1 /*
   2  * Copyright (c) 2003, 2014, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 #include "precompiled.hpp"
  26 #include "asm/macroAssembler.hpp"
  27 #include "interpreter/bytecodeHistogram.hpp"
  28 #include "interpreter/interpreter.hpp"
  29 #include "interpreter/interpreterGenerator.hpp"
  30 #include "interpreter/interpreterRuntime.hpp"
  31 #include "interpreter/interp_masm.hpp"
  32 #include "interpreter/templateTable.hpp"
  33 #include "oops/arrayOop.hpp"
  34 #include "oops/methodData.hpp"
  35 #include "oops/method.hpp"
  36 #include "oops/oop.inline.hpp"
  37 #include "prims/jvmtiExport.hpp"
  38 #include "prims/jvmtiThreadState.hpp"
  39 #include "runtime/arguments.hpp"
  40 #include "runtime/deoptimization.hpp"
  41 #include "runtime/frame.inline.hpp"
  42 #include "runtime/sharedRuntime.hpp"
  43 #include "runtime/stubRoutines.hpp"
  44 #include "runtime/synchronizer.hpp"
  45 #include "runtime/timer.hpp"
  46 #include "runtime/vframeArray.hpp"
  47 #include "utilities/debug.hpp"
  48 #include "utilities/macros.hpp"
  49 
  50 #define __ _masm->
  51 
  52 #ifndef CC_INTERP
  53 
  54 const int method_offset = frame::interpreter_frame_method_offset * wordSize;
  55 const int bcp_offset    = frame::interpreter_frame_bcp_offset    * wordSize;
  56 const int locals_offset = frame::interpreter_frame_locals_offset * wordSize;
  57 
  58 //-----------------------------------------------------------------------------
  59 
  60 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() {
  61   address entry = __ pc();
  62 
  63 #ifdef ASSERT
  64   {
  65     Label L;
  66     __ lea(rax, Address(rbp,
  67                         frame::interpreter_frame_monitor_block_top_offset *
  68                         wordSize));
  69     __ cmpptr(rax, rsp); // rax = maximal rsp for current rbp (stack
  70                          // grows negative)
  71     __ jcc(Assembler::aboveEqual, L); // check if frame is complete
  72     __ stop ("interpreter frame not set up");
  73     __ bind(L);
  74   }
  75 #endif // ASSERT
  76   // Restore bcp under the assumption that the current frame is still
  77   // interpreted
  78   __ restore_bcp();
  79 
  80   // expression stack must be empty before entering the VM if an
  81   // exception happened
  82   __ empty_expression_stack();
  83   // throw exception
  84   __ call_VM(noreg,
  85              CAST_FROM_FN_PTR(address,
  86                               InterpreterRuntime::throw_StackOverflowError));
  87   return entry;
  88 }
  89 
  90 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler(
  91         const char* name) {
  92   address entry = __ pc();
  93   // expression stack must be empty before entering the VM if an
  94   // exception happened
  95   __ empty_expression_stack();
  96   // setup parameters
  97   // ??? convention: expect aberrant index in register ebx
  98   __ lea(c_rarg1, ExternalAddress((address)name));
  99   __ call_VM(noreg,
 100              CAST_FROM_FN_PTR(address,
 101                               InterpreterRuntime::
 102                               throw_ArrayIndexOutOfBoundsException),
 103              c_rarg1, rbx);
 104   return entry;
 105 }
 106 
 107 address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
 108   address entry = __ pc();
 109 
 110   // object is at TOS
 111   __ pop(c_rarg1);
 112 
 113   // expression stack must be empty before entering the VM if an
 114   // exception happened
 115   __ empty_expression_stack();
 116 
 117   __ call_VM(noreg,
 118              CAST_FROM_FN_PTR(address,
 119                               InterpreterRuntime::
 120                               throw_ClassCastException),
 121              c_rarg1);
 122   return entry;
 123 }
 124 
 125 address TemplateInterpreterGenerator::generate_exception_handler_common(
 126         const char* name, const char* message, bool pass_oop) {
 127   assert(!pass_oop || message == NULL, "either oop or message but not both");
 128   address entry = __ pc();
 129   if (pass_oop) {
 130     // object is at TOS
 131     __ pop(c_rarg2);
 132   }
 133   // expression stack must be empty before entering the VM if an
 134   // exception happened
 135   __ empty_expression_stack();
 136   // setup parameters
 137   __ lea(c_rarg1, ExternalAddress((address)name));
 138   if (pass_oop) {
 139     __ call_VM(rax, CAST_FROM_FN_PTR(address,
 140                                      InterpreterRuntime::
 141                                      create_klass_exception),
 142                c_rarg1, c_rarg2);
 143   } else {
 144     // kind of lame ExternalAddress can't take NULL because
 145     // external_word_Relocation will assert.
 146     if (message != NULL) {
 147       __ lea(c_rarg2, ExternalAddress((address)message));
 148     } else {
 149       __ movptr(c_rarg2, NULL_WORD);
 150     }
 151     __ call_VM(rax,
 152                CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception),
 153                c_rarg1, c_rarg2);
 154   }
 155   // throw exception
 156   __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
 157   return entry;
 158 }
 159 
 160 
 161 address TemplateInterpreterGenerator::generate_continuation_for(TosState state) {
 162   address entry = __ pc();
 163   // NULL last_sp until next java call
 164   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
 165   __ dispatch_next(state);
 166   return entry;
 167 }
 168 
 169 
 170 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step, size_t index_size) {
 171   address entry = __ pc();
 172 
 173   // Restore stack bottom in case i2c adjusted stack
 174   __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
 175   // and NULL it as marker that esp is now tos until next java call
 176   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
 177 
 178   __ restore_bcp();
 179   __ restore_locals();
 180 
 181   if (state == atos) {
 182     Register mdp = rbx;
 183     Register tmp = rcx;
 184     __ profile_return_type(mdp, rax, tmp);
 185   }
 186 
 187   const Register cache = rbx;
 188   const Register index = rcx;
 189   __ get_cache_and_index_at_bcp(cache, index, 1, index_size);
 190 
 191   const Register flags = cache;
 192   __ movl(flags, Address(cache, index, Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()));
 193   __ andl(flags, ConstantPoolCacheEntry::parameter_size_mask);
 194   __ lea(rsp, Address(rsp, flags, Interpreter::stackElementScale()));
 195   __ dispatch_next(state, step);
 196 
 197   return entry;
 198 }
 199 
 200 
 201 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state,
 202                                                                int step) {
 203   address entry = __ pc();
 204   // NULL last_sp until next java call
 205   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
 206   __ restore_bcp();
 207   __ restore_locals();
 208   // handle exceptions
 209   {
 210     Label L;
 211     __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t) NULL_WORD);
 212     __ jcc(Assembler::zero, L);
 213     __ call_VM(noreg,
 214                CAST_FROM_FN_PTR(address,
 215                                 InterpreterRuntime::throw_pending_exception));
 216     __ should_not_reach_here();
 217     __ bind(L);
 218   }
 219   __ dispatch_next(state, step);
 220   return entry;
 221 }
 222 
 223 int AbstractInterpreter::BasicType_as_index(BasicType type) {
 224   int i = 0;
 225   switch (type) {
 226     case T_BOOLEAN: i = 0; break;
 227     case T_CHAR   : i = 1; break;
 228     case T_BYTE   : i = 2; break;
 229     case T_SHORT  : i = 3; break;
 230     case T_INT    : i = 4; break;
 231     case T_LONG   : i = 5; break;
 232     case T_VOID   : i = 6; break;
 233     case T_FLOAT  : i = 7; break;
 234     case T_DOUBLE : i = 8; break;
 235     case T_OBJECT : i = 9; break;
 236     case T_ARRAY  : i = 9; break;
 237     default       : ShouldNotReachHere();
 238   }
 239   assert(0 <= i && i < AbstractInterpreter::number_of_result_handlers,
 240          "index out of bounds");
 241   return i;
 242 }
 243 
 244 
 245 address TemplateInterpreterGenerator::generate_result_handler_for(
 246         BasicType type) {
 247   address entry = __ pc();
 248   switch (type) {
 249   case T_BOOLEAN: __ c2bool(rax);            break;
 250   case T_CHAR   : __ movzwl(rax, rax);       break;
 251   case T_BYTE   : __ sign_extend_byte(rax);  break;
 252   case T_SHORT  : __ sign_extend_short(rax); break;
 253   case T_INT    : /* nothing to do */        break;
 254   case T_LONG   : /* nothing to do */        break;
 255   case T_VOID   : /* nothing to do */        break;
 256   case T_FLOAT  : /* nothing to do */        break;
 257   case T_DOUBLE : /* nothing to do */        break;
 258   case T_OBJECT :
 259     // retrieve result from frame
 260     __ movptr(rax, Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize));
 261     // and verify it
 262     __ verify_oop(rax);
 263     break;
 264   default       : ShouldNotReachHere();
 265   }
 266   __ ret(0);                                   // return from result handler
 267   return entry;
 268 }
 269 
 270 address TemplateInterpreterGenerator::generate_safept_entry_for(
 271         TosState state,
 272         address runtime_entry) {
 273   address entry = __ pc();
 274   __ push(state);
 275   __ call_VM(noreg, runtime_entry);
 276   __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
 277   return entry;
 278 }
 279 
 280 
 281 
 282 // Helpers for commoning out cases in the various type of method entries.
 283 //
 284 
 285 
 286 // increment invocation count & check for overflow
 287 //
 288 // Note: checking for negative value instead of overflow
 289 //       so we have a 'sticky' overflow test
 290 //
 291 // rbx: method
 292 // ecx: invocation counter
 293 //
 294 void InterpreterGenerator::generate_counter_incr(
 295         Label* overflow,
 296         Label* profile_method,
 297         Label* profile_method_continue) {
 298   Label done;
 299   // Note: In tiered we increment either counters in Method* or in MDO depending if we're profiling or not.
 300   if (TieredCompilation) {
 301     int increment = InvocationCounter::count_increment;
 302     Label no_mdo;
 303     if (ProfileInterpreter) {
 304       // Are we profiling?
 305       __ movptr(rax, Address(rbx, Method::method_data_offset()));
 306       __ testptr(rax, rax);
 307       __ jccb(Assembler::zero, no_mdo);
 308       // Increment counter in the MDO
 309       const Address mdo_invocation_counter(rax, in_bytes(MethodData::invocation_counter_offset()) +
 310                                                 in_bytes(InvocationCounter::counter_offset()));
 311       const Address mask(rax, in_bytes(MethodData::invoke_mask_offset()));
 312       __ increment_mask_and_jump(mdo_invocation_counter, increment, mask, rcx, false, Assembler::zero, overflow);
 313       __ jmp(done);
 314     }
 315     __ bind(no_mdo);
 316     // Increment counter in MethodCounters
 317     const Address invocation_counter(rax,
 318                   MethodCounters::invocation_counter_offset() +
 319                   InvocationCounter::counter_offset());
 320     __ get_method_counters(rbx, rax, done);
 321     const Address mask(rax, in_bytes(MethodCounters::invoke_mask_offset()));
 322     __ increment_mask_and_jump(invocation_counter, increment, mask, rcx,
 323                                false, Assembler::zero, overflow);
 324     __ bind(done);
 325   } else { // not TieredCompilation
 326     const Address backedge_counter(rax,
 327                   MethodCounters::backedge_counter_offset() +
 328                   InvocationCounter::counter_offset());
 329     const Address invocation_counter(rax,
 330                   MethodCounters::invocation_counter_offset() +
 331                   InvocationCounter::counter_offset());
 332 
 333     __ get_method_counters(rbx, rax, done);
 334 
 335     if (ProfileInterpreter) {
 336       __ incrementl(Address(rax,
 337               MethodCounters::interpreter_invocation_counter_offset()));
 338     }
 339     // Update standard invocation counters
 340     __ movl(rcx, invocation_counter);
 341     __ incrementl(rcx, InvocationCounter::count_increment);
 342     __ movl(invocation_counter, rcx); // save invocation count
 343 
 344     __ movl(rax, backedge_counter);   // load backedge counter
 345     __ andl(rax, InvocationCounter::count_mask_value); // mask out the status bits
 346 
 347     __ addl(rcx, rax);                // add both counters
 348 
 349     // profile_method is non-null only for interpreted method so
 350     // profile_method != NULL == !native_call
 351 
 352     if (ProfileInterpreter && profile_method != NULL) {
 353       // Test to see if we should create a method data oop
 354       __ movptr(rax, Address(rbx, Method::method_counters_offset()));
 355       __ cmp32(rcx, Address(rax, in_bytes(MethodCounters::interpreter_profile_limit_offset())));
 356       __ jcc(Assembler::less, *profile_method_continue);
 357 
 358       // if no method data exists, go to profile_method
 359       __ test_method_data_pointer(rax, *profile_method);
 360     }
 361 
 362     __ movptr(rax, Address(rbx, Method::method_counters_offset()));
 363     __ cmp32(rcx, Address(rax, in_bytes(MethodCounters::interpreter_invocation_limit_offset())));
 364     __ jcc(Assembler::aboveEqual, *overflow);
 365     __ bind(done);
 366   }
 367 }
 368 
 369 void InterpreterGenerator::generate_counter_overflow(Label* do_continue) {
 370 
 371   // Asm interpreter on entry
 372   // r14 - locals
 373   // r13 - bcp
 374   // rbx - method
 375   // edx - cpool --- DOES NOT APPEAR TO BE TRUE
 376   // rbp - interpreter frame
 377 
 378   // On return (i.e. jump to entry_point) [ back to invocation of interpreter ]
 379   // Everything as it was on entry
 380   // rdx is not restored. Doesn't appear to really be set.
 381 
 382   // InterpreterRuntime::frequency_counter_overflow takes two
 383   // arguments, the first (thread) is passed by call_VM, the second
 384   // indicates if the counter overflow occurs at a backwards branch
 385   // (NULL bcp).  We pass zero for it.  The call returns the address
 386   // of the verified entry point for the method or NULL if the
 387   // compilation did not complete (either went background or bailed
 388   // out).
 389   __ movl(c_rarg1, 0);
 390   __ call_VM(noreg,
 391              CAST_FROM_FN_PTR(address,
 392                               InterpreterRuntime::frequency_counter_overflow),
 393              c_rarg1);
 394 
 395   __ movptr(rbx, Address(rbp, method_offset));   // restore Method*
 396   // Preserve invariant that r13/r14 contain bcp/locals of sender frame
 397   // and jump to the interpreted entry.
 398   __ jmp(*do_continue, relocInfo::none);
 399 }
 400 
 401 // See if we've got enough room on the stack for locals plus overhead.
 402 // The expression stack grows down incrementally, so the normal guard
 403 // page mechanism will work for that.
 404 //
 405 // NOTE: Since the additional locals are also always pushed (wasn't
 406 // obvious in generate_fixed_frame) so the guard should work for them
 407 // too.
 408 //
 409 // Args:
 410 //      rdx: number of additional locals this frame needs (what we must check)
 411 //      rbx: Method*
 412 //
 413 // Kills:
 414 //      rax
 415 void InterpreterGenerator::generate_stack_overflow_check(void) {
 416 
 417   // monitor entry size: see picture of stack in frame_x86.hpp
 418   const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
 419 
 420   // total overhead size: entry_size + (saved rbp through expr stack
 421   // bottom).  be sure to change this if you add/subtract anything
 422   // to/from the overhead area
 423   const int overhead_size =
 424     -(frame::interpreter_frame_initial_sp_offset * wordSize) + entry_size;
 425 
 426   const int page_size = os::vm_page_size();
 427 
 428   Label after_frame_check;
 429 
 430   // see if the frame is greater than one page in size. If so,
 431   // then we need to verify there is enough stack space remaining
 432   // for the additional locals.
 433   __ cmpl(rdx, (page_size - overhead_size) / Interpreter::stackElementSize);
 434   __ jcc(Assembler::belowEqual, after_frame_check);
 435 
 436   // compute rsp as if this were going to be the last frame on
 437   // the stack before the red zone
 438 
 439   const Address stack_base(r15_thread, Thread::stack_base_offset());
 440   const Address stack_size(r15_thread, Thread::stack_size_offset());
 441 
 442   // locals + overhead, in bytes
 443   __ mov(rax, rdx);
 444   __ shlptr(rax, Interpreter::logStackElementSize);  // 2 slots per parameter.
 445   __ addptr(rax, overhead_size);
 446 
 447 #ifdef ASSERT
 448   Label stack_base_okay, stack_size_okay;
 449   // verify that thread stack base is non-zero
 450   __ cmpptr(stack_base, (int32_t)NULL_WORD);
 451   __ jcc(Assembler::notEqual, stack_base_okay);
 452   __ stop("stack base is zero");
 453   __ bind(stack_base_okay);
 454   // verify that thread stack size is non-zero
 455   __ cmpptr(stack_size, 0);
 456   __ jcc(Assembler::notEqual, stack_size_okay);
 457   __ stop("stack size is zero");
 458   __ bind(stack_size_okay);
 459 #endif
 460 
 461   // Add stack base to locals and subtract stack size
 462   __ addptr(rax, stack_base);
 463   __ subptr(rax, stack_size);
 464 
 465   // Use the maximum number of pages we might bang.
 466   const int max_pages = StackShadowPages > (StackRedPages+StackYellowPages) ? StackShadowPages :
 467                                                                               (StackRedPages+StackYellowPages);
 468 
 469   // add in the red and yellow zone sizes
 470   __ addptr(rax, max_pages * page_size);
 471 
 472   // check against the current stack bottom
 473   __ cmpptr(rsp, rax);
 474   __ jcc(Assembler::above, after_frame_check);
 475 
 476   // Restore sender's sp as SP. This is necessary if the sender's
 477   // frame is an extended compiled frame (see gen_c2i_adapter())
 478   // and safer anyway in case of JSR292 adaptations.
 479 
 480   __ pop(rax); // return address must be moved if SP is changed
 481   __ mov(rsp, r13);
 482   __ push(rax);
 483 
 484   // Note: the restored frame is not necessarily interpreted.
 485   // Use the shared runtime version of the StackOverflowError.
 486   assert(StubRoutines::throw_StackOverflowError_entry() != NULL, "stub not yet generated");
 487   __ jump(ExternalAddress(StubRoutines::throw_StackOverflowError_entry()));
 488 
 489   // all done with frame size check
 490   __ bind(after_frame_check);
 491 }
 492 
 493 // Allocate monitor and lock method (asm interpreter)
 494 //
 495 // Args:
 496 //      rbx: Method*
 497 //      r14: locals
 498 //
 499 // Kills:
 500 //      rax
 501 //      c_rarg0, c_rarg1, c_rarg2, c_rarg3, ...(param regs)
 502 //      rscratch1, rscratch2 (scratch regs)
 503 void InterpreterGenerator::lock_method(void) {
 504   // synchronize method
 505   const Address access_flags(rbx, Method::access_flags_offset());
 506   const Address monitor_block_top(
 507         rbp,
 508         frame::interpreter_frame_monitor_block_top_offset * wordSize);
 509   const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
 510 
 511 #ifdef ASSERT
 512   {
 513     Label L;
 514     __ movl(rax, access_flags);
 515     __ testl(rax, JVM_ACC_SYNCHRONIZED);
 516     __ jcc(Assembler::notZero, L);
 517     __ stop("method doesn't need synchronization");
 518     __ bind(L);
 519   }
 520 #endif // ASSERT
 521 
 522   // get synchronization object
 523   {
 524     const int mirror_offset = in_bytes(Klass::java_mirror_offset());
 525     Label done;
 526     __ movl(rax, access_flags);
 527     __ testl(rax, JVM_ACC_STATIC);
 528     // get receiver (assume this is frequent case)
 529     __ movptr(rax, Address(r14, Interpreter::local_offset_in_bytes(0)));
 530     __ jcc(Assembler::zero, done);
 531     __ movptr(rax, Address(rbx, Method::const_offset()));
 532     __ movptr(rax, Address(rax, ConstMethod::constants_offset()));
 533     __ movptr(rax, Address(rax,
 534                            ConstantPool::pool_holder_offset_in_bytes()));
 535     __ movptr(rax, Address(rax, mirror_offset));
 536 
 537 #ifdef ASSERT
 538     {
 539       Label L;
 540       __ testptr(rax, rax);
 541       __ jcc(Assembler::notZero, L);
 542       __ stop("synchronization object is NULL");
 543       __ bind(L);
 544     }
 545 #endif // ASSERT
 546 
 547     __ bind(done);
 548   }
 549 
 550   // add space for monitor & lock
 551   __ subptr(rsp, entry_size); // add space for a monitor entry
 552   __ movptr(monitor_block_top, rsp);  // set new monitor block top
 553   // store object
 554   __ movptr(Address(rsp, BasicObjectLock::obj_offset_in_bytes()), rax);
 555   __ movptr(c_rarg1, rsp); // object address
 556   __ lock_object(c_rarg1);
 557 }
 558 
 559 // Generate a fixed interpreter frame. This is identical setup for
 560 // interpreted methods and for native methods hence the shared code.
 561 //
 562 // Args:
 563 //      rax: return address
 564 //      rbx: Method*
 565 //      r14: pointer to locals
 566 //      r13: sender sp
 567 //      rdx: cp cache
 568 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
 569   // initialize fixed part of activation frame
 570   __ push(rax);        // save return address
 571   __ enter();          // save old & set new rbp
 572   __ push(r13);        // set sender sp
 573   __ push((int)NULL_WORD); // leave last_sp as null
 574   __ movptr(r13, Address(rbx, Method::const_offset()));      // get ConstMethod*
 575   __ lea(r13, Address(r13, ConstMethod::codes_offset())); // get codebase
 576   __ push(rbx);        // save Method*
 577   if (ProfileInterpreter) {
 578     Label method_data_continue;
 579     __ movptr(rdx, Address(rbx, in_bytes(Method::method_data_offset())));
 580     __ testptr(rdx, rdx);
 581     __ jcc(Assembler::zero, method_data_continue);
 582     __ addptr(rdx, in_bytes(MethodData::data_offset()));
 583     __ bind(method_data_continue);
 584     __ push(rdx);      // set the mdp (method data pointer)
 585   } else {
 586     __ push(0);
 587   }
 588 
 589   __ movptr(rdx, Address(rbx, Method::const_offset()));
 590   __ movptr(rdx, Address(rdx, ConstMethod::constants_offset()));
 591   __ movptr(rdx, Address(rdx, ConstantPool::cache_offset_in_bytes()));
 592   __ push(rdx); // set constant pool cache
 593   __ push(r14); // set locals pointer
 594   if (native_call) {
 595     __ push(0); // no bcp
 596   } else {
 597     __ push(r13); // set bcp
 598   }
 599   __ push(0); // reserve word for pointer to expression stack bottom
 600   __ movptr(Address(rsp, 0), rsp); // set expression stack bottom
 601 }
 602 
 603 // End of helpers
 604 
 605 // Method entry for java.lang.ref.Reference.get.
 606 address InterpreterGenerator::generate_Reference_get_entry(void) {
 607 #if INCLUDE_ALL_GCS
 608   // Code: _aload_0, _getfield, _areturn
 609   // parameter size = 1
 610   //
 611   // The code that gets generated by this routine is split into 2 parts:
 612   //    1. The "intrinsified" code for G1 (or any SATB based GC),
 613   //    2. The slow path - which is an expansion of the regular method entry.
 614   //
 615   // Notes:-
 616   // * In the G1 code we do not check whether we need to block for
 617   //   a safepoint. If G1 is enabled then we must execute the specialized
 618   //   code for Reference.get (except when the Reference object is null)
 619   //   so that we can log the value in the referent field with an SATB
 620   //   update buffer.
 621   //   If the code for the getfield template is modified so that the
 622   //   G1 pre-barrier code is executed when the current method is
 623   //   Reference.get() then going through the normal method entry
 624   //   will be fine.
 625   // * The G1 code can, however, check the receiver object (the instance
 626   //   of java.lang.Reference) and jump to the slow path if null. If the
 627   //   Reference object is null then we obviously cannot fetch the referent
 628   //   and so we don't need to call the G1 pre-barrier. Thus we can use the
 629   //   regular method entry code to generate the NPE.
 630   //
 631   // rbx: Method*
 632 
 633   // r13: senderSP must preserve for slow path, set SP to it on fast path
 634 
 635   address entry = __ pc();
 636 
 637   const int referent_offset = java_lang_ref_Reference::referent_offset;
 638   guarantee(referent_offset > 0, "referent offset not initialized");
 639 
 640   if (UseG1GC) {
 641     Label slow_path;
 642     // rbx: method
 643 
 644     // Check if local 0 != NULL
 645     // If the receiver is null then it is OK to jump to the slow path.
 646     __ movptr(rax, Address(rsp, wordSize));
 647 
 648     __ testptr(rax, rax);
 649     __ jcc(Assembler::zero, slow_path);
 650 
 651     // rax: local 0
 652     // rbx: method (but can be used as scratch now)
 653     // rdx: scratch
 654     // rdi: scratch
 655 
 656     // Generate the G1 pre-barrier code to log the value of
 657     // the referent field in an SATB buffer.
 658 
 659     // Load the value of the referent field.
 660     const Address field_address(rax, referent_offset);
 661     __ load_heap_oop(rax, field_address);
 662 
 663     // Generate the G1 pre-barrier code to log the value of
 664     // the referent field in an SATB buffer.
 665     __ g1_write_barrier_pre(noreg /* obj */,
 666                             rax /* pre_val */,
 667                             r15_thread /* thread */,
 668                             rbx /* tmp */,
 669                             true /* tosca_live */,
 670                             true /* expand_call */);
 671 
 672     // _areturn
 673     __ pop(rdi);                // get return address
 674     __ mov(rsp, r13);           // set sp to sender sp
 675     __ jmp(rdi);
 676     __ ret(0);
 677 
 678     // generate a vanilla interpreter entry as the slow path
 679     __ bind(slow_path);
 680     (void) generate_normal_entry(false);
 681 
 682     return entry;
 683   }
 684 #endif // INCLUDE_ALL_GCS
 685 
 686   // If G1 is not enabled then attempt to go through the accessor entry point
 687   // Reference.get is an accessor
 688   return generate_jump_to_normal_entry();
 689 }
 690 
 691 /**
 692  * Method entry for static native methods:
 693  *   int java.util.zip.CRC32.update(int crc, int b)
 694  */
 695 address InterpreterGenerator::generate_CRC32_update_entry() {
 696   if (UseCRC32Intrinsics) {
 697     address entry = __ pc();
 698 
 699     // rbx,: Method*
 700     // r13: senderSP must preserved for slow path, set SP to it on fast path
 701     // c_rarg0: scratch (rdi on non-Win64, rcx on Win64)
 702     // c_rarg1: scratch (rsi on non-Win64, rdx on Win64)
 703 
 704     Label slow_path;
 705     // If we need a safepoint check, generate full interpreter entry.
 706     ExternalAddress state(SafepointSynchronize::address_of_state());
 707     __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
 708              SafepointSynchronize::_not_synchronized);
 709     __ jcc(Assembler::notEqual, slow_path);
 710 
 711     // We don't generate local frame and don't align stack because
 712     // we call stub code and there is no safepoint on this path.
 713 
 714     // Load parameters
 715     const Register crc = rax;  // crc
 716     const Register val = c_rarg0;  // source java byte value
 717     const Register tbl = c_rarg1;  // scratch
 718 
 719     // Arguments are reversed on java expression stack
 720     __ movl(val, Address(rsp,   wordSize)); // byte value
 721     __ movl(crc, Address(rsp, 2*wordSize)); // Initial CRC
 722 
 723     __ lea(tbl, ExternalAddress(StubRoutines::crc_table_addr()));
 724     __ notl(crc); // ~crc
 725     __ update_byte_crc32(crc, val, tbl);
 726     __ notl(crc); // ~crc
 727     // result in rax
 728 
 729     // _areturn
 730     __ pop(rdi);                // get return address
 731     __ mov(rsp, r13);           // set sp to sender sp
 732     __ jmp(rdi);
 733 
 734     // generate a vanilla native entry as the slow path
 735     __ bind(slow_path);
 736 
 737     (void) generate_native_entry(false);
 738 
 739     return entry;
 740   }
 741   return generate_native_entry(false);
 742 }
 743 
 744 /**
 745  * Method entry for static native methods:
 746  *   int java.util.zip.CRC32.updateBytes(int crc, byte[] b, int off, int len)
 747  *   int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
 748  */
 749 address InterpreterGenerator::generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
 750   if (UseCRC32Intrinsics) {
 751     address entry = __ pc();
 752 
 753     // rbx,: Method*
 754     // r13: senderSP must preserved for slow path, set SP to it on fast path
 755 
 756     Label slow_path;
 757     // If we need a safepoint check, generate full interpreter entry.
 758     ExternalAddress state(SafepointSynchronize::address_of_state());
 759     __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
 760              SafepointSynchronize::_not_synchronized);
 761     __ jcc(Assembler::notEqual, slow_path);
 762 
 763     // We don't generate local frame and don't align stack because
 764     // we call stub code and there is no safepoint on this path.
 765 
 766     // Load parameters
 767     const Register crc = c_rarg0;  // crc
 768     const Register buf = c_rarg1;  // source java byte array address
 769     const Register len = c_rarg2;  // length
 770     const Register off = len;      // offset (never overlaps with 'len')
 771 
 772     // Arguments are reversed on java expression stack
 773     // Calculate address of start element
 774     if (kind == Interpreter::java_util_zip_CRC32_updateByteBuffer) {
 775       __ movptr(buf, Address(rsp, 3*wordSize)); // long buf
 776       __ movl2ptr(off, Address(rsp, 2*wordSize)); // offset
 777       __ addq(buf, off); // + offset
 778       __ movl(crc,   Address(rsp, 5*wordSize)); // Initial CRC
 779     } else {
 780       __ movptr(buf, Address(rsp, 3*wordSize)); // byte[] array
 781       __ addptr(buf, arrayOopDesc::base_offset_in_bytes(T_BYTE)); // + header size
 782       __ movl2ptr(off, Address(rsp, 2*wordSize)); // offset
 783       __ addq(buf, off); // + offset
 784       __ movl(crc,   Address(rsp, 4*wordSize)); // Initial CRC
 785     }
 786     // Can now load 'len' since we're finished with 'off'
 787     __ movl(len, Address(rsp, wordSize)); // Length
 788 
 789     __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, StubRoutines::updateBytesCRC32()), crc, buf, len);
 790     // result in rax
 791 
 792     // _areturn
 793     __ pop(rdi);                // get return address
 794     __ mov(rsp, r13);           // set sp to sender sp
 795     __ jmp(rdi);
 796 
 797     // generate a vanilla native entry as the slow path
 798     __ bind(slow_path);
 799 
 800     (void) generate_native_entry(false);
 801 
 802     return entry;
 803   }
 804   return generate_native_entry(false);
 805 }
 806 
 807 /**
 808 * Method entry for static native methods:
 809 *   int java.util.zip.CRC32C.updateBytes(int crc, byte[] b, int off, int end)
 810 *   int java.util.zip.CRC32C.updateByteBuffer(int crc, long address, int off, int end)
 811 */
 812 address InterpreterGenerator::generate_CRC32C_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
 813   if (UseCRC32CIntrinsics) {
 814     address entry = __ pc();
 815     // Load parameters
 816     const Register crc = c_rarg0;  // crc
 817     const Register buf = c_rarg1;  // source java byte array address
 818     const Register len = c_rarg2;
 819     const Register off = c_rarg3;  // offset
 820     const Register end = len;
 821 
 822     // Arguments are reversed on java expression stack
 823     // Calculate address of start element
 824     if (kind == Interpreter::java_util_zip_CRC32C_updateDirectByteBuffer) {
 825       __ movptr(buf, Address(rsp, 3 * wordSize)); // long buf
 826       __ movl2ptr(off, Address(rsp, 2 * wordSize)); // offset
 827       __ addq(buf, off); // + offset
 828       __ movl(crc, Address(rsp, 5 * wordSize)); // Initial CRC
 829       // Note on 5 * wordSize vs. 4 * wordSize:
 830       // *   int java.util.zip.CRC32C.updateByteBuffer(int crc, long address, int off, int end)
 831       //                                                   4         2,3          1        0
 832       // end starts at SP + 8
 833       // The JavaŽ Virtual Machine Specification Java SE 7 Edition
 834       // 4.10.2.3. Values of Types long and double
 835       //    "When calculating operand stack length, values of type long and double have length two."
 836     } else {
 837       __ movptr(buf, Address(rsp, 3 * wordSize)); // byte[] array
 838       __ addptr(buf, arrayOopDesc::base_offset_in_bytes(T_BYTE)); // + header size
 839       __ movl2ptr(off, Address(rsp, 2 * wordSize)); // offset
 840       __ addq(buf, off); // + offset
 841       __ movl(crc, Address(rsp, 4 * wordSize)); // Initial CRC
 842     }
 843     __ movl(end, Address(rsp, wordSize)); // end
 844     __ subl(end, off); // end - off
 845     __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, StubRoutines::updateBytesCRC32C()), crc, buf, len);
 846     // result in rax
 847     // _areturn
 848     __ pop(rdi);                // get return address
 849     __ mov(rsp, r13);           // set sp to sender sp
 850     __ jmp(rdi);
 851 
 852     return entry;
 853   }
 854 
 855   return generate_native_entry(false);
 856 }
 857 
 858 // Interpreter stub for calling a native method. (asm interpreter)
 859 // This sets up a somewhat different looking stack for calling the
 860 // native method than the typical interpreter frame setup.
 861 address InterpreterGenerator::generate_native_entry(bool synchronized) {
 862   // determine code generation flags
 863   bool inc_counter  = UseCompiler || CountCompiledCalls;
 864 
 865   // rbx: Method*
 866   // r13: sender sp
 867 
 868   address entry_point = __ pc();
 869 
 870   const Address constMethod       (rbx, Method::const_offset());
 871   const Address access_flags      (rbx, Method::access_flags_offset());
 872   const Address size_of_parameters(rcx, ConstMethod::
 873                                         size_of_parameters_offset());
 874 
 875 
 876   // get parameter size (always needed)
 877   __ movptr(rcx, constMethod);
 878   __ load_unsigned_short(rcx, size_of_parameters);
 879 
 880   // native calls don't need the stack size check since they have no
 881   // expression stack and the arguments are already on the stack and
 882   // we only add a handful of words to the stack
 883 
 884   // rbx: Method*
 885   // rcx: size of parameters
 886   // r13: sender sp
 887   __ pop(rax);                                       // get return address
 888 
 889   // for natives the size of locals is zero
 890 
 891   // compute beginning of parameters (r14)
 892   __ lea(r14, Address(rsp, rcx, Address::times_8, -wordSize));
 893 
 894   // add 2 zero-initialized slots for native calls
 895   // initialize result_handler slot
 896   __ push((int) NULL_WORD);
 897   // slot for oop temp
 898   // (static native method holder mirror/jni oop result)
 899   __ push((int) NULL_WORD);
 900 
 901   // initialize fixed part of activation frame
 902   generate_fixed_frame(true);
 903 
 904   // make sure method is native & not abstract
 905 #ifdef ASSERT
 906   __ movl(rax, access_flags);
 907   {
 908     Label L;
 909     __ testl(rax, JVM_ACC_NATIVE);
 910     __ jcc(Assembler::notZero, L);
 911     __ stop("tried to execute non-native method as native");
 912     __ bind(L);
 913   }
 914   {
 915     Label L;
 916     __ testl(rax, JVM_ACC_ABSTRACT);
 917     __ jcc(Assembler::zero, L);
 918     __ stop("tried to execute abstract method in interpreter");
 919     __ bind(L);
 920   }
 921 #endif
 922 
 923   // Since at this point in the method invocation the exception handler
 924   // would try to exit the monitor of synchronized methods which hasn't
 925   // been entered yet, we set the thread local variable
 926   // _do_not_unlock_if_synchronized to true. The remove_activation will
 927   // check this flag.
 928 
 929   const Address do_not_unlock_if_synchronized(r15_thread,
 930         in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
 931   __ movbool(do_not_unlock_if_synchronized, true);
 932 
 933   // increment invocation count & check for overflow
 934   Label invocation_counter_overflow;
 935   if (inc_counter) {
 936     generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
 937   }
 938 
 939   Label continue_after_compile;
 940   __ bind(continue_after_compile);
 941 
 942   bang_stack_shadow_pages(true);
 943 
 944   // reset the _do_not_unlock_if_synchronized flag
 945   __ movbool(do_not_unlock_if_synchronized, false);
 946 
 947   // check for synchronized methods
 948   // Must happen AFTER invocation_counter check and stack overflow check,
 949   // so method is not locked if overflows.
 950   if (synchronized) {
 951     lock_method();
 952   } else {
 953     // no synchronization necessary
 954 #ifdef ASSERT
 955     {
 956       Label L;
 957       __ movl(rax, access_flags);
 958       __ testl(rax, JVM_ACC_SYNCHRONIZED);
 959       __ jcc(Assembler::zero, L);
 960       __ stop("method needs synchronization");
 961       __ bind(L);
 962     }
 963 #endif
 964   }
 965 
 966   // start execution
 967 #ifdef ASSERT
 968   {
 969     Label L;
 970     const Address monitor_block_top(rbp,
 971                  frame::interpreter_frame_monitor_block_top_offset * wordSize);
 972     __ movptr(rax, monitor_block_top);
 973     __ cmpptr(rax, rsp);
 974     __ jcc(Assembler::equal, L);
 975     __ stop("broken stack frame setup in interpreter");
 976     __ bind(L);
 977   }
 978 #endif
 979 
 980   // jvmti support
 981   __ notify_method_entry();
 982 
 983   // work registers
 984   const Register method = rbx;
 985   const Register t      = r11;
 986 
 987   // allocate space for parameters
 988   __ get_method(method);
 989   __ movptr(t, Address(method, Method::const_offset()));
 990   __ load_unsigned_short(t, Address(t, ConstMethod::size_of_parameters_offset()));
 991   __ shll(t, Interpreter::logStackElementSize);
 992 
 993   __ subptr(rsp, t);
 994   __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
 995   __ andptr(rsp, -16); // must be 16 byte boundary (see amd64 ABI)
 996 
 997   // get signature handler
 998   {
 999     Label L;
1000     __ movptr(t, Address(method, Method::signature_handler_offset()));
1001     __ testptr(t, t);
1002     __ jcc(Assembler::notZero, L);
1003     __ call_VM(noreg,
1004                CAST_FROM_FN_PTR(address,
1005                                 InterpreterRuntime::prepare_native_call),
1006                method);
1007     __ get_method(method);
1008     __ movptr(t, Address(method, Method::signature_handler_offset()));
1009     __ bind(L);
1010   }
1011 
1012   // call signature handler
1013   assert(InterpreterRuntime::SignatureHandlerGenerator::from() == r14,
1014          "adjust this code");
1015   assert(InterpreterRuntime::SignatureHandlerGenerator::to() == rsp,
1016          "adjust this code");
1017   assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == rscratch1,
1018           "adjust this code");
1019 
1020   // The generated handlers do not touch RBX (the method oop).
1021   // However, large signatures cannot be cached and are generated
1022   // each time here.  The slow-path generator can do a GC on return,
1023   // so we must reload it after the call.
1024   __ call(t);
1025   __ get_method(method);        // slow path can do a GC, reload RBX
1026 
1027 
1028   // result handler is in rax
1029   // set result handler
1030   __ movptr(Address(rbp,
1031                     (frame::interpreter_frame_result_handler_offset) * wordSize),
1032             rax);
1033 
1034   // pass mirror handle if static call
1035   {
1036     Label L;
1037     const int mirror_offset = in_bytes(Klass::java_mirror_offset());
1038     __ movl(t, Address(method, Method::access_flags_offset()));
1039     __ testl(t, JVM_ACC_STATIC);
1040     __ jcc(Assembler::zero, L);
1041     // get mirror
1042     __ movptr(t, Address(method, Method::const_offset()));
1043     __ movptr(t, Address(t, ConstMethod::constants_offset()));
1044     __ movptr(t, Address(t, ConstantPool::pool_holder_offset_in_bytes()));
1045     __ movptr(t, Address(t, mirror_offset));
1046     // copy mirror into activation frame
1047     __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize),
1048             t);
1049     // pass handle to mirror
1050     __ lea(c_rarg1,
1051            Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize));
1052     __ bind(L);
1053   }
1054 
1055   // get native function entry point
1056   {
1057     Label L;
1058     __ movptr(rax, Address(method, Method::native_function_offset()));
1059     ExternalAddress unsatisfied(SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
1060     __ movptr(rscratch2, unsatisfied.addr());
1061     __ cmpptr(rax, rscratch2);
1062     __ jcc(Assembler::notEqual, L);
1063     __ call_VM(noreg,
1064                CAST_FROM_FN_PTR(address,
1065                                 InterpreterRuntime::prepare_native_call),
1066                method);
1067     __ get_method(method);
1068     __ movptr(rax, Address(method, Method::native_function_offset()));
1069     __ bind(L);
1070   }
1071 
1072   // pass JNIEnv
1073   __ lea(c_rarg0, Address(r15_thread, JavaThread::jni_environment_offset()));
1074 
1075   // It is enough that the pc() points into the right code
1076   // segment. It does not have to be the correct return pc.
1077   __ set_last_Java_frame(rsp, rbp, (address) __ pc());
1078 
1079   // change thread state
1080 #ifdef ASSERT
1081   {
1082     Label L;
1083     __ movl(t, Address(r15_thread, JavaThread::thread_state_offset()));
1084     __ cmpl(t, _thread_in_Java);
1085     __ jcc(Assembler::equal, L);
1086     __ stop("Wrong thread state in native stub");
1087     __ bind(L);
1088   }
1089 #endif
1090 
1091   // Change state to native
1092 
1093   __ movl(Address(r15_thread, JavaThread::thread_state_offset()),
1094           _thread_in_native);
1095 
1096   // Call the native method.
1097   __ call(rax);
1098   // result potentially in rax or xmm0
1099 
1100   // Verify or restore cpu control state after JNI call
1101   __ restore_cpu_control_state_after_jni();
1102 
1103   // NOTE: The order of these pushes is known to frame::interpreter_frame_result
1104   // in order to extract the result of a method call. If the order of these
1105   // pushes change or anything else is added to the stack then the code in
1106   // interpreter_frame_result must also change.
1107 
1108   __ push(dtos);
1109   __ push(ltos);
1110 
1111   // change thread state
1112   __ movl(Address(r15_thread, JavaThread::thread_state_offset()),
1113           _thread_in_native_trans);
1114 
1115   if (os::is_MP()) {
1116     if (UseMembar) {
1117       // Force this write out before the read below
1118       __ membar(Assembler::Membar_mask_bits(
1119            Assembler::LoadLoad | Assembler::LoadStore |
1120            Assembler::StoreLoad | Assembler::StoreStore));
1121     } else {
1122       // Write serialization page so VM thread can do a pseudo remote membar.
1123       // We use the current thread pointer to calculate a thread specific
1124       // offset to write to within the page. This minimizes bus traffic
1125       // due to cache line collision.
1126       __ serialize_memory(r15_thread, rscratch2);
1127     }
1128   }
1129 
1130   // check for safepoint operation in progress and/or pending suspend requests
1131   {
1132     Label Continue;
1133     __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
1134              SafepointSynchronize::_not_synchronized);
1135 
1136     Label L;
1137     __ jcc(Assembler::notEqual, L);
1138     __ cmpl(Address(r15_thread, JavaThread::suspend_flags_offset()), 0);
1139     __ jcc(Assembler::equal, Continue);
1140     __ bind(L);
1141 
1142     // Don't use call_VM as it will see a possible pending exception
1143     // and forward it and never return here preventing us from
1144     // clearing _last_native_pc down below.  Also can't use
1145     // call_VM_leaf either as it will check to see if r13 & r14 are
1146     // preserved and correspond to the bcp/locals pointers. So we do a
1147     // runtime call by hand.
1148     //
1149     __ mov(c_rarg0, r15_thread);
1150     __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1151     __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1152     __ andptr(rsp, -16); // align stack as required by ABI
1153     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans)));
1154     __ mov(rsp, r12); // restore sp
1155     __ reinit_heapbase();
1156     __ bind(Continue);
1157   }
1158 
1159   // change thread state
1160   __ movl(Address(r15_thread, JavaThread::thread_state_offset()), _thread_in_Java);
1161 
1162   // reset_last_Java_frame
1163   __ reset_last_Java_frame(true, true);
1164 
1165   // reset handle block
1166   __ movptr(t, Address(r15_thread, JavaThread::active_handles_offset()));
1167   __ movl(Address(t, JNIHandleBlock::top_offset_in_bytes()), (int32_t)NULL_WORD);
1168 
1169   // If result is an oop unbox and store it in frame where gc will see it
1170   // and result handler will pick it up
1171 
1172   {
1173     Label no_oop, store_result;
1174     __ lea(t, ExternalAddress(AbstractInterpreter::result_handler(T_OBJECT)));
1175     __ cmpptr(t, Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize));
1176     __ jcc(Assembler::notEqual, no_oop);
1177     // retrieve result
1178     __ pop(ltos);
1179     __ testptr(rax, rax);
1180     __ jcc(Assembler::zero, store_result);
1181     __ movptr(rax, Address(rax, 0));
1182     __ bind(store_result);
1183     __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize), rax);
1184     // keep stack depth as expected by pushing oop which will eventually be discarde
1185     __ push(ltos);
1186     __ bind(no_oop);
1187   }
1188 
1189 
1190   {
1191     Label no_reguard;
1192     __ cmpl(Address(r15_thread, JavaThread::stack_guard_state_offset()),
1193             JavaThread::stack_guard_yellow_disabled);
1194     __ jcc(Assembler::notEqual, no_reguard);
1195 
1196     __ pusha(); // XXX only save smashed registers
1197     __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1198     __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1199     __ andptr(rsp, -16); // align stack as required by ABI
1200     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
1201     __ mov(rsp, r12); // restore sp
1202     __ popa(); // XXX only restore smashed registers
1203     __ reinit_heapbase();
1204 
1205     __ bind(no_reguard);
1206   }
1207 
1208 
1209   // The method register is junk from after the thread_in_native transition
1210   // until here.  Also can't call_VM until the bcp has been
1211   // restored.  Need bcp for throwing exception below so get it now.
1212   __ get_method(method);
1213 
1214   // restore r13 to have legal interpreter frame, i.e., bci == 0 <=>
1215   // r13 == code_base()
1216   __ movptr(r13, Address(method, Method::const_offset()));   // get ConstMethod*
1217   __ lea(r13, Address(r13, ConstMethod::codes_offset()));    // get codebase
1218   // handle exceptions (exception handling will handle unlocking!)
1219   {
1220     Label L;
1221     __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t) NULL_WORD);
1222     __ jcc(Assembler::zero, L);
1223     // Note: At some point we may want to unify this with the code
1224     // used in call_VM_base(); i.e., we should use the
1225     // StubRoutines::forward_exception code. For now this doesn't work
1226     // here because the rsp is not correctly set at this point.
1227     __ MacroAssembler::call_VM(noreg,
1228                                CAST_FROM_FN_PTR(address,
1229                                InterpreterRuntime::throw_pending_exception));
1230     __ should_not_reach_here();
1231     __ bind(L);
1232   }
1233 
1234   // do unlocking if necessary
1235   {
1236     Label L;
1237     __ movl(t, Address(method, Method::access_flags_offset()));
1238     __ testl(t, JVM_ACC_SYNCHRONIZED);
1239     __ jcc(Assembler::zero, L);
1240     // the code below should be shared with interpreter macro
1241     // assembler implementation
1242     {
1243       Label unlock;
1244       // BasicObjectLock will be first in list, since this is a
1245       // synchronized method. However, need to check that the object
1246       // has not been unlocked by an explicit monitorexit bytecode.
1247       const Address monitor(rbp,
1248                             (intptr_t)(frame::interpreter_frame_initial_sp_offset *
1249                                        wordSize - sizeof(BasicObjectLock)));
1250 
1251       // monitor expect in c_rarg1 for slow unlock path
1252       __ lea(c_rarg1, monitor); // address of first monitor
1253 
1254       __ movptr(t, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
1255       __ testptr(t, t);
1256       __ jcc(Assembler::notZero, unlock);
1257 
1258       // Entry already unlocked, need to throw exception
1259       __ MacroAssembler::call_VM(noreg,
1260                                  CAST_FROM_FN_PTR(address,
1261                    InterpreterRuntime::throw_illegal_monitor_state_exception));
1262       __ should_not_reach_here();
1263 
1264       __ bind(unlock);
1265       __ unlock_object(c_rarg1);
1266     }
1267     __ bind(L);
1268   }
1269 
1270   // jvmti support
1271   // Note: This must happen _after_ handling/throwing any exceptions since
1272   //       the exception handler code notifies the runtime of method exits
1273   //       too. If this happens before, method entry/exit notifications are
1274   //       not properly paired (was bug - gri 11/22/99).
1275   __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI);
1276 
1277   // restore potential result in edx:eax, call result handler to
1278   // restore potential result in ST0 & handle result
1279 
1280   __ pop(ltos);
1281   __ pop(dtos);
1282 
1283   __ movptr(t, Address(rbp,
1284                        (frame::interpreter_frame_result_handler_offset) * wordSize));
1285   __ call(t);
1286 
1287   // remove activation
1288   __ movptr(t, Address(rbp,
1289                        frame::interpreter_frame_sender_sp_offset *
1290                        wordSize)); // get sender sp
1291   __ leave();                                // remove frame anchor
1292   __ pop(rdi);                               // get return address
1293   __ mov(rsp, t);                            // set sp to sender sp
1294   __ jmp(rdi);
1295 
1296   if (inc_counter) {
1297     // Handle overflow of counter and compile method
1298     __ bind(invocation_counter_overflow);
1299     generate_counter_overflow(&continue_after_compile);
1300   }
1301 
1302   return entry_point;
1303 }
1304 
1305 //
1306 // Generic interpreted method entry to (asm) interpreter
1307 //
1308 address InterpreterGenerator::generate_normal_entry(bool synchronized) {
1309   // determine code generation flags
1310   bool inc_counter  = UseCompiler || CountCompiledCalls;
1311 
1312   // ebx: Method*
1313   // r13: sender sp
1314   address entry_point = __ pc();
1315 
1316   const Address constMethod(rbx, Method::const_offset());
1317   const Address access_flags(rbx, Method::access_flags_offset());
1318   const Address size_of_parameters(rdx,
1319                                    ConstMethod::size_of_parameters_offset());
1320   const Address size_of_locals(rdx, ConstMethod::size_of_locals_offset());
1321 
1322 
1323   // get parameter size (always needed)
1324   __ movptr(rdx, constMethod);
1325   __ load_unsigned_short(rcx, size_of_parameters);
1326 
1327   // rbx: Method*
1328   // rcx: size of parameters
1329   // r13: sender_sp (could differ from sp+wordSize if we were called via c2i )
1330 
1331   __ load_unsigned_short(rdx, size_of_locals); // get size of locals in words
1332   __ subl(rdx, rcx); // rdx = no. of additional locals
1333 
1334   // YYY
1335 //   __ incrementl(rdx);
1336 //   __ andl(rdx, -2);
1337 
1338   // see if we've got enough room on the stack for locals plus overhead.
1339   generate_stack_overflow_check();
1340 
1341   // get return address
1342   __ pop(rax);
1343 
1344   // compute beginning of parameters (r14)
1345   __ lea(r14, Address(rsp, rcx, Address::times_8, -wordSize));
1346 
1347   // rdx - # of additional locals
1348   // allocate space for locals
1349   // explicitly initialize locals
1350   {
1351     Label exit, loop;
1352     __ testl(rdx, rdx);
1353     __ jcc(Assembler::lessEqual, exit); // do nothing if rdx <= 0
1354     __ bind(loop);
1355     __ push((int) NULL_WORD); // initialize local variables
1356     __ decrementl(rdx); // until everything initialized
1357     __ jcc(Assembler::greater, loop);
1358     __ bind(exit);
1359   }
1360 
1361   // initialize fixed part of activation frame
1362   generate_fixed_frame(false);
1363 
1364   // make sure method is not native & not abstract
1365 #ifdef ASSERT
1366   __ movl(rax, access_flags);
1367   {
1368     Label L;
1369     __ testl(rax, JVM_ACC_NATIVE);
1370     __ jcc(Assembler::zero, L);
1371     __ stop("tried to execute native method as non-native");
1372     __ bind(L);
1373   }
1374   {
1375     Label L;
1376     __ testl(rax, JVM_ACC_ABSTRACT);
1377     __ jcc(Assembler::zero, L);
1378     __ stop("tried to execute abstract method in interpreter");
1379     __ bind(L);
1380   }
1381 #endif
1382 
1383   // Since at this point in the method invocation the exception
1384   // handler would try to exit the monitor of synchronized methods
1385   // which hasn't been entered yet, we set the thread local variable
1386   // _do_not_unlock_if_synchronized to true. The remove_activation
1387   // will check this flag.
1388 
1389   const Address do_not_unlock_if_synchronized(r15_thread,
1390         in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1391   __ movbool(do_not_unlock_if_synchronized, true);
1392 
1393   __ profile_parameters_type(rax, rcx, rdx);
1394   // increment invocation count & check for overflow
1395   Label invocation_counter_overflow;
1396   Label profile_method;
1397   Label profile_method_continue;
1398   if (inc_counter) {
1399     generate_counter_incr(&invocation_counter_overflow,
1400                           &profile_method,
1401                           &profile_method_continue);
1402     if (ProfileInterpreter) {
1403       __ bind(profile_method_continue);
1404     }
1405   }
1406 
1407   Label continue_after_compile;
1408   __ bind(continue_after_compile);
1409 
1410   // check for synchronized interpreted methods
1411   bang_stack_shadow_pages(false);
1412 
1413   // reset the _do_not_unlock_if_synchronized flag
1414   __ movbool(do_not_unlock_if_synchronized, false);
1415 
1416   // check for synchronized methods
1417   // Must happen AFTER invocation_counter check and stack overflow check,
1418   // so method is not locked if overflows.
1419   if (synchronized) {
1420     // Allocate monitor and lock method
1421     lock_method();
1422   } else {
1423     // no synchronization necessary
1424 #ifdef ASSERT
1425     {
1426       Label L;
1427       __ movl(rax, access_flags);
1428       __ testl(rax, JVM_ACC_SYNCHRONIZED);
1429       __ jcc(Assembler::zero, L);
1430       __ stop("method needs synchronization");
1431       __ bind(L);
1432     }
1433 #endif
1434   }
1435 
1436   // start execution
1437 #ifdef ASSERT
1438   {
1439     Label L;
1440      const Address monitor_block_top (rbp,
1441                  frame::interpreter_frame_monitor_block_top_offset * wordSize);
1442     __ movptr(rax, monitor_block_top);
1443     __ cmpptr(rax, rsp);
1444     __ jcc(Assembler::equal, L);
1445     __ stop("broken stack frame setup in interpreter");
1446     __ bind(L);
1447   }
1448 #endif
1449 
1450   // jvmti support
1451   __ notify_method_entry();
1452 
1453   __ dispatch_next(vtos);
1454 
1455   // invocation counter overflow
1456   if (inc_counter) {
1457     if (ProfileInterpreter) {
1458       // We have decided to profile this method in the interpreter
1459       __ bind(profile_method);
1460       __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1461       __ set_method_data_pointer_for_bcp();
1462       __ get_method(rbx);
1463       __ jmp(profile_method_continue);
1464     }
1465     // Handle overflow of counter and compile method
1466     __ bind(invocation_counter_overflow);
1467     generate_counter_overflow(&continue_after_compile);
1468   }
1469 
1470   return entry_point;
1471 }
1472 
1473 
1474 // These should never be compiled since the interpreter will prefer
1475 // the compiled version to the intrinsic version.
1476 bool AbstractInterpreter::can_be_compiled(methodHandle m) {
1477   switch (method_kind(m)) {
1478     case Interpreter::java_lang_math_sin     : // fall thru
1479     case Interpreter::java_lang_math_cos     : // fall thru
1480     case Interpreter::java_lang_math_tan     : // fall thru
1481     case Interpreter::java_lang_math_abs     : // fall thru
1482     case Interpreter::java_lang_math_log     : // fall thru
1483     case Interpreter::java_lang_math_log10   : // fall thru
1484     case Interpreter::java_lang_math_sqrt    : // fall thru
1485     case Interpreter::java_lang_math_pow     : // fall thru
1486     case Interpreter::java_lang_math_exp     :
1487       return false;
1488     default:
1489       return true;
1490   }
1491 }
1492 
1493 // How much stack a method activation needs in words.
1494 int AbstractInterpreter::size_top_interpreter_activation(Method* method) {
1495   const int entry_size = frame::interpreter_frame_monitor_size();
1496 
1497   // total overhead size: entry_size + (saved rbp thru expr stack
1498   // bottom).  be sure to change this if you add/subtract anything
1499   // to/from the overhead area
1500   const int overhead_size =
1501     -(frame::interpreter_frame_initial_sp_offset) + entry_size;
1502 
1503   const int stub_code = frame::entry_frame_after_call_words;
1504   const int method_stack = (method->max_locals() + method->max_stack()) *
1505                            Interpreter::stackElementWords;
1506   return (overhead_size + method_stack + stub_code);
1507 }
1508 
1509 //-----------------------------------------------------------------------------
1510 // Exceptions
1511 
1512 void TemplateInterpreterGenerator::generate_throw_exception() {
1513   // Entry point in previous activation (i.e., if the caller was
1514   // interpreted)
1515   Interpreter::_rethrow_exception_entry = __ pc();
1516   // Restore sp to interpreter_frame_last_sp even though we are going
1517   // to empty the expression stack for the exception processing.
1518   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
1519   // rax: exception
1520   // rdx: return address/pc that threw exception
1521   __ restore_bcp();    // r13 points to call/send
1522   __ restore_locals();
1523   __ reinit_heapbase();  // restore r12 as heapbase.
1524   // Entry point for exceptions thrown within interpreter code
1525   Interpreter::_throw_exception_entry = __ pc();
1526   // expression stack is undefined here
1527   // rax: exception
1528   // r13: exception bcp
1529   __ verify_oop(rax);
1530   __ mov(c_rarg1, rax);
1531 
1532   // expression stack must be empty before entering the VM in case of
1533   // an exception
1534   __ empty_expression_stack();
1535   // find exception handler address and preserve exception oop
1536   __ call_VM(rdx,
1537              CAST_FROM_FN_PTR(address,
1538                           InterpreterRuntime::exception_handler_for_exception),
1539              c_rarg1);
1540   // rax: exception handler entry point
1541   // rdx: preserved exception oop
1542   // r13: bcp for exception handler
1543   __ push_ptr(rdx); // push exception which is now the only value on the stack
1544   __ jmp(rax); // jump to exception handler (may be _remove_activation_entry!)
1545 
1546   // If the exception is not handled in the current frame the frame is
1547   // removed and the exception is rethrown (i.e. exception
1548   // continuation is _rethrow_exception).
1549   //
1550   // Note: At this point the bci is still the bxi for the instruction
1551   // which caused the exception and the expression stack is
1552   // empty. Thus, for any VM calls at this point, GC will find a legal
1553   // oop map (with empty expression stack).
1554 
1555   // In current activation
1556   // tos: exception
1557   // esi: exception bcp
1558 
1559   //
1560   // JVMTI PopFrame support
1561   //
1562 
1563   Interpreter::_remove_activation_preserving_args_entry = __ pc();
1564   __ empty_expression_stack();
1565   // Set the popframe_processing bit in pending_popframe_condition
1566   // indicating that we are currently handling popframe, so that
1567   // call_VMs that may happen later do not trigger new popframe
1568   // handling cycles.
1569   __ movl(rdx, Address(r15_thread, JavaThread::popframe_condition_offset()));
1570   __ orl(rdx, JavaThread::popframe_processing_bit);
1571   __ movl(Address(r15_thread, JavaThread::popframe_condition_offset()), rdx);
1572 
1573   {
1574     // Check to see whether we are returning to a deoptimized frame.
1575     // (The PopFrame call ensures that the caller of the popped frame is
1576     // either interpreted or compiled and deoptimizes it if compiled.)
1577     // In this case, we can't call dispatch_next() after the frame is
1578     // popped, but instead must save the incoming arguments and restore
1579     // them after deoptimization has occurred.
1580     //
1581     // Note that we don't compare the return PC against the
1582     // deoptimization blob's unpack entry because of the presence of
1583     // adapter frames in C2.
1584     Label caller_not_deoptimized;
1585     __ movptr(c_rarg1, Address(rbp, frame::return_addr_offset * wordSize));
1586     __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1587                                InterpreterRuntime::interpreter_contains), c_rarg1);
1588     __ testl(rax, rax);
1589     __ jcc(Assembler::notZero, caller_not_deoptimized);
1590 
1591     // Compute size of arguments for saving when returning to
1592     // deoptimized caller
1593     __ get_method(rax);
1594     __ movptr(rax, Address(rax, Method::const_offset()));
1595     __ load_unsigned_short(rax, Address(rax, in_bytes(ConstMethod::
1596                                                 size_of_parameters_offset())));
1597     __ shll(rax, Interpreter::logStackElementSize);
1598     __ restore_locals(); // XXX do we need this?
1599     __ subptr(r14, rax);
1600     __ addptr(r14, wordSize);
1601     // Save these arguments
1602     __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1603                                            Deoptimization::
1604                                            popframe_preserve_args),
1605                           r15_thread, rax, r14);
1606 
1607     __ remove_activation(vtos, rdx,
1608                          /* throw_monitor_exception */ false,
1609                          /* install_monitor_exception */ false,
1610                          /* notify_jvmdi */ false);
1611 
1612     // Inform deoptimization that it is responsible for restoring
1613     // these arguments
1614     __ movl(Address(r15_thread, JavaThread::popframe_condition_offset()),
1615             JavaThread::popframe_force_deopt_reexecution_bit);
1616 
1617     // Continue in deoptimization handler
1618     __ jmp(rdx);
1619 
1620     __ bind(caller_not_deoptimized);
1621   }
1622 
1623   __ remove_activation(vtos, rdx, /* rdx result (retaddr) is not used */
1624                        /* throw_monitor_exception */ false,
1625                        /* install_monitor_exception */ false,
1626                        /* notify_jvmdi */ false);
1627 
1628   // Finish with popframe handling
1629   // A previous I2C followed by a deoptimization might have moved the
1630   // outgoing arguments further up the stack. PopFrame expects the
1631   // mutations to those outgoing arguments to be preserved and other
1632   // constraints basically require this frame to look exactly as
1633   // though it had previously invoked an interpreted activation with
1634   // no space between the top of the expression stack (current
1635   // last_sp) and the top of stack. Rather than force deopt to
1636   // maintain this kind of invariant all the time we call a small
1637   // fixup routine to move the mutated arguments onto the top of our
1638   // expression stack if necessary.
1639   __ mov(c_rarg1, rsp);
1640   __ movptr(c_rarg2, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1641   // PC must point into interpreter here
1642   __ set_last_Java_frame(noreg, rbp, __ pc());
1643   __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), r15_thread, c_rarg1, c_rarg2);
1644   __ reset_last_Java_frame(true, true);
1645   // Restore the last_sp and null it out
1646   __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1647   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
1648 
1649   __ restore_bcp();  // XXX do we need this?
1650   __ restore_locals(); // XXX do we need this?
1651   // The method data pointer was incremented already during
1652   // call profiling. We have to restore the mdp for the current bcp.
1653   if (ProfileInterpreter) {
1654     __ set_method_data_pointer_for_bcp();
1655   }
1656 
1657   // Clear the popframe condition flag
1658   __ movl(Address(r15_thread, JavaThread::popframe_condition_offset()),
1659           JavaThread::popframe_inactive);
1660 
1661 #if INCLUDE_JVMTI
1662   {
1663     Label L_done;
1664     const Register local0 = r14;
1665 
1666     __ cmpb(Address(r13, 0), Bytecodes::_invokestatic);
1667     __ jcc(Assembler::notEqual, L_done);
1668 
1669     // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call.
1670     // Detect such a case in the InterpreterRuntime function and return the member name argument, or NULL.
1671 
1672     __ get_method(rdx);
1673     __ movptr(rax, Address(local0, 0));
1674     __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), rax, rdx, r13);
1675 
1676     __ testptr(rax, rax);
1677     __ jcc(Assembler::zero, L_done);
1678 
1679     __ movptr(Address(rbx, 0), rax);
1680     __ bind(L_done);
1681   }
1682 #endif // INCLUDE_JVMTI
1683 
1684   __ dispatch_next(vtos);
1685   // end of PopFrame support
1686 
1687   Interpreter::_remove_activation_entry = __ pc();
1688 
1689   // preserve exception over this code sequence
1690   __ pop_ptr(rax);
1691   __ movptr(Address(r15_thread, JavaThread::vm_result_offset()), rax);
1692   // remove the activation (without doing throws on illegalMonitorExceptions)
1693   __ remove_activation(vtos, rdx, false, true, false);
1694   // restore exception
1695   __ get_vm_result(rax, r15_thread);
1696 
1697   // In between activations - previous activation type unknown yet
1698   // compute continuation point - the continuation point expects the
1699   // following registers set up:
1700   //
1701   // rax: exception
1702   // rdx: return address/pc that threw exception
1703   // rsp: expression stack of caller
1704   // rbp: ebp of caller
1705   __ push(rax);                                  // save exception
1706   __ push(rdx);                                  // save return address
1707   __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1708                           SharedRuntime::exception_handler_for_return_address),
1709                         r15_thread, rdx);
1710   __ mov(rbx, rax);                              // save exception handler
1711   __ pop(rdx);                                   // restore return address
1712   __ pop(rax);                                   // restore exception
1713   // Note that an "issuing PC" is actually the next PC after the call
1714   __ jmp(rbx);                                   // jump to exception
1715                                                  // handler of caller
1716 }
1717 
1718 
1719 //
1720 // JVMTI ForceEarlyReturn support
1721 //
1722 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
1723   address entry = __ pc();
1724 
1725   __ restore_bcp();
1726   __ restore_locals();
1727   __ empty_expression_stack();
1728   __ load_earlyret_value(state);
1729 
1730   __ movptr(rdx, Address(r15_thread, JavaThread::jvmti_thread_state_offset()));
1731   Address cond_addr(rdx, JvmtiThreadState::earlyret_state_offset());
1732 
1733   // Clear the earlyret state
1734   __ movl(cond_addr, JvmtiThreadState::earlyret_inactive);
1735 
1736   __ remove_activation(state, rsi,
1737                        false, /* throw_monitor_exception */
1738                        false, /* install_monitor_exception */
1739                        true); /* notify_jvmdi */
1740   __ jmp(rsi);
1741 
1742   return entry;
1743 } // end of ForceEarlyReturn support
1744 
1745 
1746 //-----------------------------------------------------------------------------
1747 // Helper for vtos entry point generation
1748 
1749 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
1750                                                          address& bep,
1751                                                          address& cep,
1752                                                          address& sep,
1753                                                          address& aep,
1754                                                          address& iep,
1755                                                          address& lep,
1756                                                          address& fep,
1757                                                          address& dep,
1758                                                          address& vep) {
1759   assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
1760   Label L;
1761   aep = __ pc();  __ push_ptr();  __ jmp(L);
1762   fep = __ pc();  __ push_f();    __ jmp(L);
1763   dep = __ pc();  __ push_d();    __ jmp(L);
1764   lep = __ pc();  __ push_l();    __ jmp(L);
1765   bep = cep = sep =
1766   iep = __ pc();  __ push_i();
1767   vep = __ pc();
1768   __ bind(L);
1769   generate_and_dispatch(t);
1770 }
1771 
1772 
1773 //-----------------------------------------------------------------------------
1774 // Generation of individual instructions
1775 
1776 // helpers for generate_and_dispatch
1777 
1778 
1779 InterpreterGenerator::InterpreterGenerator(StubQueue* code)
1780   : TemplateInterpreterGenerator(code) {
1781    generate_all(); // down here so it can be "virtual"
1782 }
1783 
1784 //-----------------------------------------------------------------------------
1785 
1786 // Non-product code
1787 #ifndef PRODUCT
1788 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
1789   address entry = __ pc();
1790 
1791   __ push(state);
1792   __ push(c_rarg0);
1793   __ push(c_rarg1);
1794   __ push(c_rarg2);
1795   __ push(c_rarg3);
1796   __ mov(c_rarg2, rax);  // Pass itos
1797 #ifdef _WIN64
1798   __ movflt(xmm3, xmm0); // Pass ftos
1799 #endif
1800   __ call_VM(noreg,
1801              CAST_FROM_FN_PTR(address, SharedRuntime::trace_bytecode),
1802              c_rarg1, c_rarg2, c_rarg3);
1803   __ pop(c_rarg3);
1804   __ pop(c_rarg2);
1805   __ pop(c_rarg1);
1806   __ pop(c_rarg0);
1807   __ pop(state);
1808   __ ret(0);                                   // return from result handler
1809 
1810   return entry;
1811 }
1812 
1813 void TemplateInterpreterGenerator::count_bytecode() {
1814   __ incrementl(ExternalAddress((address) &BytecodeCounter::_counter_value));
1815 }
1816 
1817 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
1818   __ incrementl(ExternalAddress((address) &BytecodeHistogram::_counters[t->bytecode()]));
1819 }
1820 
1821 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
1822   __ mov32(rbx, ExternalAddress((address) &BytecodePairHistogram::_index));
1823   __ shrl(rbx, BytecodePairHistogram::log2_number_of_codes);
1824   __ orl(rbx,
1825          ((int) t->bytecode()) <<
1826          BytecodePairHistogram::log2_number_of_codes);
1827   __ mov32(ExternalAddress((address) &BytecodePairHistogram::_index), rbx);
1828   __ lea(rscratch1, ExternalAddress((address) BytecodePairHistogram::_counters));
1829   __ incrementl(Address(rscratch1, rbx, Address::times_4));
1830 }
1831 
1832 
1833 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
1834   // Call a little run-time stub to avoid blow-up for each bytecode.
1835   // The run-time runtime saves the right registers, depending on
1836   // the tosca in-state for the given template.
1837 
1838   assert(Interpreter::trace_code(t->tos_in()) != NULL,
1839          "entry must have been generated");
1840   __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1841   __ andptr(rsp, -16); // align stack as required by ABI
1842   __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in())));
1843   __ mov(rsp, r12); // restore sp
1844   __ reinit_heapbase();
1845 }
1846 
1847 
1848 void TemplateInterpreterGenerator::stop_interpreter_at() {
1849   Label L;
1850   __ cmp32(ExternalAddress((address) &BytecodeCounter::_counter_value),
1851            StopInterpreterAt);
1852   __ jcc(Assembler::notEqual, L);
1853   __ int3();
1854   __ bind(L);
1855 }
1856 #endif // !PRODUCT
1857 #endif // ! CC_INTERP