1 /*
   2  * Copyright (c) 2003, 2015, 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/interp_masm.hpp"
  29 #include "interpreter/interpreter.hpp"
  30 #include "interpreter/interpreterRuntime.hpp"
  31 #include "interpreter/templateInterpreterGenerator.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 // Global Register Names
  53 static const Register rbcp     = LP64_ONLY(r13) NOT_LP64(rsi);
  54 static const Register rlocals  = LP64_ONLY(r14) NOT_LP64(rdi);
  55 
  56 const int method_offset = frame::interpreter_frame_method_offset * wordSize;
  57 const int bcp_offset    = frame::interpreter_frame_bcp_offset    * wordSize;
  58 const int locals_offset = frame::interpreter_frame_locals_offset * wordSize;
  59 
  60 //-----------------------------------------------------------------------------
  61 
  62 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() {
  63   address entry = __ pc();
  64 
  65 #ifdef ASSERT
  66   {
  67     Label L;
  68     __ lea(rax, Address(rbp,
  69                         frame::interpreter_frame_monitor_block_top_offset *
  70                         wordSize));
  71     __ cmpptr(rax, rsp); // rax = maximal rsp for current rbp (stack
  72                          // grows negative)
  73     __ jcc(Assembler::aboveEqual, L); // check if frame is complete
  74     __ stop ("interpreter frame not set up");
  75     __ bind(L);
  76   }
  77 #endif // ASSERT
  78   // Restore bcp under the assumption that the current frame is still
  79   // interpreted
  80   __ restore_bcp();
  81 
  82   // expression stack must be empty before entering the VM if an
  83   // exception happened
  84   __ empty_expression_stack();
  85   // throw exception
  86   __ call_VM(noreg,
  87              CAST_FROM_FN_PTR(address,
  88                               InterpreterRuntime::throw_StackOverflowError));
  89   return entry;
  90 }
  91 
  92 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler(
  93         const char* name) {
  94   address entry = __ pc();
  95   // expression stack must be empty before entering the VM if an
  96   // exception happened
  97   __ empty_expression_stack();
  98   // setup parameters
  99   // ??? convention: expect aberrant index in register ebx
 100   Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
 101   __ lea(rarg, ExternalAddress((address)name));
 102   __ call_VM(noreg,
 103              CAST_FROM_FN_PTR(address,
 104                               InterpreterRuntime::
 105                               throw_ArrayIndexOutOfBoundsException),
 106              rarg, rbx);
 107   return entry;
 108 }
 109 
 110 address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
 111   address entry = __ pc();
 112 
 113   // object is at TOS
 114   Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
 115   __ pop(rarg);
 116 
 117   // expression stack must be empty before entering the VM if an
 118   // exception happened
 119   __ empty_expression_stack();
 120 
 121   __ call_VM(noreg,
 122              CAST_FROM_FN_PTR(address,
 123                               InterpreterRuntime::
 124                               throw_ClassCastException),
 125              rarg);
 126   return entry;
 127 }
 128 
 129 address TemplateInterpreterGenerator::generate_exception_handler_common(
 130         const char* name, const char* message, bool pass_oop) {
 131   assert(!pass_oop || message == NULL, "either oop or message but not both");
 132   address entry = __ pc();
 133 
 134   Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
 135   Register rarg2 = NOT_LP64(rbx) LP64_ONLY(c_rarg2);
 136 
 137   if (pass_oop) {
 138     // object is at TOS
 139     __ pop(rarg2);
 140   }
 141   // expression stack must be empty before entering the VM if an
 142   // exception happened
 143   __ empty_expression_stack();
 144   // setup parameters
 145   __ lea(rarg, ExternalAddress((address)name));
 146   if (pass_oop) {
 147     __ call_VM(rax, CAST_FROM_FN_PTR(address,
 148                                      InterpreterRuntime::
 149                                      create_klass_exception),
 150                rarg, rarg2);
 151   } else {
 152     // kind of lame ExternalAddress can't take NULL because
 153     // external_word_Relocation will assert.
 154     if (message != NULL) {
 155       __ lea(rarg2, ExternalAddress((address)message));
 156     } else {
 157       __ movptr(rarg2, NULL_WORD);
 158     }
 159     __ call_VM(rax,
 160                CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception),
 161                rarg, rarg2);
 162   }
 163   // throw exception
 164   __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
 165   return entry;
 166 }
 167 
 168 
 169 address TemplateInterpreterGenerator::generate_continuation_for(TosState state) {
 170   address entry = __ pc();
 171   // NULL last_sp until next java call
 172   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
 173   __ dispatch_next(state);
 174   return entry;
 175 }
 176 
 177 
 178 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step, size_t index_size) {
 179   address entry = __ pc();
 180 
 181 #ifndef _LP64
 182 #ifdef COMPILER2
 183   // The FPU stack is clean if UseSSE >= 2 but must be cleaned in other cases
 184   if ((state == ftos && UseSSE < 1) || (state == dtos && UseSSE < 2)) {
 185     for (int i = 1; i < 8; i++) {
 186         __ ffree(i);
 187     }
 188   } else if (UseSSE < 2) {
 189     __ empty_FPU_stack();
 190   }
 191 #endif // COMPILER2
 192   if ((state == ftos && UseSSE < 1) || (state == dtos && UseSSE < 2)) {
 193     __ MacroAssembler::verify_FPU(1, "generate_return_entry_for compiled");
 194   } else {
 195     __ MacroAssembler::verify_FPU(0, "generate_return_entry_for compiled");
 196   }
 197 
 198   if (state == ftos) {
 199     __ MacroAssembler::verify_FPU(UseSSE >= 1 ? 0 : 1, "generate_return_entry_for in interpreter");
 200   } else if (state == dtos) {
 201     __ MacroAssembler::verify_FPU(UseSSE >= 2 ? 0 : 1, "generate_return_entry_for in interpreter");
 202   }
 203 #endif // _LP64
 204 
 205   // Restore stack bottom in case i2c adjusted stack
 206   __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
 207   // and NULL it as marker that esp is now tos until next java call
 208   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
 209 
 210   __ restore_bcp();
 211   __ restore_locals();
 212 
 213   if (state == atos) {
 214     Register mdp = rbx;
 215     Register tmp = rcx;
 216     __ profile_return_type(mdp, rax, tmp);
 217   }
 218 
 219   const Register cache = rbx;
 220   const Register index = rcx;
 221   __ get_cache_and_index_at_bcp(cache, index, 1, index_size);
 222 
 223   const Register flags = cache;
 224   __ movl(flags, Address(cache, index, Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()));
 225   __ andl(flags, ConstantPoolCacheEntry::parameter_size_mask);
 226   __ lea(rsp, Address(rsp, flags, Interpreter::stackElementScale()));
 227   __ dispatch_next(state, step);
 228 
 229   return entry;
 230 }
 231 
 232 
 233 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state, int step) {
 234   address entry = __ pc();
 235 
 236 #ifndef _LP64
 237   if (state == ftos) {
 238     __ MacroAssembler::verify_FPU(UseSSE >= 1 ? 0 : 1, "generate_deopt_entry_for in interpreter");
 239   } else if (state == dtos) {
 240     __ MacroAssembler::verify_FPU(UseSSE >= 2 ? 0 : 1, "generate_deopt_entry_for in interpreter");
 241   }
 242 #endif // _LP64
 243 
 244   // NULL last_sp until next java call
 245   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
 246   __ restore_bcp();
 247   __ restore_locals();
 248   const Register thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
 249   NOT_LP64(__ get_thread(thread));
 250 #if INCLUDE_JVMCI
 251   // Check if we need to take lock at entry of synchronized method.
 252   if (UseJVMCICompiler) {
 253     Label L;
 254     __ cmpb(Address(thread, JavaThread::pending_monitorenter_offset()), 0);
 255     __ jcc(Assembler::zero, L);
 256     // Clear flag.
 257     __ movb(Address(thread, JavaThread::pending_monitorenter_offset()), 0);
 258     // Satisfy calling convention for lock_method().
 259     __ get_method(rbx);
 260     // Take lock.
 261     lock_method();
 262     __ bind(L);
 263   }
 264 #endif
 265   // handle exceptions
 266   {
 267     Label L;
 268     __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t) NULL_WORD);
 269     __ jcc(Assembler::zero, L);
 270     __ call_VM(noreg,
 271                CAST_FROM_FN_PTR(address,
 272                                 InterpreterRuntime::throw_pending_exception));
 273     __ should_not_reach_here();
 274     __ bind(L);
 275   }
 276   __ dispatch_next(state, step);
 277   return entry;
 278 }
 279 
 280 address TemplateInterpreterGenerator::generate_result_handler_for(
 281         BasicType type) {
 282   address entry = __ pc();
 283   switch (type) {
 284   case T_BOOLEAN: __ c2bool(rax);            break;
 285 #ifndef _LP64
 286   case T_CHAR   : __ andptr(rax, 0xFFFF);    break;
 287 #else
 288   case T_CHAR   : __ movzwl(rax, rax);       break;
 289 #endif // _LP64
 290   case T_BYTE   : __ sign_extend_byte(rax);  break;
 291   case T_SHORT  : __ sign_extend_short(rax); break;
 292   case T_INT    : /* nothing to do */        break;
 293   case T_LONG   : /* nothing to do */        break;
 294   case T_VOID   : /* nothing to do */        break;
 295 #ifndef _LP64
 296   case T_DOUBLE :
 297   case T_FLOAT  :
 298     { const Register t = InterpreterRuntime::SignatureHandlerGenerator::temp();
 299       __ pop(t);                            // remove return address first
 300       // Must return a result for interpreter or compiler. In SSE
 301       // mode, results are returned in xmm0 and the FPU stack must
 302       // be empty.
 303       if (type == T_FLOAT && UseSSE >= 1) {
 304         // Load ST0
 305         __ fld_d(Address(rsp, 0));
 306         // Store as float and empty fpu stack
 307         __ fstp_s(Address(rsp, 0));
 308         // and reload
 309         __ movflt(xmm0, Address(rsp, 0));
 310       } else if (type == T_DOUBLE && UseSSE >= 2 ) {
 311         __ movdbl(xmm0, Address(rsp, 0));
 312       } else {
 313         // restore ST0
 314         __ fld_d(Address(rsp, 0));
 315       }
 316       // and pop the temp
 317       __ addptr(rsp, 2 * wordSize);
 318       __ push(t);                           // restore return address
 319     }
 320     break;
 321 #else
 322   case T_FLOAT  : /* nothing to do */        break;
 323   case T_DOUBLE : /* nothing to do */        break;
 324 #endif // _LP64
 325 
 326   case T_OBJECT :
 327     // retrieve result from frame
 328     __ movptr(rax, Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize));
 329     // and verify it
 330     __ verify_oop(rax);
 331     break;
 332   default       : ShouldNotReachHere();
 333   }
 334   __ ret(0);                                   // return from result handler
 335   return entry;
 336 }
 337 
 338 address TemplateInterpreterGenerator::generate_safept_entry_for(
 339         TosState state,
 340         address runtime_entry) {
 341   address entry = __ pc();
 342   __ push(state);
 343   __ call_VM(noreg, runtime_entry);
 344   __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
 345   return entry;
 346 }
 347 
 348 
 349 
 350 // Helpers for commoning out cases in the various type of method entries.
 351 //
 352 
 353 
 354 // increment invocation count & check for overflow
 355 //
 356 // Note: checking for negative value instead of overflow
 357 //       so we have a 'sticky' overflow test
 358 //
 359 // rbx: method
 360 // rcx: invocation counter
 361 //
 362 void TemplateInterpreterGenerator::generate_counter_incr(
 363         Label* overflow,
 364         Label* profile_method,
 365         Label* profile_method_continue) {
 366   Label done;
 367   // Note: In tiered we increment either counters in Method* or in MDO depending if we're profiling or not.
 368   if (TieredCompilation) {
 369     int increment = InvocationCounter::count_increment;
 370     Label no_mdo;
 371     if (ProfileInterpreter) {
 372       // Are we profiling?
 373       __ movptr(rax, Address(rbx, Method::method_data_offset()));
 374       __ testptr(rax, rax);
 375       __ jccb(Assembler::zero, no_mdo);
 376       // Increment counter in the MDO
 377       const Address mdo_invocation_counter(rax, in_bytes(MethodData::invocation_counter_offset()) +
 378                                                 in_bytes(InvocationCounter::counter_offset()));
 379       const Address mask(rax, in_bytes(MethodData::invoke_mask_offset()));
 380       __ increment_mask_and_jump(mdo_invocation_counter, increment, mask, rcx, false, Assembler::zero, overflow);
 381       __ jmp(done);
 382     }
 383     __ bind(no_mdo);
 384     // Increment counter in MethodCounters
 385     const Address invocation_counter(rax,
 386                   MethodCounters::invocation_counter_offset() +
 387                   InvocationCounter::counter_offset());
 388     __ get_method_counters(rbx, rax, done);
 389     const Address mask(rax, in_bytes(MethodCounters::invoke_mask_offset()));
 390     __ increment_mask_and_jump(invocation_counter, increment, mask, rcx,
 391                                false, Assembler::zero, overflow);
 392     __ bind(done);
 393   } else { // not TieredCompilation
 394     const Address backedge_counter(rax,
 395                   MethodCounters::backedge_counter_offset() +
 396                   InvocationCounter::counter_offset());
 397     const Address invocation_counter(rax,
 398                   MethodCounters::invocation_counter_offset() +
 399                   InvocationCounter::counter_offset());
 400 
 401     __ get_method_counters(rbx, rax, done);
 402 
 403     if (ProfileInterpreter) {
 404       __ incrementl(Address(rax,
 405               MethodCounters::interpreter_invocation_counter_offset()));
 406     }
 407     // Update standard invocation counters
 408     __ movl(rcx, invocation_counter);
 409     __ incrementl(rcx, InvocationCounter::count_increment);
 410     __ movl(invocation_counter, rcx); // save invocation count
 411 
 412     __ movl(rax, backedge_counter);   // load backedge counter
 413     __ andl(rax, InvocationCounter::count_mask_value); // mask out the status bits
 414 
 415     __ addl(rcx, rax);                // add both counters
 416 
 417     // profile_method is non-null only for interpreted method so
 418     // profile_method != NULL == !native_call
 419 
 420     if (ProfileInterpreter && profile_method != NULL) {
 421       // Test to see if we should create a method data oop
 422       __ movptr(rax, Address(rbx, Method::method_counters_offset()));
 423       __ cmp32(rcx, Address(rax, in_bytes(MethodCounters::interpreter_profile_limit_offset())));
 424       __ jcc(Assembler::less, *profile_method_continue);
 425 
 426       // if no method data exists, go to profile_method
 427       __ test_method_data_pointer(rax, *profile_method);
 428     }
 429 
 430     __ movptr(rax, Address(rbx, Method::method_counters_offset()));
 431     __ cmp32(rcx, Address(rax, in_bytes(MethodCounters::interpreter_invocation_limit_offset())));
 432     __ jcc(Assembler::aboveEqual, *overflow);
 433     __ bind(done);
 434   }
 435 }
 436 
 437 void TemplateInterpreterGenerator::generate_counter_overflow(Label& do_continue) {
 438 
 439   // Asm interpreter on entry
 440   // r14/rdi - locals
 441   // r13/rsi - bcp
 442   // rbx - method
 443   // rdx - cpool --- DOES NOT APPEAR TO BE TRUE
 444   // rbp - interpreter frame
 445 
 446   // On return (i.e. jump to entry_point) [ back to invocation of interpreter ]
 447   // Everything as it was on entry
 448   // rdx is not restored. Doesn't appear to really be set.
 449 
 450   // InterpreterRuntime::frequency_counter_overflow takes two
 451   // arguments, the first (thread) is passed by call_VM, the second
 452   // indicates if the counter overflow occurs at a backwards branch
 453   // (NULL bcp).  We pass zero for it.  The call returns the address
 454   // of the verified entry point for the method or NULL if the
 455   // compilation did not complete (either went background or bailed
 456   // out).
 457   Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
 458   __ movl(rarg, 0);
 459   __ call_VM(noreg,
 460              CAST_FROM_FN_PTR(address,
 461                               InterpreterRuntime::frequency_counter_overflow),
 462              rarg);
 463 
 464   __ movptr(rbx, Address(rbp, method_offset));   // restore Method*
 465   // Preserve invariant that r13/r14 contain bcp/locals of sender frame
 466   // and jump to the interpreted entry.
 467   __ jmp(do_continue, relocInfo::none);
 468 }
 469 
 470 // See if we've got enough room on the stack for locals plus overhead.
 471 // The expression stack grows down incrementally, so the normal guard
 472 // page mechanism will work for that.
 473 //
 474 // NOTE: Since the additional locals are also always pushed (wasn't
 475 // obvious in generate_fixed_frame) so the guard should work for them
 476 // too.
 477 //
 478 // Args:
 479 //      rdx: number of additional locals this frame needs (what we must check)
 480 //      rbx: Method*
 481 //
 482 // Kills:
 483 //      rax
 484 void TemplateInterpreterGenerator::generate_stack_overflow_check(void) {
 485 
 486   // monitor entry size: see picture of stack in frame_x86.hpp
 487   const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
 488 
 489   // total overhead size: entry_size + (saved rbp through expr stack
 490   // bottom).  be sure to change this if you add/subtract anything
 491   // to/from the overhead area
 492   const int overhead_size =
 493     -(frame::interpreter_frame_initial_sp_offset * wordSize) + entry_size;
 494 
 495   const int page_size = os::vm_page_size();
 496 
 497   Label after_frame_check;
 498 
 499   // see if the frame is greater than one page in size. If so,
 500   // then we need to verify there is enough stack space remaining
 501   // for the additional locals.
 502   __ cmpl(rdx, (page_size - overhead_size) / Interpreter::stackElementSize);
 503   __ jcc(Assembler::belowEqual, after_frame_check);
 504 
 505   // compute rsp as if this were going to be the last frame on
 506   // the stack before the red zone
 507 
 508   Label after_frame_check_pop;
 509   const Register thread = NOT_LP64(rsi) LP64_ONLY(r15_thread);
 510 #ifndef _LP64
 511   __ push(thread);
 512   __ get_thread(thread);
 513 #endif
 514 
 515   const Address stack_base(thread, Thread::stack_base_offset());
 516   const Address stack_size(thread, Thread::stack_size_offset());
 517 
 518   // locals + overhead, in bytes
 519   __ mov(rax, rdx);
 520   __ shlptr(rax, Interpreter::logStackElementSize);  // 2 slots per parameter.
 521   __ addptr(rax, overhead_size);
 522 
 523 #ifdef ASSERT
 524   Label stack_base_okay, stack_size_okay;
 525   // verify that thread stack base is non-zero
 526   __ cmpptr(stack_base, (int32_t)NULL_WORD);
 527   __ jcc(Assembler::notEqual, stack_base_okay);
 528   __ stop("stack base is zero");
 529   __ bind(stack_base_okay);
 530   // verify that thread stack size is non-zero
 531   __ cmpptr(stack_size, 0);
 532   __ jcc(Assembler::notEqual, stack_size_okay);
 533   __ stop("stack size is zero");
 534   __ bind(stack_size_okay);
 535 #endif
 536 
 537   // Add stack base to locals and subtract stack size
 538   __ addptr(rax, stack_base);
 539   __ subptr(rax, stack_size);
 540 
 541   // Use the bigger size for banging.
 542   const int max_bang_size = (int)MAX2(JavaThread::stack_shadow_zone_size(),
 543                                       JavaThread::stack_guard_zone_size());
 544 
 545   // add in the red and yellow zone sizes
 546   __ addptr(rax, max_bang_size);
 547 
 548   // check against the current stack bottom
 549   __ cmpptr(rsp, rax);
 550 
 551   __ jcc(Assembler::above, after_frame_check_pop);
 552   NOT_LP64(__ pop(rsi));  // get saved bcp
 553 
 554   // Restore sender's sp as SP. This is necessary if the sender's
 555   // frame is an extended compiled frame (see gen_c2i_adapter())
 556   // and safer anyway in case of JSR292 adaptations.
 557 
 558   __ pop(rax); // return address must be moved if SP is changed
 559   __ mov(rsp, rbcp);
 560   __ push(rax);
 561 
 562   // Note: the restored frame is not necessarily interpreted.
 563   // Use the shared runtime version of the StackOverflowError.
 564   assert(StubRoutines::throw_StackOverflowError_entry() != NULL, "stub not yet generated");
 565   __ jump(ExternalAddress(StubRoutines::throw_StackOverflowError_entry()));
 566   // all done with frame size check
 567   __ bind(after_frame_check_pop);
 568   NOT_LP64(__ pop(rsi));
 569 
 570   // all done with frame size check
 571   __ bind(after_frame_check);
 572 }
 573 
 574 // Allocate monitor and lock method (asm interpreter)
 575 //
 576 // Args:
 577 //      rbx: Method*
 578 //      r14/rdi: locals
 579 //
 580 // Kills:
 581 //      rax
 582 //      c_rarg0, c_rarg1, c_rarg2, c_rarg3, ...(param regs)
 583 //      rscratch1, rscratch2 (scratch regs)
 584 void TemplateInterpreterGenerator::lock_method() {
 585   // synchronize method
 586   const Address access_flags(rbx, Method::access_flags_offset());
 587   const Address monitor_block_top(
 588         rbp,
 589         frame::interpreter_frame_monitor_block_top_offset * wordSize);
 590   const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
 591 
 592 #ifdef ASSERT
 593   {
 594     Label L;
 595     __ movl(rax, access_flags);
 596     __ testl(rax, JVM_ACC_SYNCHRONIZED);
 597     __ jcc(Assembler::notZero, L);
 598     __ stop("method doesn't need synchronization");
 599     __ bind(L);
 600   }
 601 #endif // ASSERT
 602 
 603   // get synchronization object
 604   {
 605     const int mirror_offset = in_bytes(Klass::java_mirror_offset());
 606     Label done;
 607     __ movl(rax, access_flags);
 608     __ testl(rax, JVM_ACC_STATIC);
 609     // get receiver (assume this is frequent case)
 610     __ movptr(rax, Address(rlocals, Interpreter::local_offset_in_bytes(0)));
 611     __ jcc(Assembler::zero, done);
 612     __ movptr(rax, Address(rbx, Method::const_offset()));
 613     __ movptr(rax, Address(rax, ConstMethod::constants_offset()));
 614     __ movptr(rax, Address(rax,
 615                            ConstantPool::pool_holder_offset_in_bytes()));
 616     __ movptr(rax, Address(rax, mirror_offset));
 617 
 618 #ifdef ASSERT
 619     {
 620       Label L;
 621       __ testptr(rax, rax);
 622       __ jcc(Assembler::notZero, L);
 623       __ stop("synchronization object is NULL");
 624       __ bind(L);
 625     }
 626 #endif // ASSERT
 627 
 628     __ bind(done);
 629   }
 630 
 631   // add space for monitor & lock
 632   __ subptr(rsp, entry_size); // add space for a monitor entry
 633   __ movptr(monitor_block_top, rsp);  // set new monitor block top
 634   // store object
 635   __ movptr(Address(rsp, BasicObjectLock::obj_offset_in_bytes()), rax);
 636   const Register lockreg = NOT_LP64(rdx) LP64_ONLY(c_rarg1);
 637   __ movptr(lockreg, rsp); // object address
 638   __ lock_object(lockreg);
 639 }
 640 
 641 // Generate a fixed interpreter frame. This is identical setup for
 642 // interpreted methods and for native methods hence the shared code.
 643 //
 644 // Args:
 645 //      rax: return address
 646 //      rbx: Method*
 647 //      r14/rdi: pointer to locals
 648 //      r13/rsi: sender sp
 649 //      rdx: cp cache
 650 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
 651   // initialize fixed part of activation frame
 652   __ push(rax);        // save return address
 653   __ enter();          // save old & set new rbp
 654   __ push(rbcp);        // set sender sp
 655   __ push((int)NULL_WORD); // leave last_sp as null
 656   __ movptr(rbcp, Address(rbx, Method::const_offset()));      // get ConstMethod*
 657   __ lea(rbcp, Address(rbcp, ConstMethod::codes_offset())); // get codebase
 658   __ push(rbx);        // save Method*
 659   if (ProfileInterpreter) {
 660     Label method_data_continue;
 661     __ movptr(rdx, Address(rbx, in_bytes(Method::method_data_offset())));
 662     __ testptr(rdx, rdx);
 663     __ jcc(Assembler::zero, method_data_continue);
 664     __ addptr(rdx, in_bytes(MethodData::data_offset()));
 665     __ bind(method_data_continue);
 666     __ push(rdx);      // set the mdp (method data pointer)
 667   } else {
 668     __ push(0);
 669   }
 670 
 671   __ movptr(rdx, Address(rbx, Method::const_offset()));
 672   __ movptr(rdx, Address(rdx, ConstMethod::constants_offset()));
 673   __ movptr(rdx, Address(rdx, ConstantPool::cache_offset_in_bytes()));
 674   __ push(rdx); // set constant pool cache
 675   __ push(rlocals); // set locals pointer
 676   if (native_call) {
 677     __ push(0); // no bcp
 678   } else {
 679     __ push(rbcp); // set bcp
 680   }
 681   __ push(0); // reserve word for pointer to expression stack bottom
 682   __ movptr(Address(rsp, 0), rsp); // set expression stack bottom
 683 }
 684 
 685 // End of helpers
 686 
 687 // Method entry for java.lang.ref.Reference.get.
 688 address TemplateInterpreterGenerator::generate_Reference_get_entry(void) {
 689 #if INCLUDE_ALL_GCS
 690   // Code: _aload_0, _getfield, _areturn
 691   // parameter size = 1
 692   //
 693   // The code that gets generated by this routine is split into 2 parts:
 694   //    1. The "intrinsified" code for G1 (or any SATB based GC),
 695   //    2. The slow path - which is an expansion of the regular method entry.
 696   //
 697   // Notes:-
 698   // * In the G1 code we do not check whether we need to block for
 699   //   a safepoint. If G1 is enabled then we must execute the specialized
 700   //   code for Reference.get (except when the Reference object is null)
 701   //   so that we can log the value in the referent field with an SATB
 702   //   update buffer.
 703   //   If the code for the getfield template is modified so that the
 704   //   G1 pre-barrier code is executed when the current method is
 705   //   Reference.get() then going through the normal method entry
 706   //   will be fine.
 707   // * The G1 code can, however, check the receiver object (the instance
 708   //   of java.lang.Reference) and jump to the slow path if null. If the
 709   //   Reference object is null then we obviously cannot fetch the referent
 710   //   and so we don't need to call the G1 pre-barrier. Thus we can use the
 711   //   regular method entry code to generate the NPE.
 712   //
 713   // rbx: Method*
 714 
 715   // r13: senderSP must preserve for slow path, set SP to it on fast path
 716 
 717   address entry = __ pc();
 718 
 719   const int referent_offset = java_lang_ref_Reference::referent_offset;
 720   guarantee(referent_offset > 0, "referent offset not initialized");
 721 
 722   if (UseG1GC) {
 723     Label slow_path;
 724     // rbx: method
 725 
 726     // Check if local 0 != NULL
 727     // If the receiver is null then it is OK to jump to the slow path.
 728     __ movptr(rax, Address(rsp, wordSize));
 729 
 730     __ testptr(rax, rax);
 731     __ jcc(Assembler::zero, slow_path);
 732 
 733     // rax: local 0
 734     // rbx: method (but can be used as scratch now)
 735     // rdx: scratch
 736     // rdi: scratch
 737 
 738     // Preserve the sender sp in case the pre-barrier
 739     // calls the runtime
 740     NOT_LP64(__ push(rsi));
 741 
 742     // Generate the G1 pre-barrier code to log the value of
 743     // the referent field in an SATB buffer.
 744 
 745     // Load the value of the referent field.
 746     const Address field_address(rax, referent_offset);
 747     __ load_heap_oop(rax, field_address);
 748 
 749     const Register sender_sp = NOT_LP64(rsi) LP64_ONLY(r13);
 750     const Register thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
 751     NOT_LP64(__ get_thread(thread));
 752 
 753     // Generate the G1 pre-barrier code to log the value of
 754     // the referent field in an SATB buffer.
 755     __ g1_write_barrier_pre(noreg /* obj */,
 756                             rax /* pre_val */,
 757                             thread /* thread */,
 758                             rbx /* tmp */,
 759                             true /* tosca_live */,
 760                             true /* expand_call */);
 761 
 762     // _areturn
 763     NOT_LP64(__ pop(rsi));      // get sender sp
 764     __ pop(rdi);                // get return address
 765     __ mov(rsp, sender_sp);     // set sp to sender sp
 766     __ jmp(rdi);
 767     __ ret(0);
 768 
 769     // generate a vanilla interpreter entry as the slow path
 770     __ bind(slow_path);
 771     __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::zerolocals));
 772     return entry;
 773   }
 774 #endif // INCLUDE_ALL_GCS
 775 
 776   // If G1 is not enabled then attempt to go through the accessor entry point
 777   // Reference.get is an accessor
 778   return NULL;
 779 }
 780 
 781 // Interpreter stub for calling a native method. (asm interpreter)
 782 // This sets up a somewhat different looking stack for calling the
 783 // native method than the typical interpreter frame setup.
 784 address TemplateInterpreterGenerator::generate_native_entry(bool synchronized) {
 785   // determine code generation flags
 786   bool inc_counter  = UseCompiler || CountCompiledCalls || LogTouchedMethods;
 787 
 788   // rbx: Method*
 789   // rbcp: sender sp
 790 
 791   address entry_point = __ pc();
 792 
 793   const Address constMethod       (rbx, Method::const_offset());
 794   const Address access_flags      (rbx, Method::access_flags_offset());
 795   const Address size_of_parameters(rcx, ConstMethod::
 796                                         size_of_parameters_offset());
 797 
 798 
 799   // get parameter size (always needed)
 800   __ movptr(rcx, constMethod);
 801   __ load_unsigned_short(rcx, size_of_parameters);
 802 
 803   // native calls don't need the stack size check since they have no
 804   // expression stack and the arguments are already on the stack and
 805   // we only add a handful of words to the stack
 806 
 807   // rbx: Method*
 808   // rcx: size of parameters
 809   // rbcp: sender sp
 810   __ pop(rax);                                       // get return address
 811 
 812   // for natives the size of locals is zero
 813 
 814   // compute beginning of parameters
 815   __ lea(rlocals, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize));
 816 
 817   // add 2 zero-initialized slots for native calls
 818   // initialize result_handler slot
 819   __ push((int) NULL_WORD);
 820   // slot for oop temp
 821   // (static native method holder mirror/jni oop result)
 822   __ push((int) NULL_WORD);
 823 
 824   // initialize fixed part of activation frame
 825   generate_fixed_frame(true);
 826 
 827   // make sure method is native & not abstract
 828 #ifdef ASSERT
 829   __ movl(rax, access_flags);
 830   {
 831     Label L;
 832     __ testl(rax, JVM_ACC_NATIVE);
 833     __ jcc(Assembler::notZero, L);
 834     __ stop("tried to execute non-native method as native");
 835     __ bind(L);
 836   }
 837   {
 838     Label L;
 839     __ testl(rax, JVM_ACC_ABSTRACT);
 840     __ jcc(Assembler::zero, L);
 841     __ stop("tried to execute abstract method in interpreter");
 842     __ bind(L);
 843   }
 844 #endif
 845 
 846   // Since at this point in the method invocation the exception handler
 847   // would try to exit the monitor of synchronized methods which hasn't
 848   // been entered yet, we set the thread local variable
 849   // _do_not_unlock_if_synchronized to true. The remove_activation will
 850   // check this flag.
 851 
 852   const Register thread1 = NOT_LP64(rax) LP64_ONLY(r15_thread);
 853   NOT_LP64(__ get_thread(thread1));
 854   const Address do_not_unlock_if_synchronized(thread1,
 855         in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
 856   __ movbool(do_not_unlock_if_synchronized, true);
 857 
 858   // increment invocation count & check for overflow
 859   Label invocation_counter_overflow;
 860   if (inc_counter) {
 861     generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
 862   }
 863 
 864   Label continue_after_compile;
 865   __ bind(continue_after_compile);
 866 
 867   bang_stack_shadow_pages(true);
 868 
 869   // reset the _do_not_unlock_if_synchronized flag
 870   NOT_LP64(__ get_thread(thread1));
 871   __ movbool(do_not_unlock_if_synchronized, false);
 872 
 873   // check for synchronized methods
 874   // Must happen AFTER invocation_counter check and stack overflow check,
 875   // so method is not locked if overflows.
 876   if (synchronized) {
 877     lock_method();
 878   } else {
 879     // no synchronization necessary
 880 #ifdef ASSERT
 881     {
 882       Label L;
 883       __ movl(rax, access_flags);
 884       __ testl(rax, JVM_ACC_SYNCHRONIZED);
 885       __ jcc(Assembler::zero, L);
 886       __ stop("method needs synchronization");
 887       __ bind(L);
 888     }
 889 #endif
 890   }
 891 
 892   // start execution
 893 #ifdef ASSERT
 894   {
 895     Label L;
 896     const Address monitor_block_top(rbp,
 897                  frame::interpreter_frame_monitor_block_top_offset * wordSize);
 898     __ movptr(rax, monitor_block_top);
 899     __ cmpptr(rax, rsp);
 900     __ jcc(Assembler::equal, L);
 901     __ stop("broken stack frame setup in interpreter");
 902     __ bind(L);
 903   }
 904 #endif
 905 
 906   // jvmti support
 907   __ notify_method_entry();
 908 
 909   // work registers
 910   const Register method = rbx;
 911   const Register thread = NOT_LP64(rdi) LP64_ONLY(r15_thread);
 912   const Register t      = NOT_LP64(rcx) LP64_ONLY(r11);
 913 
 914   // allocate space for parameters
 915   __ get_method(method);
 916   __ movptr(t, Address(method, Method::const_offset()));
 917   __ load_unsigned_short(t, Address(t, ConstMethod::size_of_parameters_offset()));
 918 
 919 #ifndef _LP64
 920   __ shlptr(t, Interpreter::logStackElementSize);
 921   __ addptr(t, 2*wordSize);     // allocate two more slots for JNIEnv and possible mirror
 922   __ subptr(rsp, t);
 923   __ andptr(rsp, -(StackAlignmentInBytes)); // gcc needs 16 byte aligned stacks to do XMM intrinsics
 924 #else
 925   __ shll(t, Interpreter::logStackElementSize);
 926 
 927   __ subptr(rsp, t);
 928   __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
 929   __ andptr(rsp, -16); // must be 16 byte boundary (see amd64 ABI)
 930 #endif // _LP64
 931 
 932   // get signature handler
 933   {
 934     Label L;
 935     __ movptr(t, Address(method, Method::signature_handler_offset()));
 936     __ testptr(t, t);
 937     __ jcc(Assembler::notZero, L);
 938     __ call_VM(noreg,
 939                CAST_FROM_FN_PTR(address,
 940                                 InterpreterRuntime::prepare_native_call),
 941                method);
 942     __ get_method(method);
 943     __ movptr(t, Address(method, Method::signature_handler_offset()));
 944     __ bind(L);
 945   }
 946 
 947   // call signature handler
 948   assert(InterpreterRuntime::SignatureHandlerGenerator::from() == rlocals,
 949          "adjust this code");
 950   assert(InterpreterRuntime::SignatureHandlerGenerator::to() == rsp,
 951          "adjust this code");
 952   assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == NOT_LP64(t) LP64_ONLY(rscratch1),
 953          "adjust this code");
 954 
 955   // The generated handlers do not touch RBX (the method oop).
 956   // However, large signatures cannot be cached and are generated
 957   // each time here.  The slow-path generator can do a GC on return,
 958   // so we must reload it after the call.
 959   __ call(t);
 960   __ get_method(method);        // slow path can do a GC, reload RBX
 961 
 962 
 963   // result handler is in rax
 964   // set result handler
 965   __ movptr(Address(rbp,
 966                     (frame::interpreter_frame_result_handler_offset) * wordSize),
 967             rax);
 968 
 969   // pass mirror handle if static call
 970   {
 971     Label L;
 972     const int mirror_offset = in_bytes(Klass::java_mirror_offset());
 973     __ movl(t, Address(method, Method::access_flags_offset()));
 974     __ testl(t, JVM_ACC_STATIC);
 975     __ jcc(Assembler::zero, L);
 976     // get mirror
 977     __ movptr(t, Address(method, Method::const_offset()));
 978     __ movptr(t, Address(t, ConstMethod::constants_offset()));
 979     __ movptr(t, Address(t, ConstantPool::pool_holder_offset_in_bytes()));
 980     __ movptr(t, Address(t, mirror_offset));
 981     // copy mirror into activation frame
 982     __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize),
 983             t);
 984     // pass handle to mirror
 985 #ifndef _LP64
 986     __ lea(t, Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize));
 987     __ movptr(Address(rsp, wordSize), t);
 988 #else
 989     __ lea(c_rarg1,
 990            Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize));
 991 #endif // _LP64
 992     __ bind(L);
 993   }
 994 
 995   // get native function entry point
 996   {
 997     Label L;
 998     __ movptr(rax, Address(method, Method::native_function_offset()));
 999     ExternalAddress unsatisfied(SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
1000     __ cmpptr(rax, unsatisfied.addr());
1001     __ jcc(Assembler::notEqual, L);
1002     __ call_VM(noreg,
1003                CAST_FROM_FN_PTR(address,
1004                                 InterpreterRuntime::prepare_native_call),
1005                method);
1006     __ get_method(method);
1007     __ movptr(rax, Address(method, Method::native_function_offset()));
1008     __ bind(L);
1009   }
1010 
1011   // pass JNIEnv
1012 #ifndef _LP64
1013    __ get_thread(thread);
1014    __ lea(t, Address(thread, JavaThread::jni_environment_offset()));
1015    __ movptr(Address(rsp, 0), t);
1016 
1017    // set_last_Java_frame_before_call
1018    // It is enough that the pc()
1019    // points into the right code segment. It does not have to be the correct return pc.
1020    __ set_last_Java_frame(thread, noreg, rbp, __ pc());
1021 #else
1022    __ lea(c_rarg0, Address(r15_thread, JavaThread::jni_environment_offset()));
1023 
1024    // It is enough that the pc() points into the right code
1025    // segment. It does not have to be the correct return pc.
1026    __ set_last_Java_frame(rsp, rbp, (address) __ pc());
1027 #endif // _LP64
1028 
1029   // change thread state
1030 #ifdef ASSERT
1031   {
1032     Label L;
1033     __ movl(t, Address(thread, JavaThread::thread_state_offset()));
1034     __ cmpl(t, _thread_in_Java);
1035     __ jcc(Assembler::equal, L);
1036     __ stop("Wrong thread state in native stub");
1037     __ bind(L);
1038   }
1039 #endif
1040 
1041   // Change state to native
1042 
1043   __ movl(Address(thread, JavaThread::thread_state_offset()),
1044           _thread_in_native);
1045 
1046   // Call the native method.
1047   __ call(rax);
1048   // 32: result potentially in rdx:rax or ST0
1049   // 64: result potentially in rax or xmm0
1050 
1051   // Verify or restore cpu control state after JNI call
1052   __ restore_cpu_control_state_after_jni();
1053 
1054   // NOTE: The order of these pushes is known to frame::interpreter_frame_result
1055   // in order to extract the result of a method call. If the order of these
1056   // pushes change or anything else is added to the stack then the code in
1057   // interpreter_frame_result must also change.
1058 
1059 #ifndef _LP64
1060   // save potential result in ST(0) & rdx:rax
1061   // (if result handler is the T_FLOAT or T_DOUBLE handler, result must be in ST0 -
1062   // the check is necessary to avoid potential Intel FPU overflow problems by saving/restoring 'empty' FPU registers)
1063   // It is safe to do this push because state is _thread_in_native and return address will be found
1064   // via _last_native_pc and not via _last_jave_sp
1065 
1066   // NOTE: the order of theses push(es) is known to frame::interpreter_frame_result.
1067   // If the order changes or anything else is added to the stack the code in
1068   // interpreter_frame_result will have to be changed.
1069 
1070   { Label L;
1071     Label push_double;
1072     ExternalAddress float_handler(AbstractInterpreter::result_handler(T_FLOAT));
1073     ExternalAddress double_handler(AbstractInterpreter::result_handler(T_DOUBLE));
1074     __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize),
1075               float_handler.addr());
1076     __ jcc(Assembler::equal, push_double);
1077     __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize),
1078               double_handler.addr());
1079     __ jcc(Assembler::notEqual, L);
1080     __ bind(push_double);
1081     __ push_d(); // FP values are returned using the FPU, so push FPU contents (even if UseSSE > 0).
1082     __ bind(L);
1083   }
1084 #else
1085   __ push(dtos);
1086 #endif // _LP64
1087 
1088   __ push(ltos);
1089 
1090   // change thread state
1091   NOT_LP64(__ get_thread(thread));
1092   __ movl(Address(thread, JavaThread::thread_state_offset()),
1093           _thread_in_native_trans);
1094 
1095   if (os::is_MP()) {
1096     if (UseMembar) {
1097       // Force this write out before the read below
1098       __ membar(Assembler::Membar_mask_bits(
1099            Assembler::LoadLoad | Assembler::LoadStore |
1100            Assembler::StoreLoad | Assembler::StoreStore));
1101     } else {
1102       // Write serialization page so VM thread can do a pseudo remote membar.
1103       // We use the current thread pointer to calculate a thread specific
1104       // offset to write to within the page. This minimizes bus traffic
1105       // due to cache line collision.
1106       __ serialize_memory(thread, rcx);
1107     }
1108   }
1109 
1110 #ifndef _LP64
1111   if (AlwaysRestoreFPU) {
1112     //  Make sure the control word is correct.
1113     __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
1114   }
1115 #endif // _LP64
1116 
1117   // check for safepoint operation in progress and/or pending suspend requests
1118   {
1119     Label Continue;
1120     __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
1121              SafepointSynchronize::_not_synchronized);
1122 
1123     Label L;
1124     __ jcc(Assembler::notEqual, L);
1125     __ cmpl(Address(thread, JavaThread::suspend_flags_offset()), 0);
1126     __ jcc(Assembler::equal, Continue);
1127     __ bind(L);
1128 
1129     // Don't use call_VM as it will see a possible pending exception
1130     // and forward it and never return here preventing us from
1131     // clearing _last_native_pc down below.  Also can't use
1132     // call_VM_leaf either as it will check to see if r13 & r14 are
1133     // preserved and correspond to the bcp/locals pointers. So we do a
1134     // runtime call by hand.
1135     //
1136 #ifndef _LP64
1137     __ push(thread);
1138     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address,
1139                                             JavaThread::check_special_condition_for_native_trans)));
1140     __ increment(rsp, wordSize);
1141     __ get_thread(thread);
1142 #else
1143     __ mov(c_rarg0, r15_thread);
1144     __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1145     __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1146     __ andptr(rsp, -16); // align stack as required by ABI
1147     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans)));
1148     __ mov(rsp, r12); // restore sp
1149     __ reinit_heapbase();
1150 #endif // _LP64
1151     __ bind(Continue);
1152   }
1153 
1154   // change thread state
1155   __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_Java);
1156 
1157   // reset_last_Java_frame
1158   __ reset_last_Java_frame(thread, true, true);
1159 
1160   // reset handle block
1161   __ movptr(t, Address(thread, JavaThread::active_handles_offset()));
1162   __ movl(Address(t, JNIHandleBlock::top_offset_in_bytes()), (int32_t)NULL_WORD);
1163 
1164   // If result is an oop unbox and store it in frame where gc will see it
1165   // and result handler will pick it up
1166 
1167   {
1168     Label no_oop, store_result;
1169     __ lea(t, ExternalAddress(AbstractInterpreter::result_handler(T_OBJECT)));
1170     __ cmpptr(t, Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize));
1171     __ jcc(Assembler::notEqual, no_oop);
1172     // retrieve result
1173     __ pop(ltos);
1174     __ testptr(rax, rax);
1175     __ jcc(Assembler::zero, store_result);
1176     __ movptr(rax, Address(rax, 0));
1177     __ bind(store_result);
1178     __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize), rax);
1179     // keep stack depth as expected by pushing oop which will eventually be discarded
1180     __ push(ltos);
1181     __ bind(no_oop);
1182   }
1183 
1184 
1185   {
1186     Label no_reguard;
1187     __ cmpl(Address(thread, JavaThread::stack_guard_state_offset()),
1188             JavaThread::stack_guard_yellow_reserved_disabled);
1189     __ jcc(Assembler::notEqual, no_reguard);
1190 
1191     __ pusha(); // XXX only save smashed registers
1192 #ifndef _LP64
1193     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
1194     __ popa();
1195 #else
1196     __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1197     __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1198     __ andptr(rsp, -16); // align stack as required by ABI
1199     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
1200     __ mov(rsp, r12); // restore sp
1201     __ popa(); // XXX only restore smashed registers
1202     __ reinit_heapbase();
1203 #endif // _LP64
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 to have legal interpreter frame, i.e., bci == 0 <=> code_base()
1215   __ movptr(rbcp, Address(method, Method::const_offset()));   // get ConstMethod*
1216   __ lea(rbcp, Address(rbcp, ConstMethod::codes_offset()));    // get codebase
1217 
1218   // handle exceptions (exception handling will handle unlocking!)
1219   {
1220     Label L;
1221     __ cmpptr(Address(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 - (int)sizeof(BasicObjectLock)));
1250 
1251       const Register regmon = NOT_LP64(rdx) LP64_ONLY(c_rarg1);
1252 
1253       // monitor expect in c_rarg1 for slow unlock path
1254       __ lea(regmon, monitor); // address of first monitor
1255 
1256       __ movptr(t, Address(regmon, BasicObjectLock::obj_offset_in_bytes()));
1257       __ testptr(t, t);
1258       __ jcc(Assembler::notZero, unlock);
1259 
1260       // Entry already unlocked, need to throw exception
1261       __ MacroAssembler::call_VM(noreg,
1262                                  CAST_FROM_FN_PTR(address,
1263                    InterpreterRuntime::throw_illegal_monitor_state_exception));
1264       __ should_not_reach_here();
1265 
1266       __ bind(unlock);
1267       __ unlock_object(regmon);
1268     }
1269     __ bind(L);
1270   }
1271 
1272   // jvmti support
1273   // Note: This must happen _after_ handling/throwing any exceptions since
1274   //       the exception handler code notifies the runtime of method exits
1275   //       too. If this happens before, method entry/exit notifications are
1276   //       not properly paired (was bug - gri 11/22/99).
1277   __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI);
1278 
1279   // restore potential result in edx:eax, call result handler to
1280   // restore potential result in ST0 & handle result
1281 
1282   __ pop(ltos);
1283   LP64_ONLY( __ pop(dtos));
1284 
1285   __ movptr(t, Address(rbp,
1286                        (frame::interpreter_frame_result_handler_offset) * wordSize));
1287   __ call(t);
1288 
1289   // remove activation
1290   __ movptr(t, Address(rbp,
1291                        frame::interpreter_frame_sender_sp_offset *
1292                        wordSize)); // get sender sp
1293   __ leave();                                // remove frame anchor
1294   __ pop(rdi);                               // get return address
1295   __ mov(rsp, t);                            // set sp to sender sp
1296   __ jmp(rdi);
1297 
1298   if (inc_counter) {
1299     // Handle overflow of counter and compile method
1300     __ bind(invocation_counter_overflow);
1301     generate_counter_overflow(continue_after_compile);
1302   }
1303 
1304   return entry_point;
1305 }
1306 
1307 //
1308 // Generic interpreted method entry to (asm) interpreter
1309 //
1310 address TemplateInterpreterGenerator::generate_normal_entry(bool synchronized) {
1311   // determine code generation flags
1312   bool inc_counter  = UseCompiler || CountCompiledCalls || LogTouchedMethods;
1313 
1314   // ebx: Method*
1315   // rbcp: sender sp
1316   address entry_point = __ pc();
1317 
1318   const Address constMethod(rbx, Method::const_offset());
1319   const Address access_flags(rbx, Method::access_flags_offset());
1320   const Address size_of_parameters(rdx,
1321                                    ConstMethod::size_of_parameters_offset());
1322   const Address size_of_locals(rdx, ConstMethod::size_of_locals_offset());
1323 
1324 
1325   // get parameter size (always needed)
1326   __ movptr(rdx, constMethod);
1327   __ load_unsigned_short(rcx, size_of_parameters);
1328 
1329   // rbx: Method*
1330   // rcx: size of parameters
1331   // rbcp: sender_sp (could differ from sp+wordSize if we were called via c2i )
1332 
1333   __ load_unsigned_short(rdx, size_of_locals); // get size of locals in words
1334   __ subl(rdx, rcx); // rdx = no. of additional locals
1335 
1336   // YYY
1337 //   __ incrementl(rdx);
1338 //   __ andl(rdx, -2);
1339 
1340   // see if we've got enough room on the stack for locals plus overhead.
1341   generate_stack_overflow_check();
1342 
1343   // get return address
1344   __ pop(rax);
1345 
1346   // compute beginning of parameters
1347   __ lea(rlocals, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize));
1348 
1349   // rdx - # of additional locals
1350   // allocate space for locals
1351   // explicitly initialize locals
1352   {
1353     Label exit, loop;
1354     __ testl(rdx, rdx);
1355     __ jcc(Assembler::lessEqual, exit); // do nothing if rdx <= 0
1356     __ bind(loop);
1357     __ push((int) NULL_WORD); // initialize local variables
1358     __ decrementl(rdx); // until everything initialized
1359     __ jcc(Assembler::greater, loop);
1360     __ bind(exit);
1361   }
1362 
1363   // initialize fixed part of activation frame
1364   generate_fixed_frame(false);
1365 
1366   // make sure method is not native & not abstract
1367 #ifdef ASSERT
1368   __ movl(rax, access_flags);
1369   {
1370     Label L;
1371     __ testl(rax, JVM_ACC_NATIVE);
1372     __ jcc(Assembler::zero, L);
1373     __ stop("tried to execute native method as non-native");
1374     __ bind(L);
1375   }
1376   {
1377     Label L;
1378     __ testl(rax, JVM_ACC_ABSTRACT);
1379     __ jcc(Assembler::zero, L);
1380     __ stop("tried to execute abstract method in interpreter");
1381     __ bind(L);
1382   }
1383 #endif
1384 
1385   // Since at this point in the method invocation the exception
1386   // handler would try to exit the monitor of synchronized methods
1387   // which hasn't been entered yet, we set the thread local variable
1388   // _do_not_unlock_if_synchronized to true. The remove_activation
1389   // will check this flag.
1390 
1391   const Register thread = NOT_LP64(rax) LP64_ONLY(r15_thread);
1392   NOT_LP64(__ get_thread(thread));
1393   const Address do_not_unlock_if_synchronized(thread,
1394         in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1395   __ movbool(do_not_unlock_if_synchronized, true);
1396 
1397   __ profile_parameters_type(rax, rcx, rdx);
1398   // increment invocation count & check for overflow
1399   Label invocation_counter_overflow;
1400   Label profile_method;
1401   Label profile_method_continue;
1402   if (inc_counter) {
1403     generate_counter_incr(&invocation_counter_overflow,
1404                           &profile_method,
1405                           &profile_method_continue);
1406     if (ProfileInterpreter) {
1407       __ bind(profile_method_continue);
1408     }
1409   }
1410 
1411   Label continue_after_compile;
1412   __ bind(continue_after_compile);
1413 
1414   // check for synchronized interpreted methods
1415   bang_stack_shadow_pages(false);
1416 
1417   // reset the _do_not_unlock_if_synchronized flag
1418   NOT_LP64(__ get_thread(thread));
1419   __ movbool(do_not_unlock_if_synchronized, false);
1420 
1421   // check for synchronized methods
1422   // Must happen AFTER invocation_counter check and stack overflow check,
1423   // so method is not locked if overflows.
1424   if (synchronized) {
1425     // Allocate monitor and lock method
1426     lock_method();
1427   } else {
1428     // no synchronization necessary
1429 #ifdef ASSERT
1430     {
1431       Label L;
1432       __ movl(rax, access_flags);
1433       __ testl(rax, JVM_ACC_SYNCHRONIZED);
1434       __ jcc(Assembler::zero, L);
1435       __ stop("method needs synchronization");
1436       __ bind(L);
1437     }
1438 #endif
1439   }
1440 
1441   // start execution
1442 #ifdef ASSERT
1443   {
1444     Label L;
1445      const Address monitor_block_top (rbp,
1446                  frame::interpreter_frame_monitor_block_top_offset * wordSize);
1447     __ movptr(rax, monitor_block_top);
1448     __ cmpptr(rax, rsp);
1449     __ jcc(Assembler::equal, L);
1450     __ stop("broken stack frame setup in interpreter");
1451     __ bind(L);
1452   }
1453 #endif
1454 
1455   // jvmti support
1456   __ notify_method_entry();
1457 
1458   __ dispatch_next(vtos);
1459 
1460   // invocation counter overflow
1461   if (inc_counter) {
1462     if (ProfileInterpreter) {
1463       // We have decided to profile this method in the interpreter
1464       __ bind(profile_method);
1465       __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1466       __ set_method_data_pointer_for_bcp();
1467       __ get_method(rbx);
1468       __ jmp(profile_method_continue);
1469     }
1470     // Handle overflow of counter and compile method
1471     __ bind(invocation_counter_overflow);
1472     generate_counter_overflow(continue_after_compile);
1473   }
1474 
1475   return entry_point;
1476 }
1477 
1478 //-----------------------------------------------------------------------------
1479 // Exceptions
1480 
1481 void TemplateInterpreterGenerator::generate_throw_exception() {
1482   // Entry point in previous activation (i.e., if the caller was
1483   // interpreted)
1484   Interpreter::_rethrow_exception_entry = __ pc();
1485   // Restore sp to interpreter_frame_last_sp even though we are going
1486   // to empty the expression stack for the exception processing.
1487   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
1488   // rax: exception
1489   // rdx: return address/pc that threw exception
1490   __ restore_bcp();    // r13/rsi points to call/send
1491   __ restore_locals();
1492   LP64_ONLY(__ reinit_heapbase());  // restore r12 as heapbase.
1493   // Entry point for exceptions thrown within interpreter code
1494   Interpreter::_throw_exception_entry = __ pc();
1495   // expression stack is undefined here
1496   // rax: exception
1497   // r13/rsi: exception bcp
1498   __ verify_oop(rax);
1499   Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
1500   LP64_ONLY(__ mov(c_rarg1, rax));
1501 
1502   // expression stack must be empty before entering the VM in case of
1503   // an exception
1504   __ empty_expression_stack();
1505   // find exception handler address and preserve exception oop
1506   __ call_VM(rdx,
1507              CAST_FROM_FN_PTR(address,
1508                           InterpreterRuntime::exception_handler_for_exception),
1509              rarg);
1510   // rax: exception handler entry point
1511   // rdx: preserved exception oop
1512   // r13/rsi: bcp for exception handler
1513   __ push_ptr(rdx); // push exception which is now the only value on the stack
1514   __ jmp(rax); // jump to exception handler (may be _remove_activation_entry!)
1515 
1516   // If the exception is not handled in the current frame the frame is
1517   // removed and the exception is rethrown (i.e. exception
1518   // continuation is _rethrow_exception).
1519   //
1520   // Note: At this point the bci is still the bxi for the instruction
1521   // which caused the exception and the expression stack is
1522   // empty. Thus, for any VM calls at this point, GC will find a legal
1523   // oop map (with empty expression stack).
1524 
1525   // In current activation
1526   // tos: exception
1527   // esi: exception bcp
1528 
1529   //
1530   // JVMTI PopFrame support
1531   //
1532 
1533   Interpreter::_remove_activation_preserving_args_entry = __ pc();
1534   __ empty_expression_stack();
1535   // Set the popframe_processing bit in pending_popframe_condition
1536   // indicating that we are currently handling popframe, so that
1537   // call_VMs that may happen later do not trigger new popframe
1538   // handling cycles.
1539   const Register thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
1540   NOT_LP64(__ get_thread(thread));
1541   __ movl(rdx, Address(thread, JavaThread::popframe_condition_offset()));
1542   __ orl(rdx, JavaThread::popframe_processing_bit);
1543   __ movl(Address(thread, JavaThread::popframe_condition_offset()), rdx);
1544 
1545   {
1546     // Check to see whether we are returning to a deoptimized frame.
1547     // (The PopFrame call ensures that the caller of the popped frame is
1548     // either interpreted or compiled and deoptimizes it if compiled.)
1549     // In this case, we can't call dispatch_next() after the frame is
1550     // popped, but instead must save the incoming arguments and restore
1551     // them after deoptimization has occurred.
1552     //
1553     // Note that we don't compare the return PC against the
1554     // deoptimization blob's unpack entry because of the presence of
1555     // adapter frames in C2.
1556     Label caller_not_deoptimized;
1557     Register rarg = NOT_LP64(rdx) LP64_ONLY(c_rarg1);
1558     __ movptr(rarg, Address(rbp, frame::return_addr_offset * wordSize));
1559     __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1560                                InterpreterRuntime::interpreter_contains), rarg);
1561     __ testl(rax, rax);
1562     __ jcc(Assembler::notZero, caller_not_deoptimized);
1563 
1564     // Compute size of arguments for saving when returning to
1565     // deoptimized caller
1566     __ get_method(rax);
1567     __ movptr(rax, Address(rax, Method::const_offset()));
1568     __ load_unsigned_short(rax, Address(rax, in_bytes(ConstMethod::
1569                                                 size_of_parameters_offset())));
1570     __ shll(rax, Interpreter::logStackElementSize);
1571     __ restore_locals();
1572     __ subptr(rlocals, rax);
1573     __ addptr(rlocals, wordSize);
1574     // Save these arguments
1575     NOT_LP64(__ get_thread(thread));
1576     __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1577                                            Deoptimization::
1578                                            popframe_preserve_args),
1579                           thread, rax, rlocals);
1580 
1581     __ remove_activation(vtos, rdx,
1582                          /* throw_monitor_exception */ false,
1583                          /* install_monitor_exception */ false,
1584                          /* notify_jvmdi */ false);
1585 
1586     // Inform deoptimization that it is responsible for restoring
1587     // these arguments
1588     NOT_LP64(__ get_thread(thread));
1589     __ movl(Address(thread, JavaThread::popframe_condition_offset()),
1590             JavaThread::popframe_force_deopt_reexecution_bit);
1591 
1592     // Continue in deoptimization handler
1593     __ jmp(rdx);
1594 
1595     __ bind(caller_not_deoptimized);
1596   }
1597 
1598   __ remove_activation(vtos, rdx, /* rdx result (retaddr) is not used */
1599                        /* throw_monitor_exception */ false,
1600                        /* install_monitor_exception */ false,
1601                        /* notify_jvmdi */ false);
1602 
1603   // Finish with popframe handling
1604   // A previous I2C followed by a deoptimization might have moved the
1605   // outgoing arguments further up the stack. PopFrame expects the
1606   // mutations to those outgoing arguments to be preserved and other
1607   // constraints basically require this frame to look exactly as
1608   // though it had previously invoked an interpreted activation with
1609   // no space between the top of the expression stack (current
1610   // last_sp) and the top of stack. Rather than force deopt to
1611   // maintain this kind of invariant all the time we call a small
1612   // fixup routine to move the mutated arguments onto the top of our
1613   // expression stack if necessary.
1614 #ifndef _LP64
1615   __ mov(rax, rsp);
1616   __ movptr(rbx, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1617   __ get_thread(thread);
1618   // PC must point into interpreter here
1619   __ set_last_Java_frame(thread, noreg, rbp, __ pc());
1620   __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), thread, rax, rbx);
1621   __ get_thread(thread);
1622 #else
1623   __ mov(c_rarg1, rsp);
1624   __ movptr(c_rarg2, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1625   // PC must point into interpreter here
1626   __ set_last_Java_frame(noreg, rbp, __ pc());
1627   __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), r15_thread, c_rarg1, c_rarg2);
1628 #endif
1629   __ reset_last_Java_frame(thread, true, true);
1630 
1631   // Restore the last_sp and null it out
1632   __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1633   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
1634 
1635   __ restore_bcp();
1636   __ restore_locals();
1637   // The method data pointer was incremented already during
1638   // call profiling. We have to restore the mdp for the current bcp.
1639   if (ProfileInterpreter) {
1640     __ set_method_data_pointer_for_bcp();
1641   }
1642 
1643   // Clear the popframe condition flag
1644   NOT_LP64(__ get_thread(thread));
1645   __ movl(Address(thread, JavaThread::popframe_condition_offset()),
1646           JavaThread::popframe_inactive);
1647 
1648 #if INCLUDE_JVMTI
1649   {
1650     Label L_done;
1651     const Register local0 = rlocals;
1652 
1653     __ cmpb(Address(rbcp, 0), Bytecodes::_invokestatic);
1654     __ jcc(Assembler::notEqual, L_done);
1655 
1656     // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call.
1657     // Detect such a case in the InterpreterRuntime function and return the member name argument, or NULL.
1658 
1659     __ get_method(rdx);
1660     __ movptr(rax, Address(local0, 0));
1661     __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), rax, rdx, rbcp);
1662 
1663     __ testptr(rax, rax);
1664     __ jcc(Assembler::zero, L_done);
1665 
1666     __ movptr(Address(rbx, 0), rax);
1667     __ bind(L_done);
1668   }
1669 #endif // INCLUDE_JVMTI
1670 
1671   __ dispatch_next(vtos);
1672   // end of PopFrame support
1673 
1674   Interpreter::_remove_activation_entry = __ pc();
1675 
1676   // preserve exception over this code sequence
1677   __ pop_ptr(rax);
1678   NOT_LP64(__ get_thread(thread));
1679   __ movptr(Address(thread, JavaThread::vm_result_offset()), rax);
1680   // remove the activation (without doing throws on illegalMonitorExceptions)
1681   __ remove_activation(vtos, rdx, false, true, false);
1682   // restore exception
1683   NOT_LP64(__ get_thread(thread));
1684   __ get_vm_result(rax, thread);
1685 
1686   // In between activations - previous activation type unknown yet
1687   // compute continuation point - the continuation point expects the
1688   // following registers set up:
1689   //
1690   // rax: exception
1691   // rdx: return address/pc that threw exception
1692   // rsp: expression stack of caller
1693   // rbp: ebp of caller
1694   __ push(rax);                                  // save exception
1695   __ push(rdx);                                  // save return address
1696   __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1697                           SharedRuntime::exception_handler_for_return_address),
1698                         thread, rdx);
1699   __ mov(rbx, rax);                              // save exception handler
1700   __ pop(rdx);                                   // restore return address
1701   __ pop(rax);                                   // restore exception
1702   // Note that an "issuing PC" is actually the next PC after the call
1703   __ jmp(rbx);                                   // jump to exception
1704                                                  // handler of caller
1705 }
1706 
1707 
1708 //
1709 // JVMTI ForceEarlyReturn support
1710 //
1711 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
1712   address entry = __ pc();
1713 
1714   __ restore_bcp();
1715   __ restore_locals();
1716   __ empty_expression_stack();
1717   __ load_earlyret_value(state);  // 32 bits returns value in rdx, so don't reuse
1718 
1719   const Register thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
1720   NOT_LP64(__ get_thread(thread));
1721   __ movptr(rcx, Address(thread, JavaThread::jvmti_thread_state_offset()));
1722   Address cond_addr(rcx, JvmtiThreadState::earlyret_state_offset());
1723 
1724   // Clear the earlyret state
1725   __ movl(cond_addr, JvmtiThreadState::earlyret_inactive);
1726 
1727   __ remove_activation(state, rsi,
1728                        false, /* throw_monitor_exception */
1729                        false, /* install_monitor_exception */
1730                        true); /* notify_jvmdi */
1731   __ jmp(rsi);
1732 
1733   return entry;
1734 } // end of ForceEarlyReturn support
1735 
1736 
1737 //-----------------------------------------------------------------------------
1738 // Helper for vtos entry point generation
1739 
1740 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
1741                                                          address& bep,
1742                                                          address& cep,
1743                                                          address& sep,
1744                                                          address& aep,
1745                                                          address& iep,
1746                                                          address& lep,
1747                                                          address& fep,
1748                                                          address& dep,
1749                                                          address& vep) {
1750   assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
1751   Label L;
1752   aep = __ pc();  __ push_ptr();   __ jmp(L);
1753 #ifndef _LP64
1754   fep = __ pc(); __ push(ftos); __ jmp(L);
1755   dep = __ pc(); __ push(dtos); __ jmp(L);
1756 #else
1757   fep = __ pc();  __ push_f(xmm0); __ jmp(L);
1758   dep = __ pc();  __ push_d(xmm0); __ jmp(L);
1759 #endif // _LP64
1760   lep = __ pc();  __ push_l();     __ jmp(L);
1761   bep = cep = sep =
1762   iep = __ pc();  __ push_i();
1763   vep = __ pc();
1764   __ bind(L);
1765   generate_and_dispatch(t);
1766 }
1767 
1768 //-----------------------------------------------------------------------------
1769 
1770 // Non-product code
1771 #ifndef PRODUCT
1772 
1773 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
1774   address entry = __ pc();
1775 
1776 #ifndef _LP64
1777   // prepare expression stack
1778   __ pop(rcx);          // pop return address so expression stack is 'pure'
1779   __ push(state);       // save tosca
1780 
1781   // pass tosca registers as arguments & call tracer
1782   __ call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::trace_bytecode), rcx, rax, rdx);
1783   __ mov(rcx, rax);     // make sure return address is not destroyed by pop(state)
1784   __ pop(state);        // restore tosca
1785 
1786   // return
1787   __ jmp(rcx);
1788 #else
1789   __ push(state);
1790   __ push(c_rarg0);
1791   __ push(c_rarg1);
1792   __ push(c_rarg2);
1793   __ push(c_rarg3);
1794   __ mov(c_rarg2, rax);  // Pass itos
1795 #ifdef _WIN64
1796   __ movflt(xmm3, xmm0); // Pass ftos
1797 #endif
1798   __ call_VM(noreg,
1799              CAST_FROM_FN_PTR(address, SharedRuntime::trace_bytecode),
1800              c_rarg1, c_rarg2, c_rarg3);
1801   __ pop(c_rarg3);
1802   __ pop(c_rarg2);
1803   __ pop(c_rarg1);
1804   __ pop(c_rarg0);
1805   __ pop(state);
1806   __ ret(0);                                   // return from result handler
1807 #endif // _LP64
1808 
1809   return entry;
1810 }
1811 
1812 void TemplateInterpreterGenerator::count_bytecode() {
1813   __ incrementl(ExternalAddress((address) &BytecodeCounter::_counter_value));
1814 }
1815 
1816 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
1817   __ incrementl(ExternalAddress((address) &BytecodeHistogram::_counters[t->bytecode()]));
1818 }
1819 
1820 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
1821   __ mov32(rbx, ExternalAddress((address) &BytecodePairHistogram::_index));
1822   __ shrl(rbx, BytecodePairHistogram::log2_number_of_codes);
1823   __ orl(rbx,
1824          ((int) t->bytecode()) <<
1825          BytecodePairHistogram::log2_number_of_codes);
1826   __ mov32(ExternalAddress((address) &BytecodePairHistogram::_index), rbx);
1827   __ lea(rscratch1, ExternalAddress((address) BytecodePairHistogram::_counters));
1828   __ incrementl(Address(rscratch1, rbx, Address::times_4));
1829 }
1830 
1831 
1832 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
1833   // Call a little run-time stub to avoid blow-up for each bytecode.
1834   // The run-time runtime saves the right registers, depending on
1835   // the tosca in-state for the given template.
1836 
1837   assert(Interpreter::trace_code(t->tos_in()) != NULL,
1838          "entry must have been generated");
1839 #ifndef _LP64
1840   __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in())));
1841 #else
1842   __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1843   __ andptr(rsp, -16); // align stack as required by ABI
1844   __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in())));
1845   __ mov(rsp, r12); // restore sp
1846   __ reinit_heapbase();
1847 #endif // _LP64
1848 }
1849 
1850 
1851 void TemplateInterpreterGenerator::stop_interpreter_at() {
1852   Label L;
1853   __ cmp32(ExternalAddress((address) &BytecodeCounter::_counter_value),
1854            StopInterpreterAt);
1855   __ jcc(Assembler::notEqual, L);
1856   __ int3();
1857   __ bind(L);
1858 }
1859 #endif // !PRODUCT