1 /*
   2  * Copyright (c) 1999, 2011, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 #include "precompiled.hpp"
  26 #include "c1/c1_Defs.hpp"
  27 #include "c1/c1_MacroAssembler.hpp"
  28 #include "c1/c1_Runtime1.hpp"
  29 #include "interpreter/interpreter.hpp"
  30 #include "nativeInst_x86.hpp"
  31 #include "oops/compiledICHolderOop.hpp"
  32 #include "oops/oop.inline.hpp"
  33 #include "prims/jvmtiExport.hpp"
  34 #include "register_x86.hpp"
  35 #include "runtime/sharedRuntime.hpp"
  36 #include "runtime/signature.hpp"
  37 #include "runtime/vframeArray.hpp"
  38 #include "vmreg_x86.inline.hpp"
  39 
  40 
  41 // Implementation of StubAssembler
  42 
  43 int StubAssembler::call_RT(Register oop_result1, Register oop_result2, address entry, int args_size) {
  44   // setup registers
  45   const Register thread = NOT_LP64(rdi) LP64_ONLY(r15_thread); // is callee-saved register (Visual C++ calling conventions)
  46   assert(!(oop_result1->is_valid() || oop_result2->is_valid()) || oop_result1 != oop_result2, "registers must be different");
  47   assert(oop_result1 != thread && oop_result2 != thread, "registers must be different");
  48   assert(args_size >= 0, "illegal args_size");
  49 
  50 #ifdef _LP64
  51   mov(c_rarg0, thread);
  52   set_num_rt_args(0); // Nothing on stack
  53 #else
  54   set_num_rt_args(1 + args_size);
  55 
  56   // push java thread (becomes first argument of C function)
  57   get_thread(thread);
  58   push(thread);
  59 #endif // _LP64
  60 
  61   set_last_Java_frame(thread, noreg, rbp, NULL);
  62 
  63   // do the call
  64   call(RuntimeAddress(entry));
  65   int call_offset = offset();
  66   // verify callee-saved register
  67 #ifdef ASSERT
  68   guarantee(thread != rax, "change this code");
  69   push(rax);
  70   { Label L;
  71     get_thread(rax);
  72     cmpptr(thread, rax);
  73     jcc(Assembler::equal, L);
  74     int3();
  75     stop("StubAssembler::call_RT: rdi not callee saved?");
  76     bind(L);
  77   }
  78   pop(rax);
  79 #endif
  80   reset_last_Java_frame(thread, true, false);
  81 
  82   // discard thread and arguments
  83   NOT_LP64(addptr(rsp, num_rt_args()*BytesPerWord));
  84 
  85   // check for pending exceptions
  86   { Label L;
  87     cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
  88     jcc(Assembler::equal, L);
  89     // exception pending => remove activation and forward to exception handler
  90     movptr(rax, Address(thread, Thread::pending_exception_offset()));
  91     // make sure that the vm_results are cleared
  92     if (oop_result1->is_valid()) {
  93       movptr(Address(thread, JavaThread::vm_result_offset()), NULL_WORD);
  94     }
  95     if (oop_result2->is_valid()) {
  96       movptr(Address(thread, JavaThread::vm_result_2_offset()), NULL_WORD);
  97     }
  98     if (frame_size() == no_frame_size) {
  99       leave();
 100       jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
 101     } else if (_stub_id == Runtime1::forward_exception_id) {
 102       should_not_reach_here();
 103     } else {
 104       jump(RuntimeAddress(Runtime1::entry_for(Runtime1::forward_exception_id)));
 105     }
 106     bind(L);
 107   }
 108   // get oop results if there are any and reset the values in the thread
 109   if (oop_result1->is_valid()) {
 110     movptr(oop_result1, Address(thread, JavaThread::vm_result_offset()));
 111     movptr(Address(thread, JavaThread::vm_result_offset()), NULL_WORD);
 112     verify_oop(oop_result1);
 113   }
 114   if (oop_result2->is_valid()) {
 115     movptr(oop_result2, Address(thread, JavaThread::vm_result_2_offset()));
 116     movptr(Address(thread, JavaThread::vm_result_2_offset()), NULL_WORD);
 117     verify_oop(oop_result2);
 118   }
 119   return call_offset;
 120 }
 121 
 122 
 123 int StubAssembler::call_RT(Register oop_result1, Register oop_result2, address entry, Register arg1) {
 124 #ifdef _LP64
 125   mov(c_rarg1, arg1);
 126 #else
 127   push(arg1);
 128 #endif // _LP64
 129   return call_RT(oop_result1, oop_result2, entry, 1);
 130 }
 131 
 132 
 133 int StubAssembler::call_RT(Register oop_result1, Register oop_result2, address entry, Register arg1, Register arg2) {
 134 #ifdef _LP64
 135   if (c_rarg1 == arg2) {
 136     if (c_rarg2 == arg1) {
 137       xchgq(arg1, arg2);
 138     } else {
 139       mov(c_rarg2, arg2);
 140       mov(c_rarg1, arg1);
 141     }
 142   } else {
 143     mov(c_rarg1, arg1);
 144     mov(c_rarg2, arg2);
 145   }
 146 #else
 147   push(arg2);
 148   push(arg1);
 149 #endif // _LP64
 150   return call_RT(oop_result1, oop_result2, entry, 2);
 151 }
 152 
 153 
 154 int StubAssembler::call_RT(Register oop_result1, Register oop_result2, address entry, Register arg1, Register arg2, Register arg3) {
 155 #ifdef _LP64
 156   // if there is any conflict use the stack
 157   if (arg1 == c_rarg2 || arg1 == c_rarg3 ||
 158       arg2 == c_rarg1 || arg1 == c_rarg3 ||
 159       arg3 == c_rarg1 || arg1 == c_rarg2) {
 160     push(arg3);
 161     push(arg2);
 162     push(arg1);
 163     pop(c_rarg1);
 164     pop(c_rarg2);
 165     pop(c_rarg3);
 166   } else {
 167     mov(c_rarg1, arg1);
 168     mov(c_rarg2, arg2);
 169     mov(c_rarg3, arg3);
 170   }
 171 #else
 172   push(arg3);
 173   push(arg2);
 174   push(arg1);
 175 #endif // _LP64
 176   return call_RT(oop_result1, oop_result2, entry, 3);
 177 }
 178 
 179 
 180 // Implementation of StubFrame
 181 
 182 class StubFrame: public StackObj {
 183  private:
 184   StubAssembler* _sasm;
 185 
 186  public:
 187   StubFrame(StubAssembler* sasm, const char* name, bool must_gc_arguments);
 188   void load_argument(int offset_in_words, Register reg);
 189 
 190   ~StubFrame();
 191 };
 192 
 193 
 194 #define __ _sasm->
 195 
 196 StubFrame::StubFrame(StubAssembler* sasm, const char* name, bool must_gc_arguments) {
 197   _sasm = sasm;
 198   __ set_info(name, must_gc_arguments);
 199   __ enter();
 200 }
 201 
 202 // load parameters that were stored with LIR_Assembler::store_parameter
 203 // Note: offsets for store_parameter and load_argument must match
 204 void StubFrame::load_argument(int offset_in_words, Register reg) {
 205   // rbp, + 0: link
 206   //     + 1: return address
 207   //     + 2: argument with offset 0
 208   //     + 3: argument with offset 1
 209   //     + 4: ...
 210 
 211   __ movptr(reg, Address(rbp, (offset_in_words + 2) * BytesPerWord));
 212 }
 213 
 214 
 215 StubFrame::~StubFrame() {
 216   __ leave();
 217   __ ret(0);
 218 }
 219 
 220 #undef __
 221 
 222 
 223 // Implementation of Runtime1
 224 
 225 #define __ sasm->
 226 
 227 const int float_regs_as_doubles_size_in_slots = pd_nof_fpu_regs_frame_map * 2;
 228 const int xmm_regs_as_doubles_size_in_slots = FrameMap::nof_xmm_regs * 2;
 229 
 230 // Stack layout for saving/restoring  all the registers needed during a runtime
 231 // call (this includes deoptimization)
 232 // Note: note that users of this frame may well have arguments to some runtime
 233 // while these values are on the stack. These positions neglect those arguments
 234 // but the code in save_live_registers will take the argument count into
 235 // account.
 236 //
 237 #ifdef _LP64
 238   #define SLOT2(x) x,
 239   #define SLOT_PER_WORD 2
 240 #else
 241   #define SLOT2(x)
 242   #define SLOT_PER_WORD 1
 243 #endif // _LP64
 244 
 245 enum reg_save_layout {
 246   // 64bit needs to keep stack 16 byte aligned. So we add some alignment dummies to make that
 247   // happen and will assert if the stack size we create is misaligned
 248 #ifdef _LP64
 249   align_dummy_0, align_dummy_1,
 250 #endif // _LP64
 251   dummy1, SLOT2(dummy1H)                                                                    // 0, 4
 252   dummy2, SLOT2(dummy2H)                                                                    // 8, 12
 253   // Two temps to be used as needed by users of save/restore callee registers
 254   temp_2_off, SLOT2(temp_2H_off)                                                            // 16, 20
 255   temp_1_off, SLOT2(temp_1H_off)                                                            // 24, 28
 256   xmm_regs_as_doubles_off,                                                                  // 32
 257   float_regs_as_doubles_off = xmm_regs_as_doubles_off + xmm_regs_as_doubles_size_in_slots,  // 160
 258   fpu_state_off = float_regs_as_doubles_off + float_regs_as_doubles_size_in_slots,          // 224
 259   // fpu_state_end_off is exclusive
 260   fpu_state_end_off = fpu_state_off + (FPUStateSizeInWords / SLOT_PER_WORD),                // 352
 261   marker = fpu_state_end_off, SLOT2(markerH)                                                // 352, 356
 262   extra_space_offset,                                                                       // 360
 263 #ifdef _LP64
 264   r15_off = extra_space_offset, r15H_off,                                                   // 360, 364
 265   r14_off, r14H_off,                                                                        // 368, 372
 266   r13_off, r13H_off,                                                                        // 376, 380
 267   r12_off, r12H_off,                                                                        // 384, 388
 268   r11_off, r11H_off,                                                                        // 392, 396
 269   r10_off, r10H_off,                                                                        // 400, 404
 270   r9_off, r9H_off,                                                                          // 408, 412
 271   r8_off, r8H_off,                                                                          // 416, 420
 272   rdi_off, rdiH_off,                                                                        // 424, 428
 273 #else
 274   rdi_off = extra_space_offset,
 275 #endif // _LP64
 276   rsi_off, SLOT2(rsiH_off)                                                                  // 432, 436
 277   rbp_off, SLOT2(rbpH_off)                                                                  // 440, 444
 278   rsp_off, SLOT2(rspH_off)                                                                  // 448, 452
 279   rbx_off, SLOT2(rbxH_off)                                                                  // 456, 460
 280   rdx_off, SLOT2(rdxH_off)                                                                  // 464, 468
 281   rcx_off, SLOT2(rcxH_off)                                                                  // 472, 476
 282   rax_off, SLOT2(raxH_off)                                                                  // 480, 484
 283   saved_rbp_off, SLOT2(saved_rbpH_off)                                                      // 488, 492
 284   return_off, SLOT2(returnH_off)                                                            // 496, 500
 285   reg_save_frame_size,  // As noted: neglects any parameters to runtime                     // 504
 286 
 287 #ifdef _WIN64
 288   c_rarg0_off = rcx_off,
 289 #else
 290   c_rarg0_off = rdi_off,
 291 #endif // WIN64
 292 
 293   // equates
 294 
 295   // illegal instruction handler
 296   continue_dest_off = temp_1_off,
 297 
 298   // deoptimization equates
 299   fp0_off = float_regs_as_doubles_off, // slot for java float/double return value
 300   xmm0_off = xmm_regs_as_doubles_off,  // slot for java float/double return value
 301   deopt_type = temp_2_off,             // slot for type of deopt in progress
 302   ret_type = temp_1_off                // slot for return type
 303 };
 304 
 305 
 306 
 307 // Save off registers which might be killed by calls into the runtime.
 308 // Tries to smart of about FP registers.  In particular we separate
 309 // saving and describing the FPU registers for deoptimization since we
 310 // have to save the FPU registers twice if we describe them and on P4
 311 // saving FPU registers which don't contain anything appears
 312 // expensive.  The deopt blob is the only thing which needs to
 313 // describe FPU registers.  In all other cases it should be sufficient
 314 // to simply save their current value.
 315 
 316 static OopMap* generate_oop_map(StubAssembler* sasm, int num_rt_args,
 317                                 bool save_fpu_registers = true) {
 318 
 319   // In 64bit all the args are in regs so there are no additional stack slots
 320   LP64_ONLY(num_rt_args = 0);
 321   LP64_ONLY(assert((reg_save_frame_size * VMRegImpl::stack_slot_size) % 16 == 0, "must be 16 byte aligned");)
 322   int frame_size_in_slots = reg_save_frame_size + num_rt_args; // args + thread
 323   sasm->set_frame_size(frame_size_in_slots / VMRegImpl::slots_per_word );
 324 
 325   // record saved value locations in an OopMap
 326   // locations are offsets from sp after runtime call; num_rt_args is number of arguments in call, including thread
 327   OopMap* map = new OopMap(frame_size_in_slots, 0);
 328   map->set_callee_saved(VMRegImpl::stack2reg(rax_off + num_rt_args), rax->as_VMReg());
 329   map->set_callee_saved(VMRegImpl::stack2reg(rcx_off + num_rt_args), rcx->as_VMReg());
 330   map->set_callee_saved(VMRegImpl::stack2reg(rdx_off + num_rt_args), rdx->as_VMReg());
 331   map->set_callee_saved(VMRegImpl::stack2reg(rbx_off + num_rt_args), rbx->as_VMReg());
 332   map->set_callee_saved(VMRegImpl::stack2reg(rsi_off + num_rt_args), rsi->as_VMReg());
 333   map->set_callee_saved(VMRegImpl::stack2reg(rdi_off + num_rt_args), rdi->as_VMReg());
 334 #ifdef _LP64
 335   map->set_callee_saved(VMRegImpl::stack2reg(r8_off + num_rt_args),  r8->as_VMReg());
 336   map->set_callee_saved(VMRegImpl::stack2reg(r9_off + num_rt_args),  r9->as_VMReg());
 337   map->set_callee_saved(VMRegImpl::stack2reg(r10_off + num_rt_args), r10->as_VMReg());
 338   map->set_callee_saved(VMRegImpl::stack2reg(r11_off + num_rt_args), r11->as_VMReg());
 339   map->set_callee_saved(VMRegImpl::stack2reg(r12_off + num_rt_args), r12->as_VMReg());
 340   map->set_callee_saved(VMRegImpl::stack2reg(r13_off + num_rt_args), r13->as_VMReg());
 341   map->set_callee_saved(VMRegImpl::stack2reg(r14_off + num_rt_args), r14->as_VMReg());
 342   map->set_callee_saved(VMRegImpl::stack2reg(r15_off + num_rt_args), r15->as_VMReg());
 343 
 344   // This is stupid but needed.
 345   map->set_callee_saved(VMRegImpl::stack2reg(raxH_off + num_rt_args), rax->as_VMReg()->next());
 346   map->set_callee_saved(VMRegImpl::stack2reg(rcxH_off + num_rt_args), rcx->as_VMReg()->next());
 347   map->set_callee_saved(VMRegImpl::stack2reg(rdxH_off + num_rt_args), rdx->as_VMReg()->next());
 348   map->set_callee_saved(VMRegImpl::stack2reg(rbxH_off + num_rt_args), rbx->as_VMReg()->next());
 349   map->set_callee_saved(VMRegImpl::stack2reg(rsiH_off + num_rt_args), rsi->as_VMReg()->next());
 350   map->set_callee_saved(VMRegImpl::stack2reg(rdiH_off + num_rt_args), rdi->as_VMReg()->next());
 351 
 352   map->set_callee_saved(VMRegImpl::stack2reg(r8H_off + num_rt_args),  r8->as_VMReg()->next());
 353   map->set_callee_saved(VMRegImpl::stack2reg(r9H_off + num_rt_args),  r9->as_VMReg()->next());
 354   map->set_callee_saved(VMRegImpl::stack2reg(r10H_off + num_rt_args), r10->as_VMReg()->next());
 355   map->set_callee_saved(VMRegImpl::stack2reg(r11H_off + num_rt_args), r11->as_VMReg()->next());
 356   map->set_callee_saved(VMRegImpl::stack2reg(r12H_off + num_rt_args), r12->as_VMReg()->next());
 357   map->set_callee_saved(VMRegImpl::stack2reg(r13H_off + num_rt_args), r13->as_VMReg()->next());
 358   map->set_callee_saved(VMRegImpl::stack2reg(r14H_off + num_rt_args), r14->as_VMReg()->next());
 359   map->set_callee_saved(VMRegImpl::stack2reg(r15H_off + num_rt_args), r15->as_VMReg()->next());
 360 #endif // _LP64
 361 
 362   if (save_fpu_registers) {
 363     if (UseSSE < 2) {
 364       int fpu_off = float_regs_as_doubles_off;
 365       for (int n = 0; n < FrameMap::nof_fpu_regs; n++) {
 366         VMReg fpu_name_0 = FrameMap::fpu_regname(n);
 367         map->set_callee_saved(VMRegImpl::stack2reg(fpu_off +     num_rt_args), fpu_name_0);
 368         // %%% This is really a waste but we'll keep things as they were for now
 369         if (true) {
 370           map->set_callee_saved(VMRegImpl::stack2reg(fpu_off + 1 + num_rt_args), fpu_name_0->next());
 371         }
 372         fpu_off += 2;
 373       }
 374       assert(fpu_off == fpu_state_off, "incorrect number of fpu stack slots");
 375     }
 376 
 377     if (UseSSE >= 2) {
 378       int xmm_off = xmm_regs_as_doubles_off;
 379       for (int n = 0; n < FrameMap::nof_xmm_regs; n++) {
 380         VMReg xmm_name_0 = as_XMMRegister(n)->as_VMReg();
 381         map->set_callee_saved(VMRegImpl::stack2reg(xmm_off +     num_rt_args), xmm_name_0);
 382         // %%% This is really a waste but we'll keep things as they were for now
 383         if (true) {
 384           map->set_callee_saved(VMRegImpl::stack2reg(xmm_off + 1 + num_rt_args), xmm_name_0->next());
 385         }
 386         xmm_off += 2;
 387       }
 388       assert(xmm_off == float_regs_as_doubles_off, "incorrect number of xmm registers");
 389 
 390     } else if (UseSSE == 1) {
 391       int xmm_off = xmm_regs_as_doubles_off;
 392       for (int n = 0; n < FrameMap::nof_xmm_regs; n++) {
 393         VMReg xmm_name_0 = as_XMMRegister(n)->as_VMReg();
 394         map->set_callee_saved(VMRegImpl::stack2reg(xmm_off +     num_rt_args), xmm_name_0);
 395         xmm_off += 2;
 396       }
 397       assert(xmm_off == float_regs_as_doubles_off, "incorrect number of xmm registers");
 398     }
 399   }
 400 
 401   return map;
 402 }
 403 
 404 static OopMap* save_live_registers(StubAssembler* sasm, int num_rt_args,
 405                                    bool save_fpu_registers = true) {
 406   __ block_comment("save_live_registers");
 407 
 408   // 64bit passes the args in regs to the c++ runtime
 409   int frame_size_in_slots = reg_save_frame_size NOT_LP64(+ num_rt_args); // args + thread
 410   // frame_size = round_to(frame_size, 4);
 411   sasm->set_frame_size(frame_size_in_slots / VMRegImpl::slots_per_word );
 412 
 413   __ pusha();         // integer registers
 414 
 415   // assert(float_regs_as_doubles_off % 2 == 0, "misaligned offset");
 416   // assert(xmm_regs_as_doubles_off % 2 == 0, "misaligned offset");
 417 
 418   __ subptr(rsp, extra_space_offset * VMRegImpl::stack_slot_size);
 419 
 420 #ifdef ASSERT
 421   __ movptr(Address(rsp, marker * VMRegImpl::stack_slot_size), (int32_t)0xfeedbeef);
 422 #endif
 423 
 424   if (save_fpu_registers) {
 425     if (UseSSE < 2) {
 426       // save FPU stack
 427       __ fnsave(Address(rsp, fpu_state_off * VMRegImpl::stack_slot_size));
 428       __ fwait();
 429 
 430 #ifdef ASSERT
 431       Label ok;
 432       __ cmpw(Address(rsp, fpu_state_off * VMRegImpl::stack_slot_size), StubRoutines::fpu_cntrl_wrd_std());
 433       __ jccb(Assembler::equal, ok);
 434       __ stop("corrupted control word detected");
 435       __ bind(ok);
 436 #endif
 437 
 438       // Reset the control word to guard against exceptions being unmasked
 439       // since fstp_d can cause FPU stack underflow exceptions.  Write it
 440       // into the on stack copy and then reload that to make sure that the
 441       // current and future values are correct.
 442       __ movw(Address(rsp, fpu_state_off * VMRegImpl::stack_slot_size), StubRoutines::fpu_cntrl_wrd_std());
 443       __ frstor(Address(rsp, fpu_state_off * VMRegImpl::stack_slot_size));
 444 
 445       // Save the FPU registers in de-opt-able form
 446       __ fstp_d(Address(rsp, float_regs_as_doubles_off * VMRegImpl::stack_slot_size +  0));
 447       __ fstp_d(Address(rsp, float_regs_as_doubles_off * VMRegImpl::stack_slot_size +  8));
 448       __ fstp_d(Address(rsp, float_regs_as_doubles_off * VMRegImpl::stack_slot_size + 16));
 449       __ fstp_d(Address(rsp, float_regs_as_doubles_off * VMRegImpl::stack_slot_size + 24));
 450       __ fstp_d(Address(rsp, float_regs_as_doubles_off * VMRegImpl::stack_slot_size + 32));
 451       __ fstp_d(Address(rsp, float_regs_as_doubles_off * VMRegImpl::stack_slot_size + 40));
 452       __ fstp_d(Address(rsp, float_regs_as_doubles_off * VMRegImpl::stack_slot_size + 48));
 453       __ fstp_d(Address(rsp, float_regs_as_doubles_off * VMRegImpl::stack_slot_size + 56));
 454     }
 455 
 456     if (UseSSE >= 2) {
 457       // save XMM registers
 458       // XMM registers can contain float or double values, but this is not known here,
 459       // so always save them as doubles.
 460       // note that float values are _not_ converted automatically, so for float values
 461       // the second word contains only garbage data.
 462       __ movdbl(Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size +  0), xmm0);
 463       __ movdbl(Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size +  8), xmm1);
 464       __ movdbl(Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 16), xmm2);
 465       __ movdbl(Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 24), xmm3);
 466       __ movdbl(Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 32), xmm4);
 467       __ movdbl(Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 40), xmm5);
 468       __ movdbl(Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 48), xmm6);
 469       __ movdbl(Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 56), xmm7);
 470 #ifdef _LP64
 471       __ movdbl(Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 64), xmm8);
 472       __ movdbl(Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 72), xmm9);
 473       __ movdbl(Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 80), xmm10);
 474       __ movdbl(Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 88), xmm11);
 475       __ movdbl(Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 96), xmm12);
 476       __ movdbl(Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 104), xmm13);
 477       __ movdbl(Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 112), xmm14);
 478       __ movdbl(Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 120), xmm15);
 479 #endif // _LP64
 480     } else if (UseSSE == 1) {
 481       // save XMM registers as float because double not supported without SSE2
 482       __ movflt(Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size +  0), xmm0);
 483       __ movflt(Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size +  8), xmm1);
 484       __ movflt(Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 16), xmm2);
 485       __ movflt(Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 24), xmm3);
 486       __ movflt(Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 32), xmm4);
 487       __ movflt(Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 40), xmm5);
 488       __ movflt(Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 48), xmm6);
 489       __ movflt(Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 56), xmm7);
 490     }
 491   }
 492 
 493   // FPU stack must be empty now
 494   __ verify_FPU(0, "save_live_registers");
 495 
 496   return generate_oop_map(sasm, num_rt_args, save_fpu_registers);
 497 }
 498 
 499 
 500 static void restore_fpu(StubAssembler* sasm, bool restore_fpu_registers = true) {
 501   if (restore_fpu_registers) {
 502     if (UseSSE >= 2) {
 503       // restore XMM registers
 504       __ movdbl(xmm0, Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size +  0));
 505       __ movdbl(xmm1, Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size +  8));
 506       __ movdbl(xmm2, Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 16));
 507       __ movdbl(xmm3, Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 24));
 508       __ movdbl(xmm4, Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 32));
 509       __ movdbl(xmm5, Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 40));
 510       __ movdbl(xmm6, Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 48));
 511       __ movdbl(xmm7, Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 56));
 512 #ifdef _LP64
 513       __ movdbl(xmm8, Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 64));
 514       __ movdbl(xmm9, Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 72));
 515       __ movdbl(xmm10, Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 80));
 516       __ movdbl(xmm11, Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 88));
 517       __ movdbl(xmm12, Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 96));
 518       __ movdbl(xmm13, Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 104));
 519       __ movdbl(xmm14, Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 112));
 520       __ movdbl(xmm15, Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 120));
 521 #endif // _LP64
 522     } else if (UseSSE == 1) {
 523       // restore XMM registers
 524       __ movflt(xmm0, Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size +  0));
 525       __ movflt(xmm1, Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size +  8));
 526       __ movflt(xmm2, Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 16));
 527       __ movflt(xmm3, Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 24));
 528       __ movflt(xmm4, Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 32));
 529       __ movflt(xmm5, Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 40));
 530       __ movflt(xmm6, Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 48));
 531       __ movflt(xmm7, Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 56));
 532     }
 533 
 534     if (UseSSE < 2) {
 535       __ frstor(Address(rsp, fpu_state_off * VMRegImpl::stack_slot_size));
 536     } else {
 537       // check that FPU stack is really empty
 538       __ verify_FPU(0, "restore_live_registers");
 539     }
 540 
 541   } else {
 542     // check that FPU stack is really empty
 543     __ verify_FPU(0, "restore_live_registers");
 544   }
 545 
 546 #ifdef ASSERT
 547   {
 548     Label ok;
 549     __ cmpptr(Address(rsp, marker * VMRegImpl::stack_slot_size), (int32_t)0xfeedbeef);
 550     __ jcc(Assembler::equal, ok);
 551     __ stop("bad offsets in frame");
 552     __ bind(ok);
 553   }
 554 #endif // ASSERT
 555 
 556   __ addptr(rsp, extra_space_offset * VMRegImpl::stack_slot_size);
 557 }
 558 
 559 
 560 static void restore_live_registers(StubAssembler* sasm, bool restore_fpu_registers = true) {
 561   __ block_comment("restore_live_registers");
 562 
 563   restore_fpu(sasm, restore_fpu_registers);
 564   __ popa();
 565 }
 566 
 567 
 568 static void restore_live_registers_except_rax(StubAssembler* sasm, bool restore_fpu_registers = true) {
 569   __ block_comment("restore_live_registers_except_rax");
 570 
 571   restore_fpu(sasm, restore_fpu_registers);
 572 
 573 #ifdef _LP64
 574   __ movptr(r15, Address(rsp, 0));
 575   __ movptr(r14, Address(rsp, wordSize));
 576   __ movptr(r13, Address(rsp, 2 * wordSize));
 577   __ movptr(r12, Address(rsp, 3 * wordSize));
 578   __ movptr(r11, Address(rsp, 4 * wordSize));
 579   __ movptr(r10, Address(rsp, 5 * wordSize));
 580   __ movptr(r9,  Address(rsp, 6 * wordSize));
 581   __ movptr(r8,  Address(rsp, 7 * wordSize));
 582   __ movptr(rdi, Address(rsp, 8 * wordSize));
 583   __ movptr(rsi, Address(rsp, 9 * wordSize));
 584   __ movptr(rbp, Address(rsp, 10 * wordSize));
 585   // skip rsp
 586   __ movptr(rbx, Address(rsp, 12 * wordSize));
 587   __ movptr(rdx, Address(rsp, 13 * wordSize));
 588   __ movptr(rcx, Address(rsp, 14 * wordSize));
 589 
 590   __ addptr(rsp, 16 * wordSize);
 591 #else
 592 
 593   __ pop(rdi);
 594   __ pop(rsi);
 595   __ pop(rbp);
 596   __ pop(rbx); // skip this value
 597   __ pop(rbx);
 598   __ pop(rdx);
 599   __ pop(rcx);
 600   __ addptr(rsp, BytesPerWord);
 601 #endif // _LP64
 602 }
 603 
 604 
 605 void Runtime1::initialize_pd() {
 606   // nothing to do
 607 }
 608 
 609 
 610 // target: the entry point of the method that creates and posts the exception oop
 611 // has_argument: true if the exception needs an argument (passed on stack because registers must be preserved)
 612 
 613 OopMapSet* Runtime1::generate_exception_throw(StubAssembler* sasm, address target, bool has_argument) {
 614   // preserve all registers
 615   int num_rt_args = has_argument ? 2 : 1;
 616   OopMap* oop_map = save_live_registers(sasm, num_rt_args);
 617 
 618   // now all registers are saved and can be used freely
 619   // verify that no old value is used accidentally
 620   __ invalidate_registers(true, true, true, true, true, true);
 621 
 622   // registers used by this stub
 623   const Register temp_reg = rbx;
 624 
 625   // load argument for exception that is passed as an argument into the stub
 626   if (has_argument) {
 627 #ifdef _LP64
 628     __ movptr(c_rarg1, Address(rbp, 2*BytesPerWord));
 629 #else
 630     __ movptr(temp_reg, Address(rbp, 2*BytesPerWord));
 631     __ push(temp_reg);
 632 #endif // _LP64
 633   }
 634   int call_offset = __ call_RT(noreg, noreg, target, num_rt_args - 1);
 635 
 636   OopMapSet* oop_maps = new OopMapSet();
 637   oop_maps->add_gc_map(call_offset, oop_map);
 638 
 639   __ stop("should not reach here");
 640 
 641   return oop_maps;
 642 }
 643 
 644 
 645 void Runtime1::generate_handle_exception(StubAssembler *sasm, OopMapSet* oop_maps, OopMap* oop_map, bool save_fpu_registers) {
 646   // incoming parameters
 647   const Register exception_oop = rax;
 648   const Register exception_pc = rdx;
 649   // other registers used in this stub
 650   const Register real_return_addr = rbx;
 651   const Register thread = NOT_LP64(rdi) LP64_ONLY(r15_thread);
 652 
 653   __ block_comment("generate_handle_exception");
 654 
 655 #ifdef TIERED
 656   // C2 can leave the fpu stack dirty
 657   if (UseSSE < 2 ) {
 658     __ empty_FPU_stack();
 659   }
 660 #endif // TIERED
 661 
 662   // verify that only rax, and rdx is valid at this time
 663   __ invalidate_registers(false, true, true, false, true, true);
 664   // verify that rax, contains a valid exception
 665   __ verify_not_null_oop(exception_oop);
 666 
 667   // load address of JavaThread object for thread-local data
 668   NOT_LP64(__ get_thread(thread);)
 669 
 670 #ifdef ASSERT
 671   // check that fields in JavaThread for exception oop and issuing pc are
 672   // empty before writing to them
 673   Label oop_empty;
 674   __ cmpptr(Address(thread, JavaThread::exception_oop_offset()), (int32_t) NULL_WORD);
 675   __ jcc(Assembler::equal, oop_empty);
 676   __ stop("exception oop already set");
 677   __ bind(oop_empty);
 678 
 679   Label pc_empty;
 680   __ cmpptr(Address(thread, JavaThread::exception_pc_offset()), 0);
 681   __ jcc(Assembler::equal, pc_empty);
 682   __ stop("exception pc already set");
 683   __ bind(pc_empty);
 684 #endif
 685 
 686   // save exception oop and issuing pc into JavaThread
 687   // (exception handler will load it from here)
 688   __ movptr(Address(thread, JavaThread::exception_oop_offset()), exception_oop);
 689   __ movptr(Address(thread, JavaThread::exception_pc_offset()), exception_pc);
 690 
 691   // save real return address (pc that called this stub)
 692   __ movptr(real_return_addr, Address(rbp, 1*BytesPerWord));
 693   __ movptr(Address(rsp, temp_1_off * VMRegImpl::stack_slot_size), real_return_addr);
 694 
 695   // patch throwing pc into return address (has bci & oop map)
 696   __ movptr(Address(rbp, 1*BytesPerWord), exception_pc);
 697 
 698   // compute the exception handler.
 699   // the exception oop and the throwing pc are read from the fields in JavaThread
 700   int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, exception_handler_for_pc));
 701   oop_maps->add_gc_map(call_offset, oop_map);
 702 
 703   // rax,: handler address
 704   //      will be the deopt blob if nmethod was deoptimized while we looked up
 705   //      handler regardless of whether handler existed in the nmethod.
 706 
 707   // only rax, is valid at this time, all other registers have been destroyed by the runtime call
 708   __ invalidate_registers(false, true, true, true, true, true);
 709 
 710 #ifdef ASSERT
 711   // Do we have an exception handler in the nmethod?
 712   Label done;
 713   __ testptr(rax, rax);
 714   __ jcc(Assembler::notZero, done);
 715   __ stop("no handler found");
 716   __ bind(done);
 717 #endif
 718 
 719   // exception handler found
 720   // patch the return address -> the stub will directly return to the exception handler
 721   __ movptr(Address(rbp, 1*BytesPerWord), rax);
 722 
 723   // restore registers
 724   restore_live_registers(sasm, save_fpu_registers);
 725 
 726   // return to exception handler
 727   __ leave();
 728   __ ret(0);
 729 
 730 }
 731 
 732 
 733 void Runtime1::generate_unwind_exception(StubAssembler *sasm) {
 734   // incoming parameters
 735   const Register exception_oop = rax;
 736   // callee-saved copy of exception_oop during runtime call
 737   const Register exception_oop_callee_saved = NOT_LP64(rsi) LP64_ONLY(r14);
 738   // other registers used in this stub
 739   const Register exception_pc = rdx;
 740   const Register handler_addr = rbx;
 741   const Register thread = NOT_LP64(rdi) LP64_ONLY(r15_thread);
 742 
 743   // verify that only rax, is valid at this time
 744   __ invalidate_registers(false, true, true, true, true, true);
 745 
 746 #ifdef ASSERT
 747   // check that fields in JavaThread for exception oop and issuing pc are empty
 748   NOT_LP64(__ get_thread(thread);)
 749   Label oop_empty;
 750   __ cmpptr(Address(thread, JavaThread::exception_oop_offset()), 0);
 751   __ jcc(Assembler::equal, oop_empty);
 752   __ stop("exception oop must be empty");
 753   __ bind(oop_empty);
 754 
 755   Label pc_empty;
 756   __ cmpptr(Address(thread, JavaThread::exception_pc_offset()), 0);
 757   __ jcc(Assembler::equal, pc_empty);
 758   __ stop("exception pc must be empty");
 759   __ bind(pc_empty);
 760 #endif
 761 
 762   // clear the FPU stack in case any FPU results are left behind
 763   __ empty_FPU_stack();
 764 
 765   // save exception_oop in callee-saved register to preserve it during runtime calls
 766   __ verify_not_null_oop(exception_oop);
 767   __ movptr(exception_oop_callee_saved, exception_oop);
 768 
 769   NOT_LP64(__ get_thread(thread);)
 770   // Get return address (is on top of stack after leave).
 771   __ movptr(exception_pc, Address(rsp, 0));
 772 
 773   // search the exception handler address of the caller (using the return address)
 774   __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), thread, exception_pc);
 775   // rax: exception handler address of the caller
 776 
 777   // Only RAX and RSI are valid at this time, all other registers have been destroyed by the call.
 778   __ invalidate_registers(false, true, true, true, false, true);
 779 
 780   // move result of call into correct register
 781   __ movptr(handler_addr, rax);
 782 
 783   // Restore exception oop to RAX (required convention of exception handler).
 784   __ movptr(exception_oop, exception_oop_callee_saved);
 785 
 786   // verify that there is really a valid exception in rax
 787   __ verify_not_null_oop(exception_oop);
 788 
 789   // get throwing pc (= return address).
 790   // rdx has been destroyed by the call, so it must be set again
 791   // the pop is also necessary to simulate the effect of a ret(0)
 792   __ pop(exception_pc);
 793 
 794   // Restore SP from BP if the exception PC is a MethodHandle call site.
 795   NOT_LP64(__ get_thread(thread);)
 796   __ cmpl(Address(thread, JavaThread::is_method_handle_return_offset()), 0);
 797   __ cmovptr(Assembler::notEqual, rsp, rbp_mh_SP_save);
 798 
 799   // continue at exception handler (return address removed)
 800   // note: do *not* remove arguments when unwinding the
 801   //       activation since the caller assumes having
 802   //       all arguments on the stack when entering the
 803   //       runtime to determine the exception handler
 804   //       (GC happens at call site with arguments!)
 805   // rax: exception oop
 806   // rdx: throwing pc
 807   // rbx: exception handler
 808   __ jmp(handler_addr);
 809 }
 810 
 811 
 812 OopMapSet* Runtime1::generate_patching(StubAssembler* sasm, address target) {
 813   // use the maximum number of runtime-arguments here because it is difficult to
 814   // distinguish each RT-Call.
 815   // Note: This number affects also the RT-Call in generate_handle_exception because
 816   //       the oop-map is shared for all calls.
 817   const int num_rt_args = 2;  // thread + dummy
 818 
 819   DeoptimizationBlob* deopt_blob = SharedRuntime::deopt_blob();
 820   assert(deopt_blob != NULL, "deoptimization blob must have been created");
 821 
 822   OopMap* oop_map = save_live_registers(sasm, num_rt_args);
 823 
 824 #ifdef _LP64
 825   const Register thread = r15_thread;
 826   // No need to worry about dummy
 827   __ mov(c_rarg0, thread);
 828 #else
 829   __ push(rax); // push dummy
 830 
 831   const Register thread = rdi; // is callee-saved register (Visual C++ calling conventions)
 832   // push java thread (becomes first argument of C function)
 833   __ get_thread(thread);
 834   __ push(thread);
 835 #endif // _LP64
 836   __ set_last_Java_frame(thread, noreg, rbp, NULL);
 837   // do the call
 838   __ call(RuntimeAddress(target));
 839   OopMapSet* oop_maps = new OopMapSet();
 840   oop_maps->add_gc_map(__ offset(), oop_map);
 841   // verify callee-saved register
 842 #ifdef ASSERT
 843   guarantee(thread != rax, "change this code");
 844   __ push(rax);
 845   { Label L;
 846     __ get_thread(rax);
 847     __ cmpptr(thread, rax);
 848     __ jcc(Assembler::equal, L);
 849     __ stop("StubAssembler::call_RT: rdi/r15 not callee saved?");
 850     __ bind(L);
 851   }
 852   __ pop(rax);
 853 #endif
 854   __ reset_last_Java_frame(thread, true, false);
 855 #ifndef _LP64
 856   __ pop(rcx); // discard thread arg
 857   __ pop(rcx); // discard dummy
 858 #endif // _LP64
 859 
 860   // check for pending exceptions
 861   { Label L;
 862     __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
 863     __ jcc(Assembler::equal, L);
 864     // exception pending => remove activation and forward to exception handler
 865 
 866     __ testptr(rax, rax);                                   // have we deoptimized?
 867     __ jump_cc(Assembler::equal,
 868                RuntimeAddress(Runtime1::entry_for(Runtime1::forward_exception_id)));
 869 
 870     // the deopt blob expects exceptions in the special fields of
 871     // JavaThread, so copy and clear pending exception.
 872 
 873     // load and clear pending exception
 874     __ movptr(rax, Address(thread, Thread::pending_exception_offset()));
 875     __ movptr(Address(thread, Thread::pending_exception_offset()), NULL_WORD);
 876 
 877     // check that there is really a valid exception
 878     __ verify_not_null_oop(rax);
 879 
 880     // load throwing pc: this is the return address of the stub
 881     __ movptr(rdx, Address(rsp, return_off * VMRegImpl::stack_slot_size));
 882 
 883 #ifdef ASSERT
 884     // check that fields in JavaThread for exception oop and issuing pc are empty
 885     Label oop_empty;
 886     __ cmpptr(Address(thread, JavaThread::exception_oop_offset()), (int32_t)NULL_WORD);
 887     __ jcc(Assembler::equal, oop_empty);
 888     __ stop("exception oop must be empty");
 889     __ bind(oop_empty);
 890 
 891     Label pc_empty;
 892     __ cmpptr(Address(thread, JavaThread::exception_pc_offset()), (int32_t)NULL_WORD);
 893     __ jcc(Assembler::equal, pc_empty);
 894     __ stop("exception pc must be empty");
 895     __ bind(pc_empty);
 896 #endif
 897 
 898     // store exception oop and throwing pc to JavaThread
 899     __ movptr(Address(thread, JavaThread::exception_oop_offset()), rax);
 900     __ movptr(Address(thread, JavaThread::exception_pc_offset()), rdx);
 901 
 902     restore_live_registers(sasm);
 903 
 904     __ leave();
 905     __ addptr(rsp, BytesPerWord);  // remove return address from stack
 906 
 907     // Forward the exception directly to deopt blob. We can blow no
 908     // registers and must leave throwing pc on the stack.  A patch may
 909     // have values live in registers so the entry point with the
 910     // exception in tls.
 911     __ jump(RuntimeAddress(deopt_blob->unpack_with_exception_in_tls()));
 912 
 913     __ bind(L);
 914   }
 915 
 916 
 917   // Runtime will return true if the nmethod has been deoptimized during
 918   // the patching process. In that case we must do a deopt reexecute instead.
 919 
 920   Label reexecuteEntry, cont;
 921 
 922   __ testptr(rax, rax);                                 // have we deoptimized?
 923   __ jcc(Assembler::equal, cont);                       // no
 924 
 925   // Will reexecute. Proper return address is already on the stack we just restore
 926   // registers, pop all of our frame but the return address and jump to the deopt blob
 927   restore_live_registers(sasm);
 928   __ leave();
 929   __ jump(RuntimeAddress(deopt_blob->unpack_with_reexecution()));
 930 
 931   __ bind(cont);
 932   restore_live_registers(sasm);
 933   __ leave();
 934   __ ret(0);
 935 
 936   return oop_maps;
 937 
 938 }
 939 
 940 
 941 OopMapSet* Runtime1::generate_code_for(StubID id, StubAssembler* sasm) {
 942 
 943   // for better readability
 944   const bool must_gc_arguments = true;
 945   const bool dont_gc_arguments = false;
 946 
 947   // default value; overwritten for some optimized stubs that are called from methods that do not use the fpu
 948   bool save_fpu_registers = true;
 949 
 950   // stub code & info for the different stubs
 951   OopMapSet* oop_maps = NULL;
 952   switch (id) {
 953     case forward_exception_id:
 954       {
 955         // we're handling an exception in the context of a compiled
 956         // frame.  The registers have been saved in the standard
 957         // places.  Perform an exception lookup in the caller and
 958         // dispatch to the handler if found.  Otherwise unwind and
 959         // dispatch to the callers exception handler.
 960 
 961         const Register thread = NOT_LP64(rdi) LP64_ONLY(r15_thread);
 962         const Register exception_oop = rax;
 963         const Register exception_pc = rdx;
 964 
 965         // load pending exception oop into rax,
 966         __ movptr(exception_oop, Address(thread, Thread::pending_exception_offset()));
 967         // clear pending exception
 968         __ movptr(Address(thread, Thread::pending_exception_offset()), NULL_WORD);
 969 
 970         // load issuing PC (the return address for this stub) into rdx
 971         __ movptr(exception_pc, Address(rbp, 1*BytesPerWord));
 972 
 973         // make sure that the vm_results are cleared (may be unnecessary)
 974         __ movptr(Address(thread, JavaThread::vm_result_offset()), NULL_WORD);
 975         __ movptr(Address(thread, JavaThread::vm_result_2_offset()), NULL_WORD);
 976 
 977         // verify that that there is really a valid exception in rax,
 978         __ verify_not_null_oop(exception_oop);
 979 
 980         oop_maps = new OopMapSet();
 981         OopMap* oop_map = generate_oop_map(sasm, 1);
 982         generate_handle_exception(sasm, oop_maps, oop_map);
 983         __ stop("should not reach here");
 984       }
 985       break;
 986 
 987     case new_instance_id:
 988     case fast_new_instance_id:
 989     case fast_new_instance_init_check_id:
 990       {
 991         Register klass = rdx; // Incoming
 992         Register obj   = rax; // Result
 993 
 994         if (id == new_instance_id) {
 995           __ set_info("new_instance", dont_gc_arguments);
 996         } else if (id == fast_new_instance_id) {
 997           __ set_info("fast new_instance", dont_gc_arguments);
 998         } else {
 999           assert(id == fast_new_instance_init_check_id, "bad StubID");
1000           __ set_info("fast new_instance init check", dont_gc_arguments);
1001         }
1002 
1003         if ((id == fast_new_instance_id || id == fast_new_instance_init_check_id) &&
1004             UseTLAB && FastTLABRefill) {
1005           Label slow_path;
1006           Register obj_size = rcx;
1007           Register t1       = rbx;
1008           Register t2       = rsi;
1009           assert_different_registers(klass, obj, obj_size, t1, t2);
1010 
1011           __ push(rdi);
1012           __ push(rbx);
1013 
1014           if (id == fast_new_instance_init_check_id) {
1015             // make sure the klass is initialized
1016             __ cmpl(Address(klass, instanceKlass::init_state_offset_in_bytes() + sizeof(oopDesc)), instanceKlass::fully_initialized);
1017             __ jcc(Assembler::notEqual, slow_path);
1018           }
1019 
1020 #ifdef ASSERT
1021           // assert object can be fast path allocated
1022           {
1023             Label ok, not_ok;
1024             __ movl(obj_size, Address(klass, Klass::layout_helper_offset_in_bytes() + sizeof(oopDesc)));
1025             __ cmpl(obj_size, 0);  // make sure it's an instance (LH > 0)
1026             __ jcc(Assembler::lessEqual, not_ok);
1027             __ testl(obj_size, Klass::_lh_instance_slow_path_bit);
1028             __ jcc(Assembler::zero, ok);
1029             __ bind(not_ok);
1030             __ stop("assert(can be fast path allocated)");
1031             __ should_not_reach_here();
1032             __ bind(ok);
1033           }
1034 #endif // ASSERT
1035 
1036           // if we got here then the TLAB allocation failed, so try
1037           // refilling the TLAB or allocating directly from eden.
1038           Label retry_tlab, try_eden;
1039           const Register thread =
1040             __ tlab_refill(retry_tlab, try_eden, slow_path); // does not destroy rdx (klass), returns rdi
1041 
1042           __ bind(retry_tlab);
1043 
1044           // get the instance size (size is postive so movl is fine for 64bit)
1045           __ movl(obj_size, Address(klass, klassOopDesc::header_size() * HeapWordSize + Klass::layout_helper_offset_in_bytes()));
1046 
1047           __ tlab_allocate(obj, obj_size, 0, t1, t2, slow_path);
1048 
1049           __ initialize_object(obj, klass, obj_size, 0, t1, t2);
1050           __ verify_oop(obj);
1051           __ pop(rbx);
1052           __ pop(rdi);
1053           __ ret(0);
1054 
1055           __ bind(try_eden);
1056           // get the instance size (size is postive so movl is fine for 64bit)
1057           __ movl(obj_size, Address(klass, klassOopDesc::header_size() * HeapWordSize + Klass::layout_helper_offset_in_bytes()));
1058 
1059           __ eden_allocate(obj, obj_size, 0, t1, slow_path);
1060           __ incr_allocated_bytes(thread, obj_size, 0);
1061 
1062           __ initialize_object(obj, klass, obj_size, 0, t1, t2);
1063           __ verify_oop(obj);
1064           __ pop(rbx);
1065           __ pop(rdi);
1066           __ ret(0);
1067 
1068           __ bind(slow_path);
1069           __ pop(rbx);
1070           __ pop(rdi);
1071         }
1072 
1073         __ enter();
1074         OopMap* map = save_live_registers(sasm, 2);
1075         int call_offset = __ call_RT(obj, noreg, CAST_FROM_FN_PTR(address, new_instance), klass);
1076         oop_maps = new OopMapSet();
1077         oop_maps->add_gc_map(call_offset, map);
1078         restore_live_registers_except_rax(sasm);
1079         __ verify_oop(obj);
1080         __ leave();
1081         __ ret(0);
1082 
1083         // rax,: new instance
1084       }
1085 
1086       break;
1087 
1088     case counter_overflow_id:
1089       {
1090         Register bci = rax, method = rbx;
1091         __ enter();
1092         OopMap* map = save_live_registers(sasm, 3);
1093         // Retrieve bci
1094         __ movl(bci, Address(rbp, 2*BytesPerWord));
1095         // And a pointer to the methodOop
1096         __ movptr(method, Address(rbp, 3*BytesPerWord));
1097         int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, counter_overflow), bci, method);
1098         oop_maps = new OopMapSet();
1099         oop_maps->add_gc_map(call_offset, map);
1100         restore_live_registers(sasm);
1101         __ leave();
1102         __ ret(0);
1103       }
1104       break;
1105 
1106     case new_type_array_id:
1107     case new_object_array_id:
1108       {
1109         Register length   = rbx; // Incoming
1110         Register klass    = rdx; // Incoming
1111         Register obj      = rax; // Result
1112 
1113         if (id == new_type_array_id) {
1114           __ set_info("new_type_array", dont_gc_arguments);
1115         } else {
1116           __ set_info("new_object_array", dont_gc_arguments);
1117         }
1118 
1119 #ifdef ASSERT
1120         // assert object type is really an array of the proper kind
1121         {
1122           Label ok;
1123           Register t0 = obj;
1124           __ movl(t0, Address(klass, Klass::layout_helper_offset_in_bytes() + sizeof(oopDesc)));
1125           __ sarl(t0, Klass::_lh_array_tag_shift);
1126           int tag = ((id == new_type_array_id)
1127                      ? Klass::_lh_array_tag_type_value
1128                      : Klass::_lh_array_tag_obj_value);
1129           __ cmpl(t0, tag);
1130           __ jcc(Assembler::equal, ok);
1131           __ stop("assert(is an array klass)");
1132           __ should_not_reach_here();
1133           __ bind(ok);
1134         }
1135 #endif // ASSERT
1136 
1137         if (UseTLAB && FastTLABRefill) {
1138           Register arr_size = rsi;
1139           Register t1       = rcx;  // must be rcx for use as shift count
1140           Register t2       = rdi;
1141           Label slow_path;
1142           assert_different_registers(length, klass, obj, arr_size, t1, t2);
1143 
1144           // check that array length is small enough for fast path.
1145           __ cmpl(length, C1_MacroAssembler::max_array_allocation_length);
1146           __ jcc(Assembler::above, slow_path);
1147 
1148           // if we got here then the TLAB allocation failed, so try
1149           // refilling the TLAB or allocating directly from eden.
1150           Label retry_tlab, try_eden;
1151           const Register thread =
1152             __ tlab_refill(retry_tlab, try_eden, slow_path); // preserves rbx & rdx, returns rdi
1153 
1154           __ bind(retry_tlab);
1155 
1156           // get the allocation size: round_up(hdr + length << (layout_helper & 0x1F))
1157           // since size is positive movl does right thing on 64bit
1158           __ movl(t1, Address(klass, klassOopDesc::header_size() * HeapWordSize + Klass::layout_helper_offset_in_bytes()));
1159           // since size is postive movl does right thing on 64bit
1160           __ movl(arr_size, length);
1161           assert(t1 == rcx, "fixed register usage");
1162           __ shlptr(arr_size /* by t1=rcx, mod 32 */);
1163           __ shrptr(t1, Klass::_lh_header_size_shift);
1164           __ andptr(t1, Klass::_lh_header_size_mask);
1165           __ addptr(arr_size, t1);
1166           __ addptr(arr_size, MinObjAlignmentInBytesMask); // align up
1167           __ andptr(arr_size, ~MinObjAlignmentInBytesMask);
1168 
1169           __ tlab_allocate(obj, arr_size, 0, t1, t2, slow_path);  // preserves arr_size
1170 
1171           __ initialize_header(obj, klass, length, t1, t2);
1172           __ movb(t1, Address(klass, klassOopDesc::header_size() * HeapWordSize + Klass::layout_helper_offset_in_bytes() + (Klass::_lh_header_size_shift / BitsPerByte)));
1173           assert(Klass::_lh_header_size_shift % BitsPerByte == 0, "bytewise");
1174           assert(Klass::_lh_header_size_mask <= 0xFF, "bytewise");
1175           __ andptr(t1, Klass::_lh_header_size_mask);
1176           __ subptr(arr_size, t1);  // body length
1177           __ addptr(t1, obj);       // body start
1178           __ initialize_body(t1, arr_size, 0, t2);
1179           __ verify_oop(obj);
1180           __ ret(0);
1181 
1182           __ bind(try_eden);
1183           // get the allocation size: round_up(hdr + length << (layout_helper & 0x1F))
1184           // since size is positive movl does right thing on 64bit
1185           __ movl(t1, Address(klass, klassOopDesc::header_size() * HeapWordSize + Klass::layout_helper_offset_in_bytes()));
1186           // since size is postive movl does right thing on 64bit
1187           __ movl(arr_size, length);
1188           assert(t1 == rcx, "fixed register usage");
1189           __ shlptr(arr_size /* by t1=rcx, mod 32 */);
1190           __ shrptr(t1, Klass::_lh_header_size_shift);
1191           __ andptr(t1, Klass::_lh_header_size_mask);
1192           __ addptr(arr_size, t1);
1193           __ addptr(arr_size, MinObjAlignmentInBytesMask); // align up
1194           __ andptr(arr_size, ~MinObjAlignmentInBytesMask);
1195 
1196           __ eden_allocate(obj, arr_size, 0, t1, slow_path);  // preserves arr_size
1197           __ incr_allocated_bytes(thread, arr_size, 0);
1198 
1199           __ initialize_header(obj, klass, length, t1, t2);
1200           __ movb(t1, Address(klass, klassOopDesc::header_size() * HeapWordSize + Klass::layout_helper_offset_in_bytes() + (Klass::_lh_header_size_shift / BitsPerByte)));
1201           assert(Klass::_lh_header_size_shift % BitsPerByte == 0, "bytewise");
1202           assert(Klass::_lh_header_size_mask <= 0xFF, "bytewise");
1203           __ andptr(t1, Klass::_lh_header_size_mask);
1204           __ subptr(arr_size, t1);  // body length
1205           __ addptr(t1, obj);       // body start
1206           __ initialize_body(t1, arr_size, 0, t2);
1207           __ verify_oop(obj);
1208           __ ret(0);
1209 
1210           __ bind(slow_path);
1211         }
1212 
1213         __ enter();
1214         OopMap* map = save_live_registers(sasm, 3);
1215         int call_offset;
1216         if (id == new_type_array_id) {
1217           call_offset = __ call_RT(obj, noreg, CAST_FROM_FN_PTR(address, new_type_array), klass, length);
1218         } else {
1219           call_offset = __ call_RT(obj, noreg, CAST_FROM_FN_PTR(address, new_object_array), klass, length);
1220         }
1221 
1222         oop_maps = new OopMapSet();
1223         oop_maps->add_gc_map(call_offset, map);
1224         restore_live_registers_except_rax(sasm);
1225 
1226         __ verify_oop(obj);
1227         __ leave();
1228         __ ret(0);
1229 
1230         // rax,: new array
1231       }
1232       break;
1233 
1234     case new_multi_array_id:
1235       { StubFrame f(sasm, "new_multi_array", dont_gc_arguments);
1236         // rax,: klass
1237         // rbx,: rank
1238         // rcx: address of 1st dimension
1239         OopMap* map = save_live_registers(sasm, 4);
1240         int call_offset = __ call_RT(rax, noreg, CAST_FROM_FN_PTR(address, new_multi_array), rax, rbx, rcx);
1241 
1242         oop_maps = new OopMapSet();
1243         oop_maps->add_gc_map(call_offset, map);
1244         restore_live_registers_except_rax(sasm);
1245 
1246         // rax,: new multi array
1247         __ verify_oop(rax);
1248       }
1249       break;
1250 
1251     case register_finalizer_id:
1252       {
1253         __ set_info("register_finalizer", dont_gc_arguments);
1254 
1255         // This is called via call_runtime so the arguments
1256         // will be place in C abi locations
1257 
1258 #ifdef _LP64
1259         __ verify_oop(c_rarg0);
1260         __ mov(rax, c_rarg0);
1261 #else
1262         // The object is passed on the stack and we haven't pushed a
1263         // frame yet so it's one work away from top of stack.
1264         __ movptr(rax, Address(rsp, 1 * BytesPerWord));
1265         __ verify_oop(rax);
1266 #endif // _LP64
1267 
1268         // load the klass and check the has finalizer flag
1269         Label register_finalizer;
1270         Register t = rsi;
1271         __ load_klass(t, rax);
1272         __ movl(t, Address(t, Klass::access_flags_offset_in_bytes() + sizeof(oopDesc)));
1273         __ testl(t, JVM_ACC_HAS_FINALIZER);
1274         __ jcc(Assembler::notZero, register_finalizer);
1275         __ ret(0);
1276 
1277         __ bind(register_finalizer);
1278         __ enter();
1279         OopMap* oop_map = save_live_registers(sasm, 2 /*num_rt_args */);
1280         int call_offset = __ call_RT(noreg, noreg,
1281                                      CAST_FROM_FN_PTR(address, SharedRuntime::register_finalizer), rax);
1282         oop_maps = new OopMapSet();
1283         oop_maps->add_gc_map(call_offset, oop_map);
1284 
1285         // Now restore all the live registers
1286         restore_live_registers(sasm);
1287 
1288         __ leave();
1289         __ ret(0);
1290       }
1291       break;
1292 
1293     case throw_range_check_failed_id:
1294       { StubFrame f(sasm, "range_check_failed", dont_gc_arguments);
1295         oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_range_check_exception), true);
1296       }
1297       break;
1298 
1299     case throw_index_exception_id:
1300       { StubFrame f(sasm, "index_range_check_failed", dont_gc_arguments);
1301         oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_index_exception), true);
1302       }
1303       break;
1304 
1305     case throw_div0_exception_id:
1306       { StubFrame f(sasm, "throw_div0_exception", dont_gc_arguments);
1307         oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_div0_exception), false);
1308       }
1309       break;
1310 
1311     case throw_null_pointer_exception_id:
1312       { StubFrame f(sasm, "throw_null_pointer_exception", dont_gc_arguments);
1313         oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_null_pointer_exception), false);
1314       }
1315       break;
1316 
1317     case handle_exception_nofpu_id:
1318       save_fpu_registers = false;
1319       // fall through
1320     case handle_exception_id:
1321       { StubFrame f(sasm, "handle_exception", dont_gc_arguments);
1322         oop_maps = new OopMapSet();
1323         OopMap* oop_map = save_live_registers(sasm, 1, save_fpu_registers);
1324         generate_handle_exception(sasm, oop_maps, oop_map, save_fpu_registers);
1325       }
1326       break;
1327 
1328     case unwind_exception_id:
1329       { __ set_info("unwind_exception", dont_gc_arguments);
1330         // note: no stubframe since we are about to leave the current
1331         //       activation and we are calling a leaf VM function only.
1332         generate_unwind_exception(sasm);
1333       }
1334       break;
1335 
1336     case throw_array_store_exception_id:
1337       { StubFrame f(sasm, "throw_array_store_exception", dont_gc_arguments);
1338         // tos + 0: link
1339         //     + 1: return address
1340         oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_array_store_exception), false);
1341       }
1342       break;
1343 
1344     case throw_class_cast_exception_id:
1345       { StubFrame f(sasm, "throw_class_cast_exception", dont_gc_arguments);
1346         oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_class_cast_exception), true);
1347       }
1348       break;
1349 
1350     case throw_incompatible_class_change_error_id:
1351       { StubFrame f(sasm, "throw_incompatible_class_cast_exception", dont_gc_arguments);
1352         oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_incompatible_class_change_error), false);
1353       }
1354       break;
1355 
1356     case slow_subtype_check_id:
1357       {
1358         // Typical calling sequence:
1359         // __ push(klass_RInfo);  // object klass or other subclass
1360         // __ push(sup_k_RInfo);  // array element klass or other superclass
1361         // __ call(slow_subtype_check);
1362         // Note that the subclass is pushed first, and is therefore deepest.
1363         // Previous versions of this code reversed the names 'sub' and 'super'.
1364         // This was operationally harmless but made the code unreadable.
1365         enum layout {
1366           rax_off, SLOT2(raxH_off)
1367           rcx_off, SLOT2(rcxH_off)
1368           rsi_off, SLOT2(rsiH_off)
1369           rdi_off, SLOT2(rdiH_off)
1370           // saved_rbp_off, SLOT2(saved_rbpH_off)
1371           return_off, SLOT2(returnH_off)
1372           sup_k_off, SLOT2(sup_kH_off)
1373           klass_off, SLOT2(superH_off)
1374           framesize,
1375           result_off = klass_off  // deepest argument is also the return value
1376         };
1377 
1378         __ set_info("slow_subtype_check", dont_gc_arguments);
1379         __ push(rdi);
1380         __ push(rsi);
1381         __ push(rcx);
1382         __ push(rax);
1383 
1384         // This is called by pushing args and not with C abi
1385         __ movptr(rsi, Address(rsp, (klass_off) * VMRegImpl::stack_slot_size)); // subclass
1386         __ movptr(rax, Address(rsp, (sup_k_off) * VMRegImpl::stack_slot_size)); // superclass
1387 
1388         Label miss;
1389         __ check_klass_subtype_slow_path(rsi, rax, rcx, rdi, NULL, &miss);
1390 
1391         // fallthrough on success:
1392         __ movptr(Address(rsp, (result_off) * VMRegImpl::stack_slot_size), 1); // result
1393         __ pop(rax);
1394         __ pop(rcx);
1395         __ pop(rsi);
1396         __ pop(rdi);
1397         __ ret(0);
1398 
1399         __ bind(miss);
1400         __ movptr(Address(rsp, (result_off) * VMRegImpl::stack_slot_size), NULL_WORD); // result
1401         __ pop(rax);
1402         __ pop(rcx);
1403         __ pop(rsi);
1404         __ pop(rdi);
1405         __ ret(0);
1406       }
1407       break;
1408 
1409     case monitorenter_nofpu_id:
1410       save_fpu_registers = false;
1411       // fall through
1412     case monitorenter_id:
1413       {
1414         StubFrame f(sasm, "monitorenter", dont_gc_arguments);
1415         OopMap* map = save_live_registers(sasm, 3, save_fpu_registers);
1416 
1417         // Called with store_parameter and not C abi
1418 
1419         f.load_argument(1, rax); // rax,: object
1420         f.load_argument(0, rbx); // rbx,: lock address
1421 
1422         int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, monitorenter), rax, rbx);
1423 
1424         oop_maps = new OopMapSet();
1425         oop_maps->add_gc_map(call_offset, map);
1426         restore_live_registers(sasm, save_fpu_registers);
1427       }
1428       break;
1429 
1430     case monitorexit_nofpu_id:
1431       save_fpu_registers = false;
1432       // fall through
1433     case monitorexit_id:
1434       {
1435         StubFrame f(sasm, "monitorexit", dont_gc_arguments);
1436         OopMap* map = save_live_registers(sasm, 2, save_fpu_registers);
1437 
1438         // Called with store_parameter and not C abi
1439 
1440         f.load_argument(0, rax); // rax,: lock address
1441 
1442         // note: really a leaf routine but must setup last java sp
1443         //       => use call_RT for now (speed can be improved by
1444         //       doing last java sp setup manually)
1445         int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, monitorexit), rax);
1446 
1447         oop_maps = new OopMapSet();
1448         oop_maps->add_gc_map(call_offset, map);
1449         restore_live_registers(sasm, save_fpu_registers);
1450 
1451       }
1452       break;
1453 
1454     case access_field_patching_id:
1455       { StubFrame f(sasm, "access_field_patching", dont_gc_arguments);
1456         // we should set up register map
1457         oop_maps = generate_patching(sasm, CAST_FROM_FN_PTR(address, access_field_patching));
1458       }
1459       break;
1460 
1461     case load_klass_patching_id:
1462       { StubFrame f(sasm, "load_klass_patching", dont_gc_arguments);
1463         // we should set up register map
1464         oop_maps = generate_patching(sasm, CAST_FROM_FN_PTR(address, move_klass_patching));
1465       }
1466       break;
1467 
1468     case jvmti_exception_throw_id:
1469       { // rax,: exception oop
1470         StubFrame f(sasm, "jvmti_exception_throw", dont_gc_arguments);
1471         // Preserve all registers across this potentially blocking call
1472         const int num_rt_args = 2;  // thread, exception oop
1473         OopMap* map = save_live_registers(sasm, num_rt_args);
1474         int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, Runtime1::post_jvmti_exception_throw), rax);
1475         oop_maps = new OopMapSet();
1476         oop_maps->add_gc_map(call_offset, map);
1477         restore_live_registers(sasm);
1478       }
1479       break;
1480 
1481     case dtrace_object_alloc_id:
1482       { // rax,: object
1483         StubFrame f(sasm, "dtrace_object_alloc", dont_gc_arguments);
1484         // we can't gc here so skip the oopmap but make sure that all
1485         // the live registers get saved.
1486         save_live_registers(sasm, 1);
1487 
1488         __ NOT_LP64(push(rax)) LP64_ONLY(mov(c_rarg0, rax));
1489         __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc)));
1490         NOT_LP64(__ pop(rax));
1491 
1492         restore_live_registers(sasm);
1493       }
1494       break;
1495 
1496     case fpu2long_stub_id:
1497       {
1498         // rax, and rdx are destroyed, but should be free since the result is returned there
1499         // preserve rsi,ecx
1500         __ push(rsi);
1501         __ push(rcx);
1502         LP64_ONLY(__ push(rdx);)
1503 
1504         // check for NaN
1505         Label return0, do_return, return_min_jlong, do_convert;
1506 
1507         Address value_high_word(rsp, wordSize + 4);
1508         Address value_low_word(rsp, wordSize);
1509         Address result_high_word(rsp, 3*wordSize + 4);
1510         Address result_low_word(rsp, 3*wordSize);
1511 
1512         __ subptr(rsp, 32);                    // more than enough on 32bit
1513         __ fst_d(value_low_word);
1514         __ movl(rax, value_high_word);
1515         __ andl(rax, 0x7ff00000);
1516         __ cmpl(rax, 0x7ff00000);
1517         __ jcc(Assembler::notEqual, do_convert);
1518         __ movl(rax, value_high_word);
1519         __ andl(rax, 0xfffff);
1520         __ orl(rax, value_low_word);
1521         __ jcc(Assembler::notZero, return0);
1522 
1523         __ bind(do_convert);
1524         __ fnstcw(Address(rsp, 0));
1525         __ movzwl(rax, Address(rsp, 0));
1526         __ orl(rax, 0xc00);
1527         __ movw(Address(rsp, 2), rax);
1528         __ fldcw(Address(rsp, 2));
1529         __ fwait();
1530         __ fistp_d(result_low_word);
1531         __ fldcw(Address(rsp, 0));
1532         __ fwait();
1533         // This gets the entire long in rax on 64bit
1534         __ movptr(rax, result_low_word);
1535         // testing of high bits
1536         __ movl(rdx, result_high_word);
1537         __ mov(rcx, rax);
1538         // What the heck is the point of the next instruction???
1539         __ xorl(rcx, 0x0);
1540         __ movl(rsi, 0x80000000);
1541         __ xorl(rsi, rdx);
1542         __ orl(rcx, rsi);
1543         __ jcc(Assembler::notEqual, do_return);
1544         __ fldz();
1545         __ fcomp_d(value_low_word);
1546         __ fnstsw_ax();
1547 #ifdef _LP64
1548         __ testl(rax, 0x4100);  // ZF & CF == 0
1549         __ jcc(Assembler::equal, return_min_jlong);
1550 #else
1551         __ sahf();
1552         __ jcc(Assembler::above, return_min_jlong);
1553 #endif // _LP64
1554         // return max_jlong
1555 #ifndef _LP64
1556         __ movl(rdx, 0x7fffffff);
1557         __ movl(rax, 0xffffffff);
1558 #else
1559         __ mov64(rax, CONST64(0x7fffffffffffffff));
1560 #endif // _LP64
1561         __ jmp(do_return);
1562 
1563         __ bind(return_min_jlong);
1564 #ifndef _LP64
1565         __ movl(rdx, 0x80000000);
1566         __ xorl(rax, rax);
1567 #else
1568         __ mov64(rax, CONST64(0x8000000000000000));
1569 #endif // _LP64
1570         __ jmp(do_return);
1571 
1572         __ bind(return0);
1573         __ fpop();
1574 #ifndef _LP64
1575         __ xorptr(rdx,rdx);
1576         __ xorptr(rax,rax);
1577 #else
1578         __ xorptr(rax, rax);
1579 #endif // _LP64
1580 
1581         __ bind(do_return);
1582         __ addptr(rsp, 32);
1583         LP64_ONLY(__ pop(rdx);)
1584         __ pop(rcx);
1585         __ pop(rsi);
1586         __ ret(0);
1587       }
1588       break;
1589 
1590 #ifndef SERIALGC
1591     case g1_pre_barrier_slow_id:
1592       {
1593         StubFrame f(sasm, "g1_pre_barrier", dont_gc_arguments);
1594         // arg0 : previous value of memory
1595 
1596         BarrierSet* bs = Universe::heap()->barrier_set();
1597         if (bs->kind() != BarrierSet::G1SATBCTLogging) {
1598           __ movptr(rax, (int)id);
1599           __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, unimplemented_entry), rax);
1600           __ should_not_reach_here();
1601           break;
1602         }
1603         __ push(rax);
1604         __ push(rdx);
1605 
1606         const Register pre_val = rax;
1607         const Register thread = NOT_LP64(rax) LP64_ONLY(r15_thread);
1608         const Register tmp = rdx;
1609 
1610         NOT_LP64(__ get_thread(thread);)
1611 
1612         Address in_progress(thread, in_bytes(JavaThread::satb_mark_queue_offset() +
1613                                              PtrQueue::byte_offset_of_active()));
1614 
1615         Address queue_index(thread, in_bytes(JavaThread::satb_mark_queue_offset() +
1616                                              PtrQueue::byte_offset_of_index()));
1617         Address buffer(thread, in_bytes(JavaThread::satb_mark_queue_offset() +
1618                                         PtrQueue::byte_offset_of_buf()));
1619 
1620 
1621         Label done;
1622         Label runtime;
1623 
1624         // Can we store original value in the thread's buffer?
1625 
1626 #ifdef _LP64
1627         __ movslq(tmp, queue_index);
1628         __ cmpq(tmp, 0);
1629 #else
1630         __ cmpl(queue_index, 0);
1631 #endif
1632         __ jcc(Assembler::equal, runtime);
1633 #ifdef _LP64
1634         __ subq(tmp, wordSize);
1635         __ movl(queue_index, tmp);
1636         __ addq(tmp, buffer);
1637 #else
1638         __ subl(queue_index, wordSize);
1639         __ movl(tmp, buffer);
1640         __ addl(tmp, queue_index);
1641 #endif
1642 
1643         // prev_val (rax)
1644         f.load_argument(0, pre_val);
1645         __ movptr(Address(tmp, 0), pre_val);
1646         __ jmp(done);
1647 
1648         __ bind(runtime);
1649         __ push(rcx);
1650 #ifdef _LP64
1651         __ push(r8);
1652         __ push(r9);
1653         __ push(r10);
1654         __ push(r11);
1655 #  ifndef _WIN64
1656         __ push(rdi);
1657         __ push(rsi);
1658 #  endif
1659 #endif
1660         // load the pre-value
1661         f.load_argument(0, rcx);
1662         __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::g1_wb_pre), rcx, thread);
1663 #ifdef _LP64
1664 #  ifndef _WIN64
1665         __ pop(rsi);
1666         __ pop(rdi);
1667 #  endif
1668         __ pop(r11);
1669         __ pop(r10);
1670         __ pop(r9);
1671         __ pop(r8);
1672 #endif
1673         __ pop(rcx);
1674         __ bind(done);
1675 
1676         __ pop(rdx);
1677         __ pop(rax);
1678       }
1679       break;
1680 
1681     case g1_post_barrier_slow_id:
1682       {
1683         StubFrame f(sasm, "g1_post_barrier", dont_gc_arguments);
1684 
1685 
1686         // arg0: store_address
1687         Address store_addr(rbp, 2*BytesPerWord);
1688 
1689         BarrierSet* bs = Universe::heap()->barrier_set();
1690         CardTableModRefBS* ct = (CardTableModRefBS*)bs;
1691         Label done;
1692         Label runtime;
1693 
1694         // At this point we know new_value is non-NULL and the new_value crosses regsion.
1695         // Must check to see if card is already dirty
1696 
1697         const Register thread = NOT_LP64(rax) LP64_ONLY(r15_thread);
1698 
1699         Address queue_index(thread, in_bytes(JavaThread::dirty_card_queue_offset() +
1700                                              PtrQueue::byte_offset_of_index()));
1701         Address buffer(thread, in_bytes(JavaThread::dirty_card_queue_offset() +
1702                                         PtrQueue::byte_offset_of_buf()));
1703 
1704         __ push(rax);
1705         __ push(rcx);
1706 
1707         NOT_LP64(__ get_thread(thread);)
1708         ExternalAddress cardtable((address)ct->byte_map_base);
1709         assert(sizeof(*ct->byte_map_base) == sizeof(jbyte), "adjust this code");
1710 
1711         const Register card_addr = rcx;
1712 #ifdef _LP64
1713         const Register tmp = rscratch1;
1714         f.load_argument(0, card_addr);
1715         __ shrq(card_addr, CardTableModRefBS::card_shift);
1716         __ lea(tmp, cardtable);
1717         // get the address of the card
1718         __ addq(card_addr, tmp);
1719 #else
1720         const Register card_index = rcx;
1721         f.load_argument(0, card_index);
1722         __ shrl(card_index, CardTableModRefBS::card_shift);
1723 
1724         Address index(noreg, card_index, Address::times_1);
1725         __ leal(card_addr, __ as_Address(ArrayAddress(cardtable, index)));
1726 #endif
1727 
1728         __ cmpb(Address(card_addr, 0), 0);
1729         __ jcc(Assembler::equal, done);
1730 
1731         // storing region crossing non-NULL, card is clean.
1732         // dirty card and log.
1733 
1734         __ movb(Address(card_addr, 0), 0);
1735 
1736         __ cmpl(queue_index, 0);
1737         __ jcc(Assembler::equal, runtime);
1738         __ subl(queue_index, wordSize);
1739 
1740         const Register buffer_addr = rbx;
1741         __ push(rbx);
1742 
1743         __ movptr(buffer_addr, buffer);
1744 
1745 #ifdef _LP64
1746         __ movslq(rscratch1, queue_index);
1747         __ addptr(buffer_addr, rscratch1);
1748 #else
1749         __ addptr(buffer_addr, queue_index);
1750 #endif
1751         __ movptr(Address(buffer_addr, 0), card_addr);
1752 
1753         __ pop(rbx);
1754         __ jmp(done);
1755 
1756         __ bind(runtime);
1757         __ push(rdx);
1758 #ifdef _LP64
1759         __ push(r8);
1760         __ push(r9);
1761         __ push(r10);
1762         __ push(r11);
1763 #  ifndef _WIN64
1764         __ push(rdi);
1765         __ push(rsi);
1766 #  endif
1767 #endif
1768         __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::g1_wb_post), card_addr, thread);
1769 #ifdef _LP64
1770 #  ifndef _WIN64
1771         __ pop(rsi);
1772         __ pop(rdi);
1773 #  endif
1774         __ pop(r11);
1775         __ pop(r10);
1776         __ pop(r9);
1777         __ pop(r8);
1778 #endif
1779         __ pop(rdx);
1780         __ bind(done);
1781 
1782         __ pop(rcx);
1783         __ pop(rax);
1784 
1785       }
1786       break;
1787 #endif // !SERIALGC
1788 
1789     default:
1790       { StubFrame f(sasm, "unimplemented entry", dont_gc_arguments);
1791         __ movptr(rax, (int)id);
1792         __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, unimplemented_entry), rax);
1793         __ should_not_reach_here();
1794       }
1795       break;
1796   }
1797   return oop_maps;
1798 }
1799 
1800 #undef __
1801 
1802 const char *Runtime1::pd_name_for_address(address entry) {
1803   return "<unknown function>";
1804 }