_sig_bt[j_arg] == T_OBJECT && out_sig_bt[c_arg] == T_LONG) {
2317         assert(out_sig_bt[c_arg+1] == T_VOID, "must be");
2318         ++c_arg; // skip over T_VOID to keep the loop indices in sync
2319       }
2320     }
2321 
2322     // Now that the volatile registers are safe, convert all the strings
2323     sid = 0;
2324 
2325     for (j_arg = first_arg_to_pass, c_arg = 0 ;
2326          j_arg < total_args_passed ; j_arg++, c_arg++ ) {
2327       if (out_sig_bt[c_arg] == T_ADDRESS) {
2328         // It's a string
2329         Address utf8_addr = Address(
2330             rsp, string_locs[sid++] * VMRegImpl::stack_slot_size);
2331         // The first string we find might still be in the original java arg
2332         // register
2333 
2334         VMReg src = in_regs[j_arg].first();
2335 
2336         // We will need to eventually save the final argument to the trap
2337         // in the von-volatile location dedicated to src. This is the offset
2338         // from fp we will use.
2339         int src_off = src->is_reg() ?
2340             fp_offset[src->value()] : reg2offset_in(src);
2341 
2342         // This is where the argument will eventually reside
2343         VMRegPair dst = out_regs[c_arg];
2344 
2345         if (src->is_reg()) {
2346           if (sid == 1) {
2347             __ movq(c_rarg0, src->as_Register());
2348           } else {
2349             __ movq(c_rarg0, utf8_addr);
2350           }
2351         } else {
2352           // arg is still in the original location
2353           __ movq(c_rarg0, Address(rbp, reg2offset_in(src)));
2354         }
2355         Label done, convert;
2356 
2357         // see if the oop is NULL
2358         __ testq(c_rarg0, c_rarg0);
2359         __ jcc(Assembler::notEqual, convert);
2360 
2361         if (dst.first()->is_reg()) {
2362           // Save the ptr to utf string in the origina src loc or the tmp
2363           // dedicated to it
2364           __ movq(Address(rbp, src_off), c_rarg0);
2365         } else {
2366           __ movq(Address(rsp, reg2offset_out(dst.first())), c_rarg0);
2367         }
2368         __ jmp(done);
2369 
2370         __ bind(convert);
2371 
2372         __ lea(c_rarg1, utf8_addr);
2373         if (dst.first()->is_reg()) {
2374           __ movq(Address(rbp, src_off), c_rarg1);
2375         } else {
2376           __ movq(Address(rsp, reg2offset_out(dst.first())), c_rarg1);
2377         }
2378         // And do the conversion
2379         __ call(RuntimeAddress(
2380                 CAST_FROM_FN_PTR(address, SharedRuntime::get_utf)));
2381 
2382         __ bind(done);
2383       }
2384       if (in_sig_bt[j_arg] == T_OBJECT && out_sig_bt[c_arg] == T_LONG) {
2385         assert(out_sig_bt[c_arg+1] == T_VOID, "must be");
2386         ++c_arg; // skip over T_VOID to keep the loop indices in sync
2387       }
2388     }
2389     // The get_utf call killed all the c_arg registers
2390     live[c_rarg0->as_VMReg()->value()] = false;
2391     live[c_rarg1->as_VMReg()->value()] = false;
2392     live[c_rarg2->as_VMReg()->value()] = false;
2393     live[c_rarg3->as_VMReg()->value()] = false;
2394     live[c_rarg4->as_VMReg()->value()] = false;
2395     live[c_rarg5->as_VMReg()->value()] = false;
2396 
2397     live[c_farg0->as_VMReg()->value()] = false;
2398     live[c_farg1->as_VMReg()->value()] = false;
2399     live[c_farg2->as_VMReg()->value()] = false;
2400     live[c_farg3->as_VMReg()->value()] = false;
2401     live[c_farg4->as_VMReg()->value()] = false;
2402     live[c_farg5->as_VMReg()->value()] = false;
2403     live[c_farg6->as_VMReg()->value()] = false;
2404     live[c_farg7->as_VMReg()->value()] = false;
2405   }
2406 
2407   // Now we can finally move the register args to their desired locations
2408 
2409   rax_is_zero = false;
2410 
2411   for (j_arg = first_arg_to_pass, c_arg = 0 ;
2412        j_arg < total_args_passed ; j_arg++, c_arg++ ) {
2413 
2414     VMRegPair src = in_regs[j_arg];
2415     VMRegPair dst = out_regs[c_arg];
2416 
2417     // Only need to look for args destined for the interger registers (since we
2418     // convert float/double args to look like int/long outbound)
2419     if (dst.first()->is_reg()) {
2420       Register r =  dst.first()->as_Register();
2421 
2422       // Check if the java arg is unsupported and thereofre useless
2423       bool useless =  in_sig_bt[j_arg] == T_ARRAY ||
2424                       (in_sig_bt[j_arg] == T_OBJECT &&
2425                        out_sig_bt[c_arg] != T_INT &&
2426                        out_sig_bt[c_arg] != T_ADDRESS &&
2427                        out_sig_bt[c_arg] != T_LONG);
2428 
2429 
2430       // If we're going to kill an existing arg save it first
2431       if (live[dst.first()->value()]) {
2432         // you can't kill yourself
2433         if (src.first() != dst.first()) {
2434           __ movq(Address(rbp, fp_offset[dst.first()->value()]), r);
2435         }
2436       }
2437       if (src.first()->is_reg()) {
2438         if (live[src.first()->value()] ) {
2439           if (in_sig_bt[j_arg] == T_FLOAT) {
2440             __ movdl(r, src.first()->as_XMMRegister());
2441           } else if (in_sig_bt[j_arg] == T_DOUBLE) {
2442             __ movdq(r, src.first()->as_XMMRegister());
2443           } else if (r != src.first()->as_Register()) {
2444             if (!useless) {
2445               __ movq(r, src.first()->as_Register());
2446             }
2447           }
2448         } else {
2449           // If the arg is an oop type we don't support don't bother to store
2450           // it
2451           if (!useless) {
2452             if (in_sig_bt[j_arg] == T_DOUBLE ||
2453                 in_sig_bt[j_arg] == T_LONG  ||
2454                 in_sig_bt[j_arg] == T_OBJECT ) {
2455               __ movq(r, Address(rbp, fp_offset[src.first()->value()]));
2456             } else {
2457               __ movl(r, Address(rbp, fp_offset[src.first()->value()]));
2458             }
2459           }
2460         }
2461         live[src.first()->value()] = false;
2462       } else if (!useless) {
2463         // full sized move even for int should be ok
2464         __ movq(r, Address(rbp, reg2offset_in(src.first())));
2465       }
2466 
2467       // At this point r has the original java arg in the final location
2468       // (assuming it wasn't useless). If the java arg was an oop
2469       // we have a bit more to do
2470 
2471       if (in_sig_bt[j_arg] == T_ARRAY || in_sig_bt[j_arg] == T_OBJECT ) {
2472         if (out_sig_bt[c_arg] == T_INT || out_sig_bt[c_arg] == T_LONG) {
2473           // need to unbox a one-word value
2474           Label skip;
2475           __ testq(r, r);
2476           __ jcc(Assembler::equal, skip);
2477           BasicType bt = out_sig_bt[c_arg];
2478           int box_offset = java_lang_boxing_object::value_offset_in_bytes(bt);
2479           Address src1(r, box_offset);
2480           if ( bt == T_LONG ) {
2481             __ movq(r, src1);
2482           } else {
2483             __ movl(r, src1);
2484           }
2485           __ bind(skip);
2486 
2487         } else if (out_sig_bt[c_arg] != T_ADDRESS) {
2488           // Convert the arg to NULL
2489           __ xorq(r, r);
2490         }
2491       }
2492 
2493       // dst can longer be holding an input value
2494       live[dst.first()->value()] = false;
2495     }
2496     if (in_sig_bt[j_arg] == T_OBJECT && out_sig_bt[c_arg] == T_LONG) {
2497       assert(out_sig_bt[c_arg+1] == T_VOID, "must be");
2498       ++c_arg; // skip over T_VOID to keep the loop indices in sync
2499     }
2500   }
2501 
2502 
2503   // Ok now we are done. Need to place the nop that dtrace wants in order to
2504   // patch in the trap
2505   int patch_offset = ((intptr_t)__ pc()) - start;
2506 
2507   __ nop();
2508 
2509 
2510   // Return
2511 
2512   __ leave();
2513   __ ret(0);
2514 
2515   __ flush();
2516 
2517   nmethod *nm = nmethod::new_dtrace_nmethod(
2518       method, masm->code(), vep_offset, patch_offset, frame_complete,
2519       stack_slots / VMRegImpl::slots_per_word);
2520   return nm;
2521 
2522 }
2523 
2524 #endif // HAVE_DTRACE_H
2525 
2526 // this function returns the adjust size (in number of words) to a c2i adapter
2527 // activation for use during deoptimization
2528 int Deoptimization::last_frame_adjust(int callee_parameters, int callee_locals ) {
2529   return (callee_locals - callee_parameters) * Interpreter::stackElementWords();
2530 }
2531 
2532 
2533 uint SharedRuntime::out_preserve_stack_slots() {
2534   return 0;
2535 }
2536 
2537 
2538 //------------------------------generate_deopt_blob----------------------------
2539 void SharedRuntime::generate_deopt_blob() {
2540   // Allocate space for the code
2541   ResourceMark rm;
2542   // Setup code generation tools
2543   CodeBuffer buffer("deopt_blob", 2048, 1024);
2544   MacroAssembler* masm = new MacroAssembler(&buffer);
2545   int frame_size_in_words;
2546   OopMap* map = NULL;
2547   OopMapSet *oop_maps = new OopMapSet();
2548 
2549   // -------------
2550   // This code enters when returning to a de-optimized nmethod.  A return
2551   // address has been pushed on the the stack, and return values are in
2552   // registers.
2553   // If we are doing a normal deopt then we were called from the patched
2554   // nmethod from the point we returned to the nmethod. So the return
2555   // address on the stack is wrong by NativeCall::instruction_size
2556   // We will adjust the value so it looks like we have the original return
2557   // address on the stack (like when we eagerly deoptimized).
2558   // In the case of an exception pending when deoptimizing, we enter
2559   // with a return address on the stack that points after the call we patched
2560   // into the exception handler. We have the following register state from,
2561   // e.g., the forward exception stub (see stubGenerator_x86_64.cpp).
2562   //    rax: exception oop
2563   //    rbx: exception handler
2564   //    rdx: throwing pc
2565   // So in this case we simply jam rdx into the useless return address and
2566   // the stack looks just like we want.
2567   //
2568   // At this point we need to de-opt.  We save the argument return
2569   // registers.  We call the first C routine, fetch_unroll_info().  This
2570   // routine captures the return values and returns a structure which
2571   // describes the current frame size and the sizes of all replacement frames.
2572   // The current frame is compiled code and may contain many inlined
2573   // functions, each with their own JVM state.  We pop the current frame, then
2574   // push all the new frames.  Then we call the C routine unpack_frames() to
2575   // populate these frames.  Finally unpack_frames() returns us the new target
2576   // address.  Notice that callee-save registers are BLOWN here; they have
2577   // already been captured in the vframeArray at the time the return PC was
2578   // patched.
2579   address start = __ pc();
2580   Label cont;
2581 
2582   // Prolog for non exception case!
2583 
2584   // Save everything in sight.
2585   map = RegisterSaver::save_live_registers(masm, 0, &frame_size_in_words);
2586 
2587   // Normal deoptimization.  Save exec mode for unpack_frames.
2588   __ movl(r14, Deoptimization::Unpack_deopt); // callee-saved
2589   __ jmp(cont);
2590 
2591   int reexecute_offset = __ pc() - start;
2592 
2593   // Reexecute case
2594   // return address is the pc describes what bci to do re-execute at
2595 
2596   // No need to update map as each call to save_live_registers will produce identical oopmap
2597   (void) RegisterSaver::save_live_registers(masm, 0, &frame_size_in_words);
2598 
2599   __ movl(r14, Deoptimization::Unpack_reexecute); // callee-saved
2600   __ jmp(cont);
2601 
2602   int exception_offset = __ pc() - start;
2603 
2604   // Prolog for exception case
2605 
2606   // all registers are dead at this entry point, except for rax, and
2607   // rdx which contain the exception oop and exception pc
2608   // respectively.  Set them in TLS and fall thru to the
2609   // unpack_with_exception_in_tls entry point.
2610 
2611   __ movptr(Address(r15_thread, JavaThread::exception_pc_offset()), rdx);
2612   __ movptr(Address(r15_thread, JavaThread::exception_oop_offset()), rax);
2613 
2614   int exception_in_tls_offset = __ pc() - start;
2615 
2616   // new implementation because exception oop is now passed in JavaThread
2617 
2618   // Prolog for exception case
2619   // All registers must be preserved because they might be used by LinearScan
2620   // Exceptiop oop and throwing PC are passed in JavaThread
2621   // tos: stack at point of call to method that threw the exception (i.e. only
2622   // args are on the stack, no return address)
2623 
2624   // make room on stack for the return address
2625   // It will be patched later with the throwing pc. The correct value is not
2626   // available now because loading it from memory would destroy registers.
2627   __ push(0);
2628 
2629   // Save everything in sight.
2630   map = RegisterSaver::save_live_registers(masm, 0, &frame_size_in_words);
2631 
2632   // Now it is safe to overwrite any register
2633 
2634   // Deopt during an exception.  Save exec mode for unpack_frames.
2635   __ movl(r14, Deoptimization::Unpack_exception); // callee-saved
2636 
2637   // load throwing pc from JavaThread and patch it as the return address
2638   // of the current frame. Then clear the field in JavaThread
2639 
2640   __ movptr(rdx, Address(r15_thread, JavaThread::exception_pc_offset()));
2641   __ movptr(Address(rbp, wordSize), rdx);
2642   __ movptr(Address(r15_thread, JavaThread::exception_pc_offset()), (int32_t)NULL_WORD);
2643 
2644 #ifdef ASSERT
2645   // verify that there is really an exception oop in JavaThread
2646   __ movptr(rax, Address(r15_thread, JavaThread::exception_oop_offset()));
2647   __ verify_oop(rax);
2648 
2649   // verify that there is no pending exception
2650   Label no_pending_exception;
2651   __ movptr(rax, Address(r15_thread, Thread::pending_exception_offset()));
2652   __ testptr(rax, rax);
2653   __ jcc(Assembler::zero, no_pending_exception);
2654   __ stop("must not have pending exception here");
2655   __ bind(no_pending_exception);
2656 #endif
2657 
2658   __ bind(cont);
2659 
2660   // Call C code.  Need thread and this frame, but NOT official VM entry
2661   // crud.  We cannot block on this call, no GC can happen.
2662   //
2663   // UnrollBlock* fetch_unroll_info(JavaThread* thread)
2664 
2665   // fetch_unroll_info needs to call last_java_frame().
2666 
2667   __ set_last_Java_frame(noreg, noreg, NULL);
2668 #ifdef ASSERT
2669   { Label L;
2670     __ cmpptr(Address(r15_thread,
2671                     JavaThread::last_Java_fp_offset()),
2672             (int32_t)0);
2673     __ jcc(Assembler::equal, L);
2674     __ stop("SharedRuntime::generate_deopt_blob: last_Java_fp not cleared");
2675     __ bind(L);
2676   }
2677 #endif // ASSERT
2678   __ mov(c_rarg0, r15_thread);
2679   __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, Deoptimization::fetch_unroll_info)));
2680 
2681   // Need to have an oopmap that tells fetch_unroll_info where to
2682   // find any register it might need.
2683   oop_maps->add_gc_map(__ pc() - start, map);
2684 
2685   __ reset_last_Java_frame(false, false);
2686 
2687   // Load UnrollBlock* into rdi
2688   __ mov(rdi, rax);
2689 
2690    Label noException;
2691   __ cmpl(r14, Deoptimization::Unpack_exception);   // Was exception pending?
2692   __ jcc(Assembler::notEqual, noException);
2693   __ movptr(rax, Address(r15_thread, JavaThread::exception_oop_offset()));
2694   // QQQ this is useless it was NULL above
2695   __ movptr(rdx, Address(r15_thread, JavaThread::exception_pc_offset()));
2696   __ movptr(Address(r15_thread, JavaThread::exception_oop_offset()), (int32_t)NULL_WORD);
2697   __ movptr(Address(r15_thread, JavaThread::exception_pc_offset()), (int32_t)NULL_WORD);
2698 
2699   __ verify_oop(rax);
2700 
2701   // Overwrite the result registers with the exception results.
2702   __ movptr(Address(rsp, RegisterSaver::rax_offset_in_bytes()), rax);
2703   // I think this is useless
2704   __ movptr(Address(rsp, RegisterSaver::rdx_offset_in_bytes()), rdx);
2705 
2706   __ bind(noException);
2707 
2708   // Only register save data is on the stack.
2709   // Now restore the result registers.  Everything else is either dead
2710   // or captured in the vframeArray.
2711   RegisterSaver::restore_result_registers(masm);
2712 
2713   // All of the register save area has been popped of the stack. Only the
2714   // return address remains.
2715 
2716   // Pop all the frames we must move/replace.
2717   //
2718   // Frame picture (youngest to oldest)
2719   // 1: self-frame (no frame link)
2720   // 2: deopting frame  (no frame link)
2721   // 3: caller of deopting frame (could be compiled/interpreted).
2722   //
2723   // Note: by leaving the return address of self-frame on the stack
2724   // and using the size of frame 2 to adjust the stack
2725   // when we are done the return to frame 3 will still be on the stack.
2726 
2727   // Pop deoptimized frame
2728   __ movl(rcx, Address(rdi, Deoptimization::UnrollBlock::size_of_deoptimized_frame_offset_in_bytes()));
2729   __ addptr(rsp, rcx);
2730 
2731   // rsp should be pointing at the return address to the caller (3)
2732 
2733   // Stack bang to make sure there's enough room for these interpreter frames.
2734   if (UseStackBanging) {
2735     __ movl(rbx, Address(rdi, Deoptimization::UnrollBlock::total_frame_sizes_offset_in_bytes()));
2736     __ bang_stack_size(rbx, rcx);
2737   }
2738 
2739   // Load address of array of frame pcs into rcx
2740   __ movptr(rcx, Address(rdi, Deoptimization::UnrollBlock::frame_pcs_offset_in_bytes()));
2741 
2742   // Trash the old pc
2743   __ addptr(rsp, wordSize);
2744 
2745   // Load address of array of frame sizes into rsi
2746   __ movptr(rsi, Address(rdi, Deoptimization::UnrollBlock::frame_sizes_offset_in_bytes()));
2747 
2748   // Load counter into rdx
2749   __ movl(rdx, Address(rdi, Deoptimization::UnrollBlock::number_of_frames_offset_in_bytes()));
2750 
2751   // Pick up the initial fp we should save
2752   __ movptr(rbp, Address(rdi, Deoptimization::UnrollBlock::initial_fp_offset_in_bytes()));
2753 
2754   // Now adjust the caller's stack to make up for the extra locals
2755   // but record the original sp so that we can save it in the skeletal interpreter
2756   // frame and the stack walking of interpreter_sender will get the unextended sp
2757   // value and not the "real" sp value.
2758 
2759   const Register sender_sp = r8;
2760 
2761   __ mov(sender_sp, rsp);
2762   __ movl(rbx, Address(rdi,
2763                        Deoptimization::UnrollBlock::
2764                        caller_adjustment_offset_in_bytes()));
2765   __ subptr(rsp, rbx);
2766 
2767   // Push interpreter frames in a loop
2768   Label loop;
2769   __ bind(loop);
2770   __ movptr(rbx, Address(rsi, 0));      // Load frame size
2771 #ifdef CC_INTERP
2772   __ subptr(rbx, 4*wordSize);           // we'll push pc and ebp by hand and
2773 #ifdef ASSERT
2774   __ push(0xDEADDEAD);                  // Make a recognizable pattern
2775   __ push(0xDEADDEAD);
2776 #else /* ASSERT */
2777   __ subptr(rsp, 2*wordSize);           // skip the "static long no_param"
2778 #endif /* ASSERT */
2779 #else
2780   __ subptr(rbx, 2*wordSize);           // We'll push pc and ebp by hand
2781 #endif // CC_INTERP
2782   __ pushptr(Address(rcx, 0));          // Save return address
2783   __ enter();                           // Save old & set new ebp
2784   __ subptr(rsp, rbx);                  // Prolog
2785 #ifdef CC_INTERP
2786   __ movptr(Address(rbp,
2787                   -(sizeof(BytecodeInterpreter)) + in_bytes(byte_offset_of(BytecodeInterpreter, _sender_sp))),
2788             sender_sp); // Make it walkable
2789 #else /* CC_INTERP */
2790   // This value is corrected by layout_activation_impl
2791   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD );
2792   __ movptr(Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize), sender_sp); // Make it walkable
2793 #endif /* CC_INTERP */
2794   __ mov(sender_sp, rsp);               // Pass sender_sp to next frame
2795   __ addptr(rsi, wordSize);             // Bump array pointer (sizes)
2796   __ addptr(rcx, wordSize);             // Bump array pointer (pcs)
2797   __ decrementl(rdx);                   // Decrement counter
2798   __ jcc(Assembler::notZero, loop);
2799   __ pushptr(Address(rcx, 0));          // Save final return address
2800 
2801   // Re-push self-frame
2802   __ enter();                           // Save old & set new ebp
2803 
2804   // Allocate a full sized register save area.
2805   // Return address and rbp are in place, so we allocate two less words.
2806   __ subptr(rsp, (frame_size_in_words - 2) * wordSize);
2807 
2808   // Restore frame locals after moving the frame
2809   __ movdbl(Address(rsp, RegisterSaver::xmm0_offset_in_bytes()), xmm0);
2810   __ movptr(Address(rsp, RegisterSaver::rax_offset_in_bytes()), rax);
2811 
2812   // Call C code.  Need thread but NOT official VM entry
2813   // crud.  We cannot block on this call, no GC can happen.  Call should
2814   // restore return values to their stack-slots with the new SP.
2815   //
2816   // void Deoptimization::unpack_frames(JavaThread* thread, int exec_mode)
2817 
2818   // Use rbp because the frames look interpreted now
2819   __ set_last_Java_frame(noreg, rbp, NULL);
2820 
2821   __ mov(c_rarg0, r15_thread);
2822   __ movl(c_rarg1, r14); // second arg: exec_mode
2823   __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, Deoptimization::unpack_frames)));
2824 
2825   // Set an oopmap for the call site
2826   oop_maps->add_gc_map(__ pc() - start,
2827                        new OopMap( frame_size_in_words, 0 ));
2828 
2829   __ reset_last_Java_frame(true, false);
2830 
2831   // Collect return values
2832   __ movdbl(xmm0, Address(rsp, RegisterSaver::xmm0_offset_in_bytes()));
2833   __ movptr(rax, Address(rsp, RegisterSaver::rax_offset_in_bytes()));
2834   // I think this is useless (throwing pc?)
2835   __ movptr(rdx, Address(rsp, RegisterSaver::rdx_offset_in_bytes()));
2836 
2837   // Pop self-frame.
2838   __ leave();                           // Epilog
2839 
2840   // Jump to interpreter
2841   __ ret(0);
2842 
2843   // Make sure all code is generated
2844   masm->flush();
2845 
2846   _deopt_blob = DeoptimizationBlob::create(&buffer, oop_maps, 0, exception_offset, reexecute_offset, frame_size_in_words);
2847   _deopt_blob->set_unpack_with_exception_in_tls_offset(exception_in_tls_offset);
2848 }
2849 
2850 #ifdef COMPILER2
2851 //------------------------------generate_uncommon_trap_blob--------------------
2852 void SharedRuntime::generate_uncommon_trap_blob() {
2853   // Allocate space for the code
2854   ResourceMark rm;
2855   // Setup code generation tools
2856   CodeBuffer buffer("uncommon_trap_blob", 2048, 1024);
2857   MacroAssembler* masm = new MacroAssembler(&buffer);
2858 
2859   assert(SimpleRuntimeFrame::framesize % 4 == 0, "sp not 16-byte aligned");
2860 
2861   address start = __ pc();
2862 
2863   // Push self-frame.  We get here with a return address on the
2864   // stack, so rsp is 8-byte aligned until we allocate our frame.
2865   __ subptr(rsp, SimpleRuntimeFrame::return_off << LogBytesPerInt); // Epilog!
2866 
2867   // No callee saved registers. rbp is assumed implicitly saved
2868   __ movptr(Address(rsp, SimpleRuntimeFrame::rbp_off << LogBytesPerInt), rbp);
2869 
2870   // compiler left unloaded_class_index in j_rarg0 move to where the
2871   // runtime expects it.
2872   __ movl(c_rarg1, j_rarg0);
2873 
2874   __ set_last_Java_frame(noreg, noreg, NULL);
2875 
2876   // Call C code.  Need thread but NOT official VM entry
2877   // crud.  We cannot block on this call, no GC can happen.  Call should
2878   // capture callee-saved registers as well as return values.
2879   // Thread is in rdi already.
2880   //
2881   // UnrollBlock* uncommon_trap(JavaThread* thread, jint unloaded_class_index);
2882 
2883   __ mov(c_rarg0, r15_thread);
2884   __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, Deoptimization::uncommon_trap)));
2885 
2886   // Set an oopmap for the call site
2887   OopMapSet* oop_maps = new OopMapSet();
2888   OopMap* map = new OopMap(SimpleRuntimeFrame::framesize, 0);
2889 
2890   // location of rbp is known implicitly by the frame sender code
2891 
2892   oop_maps->add_gc_map(__ pc() - start, map);
2893 
2894   __ reset_last_Java_frame(false, false);
2895 
2896   // Load UnrollBlock* into rdi
2897   __ mov(rdi, rax);
2898 
2899   // Pop all the frames we must move/replace.
2900   //
2901   // Frame picture (youngest to oldest)
2902   // 1: self-frame (no frame link)
2903   // 2: deopting frame  (no frame link)
2904   // 3: caller of deopting frame (could be compiled/interpreted).
2905 
2906   // Pop self-frame.  We have no frame, and must rely only on rax and rsp.
2907   __ addptr(rsp, (SimpleRuntimeFrame::framesize - 2) << LogBytesPerInt); // Epilog!
2908 
2909   // Pop deoptimized frame (int)
2910   __ movl(rcx, Address(rdi,
2911                        Deoptimization::UnrollBlock::
2912                        size_of_deoptimized_frame_offset_in_bytes()));
2913   __ addptr(rsp, rcx);
2914 
2915   // rsp should be pointing at the return address to the caller (3)
2916 
2917   // Stack bang to make sure there's enough room for these interpreter frames.
2918   if (UseStackBanging) {
2919     __ movl(rbx, Address(rdi ,Deoptimization::UnrollBlock::total_frame_sizes_offset_in_bytes()));
2920     __ bang_stack_size(rbx, rcx);
2921   }
2922 
2923   // Load address of array of frame pcs into rcx (address*)
2924   __ movptr(rcx,
2925             Address(rdi,
2926                     Deoptimization::UnrollBlock::frame_pcs_offset_in_bytes()));
2927 
2928   // Trash the return pc
2929   __ addptr(rsp, wordSize);
2930 
2931   // Load address of array of frame sizes into rsi (intptr_t*)
2932   __ movptr(rsi, Address(rdi,
2933                          Deoptimization::UnrollBlock::
2934                          frame_sizes_offset_in_bytes()));
2935 
2936   // Counter
2937   __ movl(rdx, Address(rdi,
2938                        Deoptimization::UnrollBlock::
2939                        number_of_frames_offset_in_bytes())); // (int)
2940 
2941   // Pick up the initial fp we should save
2942   __ movptr(rbp,
2943             Address(rdi,
2944                     Deoptimization::UnrollBlock::initial_fp_offset_in_bytes()));
2945 
2946   // Now adjust the caller's stack to make up for the extra locals but
2947   // record the original sp so that we can save it in the skeletal
2948   // interpreter frame and the stack walking of interpreter_sender
2949   // will get the unextended sp value and not the "real" sp value.
2950 
2951   const Register sender_sp = r8;
2952 
2953   __ mov(sender_sp, rsp);
2954   __ movl(rbx, Address(rdi,
2955                        Deoptimization::UnrollBlock::
2956                        caller_adjustment_offset_in_bytes())); // (int)
2957   __ subptr(rsp, rbx);
2958 
2959   // Push interpreter frames in a loop
2960   Label loop;
2961   __ bind(loop);
2962   __ movptr(rbx, Address(rsi, 0)); // Load frame size
2963   __ subptr(rbx, 2 * wordSize);    // We'll push pc and rbp by hand
2964   __ pushptr(Address(rcx, 0));     // Save return address
2965   __ enter();                      // Save old & set new rbp
2966   __ subptr(rsp, rbx);             // Prolog
2967 #ifdef CC_INTERP
2968   __ movptr(Address(rbp,
2969                   -(sizeof(BytecodeInterpreter)) + in_bytes(byte_offset_of(BytecodeInterpreter, _sender_sp))),
2970             sender_sp); // Make it walkable
2971 #else // CC_INTERP
2972   __ movptr(Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize),
2973             sender_sp);            // Make it walkable
2974   // This value is corrected by layout_activation_impl
2975   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD );
2976 #endif // CC_INTERP
2977   __ mov(sender_sp, rsp);          // Pass sender_sp to next frame
2978   __ addptr(rsi, wordSize);        // Bump array pointer (sizes)
2979   __ addptr(rcx, wordSize);        // Bump array pointer (pcs)
2980   __ decrementl(rdx);              // Decrement counter
2981   __ jcc(Assembler::notZero, loop);
2982   __ pushptr(Address(rcx, 0));     // Save final return address
2983 
2984   // Re-push self-frame
2985   __ enter();                 // Save old & set new rbp
2986   __ subptr(rsp, (SimpleRuntimeFrame::framesize - 4) << LogBytesPerInt);
2987                               // Prolog
2988 
2989   // Use rbp because the frames look interpreted now
2990   __ set_last_Java_frame(noreg, rbp, NULL);
2991 
2992   // Call C code.  Need thread but NOT official VM entry
2993   // crud.  We cannot block on this call, no GC can happen.  Call should
2994   // restore return values to their stack-slots with the new SP.
2995   // Thread is in rdi already.
2996   //
2997   // BasicType unpack_frames(JavaThread* thread, int exec_mode);
2998 
2999   __ mov(c_rarg0, r15_thread);
3000   __ movl(c_rarg1, Deoptimization::Unpack_uncommon_trap);
3001   __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, Deoptimization::unpack_frames)));
3002 
3003   // Set an oopmap for the call site
3004   oop_maps->add_gc_map(__ pc() - start, new OopMap(SimpleRuntimeFrame::framesize, 0));
3005 
3006   __ reset_last_Java_frame(true, false);
3007 
3008   // Pop self-frame.
3009   __ leave();                 // Epilog
3010 
3011   // Jump to interpreter
3012   __ ret(0);
3013 
3014   // Make sure all code is generated
3015   masm->flush();
3016 
3017   _uncommon_trap_blob =  UncommonTrapBlob::create(&buffer, oop_maps,
3018                                                  SimpleRuntimeFrame::framesize >> 1);
3019 }
3020 #endif // COMPILER2
3021 
3022 
3023 //------------------------------generate_handler_blob------
3024 //
3025 // Generate a special Compile2Runtime blob that saves all registers,
3026 // and setup oopmap.
3027 //
3028 static SafepointBlob* generate_handler_blob(address call_ptr, bool cause_return) {
3029   assert(StubRoutines::forward_exception_entry() != NULL,
3030          "must be generated before");
3031 
3032   ResourceMark rm;
3033   OopMapSet *oop_maps = new OopMapSet();
3034   OopMap* map;
3035 
3036   // Allocate space for the code.  Setup code generation tools.
3037   CodeBuffer buffer("handler_blob", 2048, 1024);
3038   MacroAssembler* masm = new MacroAssembler(&buffer);
3039 
3040   address start   = __ pc();
3041   address call_pc = NULL;
3042   int frame_size_in_words;
3043 
3044   // Make room for return address (or push it again)
3045   if (!cause_return) {
3046     __ push(rbx);
3047   }
3048 
3049   // Save registers, fpu state, and flags
3050   map = RegisterSaver::save_live_registers(masm, 0, &frame_size_in_words);
3051 
3052   // The following is basically a call_VM.  However, we need the precise
3053   // address of the call in order to generate an oopmap. Hence, we do all the
3054   // work outselves.
3055 
3056   __ set_last_Java_frame(noreg, noreg, NULL);
3057 
3058   // The return address must always be correct so that frame constructor never
3059   // sees an invalid pc.
3060 
3061   if (!cause_return) {
3062     // overwrite the dummy value we pushed on entry
3063     __ movptr(c_rarg0, Address(r15_thread, JavaThread::saved_exception_pc_offset()));
3064     __ movptr(Address(rbp, wordSize), c_rarg0);
3065   }
3066 
3067   // Do the call
3068   __ mov(c_rarg0, r15_thread);
3069   __ call(RuntimeAddress(call_ptr));
3070 
3071   // Set an oopmap for the call site.  This oopmap will map all
3072   // oop-registers and debug-info registers as callee-saved.  This
3073   // will allow deoptimization at this safepoint to find all possible
3074   // debug-info recordings, as well as let GC find all oops.
3075 
3076   oop_maps->add_gc_map( __ pc() - start, map);
3077 
3078   Label noException;
3079 
3080   __ reset_last_Java_frame(false, false);
3081 
3082   __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
3083   __ jcc(Assembler::equal, noException);
3084 
3085   // Exception pending
3086 
3087   RegisterSaver::restore_live_registers(masm);
3088 
3089   __ jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
3090 
3091   // No exception case
3092   __ bind(noException);
3093 
3094   // Normal exit, restore registers and exit.
3095   RegisterSaver::restore_live_registers(masm);
3096 
3097   __ ret(0);
3098 
3099   // Make sure all code is generated
3100   masm->flush();
3101 
3102   // Fill-out other meta info
3103   return SafepointBlob::create(&buffer, oop_maps, frame_size_in_words);
3104 }
3105 
3106 //
3107 // generate_resolve_blob - call resolution (static/virtual/opt-virtual/ic-miss
3108 //
3109 // Generate a stub that calls into vm to find out the proper destination
3110 // of a java call. All the argument registers are live at this point
3111 // but since this is generic code we don't know what they are and the caller
3112 // must do any gc of the args.
3113 //
3114 static RuntimeStub* generate_resolve_blob(address destination, const char* name) {
3115   assert (StubRoutines::forward_exception_entry() != NULL, "must be generated before");
3116 
3117   // allocate space for the code
3118   ResourceMark rm;
3119 
3120   CodeBuffer buffer(name, 1000, 512);
3121   MacroAssembler* masm                = new MacroAssembler(&buffer);
3122 
3123   int frame_size_in_words;
3124 
3125   OopMapSet *oop_maps = new OopMapSet();
3126   OopMap* map = NULL;
3127 
3128   int start = __ offset();
3129 
3130   map = RegisterSaver::save_live_registers(masm, 0, &frame_size_in_words);
3131 
3132   int frame_complete = __ offset();
3133 
3134   __ set_last_Java_frame(noreg, noreg, NULL);
3135 
3136   __ mov(c_rarg0, r15_thread);
3137 
3138   __ call(RuntimeAddress(destination));
3139 
3140 
3141   // Set an oopmap for the call site.
3142   // We need this not only for callee-saved registers, but also for volatile
3143   // registers that the compiler might be keeping live across a safepoint.
3144 
3145   oop_maps->add_gc_map( __ offset() - start, map);
3146 
3147   // rax contains the address we are going to jump to assuming no exception got installed
3148 
3149   // clear last_Java_sp
3150   __ reset_last_Java_frame(false, false);
3151   // check for pending exceptions
3152   Label pending;
3153   __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
3154   __ jcc(Assembler::notEqual, pending);
3155 
3156   // get the returned methodOop
3157   __ movptr(rbx, Address(r15_thread, JavaThread::vm_result_offset()));
3158   __ movptr(Address(rsp, RegisterSaver::rbx_offset_in_bytes()), rbx);
3159 
3160   __ movptr(Address(rsp, RegisterSaver::rax_offset_in_bytes()), rax);
3161 
3162   RegisterSaver::restore_live_registers(masm);
3163 
3164   // We are back the the original state on entry and ready to go.
3165 
3166   __ jmp(rax);
3167 
3168   // Pending exception after the safepoint
3169 
3170   __ bind(pending);
3171 
3172   RegisterSaver::restore_live_registers(masm);
3173 
3174   // exception pending => remove activation and forward to exception handler
3175 
3176   __ movptr(Address(r15_thread, JavaThread::vm_result_offset()), (int)NULL_WORD);
3177 
3178   __ movptr(rax, Address(r15_thread, Thread::pending_exception_offset()));
3179   __ jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
3180 
3181   // -------------
3182   // make sure all code is generated
3183   masm->flush();
3184 
3185   // return the  blob
3186   // frame_size_words or bytes??
3187   return RuntimeStub::new_runtime_stub(name, &buffer, frame_complete, frame_size_in_words, oop_maps, true);
3188 }
3189 
3190 
3191 void SharedRuntime::generate_stubs() {
3192 
3193   _wrong_method_blob = generate_resolve_blob(CAST_FROM_FN_PTR(address, SharedRuntime::handle_wrong_method),
3194                                         "wrong_method_stub");
3195   _ic_miss_blob =      generate_resolve_blob(CAST_FROM_FN_PTR(address, SharedRuntime::handle_wrong_method_ic_miss),
3196                                         "ic_miss_stub");
3197   _resolve_opt_virtual_call_blob = generate_resolve_blob(CAST_FROM_FN_PTR(address, SharedRuntime::resolve_opt_virtual_call_C),
3198                                         "resolve_opt_virtual_call");
3199 
3200   _resolve_virtual_call_blob = generate_resolve_blob(CAST_FROM_FN_PTR(address, SharedRuntime::resolve_virtual_call_C),
3201                                         "resolve_virtual_call");
3202 
3203   _resolve_static_call_blob = generate_resolve_blob(CAST_FROM_FN_PTR(address, SharedRuntime::resolve_static_call_C),
3204                                         "resolve_static_call");
3205   _polling_page_safepoint_handler_blob =
3206     generate_handler_blob(CAST_FROM_FN_PTR(address,
3207                    SafepointSynchronize::handle_polling_page_exception), false);
3208 
3209   _polling_page_return_handler_blob =
3210     generate_handler_blob(CAST_FROM_FN_PTR(address,
3211                    SafepointSynchronize::handle_polling_page_exception), true);
3212 
3213   generate_deopt_blob();
3214 
3215 #ifdef COMPILER2
3216   generate_uncommon_trap_blob();
3217 #endif // COMPILER2
3218 }
3219 
3220 
3221 #ifdef COMPILER2
3222 // This is here instead of runtime_x86_64.cpp because it uses SimpleRuntimeFrame
3223 //
3224 //------------------------------generate_exception_blob---------------------------
3225 // creates exception blob at the end
3226 // Using exception blob, this code is jumped from a compiled method.
3227 // (see emit_exception_handler in x86_64.ad file)
3228 //
3229 // Given an exception pc at a call we call into the runtime for the
3230 // handler in this method. This handler might merely restore state
3231 // (i.e. callee save registers) unwind the frame and jump to the
3232 // exception handler for the nmethod if there is no Java level handler
3233 // for the nmethod.
3234 //
3235 // This code is entered with a jmp.
3236 //
3237 // Arguments:
3238 //   rax: exception oop
3239 //   rdx: exception pc
3240 //
3241 // Results:
3242 //   rax: exception oop
3243 //   rdx: exception pc in caller or ???
3244 //   destination: exception handler of caller
3245 //
3246 // Note: the exception pc MUST be at a call (precise debug information)
3247 //       Registers rax, rdx, rcx, rsi, rdi, r8-r11 are not callee saved.
3248 //
3249 
3250 void OptoRuntime::generate_exception_blob() {
3251   assert(!OptoRuntime::is_callee_saved_register(RDX_num), "");
3252   assert(!OptoRuntime::is_callee_saved_register(RAX_num), "");
3253   assert(!OptoRuntime::is_callee_saved_register(RCX_num), "");
3254 
3255   assert(SimpleRuntimeFrame::framesize % 4 == 0, "sp not 16-byte aligned");
3256 
3257   // Allocate space for the code
3258   ResourceMark rm;
3259   // Setup code generation tools
3260   CodeBuffer buffer("exception_blob", 2048, 1024);
3261   MacroAssembler* masm = new MacroAssembler(&buffer);
3262 
3263 
3264   address start = __ pc();
3265 
3266   // Exception pc is 'return address' for stack walker
3267   __ push(rdx);
3268   __ subptr(rsp, SimpleRuntimeFrame::return_off << LogBytesPerInt); // Prolog
3269 
3270   // Save callee-saved registers.  See x86_64.ad.
3271 
3272   // rbp is an implicitly saved callee saved register (i.e. the calling
3273   // convention will save restore it in prolog/epilog) Other than that
3274   // there are no callee save registers now that adapter frames are gone.
3275 
3276   __ movptr(Address(rsp, SimpleRuntimeFrame::rbp_off << LogBytesPerInt), rbp);
3277 
3278   // Store exception in Thread object. We cannot pass any arguments to the
3279   // handle_exception call, since we do not want to make any assumption
3280   // about the size of the frame where the exception happened in.
3281   // c_rarg0 is either rdi (Linux) or rcx (Windows).
3282   __ movptr(Address(r15_thread, JavaThread::exception_oop_offset()),rax);
3283   __ movptr(Address(r15_thread, JavaThread::exception_pc_offset()), rdx);
3284 
3285   // This call does all the hard work.  It checks if an exception handler
3286   // exists in the method.
3287   // If so, it returns the handler address.
3288   // If not, it prepares for stack-unwinding, restoring the callee-save
3289   // registers of the frame being removed.
3290   //
3291   // address OptoRuntime::handle_exception_C(JavaThread* thread)
3292 
3293   __ set_last_Java_frame(noreg, noreg, NULL);
3294   __ mov(c_rarg0, r15_thread);
3295   __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, OptoRuntime::handle_exception_C)));
3296 
3297   // Set an oopmap for the call site.  This oopmap will only be used if we
3298   // are unwinding the stack.  Hence, all locations will be dead.
3299   // Callee-saved registers will be the same as the frame above (i.e.,
3300   // handle_exception_stub), since they were restored when we got the
3301   // exception.
3302 
3303   OopMapSet* oop_maps = new OopMapSet();
3304 
3305   oop_maps->add_gc_map( __ pc()-start, new OopMap(SimpleRuntimeFrame::framesize, 0));
3306 
3307   __ reset_last_Java_frame(false, false);
3308 
3309   // Restore callee-saved registers
3310 
3311   // rbp is an implicitly saved callee saved register (i.e. the calling
3312   // convention will save restore it in prolog/epilog) Other than that
3313   // there are no callee save registers no that adapter frames are gone.
3314 
3315   __ movptr(rbp, Address(rsp, SimpleRuntimeFrame::rbp_off << LogBytesPerInt));
3316 
3317   __ addptr(rsp, SimpleRuntimeFrame::return_off << LogBytesPerInt); // Epilog
3318   __ pop(rdx);                  // No need for exception pc anymore
3319 
3320   // rax: exception handler
3321 
3322   // We have a handler in rax (could be deopt blob).
3323   __ mov(r8, rax);
3324 
3325   // Get the exception oop
3326   __ movptr(rax, Address(r15_thread, JavaThread::exception_oop_offset()));
3327   // Get the exception pc in case we are deoptimized
3328   __ movptr(rdx, Address(r15_thread, JavaThread::exception_pc_offset()));
3329 #ifdef ASSERT
3330   __ movptr(Address(r15_thread, JavaThread::exception_handler_pc_offset()), (int)NULL_WORD);
3331   __ movptr(Address(r15_thread, JavaThread::exception_pc_offset()), (int)NULL_WORD);
3332 #endif
3333   // Clear the exception oop so GC no longer processes it as a root.
3334   __ movptr(Address(r15_thread, JavaThread::exception_oop_offset()), (int)NULL_WORD);
3335 
3336   // rax: exception oop
3337   // r8:  exception handler
3338   // rdx: exception pc
3339   // Jump to handler
3340 
3341   __ jmp(r8);
3342 
3343   // Make sure all code is generated
3344   masm->flush();
3345 
3346   // Set exception blob
3347   _exception_blob =  ExceptionBlob::create(&buffer, oop_maps, SimpleRuntimeFrame::framesize >> 1);
3348 }
3349 #endif // COMPILER2