1 /*
   2  * Copyright (c) 2003, 2022, 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 "classfile/javaClasses.hpp"
  28 #include "compiler/compiler_globals.hpp"
  29 #include "compiler/disassembler.hpp"
  30 #include "gc/shared/barrierSetAssembler.hpp"
  31 #include "interpreter/bytecodeHistogram.hpp"
  32 #include "interpreter/interp_masm.hpp"
  33 #include "interpreter/interpreter.hpp"
  34 #include "interpreter/interpreterRuntime.hpp"
  35 #include "interpreter/templateInterpreterGenerator.hpp"
  36 #include "interpreter/templateTable.hpp"
  37 #include "oops/arrayOop.hpp"
  38 #include "oops/methodData.hpp"
  39 #include "oops/method.hpp"
  40 #include "oops/oop.inline.hpp"
  41 #include "oops/inlineKlass.hpp"
  42 #include "prims/jvmtiExport.hpp"
  43 #include "prims/jvmtiThreadState.hpp"
  44 #include "runtime/continuation.hpp"
  45 #include "runtime/deoptimization.hpp"
  46 #include "runtime/frame.inline.hpp"
  47 #include "runtime/globals.hpp"
  48 #include "runtime/jniHandles.hpp"
  49 #include "runtime/sharedRuntime.hpp"
  50 #include "runtime/stubRoutines.hpp"
  51 #include "runtime/synchronizer.hpp"
  52 #include "runtime/timer.hpp"
  53 #include "runtime/vframeArray.hpp"
  54 #include "utilities/debug.hpp"
  55 #include "utilities/macros.hpp"
  56 
  57 #define __ Disassembler::hook<InterpreterMacroAssembler>(__FILE__, __LINE__, _masm)->
  58 
  59 // Size of interpreter code.  Increase if too small.  Interpreter will
  60 // fail with a guarantee ("not enough space for interpreter generation");
  61 // if too small.
  62 // Run with +PrintInterpreter to get the VM to print out the size.
  63 // Max size with JVMTI
  64 #ifdef AMD64
  65 int TemplateInterpreter::InterpreterCodeSize = JVMCI_ONLY(280) NOT_JVMCI(268) * 1024;
  66 #else
  67 int TemplateInterpreter::InterpreterCodeSize = 224 * 1024;
  68 #endif // AMD64
  69 
  70 // Global Register Names
  71 static const Register rbcp     = LP64_ONLY(r13) NOT_LP64(rsi);
  72 static const Register rlocals  = LP64_ONLY(r14) NOT_LP64(rdi);
  73 
  74 const int method_offset = frame::interpreter_frame_method_offset * wordSize;
  75 const int bcp_offset    = frame::interpreter_frame_bcp_offset    * wordSize;
  76 const int locals_offset = frame::interpreter_frame_locals_offset * wordSize;
  77 
  78 
  79 //-----------------------------------------------------------------------------
  80 
  81 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() {
  82   address entry = __ pc();
  83 
  84 #ifdef ASSERT
  85   {
  86     Label L;
  87     __ lea(rax, Address(rbp,
  88                         frame::interpreter_frame_monitor_block_top_offset *
  89                         wordSize));
  90     __ cmpptr(rax, rsp); // rax = maximal rsp for current rbp (stack
  91                          // grows negative)
  92     __ jcc(Assembler::aboveEqual, L); // check if frame is complete
  93     __ stop ("interpreter frame not set up");
  94     __ bind(L);
  95   }
  96 #endif // ASSERT
  97   // Restore bcp under the assumption that the current frame is still
  98   // interpreted
  99   __ restore_bcp();
 100 
 101   // expression stack must be empty before entering the VM if an
 102   // exception happened
 103   __ empty_expression_stack();
 104   // throw exception
 105   __ call_VM(noreg,
 106              CAST_FROM_FN_PTR(address,
 107                               InterpreterRuntime::throw_StackOverflowError));
 108   return entry;
 109 }
 110 
 111 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler() {
 112   address entry = __ pc();
 113   // The expression stack must be empty before entering the VM if an
 114   // exception happened.
 115   __ empty_expression_stack();
 116 
 117   // Setup parameters.
 118   // ??? convention: expect aberrant index in register ebx/rbx.
 119   // Pass array to create more detailed exceptions.
 120   Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
 121   __ call_VM(noreg,
 122              CAST_FROM_FN_PTR(address,
 123                               InterpreterRuntime::
 124                               throw_ArrayIndexOutOfBoundsException),
 125              rarg, rbx);
 126   return entry;
 127 }
 128 
 129 address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
 130   address entry = __ pc();
 131 
 132   // object is at TOS
 133   Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
 134   __ pop(rarg);
 135 
 136   // expression stack must be empty before entering the VM if an
 137   // exception happened
 138   __ empty_expression_stack();
 139 
 140   __ call_VM(noreg,
 141              CAST_FROM_FN_PTR(address,
 142                               InterpreterRuntime::
 143                               throw_ClassCastException),
 144              rarg);
 145   return entry;
 146 }
 147 
 148 address TemplateInterpreterGenerator::generate_exception_handler_common(
 149         const char* name, const char* message, bool pass_oop) {
 150   assert(!pass_oop || message == NULL, "either oop or message but not both");
 151   address entry = __ pc();
 152 
 153   Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
 154   Register rarg2 = NOT_LP64(rbx) LP64_ONLY(c_rarg2);
 155 
 156   if (pass_oop) {
 157     // object is at TOS
 158     __ pop(rarg2);
 159   }
 160   // expression stack must be empty before entering the VM if an
 161   // exception happened
 162   __ empty_expression_stack();
 163   // setup parameters
 164   __ lea(rarg, ExternalAddress((address)name));
 165   if (pass_oop) {
 166     __ call_VM(rax, CAST_FROM_FN_PTR(address,
 167                                      InterpreterRuntime::
 168                                      create_klass_exception),
 169                rarg, rarg2);
 170   } else {
 171     __ lea(rarg2, ExternalAddress((address)message));
 172     __ call_VM(rax,
 173                CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception),
 174                rarg, rarg2);
 175   }
 176   // throw exception
 177   __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
 178   return entry;
 179 }
 180 
 181 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step, size_t index_size) {
 182   address entry = __ pc();
 183 
 184 #ifndef _LP64
 185 #ifdef COMPILER2
 186   // The FPU stack is clean if UseSSE >= 2 but must be cleaned in other cases
 187   if ((state == ftos && UseSSE < 1) || (state == dtos && UseSSE < 2)) {
 188     for (int i = 1; i < 8; i++) {
 189         __ ffree(i);
 190     }
 191   } else if (UseSSE < 2) {
 192     __ empty_FPU_stack();
 193   }
 194 #endif // COMPILER2
 195   if ((state == ftos && UseSSE < 1) || (state == dtos && UseSSE < 2)) {
 196     __ MacroAssembler::verify_FPU(1, "generate_return_entry_for compiled");
 197   } else {
 198     __ MacroAssembler::verify_FPU(0, "generate_return_entry_for compiled");
 199   }
 200 
 201   if (state == ftos) {
 202     __ MacroAssembler::verify_FPU(UseSSE >= 1 ? 0 : 1, "generate_return_entry_for in interpreter");
 203   } else if (state == dtos) {
 204     __ MacroAssembler::verify_FPU(UseSSE >= 2 ? 0 : 1, "generate_return_entry_for in interpreter");
 205   }
 206 #endif // _LP64
 207 
 208   // Restore stack bottom in case i2c adjusted stack
 209   __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
 210   // and NULL it as marker that esp is now tos until next java call
 211   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
 212 
 213   if (state == atos && InlineTypeReturnedAsFields) {
 214     __ store_inline_type_fields_to_buf(NULL);
 215   }
 216 
 217   __ restore_bcp();
 218   __ restore_locals();
 219 
 220   if (state == atos) {
 221     Register mdp = rbx;
 222     Register tmp = rcx;
 223     __ profile_return_type(mdp, rax, tmp);
 224   }
 225 
 226   const Register cache = rbx;
 227   const Register index = rcx;
 228   __ get_cache_and_index_at_bcp(cache, index, 1, index_size);
 229 
 230   const Register flags = cache;
 231   __ movl(flags, Address(cache, index, Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()));
 232   __ andl(flags, ConstantPoolCacheEntry::parameter_size_mask);
 233   __ lea(rsp, Address(rsp, flags, Interpreter::stackElementScale()));
 234 
 235    const Register java_thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
 236    if (JvmtiExport::can_pop_frame()) {
 237      NOT_LP64(__ get_thread(java_thread));
 238      __ check_and_handle_popframe(java_thread);
 239    }
 240    if (JvmtiExport::can_force_early_return()) {
 241      NOT_LP64(__ get_thread(java_thread));
 242      __ check_and_handle_earlyret(java_thread);
 243    }
 244 
 245   __ dispatch_next(state, step);
 246 
 247   return entry;
 248 }
 249 
 250 
 251 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state, int step, address continuation) {
 252   address entry = __ pc();
 253 
 254 #ifndef _LP64
 255   if (state == ftos) {
 256     __ MacroAssembler::verify_FPU(UseSSE >= 1 ? 0 : 1, "generate_deopt_entry_for in interpreter");
 257   } else if (state == dtos) {
 258     __ MacroAssembler::verify_FPU(UseSSE >= 2 ? 0 : 1, "generate_deopt_entry_for in interpreter");
 259   }
 260 #endif // _LP64
 261 
 262   // NULL last_sp until next java call
 263   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
 264   __ restore_bcp();
 265   __ restore_locals();
 266   const Register thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
 267   NOT_LP64(__ get_thread(thread));
 268 #if INCLUDE_JVMCI
 269   // Check if we need to take lock at entry of synchronized method.  This can
 270   // only occur on method entry so emit it only for vtos with step 0.
 271   if (EnableJVMCI && state == vtos && step == 0) {
 272     Label L;
 273     __ cmpb(Address(thread, JavaThread::pending_monitorenter_offset()), 0);
 274     __ jcc(Assembler::zero, L);
 275     // Clear flag.
 276     __ movb(Address(thread, JavaThread::pending_monitorenter_offset()), 0);
 277     // Satisfy calling convention for lock_method().
 278     __ get_method(rbx);
 279     // Take lock.
 280     lock_method();
 281     __ bind(L);
 282   } else {
 283 #ifdef ASSERT
 284     if (EnableJVMCI) {
 285       Label L;
 286       __ cmpb(Address(r15_thread, JavaThread::pending_monitorenter_offset()), 0);
 287       __ jcc(Assembler::zero, L);
 288       __ stop("unexpected pending monitor in deopt entry");
 289       __ bind(L);
 290     }
 291 #endif
 292   }
 293 #endif
 294   // handle exceptions
 295   {
 296     Label L;
 297     __ cmpptr(Address(thread, Thread::pending_exception_offset()), NULL_WORD);
 298     __ jcc(Assembler::zero, L);
 299     __ call_VM(noreg,
 300                CAST_FROM_FN_PTR(address,
 301                                 InterpreterRuntime::throw_pending_exception));
 302     __ should_not_reach_here();
 303     __ bind(L);
 304   }
 305   if (continuation == NULL) {
 306     __ dispatch_next(state, step);
 307   } else {
 308     __ jump_to_entry(continuation);
 309   }
 310   return entry;
 311 }
 312 
 313 address TemplateInterpreterGenerator::generate_result_handler_for(
 314         BasicType type) {
 315   address entry = __ pc();
 316   switch (type) {
 317   case T_BOOLEAN: __ c2bool(rax);            break;
 318 #ifndef _LP64
 319   case T_CHAR   : __ andptr(rax, 0xFFFF);    break;
 320 #else
 321   case T_CHAR   : __ movzwl(rax, rax);       break;
 322 #endif // _LP64
 323   case T_BYTE   : __ sign_extend_byte(rax);  break;
 324   case T_SHORT  : __ sign_extend_short(rax); break;
 325   case T_INT    : /* nothing to do */        break;
 326   case T_LONG   : /* nothing to do */        break;
 327   case T_VOID   : /* nothing to do */        break;
 328 #ifndef _LP64
 329   case T_DOUBLE :
 330   case T_FLOAT  :
 331     { const Register t = InterpreterRuntime::SignatureHandlerGenerator::temp();
 332       __ pop(t);                            // remove return address first
 333       // Must return a result for interpreter or compiler. In SSE
 334       // mode, results are returned in xmm0 and the FPU stack must
 335       // be empty.
 336       if (type == T_FLOAT && UseSSE >= 1) {
 337         // Load ST0
 338         __ fld_d(Address(rsp, 0));
 339         // Store as float and empty fpu stack
 340         __ fstp_s(Address(rsp, 0));
 341         // and reload
 342         __ movflt(xmm0, Address(rsp, 0));
 343       } else if (type == T_DOUBLE && UseSSE >= 2 ) {
 344         __ movdbl(xmm0, Address(rsp, 0));
 345       } else {
 346         // restore ST0
 347         __ fld_d(Address(rsp, 0));
 348       }
 349       // and pop the temp
 350       __ addptr(rsp, 2 * wordSize);
 351       __ push(t);                           // restore return address
 352     }
 353     break;
 354 #else
 355   case T_FLOAT  : /* nothing to do */        break;
 356   case T_DOUBLE : /* nothing to do */        break;
 357 #endif // _LP64
 358 
 359   case T_PRIMITIVE_OBJECT: // fall through (inline types are handled with oops)
 360   case T_OBJECT :
 361     // retrieve result from frame
 362     __ movptr(rax, Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize));
 363     // and verify it
 364     __ verify_oop(rax);
 365     break;
 366   default       : ShouldNotReachHere();
 367   }
 368   __ ret(0);                                   // return from result handler
 369   return entry;
 370 }
 371 
 372 address TemplateInterpreterGenerator::generate_safept_entry_for(
 373         TosState state,
 374         address runtime_entry) {
 375   address entry = __ pc();
 376 
 377   __ push(state);
 378   __ push_cont_fastpath();
 379   __ call_VM(noreg, runtime_entry);
 380   __ pop_cont_fastpath();
 381 
 382   __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
 383   return entry;
 384 }
 385 
 386 
 387 
 388 // Helpers for commoning out cases in the various type of method entries.
 389 //
 390 
 391 
 392 // increment invocation count & check for overflow
 393 //
 394 // Note: checking for negative value instead of overflow
 395 //       so we have a 'sticky' overflow test
 396 //
 397 // rbx: method
 398 // rcx: invocation counter
 399 //
 400 void TemplateInterpreterGenerator::generate_counter_incr(Label* overflow) {
 401   Label done;
 402   // Note: In tiered we increment either counters in Method* or in MDO depending if we're profiling or not.
 403   Label no_mdo;
 404   if (ProfileInterpreter) {
 405     // Are we profiling?
 406     __ movptr(rax, Address(rbx, Method::method_data_offset()));
 407     __ testptr(rax, rax);
 408     __ jccb(Assembler::zero, no_mdo);
 409     // Increment counter in the MDO
 410     const Address mdo_invocation_counter(rax, in_bytes(MethodData::invocation_counter_offset()) +
 411         in_bytes(InvocationCounter::counter_offset()));
 412     const Address mask(rax, in_bytes(MethodData::invoke_mask_offset()));
 413     __ increment_mask_and_jump(mdo_invocation_counter, mask, rcx, overflow);
 414     __ jmp(done);
 415   }
 416   __ bind(no_mdo);
 417   // Increment counter in MethodCounters
 418   const Address invocation_counter(rax,
 419       MethodCounters::invocation_counter_offset() +
 420       InvocationCounter::counter_offset());
 421   __ get_method_counters(rbx, rax, done);
 422   const Address mask(rax, in_bytes(MethodCounters::invoke_mask_offset()));
 423   __ increment_mask_and_jump(invocation_counter, mask, rcx, overflow);
 424   __ bind(done);
 425 }
 426 
 427 void TemplateInterpreterGenerator::generate_counter_overflow(Label& do_continue) {
 428 
 429   // Asm interpreter on entry
 430   // r14/rdi - locals
 431   // r13/rsi - bcp
 432   // rbx - method
 433   // rdx - cpool --- DOES NOT APPEAR TO BE TRUE
 434   // rbp - interpreter frame
 435 
 436   // On return (i.e. jump to entry_point) [ back to invocation of interpreter ]
 437   // Everything as it was on entry
 438   // rdx is not restored. Doesn't appear to really be set.
 439 
 440   // InterpreterRuntime::frequency_counter_overflow takes two
 441   // arguments, the first (thread) is passed by call_VM, the second
 442   // indicates if the counter overflow occurs at a backwards branch
 443   // (NULL bcp).  We pass zero for it.  The call returns the address
 444   // of the verified entry point for the method or NULL if the
 445   // compilation did not complete (either went background or bailed
 446   // out).
 447   Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
 448   __ movl(rarg, 0);
 449   __ call_VM(noreg,
 450              CAST_FROM_FN_PTR(address,
 451                               InterpreterRuntime::frequency_counter_overflow),
 452              rarg);
 453 
 454   __ movptr(rbx, Address(rbp, method_offset));   // restore Method*
 455   // Preserve invariant that r13/r14 contain bcp/locals of sender frame
 456   // and jump to the interpreted entry.
 457   __ jmp(do_continue, relocInfo::none);
 458 }
 459 
 460 // See if we've got enough room on the stack for locals plus overhead below
 461 // JavaThread::stack_overflow_limit(). If not, throw a StackOverflowError
 462 // without going through the signal handler, i.e., reserved and yellow zones
 463 // will not be made usable. The shadow zone must suffice to handle the
 464 // overflow.
 465 // The expression stack grows down incrementally, so the normal guard
 466 // page mechanism will work for that.
 467 //
 468 // NOTE: Since the additional locals are also always pushed (wasn't
 469 // obvious in generate_fixed_frame) so the guard should work for them
 470 // too.
 471 //
 472 // Args:
 473 //      rdx: number of additional locals this frame needs (what we must check)
 474 //      rbx: Method*
 475 //
 476 // Kills:
 477 //      rax
 478 void TemplateInterpreterGenerator::generate_stack_overflow_check(void) {
 479 
 480   // monitor entry size: see picture of stack in frame_x86.hpp
 481   const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
 482 
 483   // total overhead size: entry_size + (saved rbp through expr stack
 484   // bottom).  be sure to change this if you add/subtract anything
 485   // to/from the overhead area
 486   const int overhead_size =
 487     -(frame::interpreter_frame_initial_sp_offset * wordSize) + entry_size;
 488 
 489   const int page_size = os::vm_page_size();
 490 
 491   Label after_frame_check;
 492 
 493   // see if the frame is greater than one page in size. If so,
 494   // then we need to verify there is enough stack space remaining
 495   // for the additional locals.
 496   __ cmpl(rdx, (page_size - overhead_size) / Interpreter::stackElementSize);
 497   __ jcc(Assembler::belowEqual, after_frame_check);
 498 
 499   // compute rsp as if this were going to be the last frame on
 500   // the stack before the red zone
 501 
 502   Label after_frame_check_pop;
 503   const Register thread = NOT_LP64(rsi) LP64_ONLY(r15_thread);
 504 #ifndef _LP64
 505   __ push(thread);
 506   __ get_thread(thread);
 507 #endif
 508 
 509   const Address stack_limit(thread, JavaThread::stack_overflow_limit_offset());
 510 
 511   // locals + overhead, in bytes
 512   __ mov(rax, rdx);
 513   __ shlptr(rax, Interpreter::logStackElementSize); // Convert parameter count to bytes.
 514   __ addptr(rax, overhead_size);
 515 
 516 #ifdef ASSERT
 517   Label limit_okay;
 518   // Verify that thread stack overflow limit is non-zero.
 519   __ cmpptr(stack_limit, NULL_WORD);
 520   __ jcc(Assembler::notEqual, limit_okay);
 521   __ stop("stack overflow limit is zero");
 522   __ bind(limit_okay);
 523 #endif
 524 
 525   // Add locals/frame size to stack limit.
 526   __ addptr(rax, stack_limit);
 527 
 528   // Check against the current stack bottom.
 529   __ cmpptr(rsp, rax);
 530 
 531   __ jcc(Assembler::above, after_frame_check_pop);
 532   NOT_LP64(__ pop(rsi));  // get saved bcp
 533 
 534   // Restore sender's sp as SP. This is necessary if the sender's
 535   // frame is an extended compiled frame (see gen_c2i_adapter())
 536   // and safer anyway in case of JSR292 adaptations.
 537 
 538   __ pop(rax); // return address must be moved if SP is changed
 539   __ mov(rsp, rbcp);
 540   __ push(rax);
 541 
 542   // Note: the restored frame is not necessarily interpreted.
 543   // Use the shared runtime version of the StackOverflowError.
 544   assert(StubRoutines::throw_StackOverflowError_entry() != NULL, "stub not yet generated");
 545   __ jump(ExternalAddress(StubRoutines::throw_StackOverflowError_entry()));
 546   // all done with frame size check
 547   __ bind(after_frame_check_pop);
 548   NOT_LP64(__ pop(rsi));
 549 
 550   // all done with frame size check
 551   __ bind(after_frame_check);
 552 }
 553 
 554 // Allocate monitor and lock method (asm interpreter)
 555 //
 556 // Args:
 557 //      rbx: Method*
 558 //      r14/rdi: locals
 559 //
 560 // Kills:
 561 //      rax
 562 //      c_rarg0, c_rarg1, c_rarg2, c_rarg3, ...(param regs)
 563 //      rscratch1, rscratch2 (scratch regs)
 564 void TemplateInterpreterGenerator::lock_method() {
 565   // synchronize method
 566   const Address access_flags(rbx, Method::access_flags_offset());
 567   const Address monitor_block_top(
 568         rbp,
 569         frame::interpreter_frame_monitor_block_top_offset * wordSize);
 570   const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
 571 
 572 #ifdef ASSERT
 573   {
 574     Label L;
 575     __ movl(rax, access_flags);
 576     __ testl(rax, JVM_ACC_SYNCHRONIZED);
 577     __ jcc(Assembler::notZero, L);
 578     __ stop("method doesn't need synchronization");
 579     __ bind(L);
 580   }
 581 #endif // ASSERT
 582 
 583   // get synchronization object
 584   {
 585     Label done;
 586     __ movl(rax, access_flags);
 587     __ testl(rax, JVM_ACC_STATIC);
 588     // get receiver (assume this is frequent case)
 589     __ movptr(rax, Address(rlocals, Interpreter::local_offset_in_bytes(0)));
 590     __ jcc(Assembler::zero, done);
 591     __ load_mirror(rax, rbx, rscratch2);
 592 
 593 #ifdef ASSERT
 594     {
 595       Label L;
 596       __ testptr(rax, rax);
 597       __ jcc(Assembler::notZero, L);
 598       __ stop("synchronization object is NULL");
 599       __ bind(L);
 600     }
 601 #endif // ASSERT
 602 
 603     __ bind(done);
 604   }
 605 
 606   // add space for monitor & lock
 607   __ subptr(rsp, entry_size); // add space for a monitor entry
 608   __ movptr(monitor_block_top, rsp);  // set new monitor block top
 609   // store object
 610   __ movptr(Address(rsp, BasicObjectLock::obj_offset_in_bytes()), rax);
 611   const Register lockreg = NOT_LP64(rdx) LP64_ONLY(c_rarg1);
 612   __ movptr(lockreg, rsp); // object address
 613   __ lock_object(lockreg);
 614 }
 615 
 616 // Generate a fixed interpreter frame. This is identical setup for
 617 // interpreted methods and for native methods hence the shared code.
 618 //
 619 // Args:
 620 //      rax: return address
 621 //      rbx: Method*
 622 //      r14/rdi: pointer to locals
 623 //      r13/rsi: sender sp
 624 //      rdx: cp cache
 625 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
 626   // initialize fixed part of activation frame
 627   __ push(rax);        // save return address
 628   __ enter();          // save old & set new rbp
 629   __ push(rbcp);        // set sender sp
 630   __ push(NULL_WORD); // leave last_sp as null
 631   __ movptr(rbcp, Address(rbx, Method::const_offset()));      // get ConstMethod*
 632   __ lea(rbcp, Address(rbcp, ConstMethod::codes_offset())); // get codebase
 633   __ push(rbx);        // save Method*
 634   // Get mirror and store it in the frame as GC root for this Method*
 635   __ load_mirror(rdx, rbx, rscratch2);
 636   __ push(rdx);
 637   if (ProfileInterpreter) {
 638     Label method_data_continue;
 639     __ movptr(rdx, Address(rbx, in_bytes(Method::method_data_offset())));
 640     __ testptr(rdx, rdx);
 641     __ jcc(Assembler::zero, method_data_continue);
 642     __ addptr(rdx, in_bytes(MethodData::data_offset()));
 643     __ bind(method_data_continue);
 644     __ push(rdx);      // set the mdp (method data pointer)
 645   } else {
 646     __ push(0);
 647   }
 648 
 649   __ movptr(rdx, Address(rbx, Method::const_offset()));
 650   __ movptr(rdx, Address(rdx, ConstMethod::constants_offset()));
 651   __ movptr(rdx, Address(rdx, ConstantPool::cache_offset_in_bytes()));
 652   __ push(rdx); // set constant pool cache
 653   __ push(rlocals); // set locals pointer
 654   if (native_call) {
 655     __ push(0); // no bcp
 656   } else {
 657     __ push(rbcp); // set bcp
 658   }
 659   __ push(0); // reserve word for pointer to expression stack bottom
 660   __ movptr(Address(rsp, 0), rsp); // set expression stack bottom
 661 }
 662 
 663 // End of helpers
 664 
 665 // Method entry for java.lang.ref.Reference.get.
 666 address TemplateInterpreterGenerator::generate_Reference_get_entry(void) {
 667   // Code: _aload_0, _getfield, _areturn
 668   // parameter size = 1
 669   //
 670   // The code that gets generated by this routine is split into 2 parts:
 671   //    1. The "intrinsified" code performing an ON_WEAK_OOP_REF load,
 672   //    2. The slow path - which is an expansion of the regular method entry.
 673   //
 674   // Notes:-
 675   // * An intrinsic is always executed, where an ON_WEAK_OOP_REF load is performed.
 676   // * We may jump to the slow path iff the receiver is null. If the
 677   //   Reference object is null then we no longer perform an ON_WEAK_OOP_REF load
 678   //   Thus we can use the regular method entry code to generate the NPE.
 679   //
 680   // rbx: Method*
 681 
 682   // r13: senderSP must preserve for slow path, set SP to it on fast path
 683 
 684   address entry = __ pc();
 685 
 686   const int referent_offset = java_lang_ref_Reference::referent_offset();
 687 
 688   Label slow_path;
 689   // rbx: method
 690 
 691   // Check if local 0 != NULL
 692   // If the receiver is null then it is OK to jump to the slow path.
 693   __ movptr(rax, Address(rsp, wordSize));
 694 
 695   __ testptr(rax, rax);
 696   __ jcc(Assembler::zero, slow_path);
 697 
 698   // rax: local 0
 699   // rbx: method (but can be used as scratch now)
 700   // rdx: scratch
 701   // rdi: scratch
 702 
 703   // Preserve the sender sp in case the load barrier
 704   // calls the runtime
 705   NOT_LP64(__ push(rsi));
 706 
 707   // Load the value of the referent field.
 708   const Address field_address(rax, referent_offset);
 709   __ load_heap_oop(rax, field_address, /*tmp1*/ rbx, /*tmp_thread*/ rdx, ON_WEAK_OOP_REF);
 710 
 711   // _areturn
 712   const Register sender_sp = NOT_LP64(rsi) LP64_ONLY(r13);
 713   NOT_LP64(__ pop(rsi));      // get sender sp
 714   __ pop(rdi);                // get return address
 715   __ mov(rsp, sender_sp);     // set sp to sender sp
 716   __ jmp(rdi);
 717   __ ret(0);
 718 
 719   // generate a vanilla interpreter entry as the slow path
 720   __ bind(slow_path);
 721   __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::zerolocals));
 722   return entry;
 723 }
 724 
 725 void TemplateInterpreterGenerator::bang_stack_shadow_pages(bool native_call) {
 726   // See more discussion in stackOverflow.hpp.
 727 
 728   // Note that we do the banging after the frame is setup, since the exception
 729   // handling code expects to find a valid interpreter frame on the stack.
 730   // Doing the banging earlier fails if the caller frame is not an interpreter
 731   // frame.
 732   // (Also, the exception throwing code expects to unlock any synchronized
 733   // method receiver, so do the banging after locking the receiver.)
 734 
 735   const int shadow_zone_size = checked_cast<int>(StackOverflow::stack_shadow_zone_size());
 736   const int page_size = os::vm_page_size();
 737   const int n_shadow_pages = shadow_zone_size / page_size;
 738 
 739   const Register thread = NOT_LP64(rsi) LP64_ONLY(r15_thread);
 740 #ifndef _LP64
 741   __ push(thread);
 742   __ get_thread(thread);
 743 #endif
 744 
 745 #ifdef ASSERT
 746   Label L_good_limit;
 747   __ cmpptr(Address(thread, JavaThread::shadow_zone_safe_limit()), NULL_WORD);
 748   __ jcc(Assembler::notEqual, L_good_limit);
 749   __ stop("shadow zone safe limit is not initialized");
 750   __ bind(L_good_limit);
 751 
 752   Label L_good_watermark;
 753   __ cmpptr(Address(thread, JavaThread::shadow_zone_growth_watermark()), NULL_WORD);
 754   __ jcc(Assembler::notEqual, L_good_watermark);
 755   __ stop("shadow zone growth watermark is not initialized");
 756   __ bind(L_good_watermark);
 757 #endif
 758 
 759   Label L_done;
 760 
 761   __ cmpptr(rsp, Address(thread, JavaThread::shadow_zone_growth_watermark()));
 762   __ jcc(Assembler::above, L_done);
 763 
 764   for (int p = 1; p <= n_shadow_pages; p++) {
 765     __ bang_stack_with_offset(p*page_size);
 766   }
 767 
 768   // Record the new watermark, but only if update is above the safe limit.
 769   // Otherwise, the next time around the check above would pass the safe limit.
 770   __ cmpptr(rsp, Address(thread, JavaThread::shadow_zone_safe_limit()));
 771   __ jccb(Assembler::belowEqual, L_done);
 772   __ movptr(Address(thread, JavaThread::shadow_zone_growth_watermark()), rsp);
 773 
 774   __ bind(L_done);
 775 
 776 #ifndef _LP64
 777   __ pop(thread);
 778 #endif
 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;
 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(NULL_WORD);
 820   // slot for oop temp
 821   // (static native method holder mirror/jni oop result)
 822   __ push(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);
 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 5");
 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); // Convert parameter count to bytes.
 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).
 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     __ movl(t, Address(method, Method::access_flags_offset()));
 973     __ testl(t, JVM_ACC_STATIC);
 974     __ jcc(Assembler::zero, L);
 975     // get mirror
 976     __ load_mirror(t, method, rax);
 977     // copy mirror into activation frame
 978     __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize),
 979             t);
 980     // pass handle to mirror
 981 #ifndef _LP64
 982     __ lea(t, Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize));
 983     __ movptr(Address(rsp, wordSize), t);
 984 #else
 985     __ lea(c_rarg1,
 986            Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize));
 987 #endif // _LP64
 988     __ bind(L);
 989   }
 990 
 991   // get native function entry point
 992   {
 993     Label L;
 994     __ movptr(rax, Address(method, Method::native_function_offset()));
 995     ExternalAddress unsatisfied(SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
 996     __ cmpptr(rax, unsatisfied.addr(), rscratch1);
 997     __ jcc(Assembler::notEqual, L);
 998     __ call_VM(noreg,
 999                CAST_FROM_FN_PTR(address,
1000                                 InterpreterRuntime::prepare_native_call),
1001                method);
1002     __ get_method(method);
1003     __ movptr(rax, Address(method, Method::native_function_offset()));
1004     __ bind(L);
1005   }
1006 
1007   // pass JNIEnv
1008 #ifndef _LP64
1009    __ get_thread(thread);
1010    __ lea(t, Address(thread, JavaThread::jni_environment_offset()));
1011    __ movptr(Address(rsp, 0), t);
1012 
1013    // set_last_Java_frame_before_call
1014    // It is enough that the pc()
1015    // points into the right code segment. It does not have to be the correct return pc.
1016    __ set_last_Java_frame(thread, noreg, rbp, __ pc(), noreg);
1017 #else
1018    __ lea(c_rarg0, Address(r15_thread, JavaThread::jni_environment_offset()));
1019 
1020    // It is enough that the pc() points into the right code
1021    // segment. It does not have to be the correct return pc.
1022    __ set_last_Java_frame(rsp, rbp, (address) __ pc(), rscratch1);
1023 #endif // _LP64
1024 
1025   // change thread state
1026 #ifdef ASSERT
1027   {
1028     Label L;
1029     __ movl(t, Address(thread, JavaThread::thread_state_offset()));
1030     __ cmpl(t, _thread_in_Java);
1031     __ jcc(Assembler::equal, L);
1032     __ stop("Wrong thread state in native stub");
1033     __ bind(L);
1034   }
1035 #endif
1036 
1037   // Change state to native
1038 
1039   __ movl(Address(thread, JavaThread::thread_state_offset()),
1040           _thread_in_native);
1041 
1042   // Call the native method.
1043   __ call(rax);
1044   // 32: result potentially in rdx:rax or ST0
1045   // 64: result potentially in rax or xmm0
1046 
1047   // Verify or restore cpu control state after JNI call
1048   __ restore_cpu_control_state_after_jni(rscratch1);
1049 
1050   // NOTE: The order of these pushes is known to frame::interpreter_frame_result
1051   // in order to extract the result of a method call. If the order of these
1052   // pushes change or anything else is added to the stack then the code in
1053   // interpreter_frame_result must also change.
1054 
1055 #ifndef _LP64
1056   // save potential result in ST(0) & rdx:rax
1057   // (if result handler is the T_FLOAT or T_DOUBLE handler, result must be in ST0 -
1058   // the check is necessary to avoid potential Intel FPU overflow problems by saving/restoring 'empty' FPU registers)
1059   // It is safe to do this push because state is _thread_in_native and return address will be found
1060   // via _last_native_pc and not via _last_jave_sp
1061 
1062   // NOTE: the order of these push(es) is known to frame::interpreter_frame_result.
1063   // If the order changes or anything else is added to the stack the code in
1064   // interpreter_frame_result will have to be changed.
1065 
1066   { Label L;
1067     Label push_double;
1068     ExternalAddress float_handler(AbstractInterpreter::result_handler(T_FLOAT));
1069     ExternalAddress double_handler(AbstractInterpreter::result_handler(T_DOUBLE));
1070     __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize),
1071               float_handler.addr(), noreg);
1072     __ jcc(Assembler::equal, push_double);
1073     __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize),
1074               double_handler.addr(), noreg);
1075     __ jcc(Assembler::notEqual, L);
1076     __ bind(push_double);
1077     __ push_d(); // FP values are returned using the FPU, so push FPU contents (even if UseSSE > 0).
1078     __ bind(L);
1079   }
1080 #else
1081   __ push(dtos);
1082 #endif // _LP64
1083 
1084   __ push(ltos);
1085 
1086   // change thread state
1087   NOT_LP64(__ get_thread(thread));
1088   __ movl(Address(thread, JavaThread::thread_state_offset()),
1089           _thread_in_native_trans);
1090 
1091   // Force this write out before the read below
1092   if (!UseSystemMemoryBarrier) {
1093     __ membar(Assembler::Membar_mask_bits(
1094                 Assembler::LoadLoad | Assembler::LoadStore |
1095                 Assembler::StoreLoad | Assembler::StoreStore));
1096   }
1097 #ifndef _LP64
1098   if (AlwaysRestoreFPU) {
1099     //  Make sure the control word is correct.
1100     __ fldcw(ExternalAddress(StubRoutines::x86::addr_fpu_cntrl_wrd_std()));
1101   }
1102 #endif // _LP64
1103 
1104   // check for safepoint operation in progress and/or pending suspend requests
1105   {
1106     Label Continue;
1107     Label slow_path;
1108 
1109     __ safepoint_poll(slow_path, thread, true /* at_return */, false /* in_nmethod */);
1110 
1111     __ cmpl(Address(thread, JavaThread::suspend_flags_offset()), 0);
1112     __ jcc(Assembler::equal, Continue);
1113     __ bind(slow_path);
1114 
1115     // Don't use call_VM as it will see a possible pending exception
1116     // and forward it and never return here preventing us from
1117     // clearing _last_native_pc down below.  Also can't use
1118     // call_VM_leaf either as it will check to see if r13 & r14 are
1119     // preserved and correspond to the bcp/locals pointers. So we do a
1120     // runtime call by hand.
1121     //
1122 #ifndef _LP64
1123     __ push(thread);
1124     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address,
1125                                             JavaThread::check_special_condition_for_native_trans)));
1126     __ increment(rsp, wordSize);
1127     __ get_thread(thread);
1128 #else
1129     __ mov(c_rarg0, r15_thread);
1130     __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1131     __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1132     __ andptr(rsp, -16); // align stack as required by ABI
1133     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans)));
1134     __ mov(rsp, r12); // restore sp
1135     __ reinit_heapbase();
1136 #endif // _LP64
1137     __ bind(Continue);
1138   }
1139 
1140   // change thread state
1141   __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_Java);
1142 
1143   // reset_last_Java_frame
1144   __ reset_last_Java_frame(thread, true);
1145 
1146   if (CheckJNICalls) {
1147     // clear_pending_jni_exception_check
1148     __ movptr(Address(thread, JavaThread::pending_jni_exception_check_fn_offset()), NULL_WORD);
1149   }
1150 
1151   // reset handle block
1152   __ movptr(t, Address(thread, JavaThread::active_handles_offset()));
1153   __ movl(Address(t, JNIHandleBlock::top_offset_in_bytes()), NULL_WORD);
1154 
1155   // If result is an oop unbox and store it in frame where gc will see it
1156   // and result handler will pick it up
1157 
1158   {
1159     Label no_oop;
1160     __ lea(t, ExternalAddress(AbstractInterpreter::result_handler(T_OBJECT)));
1161     __ cmpptr(t, Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize));
1162     __ jcc(Assembler::notEqual, no_oop);
1163     // retrieve result
1164     __ pop(ltos);
1165     // Unbox oop result, e.g. JNIHandles::resolve value.
1166     __ resolve_jobject(rax /* value */,
1167                        thread /* thread */,
1168                        t /* tmp */);
1169     __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize), rax);
1170     // keep stack depth as expected by pushing oop which will eventually be discarded
1171     __ push(ltos);
1172     __ bind(no_oop);
1173   }
1174 
1175 
1176   {
1177     Label no_reguard;
1178     __ cmpl(Address(thread, JavaThread::stack_guard_state_offset()),
1179             StackOverflow::stack_guard_yellow_reserved_disabled);
1180     __ jcc(Assembler::notEqual, no_reguard);
1181 
1182     __ pusha(); // XXX only save smashed registers
1183 #ifndef _LP64
1184     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
1185     __ popa();
1186 #else
1187     __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1188     __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1189     __ andptr(rsp, -16); // align stack as required by ABI
1190     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
1191     __ mov(rsp, r12); // restore sp
1192     __ popa(); // XXX only restore smashed registers
1193     __ reinit_heapbase();
1194 #endif // _LP64
1195 
1196     __ bind(no_reguard);
1197   }
1198 
1199 
1200   // The method register is junk from after the thread_in_native transition
1201   // until here.  Also can't call_VM until the bcp has been
1202   // restored.  Need bcp for throwing exception below so get it now.
1203   __ get_method(method);
1204 
1205   // restore to have legal interpreter frame, i.e., bci == 0 <=> code_base()
1206   __ movptr(rbcp, Address(method, Method::const_offset()));   // get ConstMethod*
1207   __ lea(rbcp, Address(rbcp, ConstMethod::codes_offset()));    // get codebase
1208 
1209   // handle exceptions (exception handling will handle unlocking!)
1210   {
1211     Label L;
1212     __ cmpptr(Address(thread, Thread::pending_exception_offset()), NULL_WORD);
1213     __ jcc(Assembler::zero, L);
1214     // Note: At some point we may want to unify this with the code
1215     // used in call_VM_base(); i.e., we should use the
1216     // StubRoutines::forward_exception code. For now this doesn't work
1217     // here because the rsp is not correctly set at this point.
1218     __ MacroAssembler::call_VM(noreg,
1219                                CAST_FROM_FN_PTR(address,
1220                                InterpreterRuntime::throw_pending_exception));
1221     __ should_not_reach_here();
1222     __ bind(L);
1223   }
1224 
1225   // do unlocking if necessary
1226   {
1227     Label L;
1228     __ movl(t, Address(method, Method::access_flags_offset()));
1229     __ testl(t, JVM_ACC_SYNCHRONIZED);
1230     __ jcc(Assembler::zero, L);
1231     // the code below should be shared with interpreter macro
1232     // assembler implementation
1233     {
1234       Label unlock;
1235       // BasicObjectLock will be first in list, since this is a
1236       // synchronized method. However, need to check that the object
1237       // has not been unlocked by an explicit monitorexit bytecode.
1238       const Address monitor(rbp,
1239                             (intptr_t)(frame::interpreter_frame_initial_sp_offset *
1240                                        wordSize - (int)sizeof(BasicObjectLock)));
1241 
1242       const Register regmon = NOT_LP64(rdx) LP64_ONLY(c_rarg1);
1243 
1244       // monitor expect in c_rarg1 for slow unlock path
1245       __ lea(regmon, monitor); // address of first monitor
1246 
1247       __ movptr(t, Address(regmon, BasicObjectLock::obj_offset_in_bytes()));
1248       __ testptr(t, t);
1249       __ jcc(Assembler::notZero, unlock);
1250 
1251       // Entry already unlocked, need to throw exception
1252       __ MacroAssembler::call_VM(noreg,
1253                                  CAST_FROM_FN_PTR(address,
1254                    InterpreterRuntime::throw_illegal_monitor_state_exception));
1255       __ should_not_reach_here();
1256 
1257       __ bind(unlock);
1258       __ unlock_object(regmon);
1259     }
1260     __ bind(L);
1261   }
1262 
1263   // jvmti support
1264   // Note: This must happen _after_ handling/throwing any exceptions since
1265   //       the exception handler code notifies the runtime of method exits
1266   //       too. If this happens before, method entry/exit notifications are
1267   //       not properly paired (was bug - gri 11/22/99).
1268   __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI);
1269 
1270   // restore potential result in edx:eax, call result handler to
1271   // restore potential result in ST0 & handle result
1272 
1273   __ pop(ltos);
1274   LP64_ONLY( __ pop(dtos));
1275 
1276   __ movptr(t, Address(rbp,
1277                        (frame::interpreter_frame_result_handler_offset) * wordSize));
1278   __ call(t);
1279 
1280   // remove activation
1281   __ movptr(t, Address(rbp,
1282                        frame::interpreter_frame_sender_sp_offset *
1283                        wordSize)); // get sender sp
1284   __ leave();                                // remove frame anchor
1285   __ pop(rdi);                               // get return address
1286   __ mov(rsp, t);                            // set sp to sender sp
1287   __ jmp(rdi);
1288 
1289   if (inc_counter) {
1290     // Handle overflow of counter and compile method
1291     __ bind(invocation_counter_overflow);
1292     generate_counter_overflow(continue_after_compile);
1293   }
1294 
1295   return entry_point;
1296 }
1297 
1298 // Abstract method entry
1299 // Attempt to execute abstract method. Throw exception
1300 address TemplateInterpreterGenerator::generate_abstract_entry(void) {
1301 
1302   address entry_point = __ pc();
1303 
1304   // abstract method entry
1305 
1306   //  pop return address, reset last_sp to NULL
1307   __ empty_expression_stack();
1308   __ restore_bcp();      // rsi must be correct for exception handler   (was destroyed)
1309   __ restore_locals();   // make sure locals pointer is correct as well (was destroyed)
1310 
1311   // throw exception
1312   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodErrorWithMethod), rbx);
1313   // the call_VM checks for exception, so we should never return here.
1314   __ should_not_reach_here();
1315 
1316   return entry_point;
1317 }
1318 
1319 //
1320 // Generic interpreted method entry to (asm) interpreter
1321 //
1322 address TemplateInterpreterGenerator::generate_normal_entry(bool synchronized) {
1323   // determine code generation flags
1324   bool inc_counter  = UseCompiler || CountCompiledCalls;
1325 
1326   // ebx: Method*
1327   // rbcp: sender sp (set in InterpreterMacroAssembler::prepare_to_jump_from_interpreted / generate_call_stub)
1328   address entry_point = __ pc();
1329 
1330   const Address constMethod(rbx, Method::const_offset());
1331   const Address access_flags(rbx, Method::access_flags_offset());
1332   const Address size_of_parameters(rdx,
1333                                    ConstMethod::size_of_parameters_offset());
1334   const Address size_of_locals(rdx, ConstMethod::size_of_locals_offset());
1335 
1336 
1337   // get parameter size (always needed)
1338   __ movptr(rdx, constMethod);
1339   __ load_unsigned_short(rcx, size_of_parameters);
1340 
1341   // rbx: Method*
1342   // rcx: size of parameters
1343   // rbcp: sender_sp (could differ from sp+wordSize if we were called via c2i )
1344 
1345   __ load_unsigned_short(rdx, size_of_locals); // get size of locals in words
1346   __ subl(rdx, rcx); // rdx = no. of additional locals
1347 
1348   // YYY
1349 //   __ incrementl(rdx);
1350 //   __ andl(rdx, -2);
1351 
1352   // see if we've got enough room on the stack for locals plus overhead.
1353   generate_stack_overflow_check();
1354 
1355   // get return address
1356   __ pop(rax);
1357 
1358   // compute beginning of parameters
1359   __ lea(rlocals, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize));
1360 
1361   // rdx - # of additional locals
1362   // allocate space for locals
1363   // explicitly initialize locals
1364   {
1365     Label exit, loop;
1366     __ testl(rdx, rdx);
1367     __ jcc(Assembler::lessEqual, exit); // do nothing if rdx <= 0
1368     __ bind(loop);
1369     __ push(NULL_WORD); // initialize local variables
1370     __ decrementl(rdx); // until everything initialized
1371     __ jcc(Assembler::greater, loop);
1372     __ bind(exit);
1373   }
1374 
1375   // initialize fixed part of activation frame
1376   generate_fixed_frame(false);
1377 
1378   // make sure method is not native & not abstract
1379 #ifdef ASSERT
1380   __ movl(rax, access_flags);
1381   {
1382     Label L;
1383     __ testl(rax, JVM_ACC_NATIVE);
1384     __ jcc(Assembler::zero, L);
1385     __ stop("tried to execute native method as non-native");
1386     __ bind(L);
1387   }
1388   {
1389     Label L;
1390     __ testl(rax, JVM_ACC_ABSTRACT);
1391     __ jcc(Assembler::zero, L);
1392     __ stop("tried to execute abstract method in interpreter");
1393     __ bind(L);
1394   }
1395 #endif
1396 
1397   // Since at this point in the method invocation the exception
1398   // handler would try to exit the monitor of synchronized methods
1399   // which hasn't been entered yet, we set the thread local variable
1400   // _do_not_unlock_if_synchronized to true. The remove_activation
1401   // will check this flag.
1402 
1403   const Register thread = NOT_LP64(rax) LP64_ONLY(r15_thread);
1404   NOT_LP64(__ get_thread(thread));
1405   const Address do_not_unlock_if_synchronized(thread,
1406         in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1407   __ movbool(do_not_unlock_if_synchronized, true);
1408 
1409   __ profile_parameters_type(rax, rcx, rdx);
1410   // increment invocation count & check for overflow
1411   Label invocation_counter_overflow;
1412   if (inc_counter) {
1413     generate_counter_incr(&invocation_counter_overflow);
1414   }
1415 
1416   Label continue_after_compile;
1417   __ bind(continue_after_compile);
1418 
1419   // check for synchronized interpreted methods
1420   bang_stack_shadow_pages(false);
1421 
1422   // reset the _do_not_unlock_if_synchronized flag
1423   NOT_LP64(__ get_thread(thread));
1424   __ movbool(do_not_unlock_if_synchronized, false);
1425 
1426   // check for synchronized methods
1427   // Must happen AFTER invocation_counter check and stack overflow check,
1428   // so method is not locked if overflows.
1429   if (synchronized) {
1430     // Allocate monitor and lock method
1431     lock_method();
1432   } else {
1433     // no synchronization necessary
1434 #ifdef ASSERT
1435     {
1436       Label L;
1437       __ movl(rax, access_flags);
1438       __ testl(rax, JVM_ACC_SYNCHRONIZED);
1439       __ jcc(Assembler::zero, L);
1440       __ stop("method needs synchronization");
1441       __ bind(L);
1442     }
1443 #endif
1444   }
1445 
1446   // start execution
1447 #ifdef ASSERT
1448   {
1449     Label L;
1450      const Address monitor_block_top (rbp,
1451                  frame::interpreter_frame_monitor_block_top_offset * wordSize);
1452     __ movptr(rax, monitor_block_top);
1453     __ cmpptr(rax, rsp);
1454     __ jcc(Assembler::equal, L);
1455     __ stop("broken stack frame setup in interpreter 6");
1456     __ bind(L);
1457   }
1458 #endif
1459 
1460   // jvmti support
1461   __ notify_method_entry();
1462 
1463   __ dispatch_next(vtos);
1464 
1465   // invocation counter overflow
1466   if (inc_counter) {
1467     // Handle overflow of counter and compile method
1468     __ bind(invocation_counter_overflow);
1469     generate_counter_overflow(continue_after_compile);
1470   }
1471 
1472   return entry_point;
1473 }
1474 
1475 //-----------------------------------------------------------------------------
1476 // Exceptions
1477 
1478 void TemplateInterpreterGenerator::generate_throw_exception() {
1479   // Entry point in previous activation (i.e., if the caller was
1480   // interpreted)
1481   Interpreter::_rethrow_exception_entry = __ pc();
1482   // Restore sp to interpreter_frame_last_sp even though we are going
1483   // to empty the expression stack for the exception processing.
1484   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
1485   // rax: exception
1486   // rdx: return address/pc that threw exception
1487   __ restore_bcp();    // r13/rsi points to call/send
1488   __ restore_locals();
1489   LP64_ONLY(__ reinit_heapbase());  // restore r12 as heapbase.
1490   // Entry point for exceptions thrown within interpreter code
1491   Interpreter::_throw_exception_entry = __ pc();
1492   // expression stack is undefined here
1493   // rax: exception
1494   // r13/rsi: exception bcp
1495   __ verify_oop(rax);
1496   Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
1497   LP64_ONLY(__ mov(c_rarg1, rax));
1498 
1499   // expression stack must be empty before entering the VM in case of
1500   // an exception
1501   __ empty_expression_stack();
1502   // find exception handler address and preserve exception oop
1503   __ call_VM(rdx,
1504              CAST_FROM_FN_PTR(address,
1505                           InterpreterRuntime::exception_handler_for_exception),
1506              rarg);
1507   // rax: exception handler entry point
1508   // rdx: preserved exception oop
1509   // r13/rsi: bcp for exception handler
1510   __ push_ptr(rdx); // push exception which is now the only value on the stack
1511   __ jmp(rax); // jump to exception handler (may be _remove_activation_entry!)
1512 
1513   // If the exception is not handled in the current frame the frame is
1514   // removed and the exception is rethrown (i.e. exception
1515   // continuation is _rethrow_exception).
1516   //
1517   // Note: At this point the bci is still the bxi for the instruction
1518   // which caused the exception and the expression stack is
1519   // empty. Thus, for any VM calls at this point, GC will find a legal
1520   // oop map (with empty expression stack).
1521 
1522   // In current activation
1523   // tos: exception
1524   // esi: exception bcp
1525 
1526   //
1527   // JVMTI PopFrame support
1528   //
1529 
1530   Interpreter::_remove_activation_preserving_args_entry = __ pc();
1531   __ empty_expression_stack();
1532   // Set the popframe_processing bit in pending_popframe_condition
1533   // indicating that we are currently handling popframe, so that
1534   // call_VMs that may happen later do not trigger new popframe
1535   // handling cycles.
1536   const Register thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
1537   NOT_LP64(__ get_thread(thread));
1538   __ movl(rdx, Address(thread, JavaThread::popframe_condition_offset()));
1539   __ orl(rdx, JavaThread::popframe_processing_bit);
1540   __ movl(Address(thread, JavaThread::popframe_condition_offset()), rdx);
1541 
1542   {
1543     // Check to see whether we are returning to a deoptimized frame.
1544     // (The PopFrame call ensures that the caller of the popped frame is
1545     // either interpreted or compiled and deoptimizes it if compiled.)
1546     // In this case, we can't call dispatch_next() after the frame is
1547     // popped, but instead must save the incoming arguments and restore
1548     // them after deoptimization has occurred.
1549     //
1550     // Note that we don't compare the return PC against the
1551     // deoptimization blob's unpack entry because of the presence of
1552     // adapter frames in C2.
1553     Label caller_not_deoptimized;
1554     Register rarg = NOT_LP64(rdx) LP64_ONLY(c_rarg1);
1555     __ movptr(rarg, Address(rbp, frame::return_addr_offset * wordSize));
1556     __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1557                                InterpreterRuntime::interpreter_contains), rarg);
1558     __ testl(rax, rax);
1559     __ jcc(Assembler::notZero, caller_not_deoptimized);
1560 
1561     // Compute size of arguments for saving when returning to
1562     // deoptimized caller
1563     __ get_method(rax);
1564     __ movptr(rax, Address(rax, Method::const_offset()));
1565     __ load_unsigned_short(rax, Address(rax, in_bytes(ConstMethod::
1566                                                 size_of_parameters_offset())));
1567     __ shll(rax, Interpreter::logStackElementSize);
1568     __ restore_locals();
1569     __ subptr(rlocals, rax);
1570     __ addptr(rlocals, wordSize);
1571     // Save these arguments
1572     NOT_LP64(__ get_thread(thread));
1573     __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1574                                            Deoptimization::
1575                                            popframe_preserve_args),
1576                           thread, rax, rlocals);
1577 
1578     __ remove_activation(vtos, rdx,
1579                          /* throw_monitor_exception */ false,
1580                          /* install_monitor_exception */ false,
1581                          /* notify_jvmdi */ false);
1582 
1583     // Inform deoptimization that it is responsible for restoring
1584     // these arguments
1585     NOT_LP64(__ get_thread(thread));
1586     __ movl(Address(thread, JavaThread::popframe_condition_offset()),
1587             JavaThread::popframe_force_deopt_reexecution_bit);
1588 
1589     // Continue in deoptimization handler
1590     __ jmp(rdx);
1591 
1592     __ bind(caller_not_deoptimized);
1593   }
1594 
1595   __ remove_activation(vtos, rdx, /* rdx result (retaddr) is not used */
1596                        /* throw_monitor_exception */ false,
1597                        /* install_monitor_exception */ false,
1598                        /* notify_jvmdi */ false);
1599 
1600   // Finish with popframe handling
1601   // A previous I2C followed by a deoptimization might have moved the
1602   // outgoing arguments further up the stack. PopFrame expects the
1603   // mutations to those outgoing arguments to be preserved and other
1604   // constraints basically require this frame to look exactly as
1605   // though it had previously invoked an interpreted activation with
1606   // no space between the top of the expression stack (current
1607   // last_sp) and the top of stack. Rather than force deopt to
1608   // maintain this kind of invariant all the time we call a small
1609   // fixup routine to move the mutated arguments onto the top of our
1610   // expression stack if necessary.
1611 #ifndef _LP64
1612   __ mov(rax, rsp);
1613   __ movptr(rbx, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1614   __ get_thread(thread);
1615   // PC must point into interpreter here
1616   __ set_last_Java_frame(thread, noreg, rbp, __ pc(), noreg);
1617   __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), thread, rax, rbx);
1618   __ get_thread(thread);
1619 #else
1620   __ mov(c_rarg1, rsp);
1621   __ movptr(c_rarg2, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1622   // PC must point into interpreter here
1623   __ set_last_Java_frame(noreg, rbp, __ pc(), rscratch1);
1624   __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), r15_thread, c_rarg1, c_rarg2);
1625 #endif
1626   __ reset_last_Java_frame(thread, true);
1627 
1628   // Restore the last_sp and null it out
1629   __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1630   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
1631 
1632   __ restore_bcp();
1633   __ restore_locals();
1634   // The method data pointer was incremented already during
1635   // call profiling. We have to restore the mdp for the current bcp.
1636   if (ProfileInterpreter) {
1637     __ set_method_data_pointer_for_bcp();
1638   }
1639 
1640   // Clear the popframe condition flag
1641   NOT_LP64(__ get_thread(thread));
1642   __ movl(Address(thread, JavaThread::popframe_condition_offset()),
1643           JavaThread::popframe_inactive);
1644 
1645 #if INCLUDE_JVMTI
1646   {
1647     Label L_done;
1648     const Register local0 = rlocals;
1649 
1650     __ cmpb(Address(rbcp, 0), Bytecodes::_invokestatic);
1651     __ jcc(Assembler::notEqual, L_done);
1652 
1653     // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call.
1654     // Detect such a case in the InterpreterRuntime function and return the member name argument, or NULL.
1655 
1656     __ get_method(rdx);
1657     __ movptr(rax, Address(local0, 0));
1658     __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), rax, rdx, rbcp);
1659 
1660     __ testptr(rax, rax);
1661     __ jcc(Assembler::zero, L_done);
1662 
1663     __ movptr(Address(rbx, 0), rax);
1664     __ bind(L_done);
1665   }
1666 #endif // INCLUDE_JVMTI
1667 
1668   __ dispatch_next(vtos);
1669   // end of PopFrame support
1670 
1671   Interpreter::_remove_activation_entry = __ pc();
1672 
1673   // preserve exception over this code sequence
1674   __ pop_ptr(rax);
1675   NOT_LP64(__ get_thread(thread));
1676   __ movptr(Address(thread, JavaThread::vm_result_offset()), rax);
1677   // remove the activation (without doing throws on illegalMonitorExceptions)
1678   __ remove_activation(vtos, rdx, false, true, false);
1679   // restore exception
1680   NOT_LP64(__ get_thread(thread));
1681   __ get_vm_result(rax, thread);
1682 
1683   // In between activations - previous activation type unknown yet
1684   // compute continuation point - the continuation point expects the
1685   // following registers set up:
1686   //
1687   // rax: exception
1688   // rdx: return address/pc that threw exception
1689   // rsp: expression stack of caller
1690   // rbp: ebp of caller
1691   __ push(rax);                                  // save exception
1692   __ push(rdx);                                  // save return address
1693   __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1694                           SharedRuntime::exception_handler_for_return_address),
1695                         thread, rdx);
1696   __ mov(rbx, rax);                              // save exception handler
1697   __ pop(rdx);                                   // restore return address
1698   __ pop(rax);                                   // restore exception
1699   // Note that an "issuing PC" is actually the next PC after the call
1700   __ jmp(rbx);                                   // jump to exception
1701                                                  // handler of caller
1702 }
1703 
1704 
1705 //
1706 // JVMTI ForceEarlyReturn support
1707 //
1708 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
1709   address entry = __ pc();
1710 
1711   __ restore_bcp();
1712   __ restore_locals();
1713   __ empty_expression_stack();
1714   __ load_earlyret_value(state);  // 32 bits returns value in rdx, so don't reuse
1715 
1716   const Register thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
1717   NOT_LP64(__ get_thread(thread));
1718   __ movptr(rcx, Address(thread, JavaThread::jvmti_thread_state_offset()));
1719   Address cond_addr(rcx, JvmtiThreadState::earlyret_state_offset());
1720 
1721   // Clear the earlyret state
1722   __ movl(cond_addr, JvmtiThreadState::earlyret_inactive);
1723 
1724   __ remove_activation(state, rsi,
1725                        false, /* throw_monitor_exception */
1726                        false, /* install_monitor_exception */
1727                        true); /* notify_jvmdi */
1728   __ jmp(rsi);
1729 
1730   return entry;
1731 } // end of ForceEarlyReturn support
1732 
1733 
1734 //-----------------------------------------------------------------------------
1735 // Helper for vtos entry point generation
1736 
1737 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
1738                                                          address& bep,
1739                                                          address& cep,
1740                                                          address& sep,
1741                                                          address& aep,
1742                                                          address& iep,
1743                                                          address& lep,
1744                                                          address& fep,
1745                                                          address& dep,
1746                                                          address& vep) {
1747   assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
1748   Label L;
1749 #ifndef _LP64
1750   fep = __ pc();     // ftos entry point
1751       __ push(ftos);
1752       __ jmpb(L);
1753   dep = __ pc();     // dtos entry point
1754       __ push(dtos);
1755       __ jmpb(L);
1756 #else
1757   fep = __ pc();     // ftos entry point
1758       __ push_f(xmm0);
1759       __ jmpb(L);
1760   dep = __ pc();     // dtos entry point
1761       __ push_d(xmm0);
1762       __ jmpb(L);
1763 #endif // _LP64
1764   lep = __ pc();     // ltos entry point
1765       __ push_l();
1766       __ jmpb(L);
1767   aep = bep = cep = sep = iep = __ pc();      // [abcsi]tos entry point
1768       __ push_i_or_ptr();
1769   vep = __ pc();    // vtos entry point
1770   __ bind(L);
1771   generate_and_dispatch(t);
1772 }
1773 
1774 //-----------------------------------------------------------------------------
1775 
1776 // Non-product code
1777 #ifndef PRODUCT
1778 
1779 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
1780   address entry = __ pc();
1781 
1782 #ifndef _LP64
1783   // prepare expression stack
1784   __ pop(rcx);          // pop return address so expression stack is 'pure'
1785   __ push(state);       // save tosca
1786 
1787   // pass tosca registers as arguments & call tracer
1788   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::trace_bytecode), rcx, rax, rdx);
1789   __ mov(rcx, rax);     // make sure return address is not destroyed by pop(state)
1790   __ pop(state);        // restore tosca
1791 
1792   // return
1793   __ jmp(rcx);
1794 #else
1795   __ push(state);
1796   __ push(c_rarg0);
1797   __ push(c_rarg1);
1798   __ push(c_rarg2);
1799   __ push(c_rarg3);
1800   __ mov(c_rarg2, rax);  // Pass itos
1801 #ifdef _WIN64
1802   __ movflt(xmm3, xmm0); // Pass ftos
1803 #endif
1804   __ call_VM(noreg,
1805              CAST_FROM_FN_PTR(address, InterpreterRuntime::trace_bytecode),
1806              c_rarg1, c_rarg2, c_rarg3);
1807   __ pop(c_rarg3);
1808   __ pop(c_rarg2);
1809   __ pop(c_rarg1);
1810   __ pop(c_rarg0);
1811   __ pop(state);
1812   __ ret(0);                                   // return from result handler
1813 #endif // _LP64
1814 
1815   return entry;
1816 }
1817 
1818 void TemplateInterpreterGenerator::count_bytecode() {
1819   __ incrementl(ExternalAddress((address) &BytecodeCounter::_counter_value), rscratch1);
1820 }
1821 
1822 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
1823   __ incrementl(ExternalAddress((address) &BytecodeHistogram::_counters[t->bytecode()]), rscratch1);
1824 }
1825 
1826 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
1827   __ mov32(rbx, ExternalAddress((address) &BytecodePairHistogram::_index));
1828   __ shrl(rbx, BytecodePairHistogram::log2_number_of_codes);
1829   __ orl(rbx,
1830          ((int) t->bytecode()) <<
1831          BytecodePairHistogram::log2_number_of_codes);
1832   __ mov32(ExternalAddress((address) &BytecodePairHistogram::_index), rbx, rscratch1);
1833   __ lea(rscratch1, ExternalAddress((address) BytecodePairHistogram::_counters));
1834   __ incrementl(Address(rscratch1, rbx, Address::times_4));
1835 }
1836 
1837 
1838 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
1839   // Call a little run-time stub to avoid blow-up for each bytecode.
1840   // The run-time runtime saves the right registers, depending on
1841   // the tosca in-state for the given template.
1842 
1843   assert(Interpreter::trace_code(t->tos_in()) != NULL,
1844          "entry must have been generated");
1845 #ifndef _LP64
1846   __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in())));
1847 #else
1848   __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1849   __ andptr(rsp, -16); // align stack as required by ABI
1850   __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in())));
1851   __ mov(rsp, r12); // restore sp
1852   __ reinit_heapbase();
1853 #endif // _LP64
1854 }
1855 
1856 
1857 void TemplateInterpreterGenerator::stop_interpreter_at() {
1858   Label L;
1859   __ cmp32(ExternalAddress((address) &BytecodeCounter::_counter_value),
1860            StopInterpreterAt,
1861            rscratch1);
1862   __ jcc(Assembler::notEqual, L);
1863   __ int3();
1864   __ bind(L);
1865 }
1866 #endif // !PRODUCT