1 /*
   2  * Copyright (c) 2003, 2023, 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/methodCounters.hpp"
  39 #include "oops/methodData.hpp"
  40 #include "oops/method.hpp"
  41 #include "oops/oop.inline.hpp"
  42 #include "oops/inlineKlass.hpp"
  43 #include "oops/resolvedIndyEntry.hpp"
  44 #include "oops/resolvedMethodEntry.hpp"
  45 #include "prims/jvmtiExport.hpp"
  46 #include "prims/jvmtiThreadState.hpp"
  47 #include "runtime/continuation.hpp"
  48 #include "runtime/deoptimization.hpp"
  49 #include "runtime/frame.inline.hpp"
  50 #include "runtime/globals.hpp"
  51 #include "runtime/jniHandles.hpp"
  52 #include "runtime/sharedRuntime.hpp"
  53 #include "runtime/stubRoutines.hpp"
  54 #include "runtime/synchronizer.hpp"
  55 #include "runtime/timer.hpp"
  56 #include "runtime/vframeArray.hpp"
  57 #include "utilities/checkedCast.hpp"
  58 #include "utilities/debug.hpp"
  59 #include "utilities/macros.hpp"
  60 
  61 #define __ Disassembler::hook<InterpreterMacroAssembler>(__FILE__, __LINE__, _masm)->
  62 
  63 // Size of interpreter code.  Increase if too small.  Interpreter will
  64 // fail with a guarantee ("not enough space for interpreter generation");
  65 // if too small.
  66 // Run with +PrintInterpreter to get the VM to print out the size.
  67 // Max size with JVMTI
  68 #ifdef AMD64
  69 int TemplateInterpreter::InterpreterCodeSize = JVMCI_ONLY(280) NOT_JVMCI(268) * 1024;
  70 #else
  71 int TemplateInterpreter::InterpreterCodeSize = 224 * 1024;
  72 #endif // AMD64
  73 
  74 // Global Register Names
  75 static const Register rbcp     = LP64_ONLY(r13) NOT_LP64(rsi);
  76 static const Register rlocals  = LP64_ONLY(r14) NOT_LP64(rdi);
  77 
  78 const int method_offset = frame::interpreter_frame_method_offset * wordSize;
  79 const int bcp_offset    = frame::interpreter_frame_bcp_offset    * wordSize;
  80 const int locals_offset = frame::interpreter_frame_locals_offset * wordSize;
  81 
  82 
  83 //-----------------------------------------------------------------------------
  84 
  85 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() {
  86   address entry = __ pc();
  87 
  88 #ifdef ASSERT
  89   {
  90     Label L;
  91     __ movptr(rax, Address(rbp,
  92                            frame::interpreter_frame_monitor_block_top_offset *
  93                            wordSize));
  94     __ lea(rax, Address(rbp, rax, Address::times_ptr));
  95     __ cmpptr(rax, rsp); // rax = maximal rsp for current rbp (stack
  96                          // grows negative)
  97     __ jcc(Assembler::aboveEqual, L); // check if frame is complete
  98     __ stop ("interpreter frame not set up");
  99     __ bind(L);
 100   }
 101 #endif // ASSERT
 102   // Restore bcp under the assumption that the current frame is still
 103   // interpreted
 104   __ restore_bcp();
 105 
 106   // expression stack must be empty before entering the VM if an
 107   // exception happened
 108   __ empty_expression_stack();
 109   // throw exception
 110   __ call_VM(noreg,
 111              CAST_FROM_FN_PTR(address,
 112                               InterpreterRuntime::throw_StackOverflowError));
 113   return entry;
 114 }
 115 
 116 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler() {
 117   address entry = __ pc();
 118   // The expression stack must be empty before entering the VM if an
 119   // exception happened.
 120   __ empty_expression_stack();
 121 
 122   // Setup parameters.
 123   // ??? convention: expect aberrant index in register ebx/rbx.
 124   // Pass array to create more detailed exceptions.
 125   Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
 126   __ call_VM(noreg,
 127              CAST_FROM_FN_PTR(address,
 128                               InterpreterRuntime::
 129                               throw_ArrayIndexOutOfBoundsException),
 130              rarg, rbx);
 131   return entry;
 132 }
 133 
 134 address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
 135   address entry = __ pc();
 136 
 137   // object is at TOS
 138   Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
 139   __ pop(rarg);
 140 
 141   // expression stack must be empty before entering the VM if an
 142   // exception happened
 143   __ empty_expression_stack();
 144 
 145   __ call_VM(noreg,
 146              CAST_FROM_FN_PTR(address,
 147                               InterpreterRuntime::
 148                               throw_ClassCastException),
 149              rarg);
 150   return entry;
 151 }
 152 
 153 address TemplateInterpreterGenerator::generate_exception_handler_common(
 154         const char* name, const char* message, bool pass_oop) {
 155   assert(!pass_oop || message == nullptr, "either oop or message but not both");
 156   address entry = __ pc();
 157 
 158   Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
 159   Register rarg2 = NOT_LP64(rbx) LP64_ONLY(c_rarg2);
 160 
 161   if (pass_oop) {
 162     // object is at TOS
 163     __ pop(rarg2);
 164   }
 165   // expression stack must be empty before entering the VM if an
 166   // exception happened
 167   __ empty_expression_stack();
 168   // setup parameters
 169   __ lea(rarg, ExternalAddress((address)name));
 170   if (pass_oop) {
 171     __ call_VM(rax, CAST_FROM_FN_PTR(address,
 172                                      InterpreterRuntime::
 173                                      create_klass_exception),
 174                rarg, rarg2);
 175   } else {
 176     __ lea(rarg2, ExternalAddress((address)message));
 177     __ call_VM(rax,
 178                CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception),
 179                rarg, rarg2);
 180   }
 181   // throw exception
 182   __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
 183   return entry;
 184 }
 185 
 186 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step, size_t index_size) {
 187   address entry = __ pc();
 188 
 189 #ifndef _LP64
 190 #ifdef COMPILER2
 191   // The FPU stack is clean if UseSSE >= 2 but must be cleaned in other cases
 192   if ((state == ftos && UseSSE < 1) || (state == dtos && UseSSE < 2)) {
 193     for (int i = 1; i < 8; i++) {
 194         __ ffree(i);
 195     }
 196   } else if (UseSSE < 2) {
 197     __ empty_FPU_stack();
 198   }
 199 #endif // COMPILER2
 200   if ((state == ftos && UseSSE < 1) || (state == dtos && UseSSE < 2)) {
 201     __ MacroAssembler::verify_FPU(1, "generate_return_entry_for compiled");
 202   } else {
 203     __ MacroAssembler::verify_FPU(0, "generate_return_entry_for compiled");
 204   }
 205 
 206   if (state == ftos) {
 207     __ MacroAssembler::verify_FPU(UseSSE >= 1 ? 0 : 1, "generate_return_entry_for in interpreter");
 208   } else if (state == dtos) {
 209     __ MacroAssembler::verify_FPU(UseSSE >= 2 ? 0 : 1, "generate_return_entry_for in interpreter");
 210   }
 211 #endif // _LP64
 212 
 213   // Restore stack bottom in case i2c adjusted stack
 214   __ movptr(rscratch1, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
 215   __ lea(rsp, Address(rbp, rscratch1, Address::times_ptr));
 216   // and null it as marker that esp is now tos until next java call
 217   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
 218 
 219   if (state == atos && InlineTypeReturnedAsFields) {
 220     __ store_inline_type_fields_to_buf(NULL);
 221   }
 222 
 223   __ restore_bcp();
 224   __ restore_locals();
 225 
 226   if (state == atos) {
 227     Register mdp = rbx;
 228     Register tmp = rcx;
 229     __ profile_return_type(mdp, rax, tmp);
 230   }
 231 
 232   const Register cache = rbx;
 233   const Register index = rcx;
 234   if (index_size == sizeof(u4)) {
 235     __ load_resolved_indy_entry(cache, index);
 236     __ load_unsigned_short(cache, Address(cache, in_bytes(ResolvedIndyEntry::num_parameters_offset())));
 237     __ lea(rsp, Address(rsp, cache, Interpreter::stackElementScale()));
 238   } else {
 239     assert(index_size == sizeof(u2), "Can only be u2");
 240     __ load_method_entry(cache, index);
 241     __ load_unsigned_short(cache, Address(cache, in_bytes(ResolvedMethodEntry::num_parameters_offset())));
 242     __ lea(rsp, Address(rsp, cache, Interpreter::stackElementScale()));
 243   }
 244 
 245    const Register java_thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
 246    if (JvmtiExport::can_pop_frame()) {
 247      NOT_LP64(__ get_thread(java_thread));
 248      __ check_and_handle_popframe(java_thread);
 249    }
 250    if (JvmtiExport::can_force_early_return()) {
 251      NOT_LP64(__ get_thread(java_thread));
 252      __ check_and_handle_earlyret(java_thread);
 253    }
 254 
 255   __ dispatch_next(state, step);
 256 
 257   return entry;
 258 }
 259 
 260 
 261 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state, int step, address continuation) {
 262   address entry = __ pc();
 263 
 264 #ifndef _LP64
 265   if (state == ftos) {
 266     __ MacroAssembler::verify_FPU(UseSSE >= 1 ? 0 : 1, "generate_deopt_entry_for in interpreter");
 267   } else if (state == dtos) {
 268     __ MacroAssembler::verify_FPU(UseSSE >= 2 ? 0 : 1, "generate_deopt_entry_for in interpreter");
 269   }
 270 #endif // _LP64
 271 
 272   // null last_sp until next java call
 273   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
 274   __ restore_bcp();
 275   __ restore_locals();
 276   const Register thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
 277   NOT_LP64(__ get_thread(thread));
 278 #if INCLUDE_JVMCI
 279   // Check if we need to take lock at entry of synchronized method.  This can
 280   // only occur on method entry so emit it only for vtos with step 0.
 281   if (EnableJVMCI && state == vtos && step == 0) {
 282     Label L;
 283     __ cmpb(Address(thread, JavaThread::pending_monitorenter_offset()), 0);
 284     __ jcc(Assembler::zero, L);
 285     // Clear flag.
 286     __ movb(Address(thread, JavaThread::pending_monitorenter_offset()), 0);
 287     // Satisfy calling convention for lock_method().
 288     __ get_method(rbx);
 289     // Take lock.
 290     lock_method();
 291     __ bind(L);
 292   } else {
 293 #ifdef ASSERT
 294     if (EnableJVMCI) {
 295       Label L;
 296       __ cmpb(Address(r15_thread, JavaThread::pending_monitorenter_offset()), 0);
 297       __ jcc(Assembler::zero, L);
 298       __ stop("unexpected pending monitor in deopt entry");
 299       __ bind(L);
 300     }
 301 #endif
 302   }
 303 #endif
 304   // handle exceptions
 305   {
 306     Label L;
 307     __ cmpptr(Address(thread, Thread::pending_exception_offset()), NULL_WORD);
 308     __ jcc(Assembler::zero, L);
 309     __ call_VM(noreg,
 310                CAST_FROM_FN_PTR(address,
 311                                 InterpreterRuntime::throw_pending_exception));
 312     __ should_not_reach_here();
 313     __ bind(L);
 314   }
 315   if (continuation == nullptr) {
 316     __ dispatch_next(state, step);
 317   } else {
 318     __ jump_to_entry(continuation);
 319   }
 320   return entry;
 321 }
 322 
 323 address TemplateInterpreterGenerator::generate_result_handler_for(
 324         BasicType type) {
 325   address entry = __ pc();
 326   switch (type) {
 327   case T_BOOLEAN: __ c2bool(rax);            break;
 328 #ifndef _LP64
 329   case T_CHAR   : __ andptr(rax, 0xFFFF);    break;
 330 #else
 331   case T_CHAR   : __ movzwl(rax, rax);       break;
 332 #endif // _LP64
 333   case T_BYTE   : __ sign_extend_byte(rax);  break;
 334   case T_SHORT  : __ sign_extend_short(rax); break;
 335   case T_INT    : /* nothing to do */        break;
 336   case T_LONG   : /* nothing to do */        break;
 337   case T_VOID   : /* nothing to do */        break;
 338 #ifndef _LP64
 339   case T_DOUBLE :
 340   case T_FLOAT  :
 341     { const Register t = InterpreterRuntime::SignatureHandlerGenerator::temp();
 342       __ pop(t);                            // remove return address first
 343       // Must return a result for interpreter or compiler. In SSE
 344       // mode, results are returned in xmm0 and the FPU stack must
 345       // be empty.
 346       if (type == T_FLOAT && UseSSE >= 1) {
 347         // Load ST0
 348         __ fld_d(Address(rsp, 0));
 349         // Store as float and empty fpu stack
 350         __ fstp_s(Address(rsp, 0));
 351         // and reload
 352         __ movflt(xmm0, Address(rsp, 0));
 353       } else if (type == T_DOUBLE && UseSSE >= 2 ) {
 354         __ movdbl(xmm0, Address(rsp, 0));
 355       } else {
 356         // restore ST0
 357         __ fld_d(Address(rsp, 0));
 358       }
 359       // and pop the temp
 360       __ addptr(rsp, 2 * wordSize);
 361       __ push(t);                           // restore return address
 362     }
 363     break;
 364 #else
 365   case T_FLOAT  : /* nothing to do */        break;
 366   case T_DOUBLE : /* nothing to do */        break;
 367 #endif // _LP64
 368 
 369   case T_OBJECT :
 370     // retrieve result from frame
 371     __ movptr(rax, Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize));
 372     // and verify it
 373     __ verify_oop(rax);
 374     break;
 375   default       : ShouldNotReachHere();
 376   }
 377   __ ret(0);                                   // return from result handler
 378   return entry;
 379 }
 380 
 381 address TemplateInterpreterGenerator::generate_safept_entry_for(
 382         TosState state,
 383         address runtime_entry) {
 384   address entry = __ pc();
 385 
 386   __ push(state);
 387   __ push_cont_fastpath();
 388   __ call_VM(noreg, runtime_entry);
 389   __ pop_cont_fastpath();
 390 
 391   __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
 392   return entry;
 393 }
 394 
 395 
 396 
 397 // Helpers for commoning out cases in the various type of method entries.
 398 //
 399 
 400 
 401 // increment invocation count & check for overflow
 402 //
 403 // Note: checking for negative value instead of overflow
 404 //       so we have a 'sticky' overflow test
 405 //
 406 // rbx: method
 407 // rcx: invocation counter
 408 //
 409 void TemplateInterpreterGenerator::generate_counter_incr(Label* overflow) {
 410   Label done;
 411   // Note: In tiered we increment either counters in Method* or in MDO depending if we're profiling or not.
 412   Label no_mdo;
 413   if (ProfileInterpreter) {
 414     // Are we profiling?
 415     __ movptr(rax, Address(rbx, Method::method_data_offset()));
 416     __ testptr(rax, rax);
 417     __ jccb(Assembler::zero, no_mdo);
 418     // Increment counter in the MDO
 419     const Address mdo_invocation_counter(rax, in_bytes(MethodData::invocation_counter_offset()) +
 420         in_bytes(InvocationCounter::counter_offset()));
 421     const Address mask(rax, in_bytes(MethodData::invoke_mask_offset()));
 422     __ increment_mask_and_jump(mdo_invocation_counter, mask, rcx, overflow);
 423     __ jmp(done);
 424   }
 425   __ bind(no_mdo);
 426   // Increment counter in MethodCounters
 427   const Address invocation_counter(rax,
 428       MethodCounters::invocation_counter_offset() +
 429       InvocationCounter::counter_offset());
 430   __ get_method_counters(rbx, rax, done);
 431   const Address mask(rax, in_bytes(MethodCounters::invoke_mask_offset()));
 432   __ increment_mask_and_jump(invocation_counter, mask, rcx, overflow);
 433   __ bind(done);
 434 }
 435 
 436 void TemplateInterpreterGenerator::generate_counter_overflow(Label& do_continue) {
 437 
 438   // Asm interpreter on entry
 439   // r14/rdi - locals
 440   // r13/rsi - bcp
 441   // rbx - method
 442   // rdx - cpool --- DOES NOT APPEAR TO BE TRUE
 443   // rbp - interpreter frame
 444 
 445   // On return (i.e. jump to entry_point) [ back to invocation of interpreter ]
 446   // Everything as it was on entry
 447   // rdx is not restored. Doesn't appear to really be set.
 448 
 449   // InterpreterRuntime::frequency_counter_overflow takes two
 450   // arguments, the first (thread) is passed by call_VM, the second
 451   // indicates if the counter overflow occurs at a backwards branch
 452   // (null bcp).  We pass zero for it.  The call returns the address
 453   // of the verified entry point for the method or null if the
 454   // compilation did not complete (either went background or bailed
 455   // out).
 456   Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
 457   __ movl(rarg, 0);
 458   __ call_VM(noreg,
 459              CAST_FROM_FN_PTR(address,
 460                               InterpreterRuntime::frequency_counter_overflow),
 461              rarg);
 462 
 463   __ movptr(rbx, Address(rbp, method_offset));   // restore Method*
 464   // Preserve invariant that r13/r14 contain bcp/locals of sender frame
 465   // and jump to the interpreted entry.
 466   __ jmp(do_continue, relocInfo::none);
 467 }
 468 
 469 // See if we've got enough room on the stack for locals plus overhead below
 470 // JavaThread::stack_overflow_limit(). If not, throw a StackOverflowError
 471 // without going through the signal handler, i.e., reserved and yellow zones
 472 // will not be made usable. The shadow zone must suffice to handle the
 473 // overflow.
 474 // The expression stack grows down incrementally, so the normal guard
 475 // page mechanism will work for that.
 476 //
 477 // NOTE: Since the additional locals are also always pushed (wasn't
 478 // obvious in generate_fixed_frame) so the guard should work for them
 479 // too.
 480 //
 481 // Args:
 482 //      rdx: number of additional locals this frame needs (what we must check)
 483 //      rbx: Method*
 484 //
 485 // Kills:
 486 //      rax
 487 void TemplateInterpreterGenerator::generate_stack_overflow_check(void) {
 488 
 489   // monitor entry size: see picture of stack in frame_x86.hpp
 490   const int entry_size = frame::interpreter_frame_monitor_size_in_bytes();
 491 
 492   // total overhead size: entry_size + (saved rbp through expr stack
 493   // bottom).  be sure to change this if you add/subtract anything
 494   // to/from the overhead area
 495   const int overhead_size =
 496     -(frame::interpreter_frame_initial_sp_offset * wordSize) + entry_size;
 497 
 498   const int page_size = (int)os::vm_page_size();
 499 
 500   Label after_frame_check;
 501 
 502   // see if the frame is greater than one page in size. If so,
 503   // then we need to verify there is enough stack space remaining
 504   // for the additional locals.
 505   __ cmpl(rdx, (page_size - overhead_size) / Interpreter::stackElementSize);
 506   __ jcc(Assembler::belowEqual, after_frame_check);
 507 
 508   // compute rsp as if this were going to be the last frame on
 509   // the stack before the red zone
 510 
 511   Label after_frame_check_pop;
 512   const Register thread = NOT_LP64(rsi) LP64_ONLY(r15_thread);
 513 #ifndef _LP64
 514   __ push(thread);
 515   __ get_thread(thread);
 516 #endif
 517 
 518   const Address stack_limit(thread, JavaThread::stack_overflow_limit_offset());
 519 
 520   // locals + overhead, in bytes
 521   __ mov(rax, rdx);
 522   __ shlptr(rax, Interpreter::logStackElementSize); // Convert parameter count to bytes.
 523   __ addptr(rax, overhead_size);
 524 
 525 #ifdef ASSERT
 526   Label limit_okay;
 527   // Verify that thread stack overflow limit is non-zero.
 528   __ cmpptr(stack_limit, NULL_WORD);
 529   __ jcc(Assembler::notEqual, limit_okay);
 530   __ stop("stack overflow limit is zero");
 531   __ bind(limit_okay);
 532 #endif
 533 
 534   // Add locals/frame size to stack limit.
 535   __ addptr(rax, stack_limit);
 536 
 537   // Check against the current stack bottom.
 538   __ cmpptr(rsp, rax);
 539 
 540   __ jcc(Assembler::above, after_frame_check_pop);
 541   NOT_LP64(__ pop(rsi));  // get saved bcp
 542 
 543   // Restore sender's sp as SP. This is necessary if the sender's
 544   // frame is an extended compiled frame (see gen_c2i_adapter())
 545   // and safer anyway in case of JSR292 adaptations.
 546 
 547   __ pop(rax); // return address must be moved if SP is changed
 548   __ mov(rsp, rbcp);
 549   __ push(rax);
 550 
 551   // Note: the restored frame is not necessarily interpreted.
 552   // Use the shared runtime version of the StackOverflowError.
 553   assert(StubRoutines::throw_StackOverflowError_entry() != nullptr, "stub not yet generated");
 554   __ jump(ExternalAddress(StubRoutines::throw_StackOverflowError_entry()));
 555   // all done with frame size check
 556   __ bind(after_frame_check_pop);
 557   NOT_LP64(__ pop(rsi));
 558 
 559   // all done with frame size check
 560   __ bind(after_frame_check);
 561 }
 562 
 563 // Allocate monitor and lock method (asm interpreter)
 564 //
 565 // Args:
 566 //      rbx: Method*
 567 //      r14/rdi: locals
 568 //
 569 // Kills:
 570 //      rax
 571 //      c_rarg0, c_rarg1, c_rarg2, c_rarg3, ...(param regs)
 572 //      rscratch1, rscratch2 (scratch regs)
 573 void TemplateInterpreterGenerator::lock_method() {
 574   // synchronize method
 575   const Address access_flags(rbx, Method::access_flags_offset());
 576   const Address monitor_block_top(
 577         rbp,
 578         frame::interpreter_frame_monitor_block_top_offset * wordSize);
 579   const int entry_size = frame::interpreter_frame_monitor_size_in_bytes();
 580 
 581 #ifdef ASSERT
 582   {
 583     Label L;
 584     __ movl(rax, access_flags);
 585     __ testl(rax, JVM_ACC_SYNCHRONIZED);
 586     __ jcc(Assembler::notZero, L);
 587     __ stop("method doesn't need synchronization");
 588     __ bind(L);
 589   }
 590 #endif // ASSERT
 591 
 592   // get synchronization object
 593   {
 594     Label done;
 595     __ movl(rax, access_flags);
 596     __ testl(rax, JVM_ACC_STATIC);
 597     // get receiver (assume this is frequent case)
 598     __ movptr(rax, Address(rlocals, Interpreter::local_offset_in_bytes(0)));
 599     __ jcc(Assembler::zero, done);
 600     __ load_mirror(rax, rbx, rscratch2);
 601 
 602 #ifdef ASSERT
 603     {
 604       Label L;
 605       __ testptr(rax, rax);
 606       __ jcc(Assembler::notZero, L);
 607       __ stop("synchronization object is null");
 608       __ bind(L);
 609     }
 610 #endif // ASSERT
 611 
 612     __ bind(done);
 613   }
 614 
 615   // add space for monitor & lock
 616   __ subptr(rsp, entry_size); // add space for a monitor entry
 617   __ subptr(monitor_block_top, entry_size / wordSize); // set new monitor block top
 618   // store object
 619   __ movptr(Address(rsp, BasicObjectLock::obj_offset()), rax);
 620   const Register lockreg = NOT_LP64(rdx) LP64_ONLY(c_rarg1);
 621   __ movptr(lockreg, rsp); // object address
 622   __ lock_object(lockreg);
 623 }
 624 
 625 // Generate a fixed interpreter frame. This is identical setup for
 626 // interpreted methods and for native methods hence the shared code.
 627 //
 628 // Args:
 629 //      rax: return address
 630 //      rbx: Method*
 631 //      r14/rdi: pointer to locals
 632 //      r13/rsi: sender sp
 633 //      rdx: cp cache
 634 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
 635   // initialize fixed part of activation frame
 636   __ push(rax);        // save return address
 637   __ enter();          // save old & set new rbp
 638   __ push(rbcp);        // set sender sp
 639   __ push(NULL_WORD); // leave last_sp as null
 640   __ movptr(rbcp, Address(rbx, Method::const_offset()));      // get ConstMethod*
 641   __ lea(rbcp, Address(rbcp, ConstMethod::codes_offset())); // get codebase
 642   __ push(rbx);        // save Method*
 643   // Get mirror and store it in the frame as GC root for this Method*
 644   __ load_mirror(rdx, rbx, rscratch2);
 645   __ push(rdx);
 646   if (ProfileInterpreter) {
 647     Label method_data_continue;
 648     __ movptr(rdx, Address(rbx, in_bytes(Method::method_data_offset())));
 649     __ testptr(rdx, rdx);
 650     __ jcc(Assembler::zero, method_data_continue);
 651     __ addptr(rdx, in_bytes(MethodData::data_offset()));
 652     __ bind(method_data_continue);
 653     __ push(rdx);      // set the mdp (method data pointer)
 654   } else {
 655     __ push(0);
 656   }
 657 
 658   __ movptr(rdx, Address(rbx, Method::const_offset()));
 659   __ movptr(rdx, Address(rdx, ConstMethod::constants_offset()));
 660   __ movptr(rdx, Address(rdx, ConstantPool::cache_offset()));
 661   __ push(rdx); // set constant pool cache
 662 
 663   __ movptr(rax, rlocals);
 664   __ subptr(rax, rbp);
 665   __ shrptr(rax, Interpreter::logStackElementSize);  // rax = rlocals - fp();
 666   __ push(rax); // set relativized rlocals, see frame::interpreter_frame_locals()
 667 
 668   if (native_call) {
 669     __ push(0); // no bcp
 670   } else {
 671     __ push(rbcp); // set bcp
 672   }
 673   // initialize relativized pointer to expression stack bottom
 674   __ push(frame::interpreter_frame_initial_sp_offset);
 675 }
 676 
 677 // End of helpers
 678 
 679 // Method entry for java.lang.ref.Reference.get.
 680 address TemplateInterpreterGenerator::generate_Reference_get_entry(void) {
 681   // Code: _aload_0, _getfield, _areturn
 682   // parameter size = 1
 683   //
 684   // The code that gets generated by this routine is split into 2 parts:
 685   //    1. The "intrinsified" code performing an ON_WEAK_OOP_REF load,
 686   //    2. The slow path - which is an expansion of the regular method entry.
 687   //
 688   // Notes:-
 689   // * An intrinsic is always executed, where an ON_WEAK_OOP_REF load is performed.
 690   // * We may jump to the slow path iff the receiver is null. If the
 691   //   Reference object is null then we no longer perform an ON_WEAK_OOP_REF load
 692   //   Thus we can use the regular method entry code to generate the NPE.
 693   //
 694   // rbx: Method*
 695 
 696   // r13: senderSP must preserve for slow path, set SP to it on fast path
 697 
 698   address entry = __ pc();
 699 
 700   const int referent_offset = java_lang_ref_Reference::referent_offset();
 701 
 702   Label slow_path;
 703   // rbx: method
 704 
 705   // Check if local 0 != null
 706   // If the receiver is null then it is OK to jump to the slow path.
 707   __ movptr(rax, Address(rsp, wordSize));
 708 
 709   __ testptr(rax, rax);
 710   __ jcc(Assembler::zero, slow_path);
 711 
 712   // rax: local 0
 713   // rbx: method (but can be used as scratch now)
 714   // rdx: scratch
 715   // rdi: scratch
 716 
 717   // Preserve the sender sp in case the load barrier
 718   // calls the runtime
 719   NOT_LP64(__ push(rsi));
 720 
 721   // Load the value of the referent field.
 722   const Address field_address(rax, referent_offset);
 723   __ load_heap_oop(rax, field_address, /*tmp1*/ rbx, /*tmp_thread*/ rdx, ON_WEAK_OOP_REF);
 724 
 725   // _areturn
 726   const Register sender_sp = NOT_LP64(rsi) LP64_ONLY(r13);
 727   NOT_LP64(__ pop(rsi));      // get sender sp
 728   __ pop(rdi);                // get return address
 729   __ mov(rsp, sender_sp);     // set sp to sender sp
 730   __ jmp(rdi);
 731   __ ret(0);
 732 
 733   // generate a vanilla interpreter entry as the slow path
 734   __ bind(slow_path);
 735   __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::zerolocals));
 736   return entry;
 737 }
 738 
 739 void TemplateInterpreterGenerator::bang_stack_shadow_pages(bool native_call) {
 740   // See more discussion in stackOverflow.hpp.
 741 
 742   // Note that we do the banging after the frame is setup, since the exception
 743   // handling code expects to find a valid interpreter frame on the stack.
 744   // Doing the banging earlier fails if the caller frame is not an interpreter
 745   // frame.
 746   // (Also, the exception throwing code expects to unlock any synchronized
 747   // method receiver, so do the banging after locking the receiver.)
 748 
 749   const int shadow_zone_size = checked_cast<int>(StackOverflow::stack_shadow_zone_size());
 750   const int page_size = (int)os::vm_page_size();
 751   const int n_shadow_pages = shadow_zone_size / page_size;
 752 
 753   const Register thread = NOT_LP64(rsi) LP64_ONLY(r15_thread);
 754 #ifndef _LP64
 755   __ push(thread);
 756   __ get_thread(thread);
 757 #endif
 758 
 759 #ifdef ASSERT
 760   Label L_good_limit;
 761   __ cmpptr(Address(thread, JavaThread::shadow_zone_safe_limit()), NULL_WORD);
 762   __ jcc(Assembler::notEqual, L_good_limit);
 763   __ stop("shadow zone safe limit is not initialized");
 764   __ bind(L_good_limit);
 765 
 766   Label L_good_watermark;
 767   __ cmpptr(Address(thread, JavaThread::shadow_zone_growth_watermark()), NULL_WORD);
 768   __ jcc(Assembler::notEqual, L_good_watermark);
 769   __ stop("shadow zone growth watermark is not initialized");
 770   __ bind(L_good_watermark);
 771 #endif
 772 
 773   Label L_done;
 774 
 775   __ cmpptr(rsp, Address(thread, JavaThread::shadow_zone_growth_watermark()));
 776   __ jcc(Assembler::above, L_done);
 777 
 778   for (int p = 1; p <= n_shadow_pages; p++) {
 779     __ bang_stack_with_offset(p*page_size);
 780   }
 781 
 782   // Record the new watermark, but only if update is above the safe limit.
 783   // Otherwise, the next time around the check above would pass the safe limit.
 784   __ cmpptr(rsp, Address(thread, JavaThread::shadow_zone_safe_limit()));
 785   __ jccb(Assembler::belowEqual, L_done);
 786   __ movptr(Address(thread, JavaThread::shadow_zone_growth_watermark()), rsp);
 787 
 788   __ bind(L_done);
 789 
 790 #ifndef _LP64
 791   __ pop(thread);
 792 #endif
 793 }
 794 
 795 // Interpreter stub for calling a native method. (asm interpreter)
 796 // This sets up a somewhat different looking stack for calling the
 797 // native method than the typical interpreter frame setup.
 798 address TemplateInterpreterGenerator::generate_native_entry(bool synchronized) {
 799   // determine code generation flags
 800   bool inc_counter  = UseCompiler || CountCompiledCalls;
 801 
 802   // rbx: Method*
 803   // rbcp: sender sp
 804 
 805   address entry_point = __ pc();
 806 
 807   const Address constMethod       (rbx, Method::const_offset());
 808   const Address access_flags      (rbx, Method::access_flags_offset());
 809   const Address size_of_parameters(rcx, ConstMethod::
 810                                         size_of_parameters_offset());
 811 
 812 
 813   // get parameter size (always needed)
 814   __ movptr(rcx, constMethod);
 815   __ load_unsigned_short(rcx, size_of_parameters);
 816 
 817   // native calls don't need the stack size check since they have no
 818   // expression stack and the arguments are already on the stack and
 819   // we only add a handful of words to the stack
 820 
 821   // rbx: Method*
 822   // rcx: size of parameters
 823   // rbcp: sender sp
 824   __ pop(rax);                                       // get return address
 825 
 826   // for natives the size of locals is zero
 827 
 828   // compute beginning of parameters
 829   __ lea(rlocals, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize));
 830 
 831   // add 2 zero-initialized slots for native calls
 832   // initialize result_handler slot
 833   __ push(NULL_WORD);
 834   // slot for oop temp
 835   // (static native method holder mirror/jni oop result)
 836   __ push(NULL_WORD);
 837 
 838   // initialize fixed part of activation frame
 839   generate_fixed_frame(true);
 840 
 841   // make sure method is native & not abstract
 842 #ifdef ASSERT
 843   __ movl(rax, access_flags);
 844   {
 845     Label L;
 846     __ testl(rax, JVM_ACC_NATIVE);
 847     __ jcc(Assembler::notZero, L);
 848     __ stop("tried to execute non-native method as native");
 849     __ bind(L);
 850   }
 851   {
 852     Label L;
 853     __ testl(rax, JVM_ACC_ABSTRACT);
 854     __ jcc(Assembler::zero, L);
 855     __ stop("tried to execute abstract method in interpreter");
 856     __ bind(L);
 857   }
 858 #endif
 859 
 860   // Since at this point in the method invocation the exception handler
 861   // would try to exit the monitor of synchronized methods which hasn't
 862   // been entered yet, we set the thread local variable
 863   // _do_not_unlock_if_synchronized to true. The remove_activation will
 864   // check this flag.
 865 
 866   const Register thread1 = NOT_LP64(rax) LP64_ONLY(r15_thread);
 867   NOT_LP64(__ get_thread(thread1));
 868   const Address do_not_unlock_if_synchronized(thread1,
 869         in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
 870   __ movbool(do_not_unlock_if_synchronized, true);
 871 
 872   // increment invocation count & check for overflow
 873   Label invocation_counter_overflow;
 874   if (inc_counter) {
 875     generate_counter_incr(&invocation_counter_overflow);
 876   }
 877 
 878   Label continue_after_compile;
 879   __ bind(continue_after_compile);
 880 
 881   bang_stack_shadow_pages(true);
 882 
 883   // reset the _do_not_unlock_if_synchronized flag
 884   NOT_LP64(__ get_thread(thread1));
 885   __ movbool(do_not_unlock_if_synchronized, false);
 886 
 887   // check for synchronized methods
 888   // Must happen AFTER invocation_counter check and stack overflow check,
 889   // so method is not locked if overflows.
 890   if (synchronized) {
 891     lock_method();
 892   } else {
 893     // no synchronization necessary
 894 #ifdef ASSERT
 895     {
 896       Label L;
 897       __ movl(rax, access_flags);
 898       __ testl(rax, JVM_ACC_SYNCHRONIZED);
 899       __ jcc(Assembler::zero, L);
 900       __ stop("method needs synchronization");
 901       __ bind(L);
 902     }
 903 #endif
 904   }
 905 
 906   // start execution
 907 #ifdef ASSERT
 908   {
 909     Label L;
 910     const Address monitor_block_top(rbp,
 911                  frame::interpreter_frame_monitor_block_top_offset * wordSize);
 912     __ movptr(rax, monitor_block_top);
 913     __ lea(rax, Address(rbp, rax, Address::times_ptr));
 914     __ cmpptr(rax, rsp);
 915     __ jcc(Assembler::equal, L);
 916     __ stop("broken stack frame setup in interpreter 5");
 917     __ bind(L);
 918   }
 919 #endif
 920 
 921   // jvmti support
 922   __ notify_method_entry();
 923 
 924   // work registers
 925   const Register method = rbx;
 926   const Register thread = NOT_LP64(rdi) LP64_ONLY(r15_thread);
 927   const Register t      = NOT_LP64(rcx) LP64_ONLY(r11);
 928 
 929   // allocate space for parameters
 930   __ get_method(method);
 931   __ movptr(t, Address(method, Method::const_offset()));
 932   __ load_unsigned_short(t, Address(t, ConstMethod::size_of_parameters_offset()));
 933 
 934 #ifndef _LP64
 935   __ shlptr(t, Interpreter::logStackElementSize); // Convert parameter count to bytes.
 936   __ addptr(t, 2*wordSize);     // allocate two more slots for JNIEnv and possible mirror
 937   __ subptr(rsp, t);
 938   __ andptr(rsp, -(StackAlignmentInBytes)); // gcc needs 16 byte aligned stacks to do XMM intrinsics
 939 #else
 940   __ shll(t, Interpreter::logStackElementSize);
 941 
 942   __ subptr(rsp, t);
 943   __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
 944   __ andptr(rsp, -16); // must be 16 byte boundary (see amd64 ABI)
 945 #endif // _LP64
 946 
 947   // get signature handler
 948   {
 949     Label L;
 950     __ movptr(t, Address(method, Method::signature_handler_offset()));
 951     __ testptr(t, t);
 952     __ jcc(Assembler::notZero, L);
 953     __ call_VM(noreg,
 954                CAST_FROM_FN_PTR(address,
 955                                 InterpreterRuntime::prepare_native_call),
 956                method);
 957     __ get_method(method);
 958     __ movptr(t, Address(method, Method::signature_handler_offset()));
 959     __ bind(L);
 960   }
 961 
 962   // call signature handler
 963   assert(InterpreterRuntime::SignatureHandlerGenerator::from() == rlocals,
 964          "adjust this code");
 965   assert(InterpreterRuntime::SignatureHandlerGenerator::to() == rsp,
 966          "adjust this code");
 967   assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == NOT_LP64(t) LP64_ONLY(rscratch1),
 968          "adjust this code");
 969 
 970   // The generated handlers do not touch RBX (the method).
 971   // However, large signatures cannot be cached and are generated
 972   // each time here.  The slow-path generator can do a GC on return,
 973   // so we must reload it after the call.
 974   __ call(t);
 975   __ get_method(method);        // slow path can do a GC, reload RBX
 976 
 977 
 978   // result handler is in rax
 979   // set result handler
 980   __ movptr(Address(rbp,
 981                     (frame::interpreter_frame_result_handler_offset) * wordSize),
 982             rax);
 983 
 984   // pass mirror handle if static call
 985   {
 986     Label L;
 987     __ movl(t, Address(method, Method::access_flags_offset()));
 988     __ testl(t, JVM_ACC_STATIC);
 989     __ jcc(Assembler::zero, L);
 990     // get mirror
 991     __ load_mirror(t, method, rax);
 992     // copy mirror into activation frame
 993     __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize),
 994             t);
 995     // pass handle to mirror
 996 #ifndef _LP64
 997     __ lea(t, Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize));
 998     __ movptr(Address(rsp, wordSize), t);
 999 #else
1000     __ lea(c_rarg1,
1001            Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize));
1002 #endif // _LP64
1003     __ bind(L);
1004   }
1005 
1006   // get native function entry point
1007   {
1008     Label L;
1009     __ movptr(rax, Address(method, Method::native_function_offset()));
1010     ExternalAddress unsatisfied(SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
1011     __ cmpptr(rax, unsatisfied.addr(), rscratch1);
1012     __ jcc(Assembler::notEqual, L);
1013     __ call_VM(noreg,
1014                CAST_FROM_FN_PTR(address,
1015                                 InterpreterRuntime::prepare_native_call),
1016                method);
1017     __ get_method(method);
1018     __ movptr(rax, Address(method, Method::native_function_offset()));
1019     __ bind(L);
1020   }
1021 
1022   // pass JNIEnv
1023 #ifndef _LP64
1024    __ get_thread(thread);
1025    __ lea(t, Address(thread, JavaThread::jni_environment_offset()));
1026    __ movptr(Address(rsp, 0), t);
1027 
1028    // set_last_Java_frame_before_call
1029    // It is enough that the pc()
1030    // points into the right code segment. It does not have to be the correct return pc.
1031    __ set_last_Java_frame(thread, noreg, rbp, __ pc(), noreg);
1032 #else
1033    __ lea(c_rarg0, Address(r15_thread, JavaThread::jni_environment_offset()));
1034 
1035    // It is enough that the pc() points into the right code
1036    // segment. It does not have to be the correct return pc.
1037    __ set_last_Java_frame(rsp, rbp, (address) __ pc(), rscratch1);
1038 #endif // _LP64
1039 
1040   // change thread state
1041 #ifdef ASSERT
1042   {
1043     Label L;
1044     __ movl(t, Address(thread, JavaThread::thread_state_offset()));
1045     __ cmpl(t, _thread_in_Java);
1046     __ jcc(Assembler::equal, L);
1047     __ stop("Wrong thread state in native stub");
1048     __ bind(L);
1049   }
1050 #endif
1051 
1052   // Change state to native
1053 
1054   __ movl(Address(thread, JavaThread::thread_state_offset()),
1055           _thread_in_native);
1056 
1057   // Call the native method.
1058   __ call(rax);
1059   // 32: result potentially in rdx:rax or ST0
1060   // 64: result potentially in rax or xmm0
1061 
1062   // Verify or restore cpu control state after JNI call
1063   __ restore_cpu_control_state_after_jni(rscratch1);
1064 
1065   // NOTE: The order of these pushes is known to frame::interpreter_frame_result
1066   // in order to extract the result of a method call. If the order of these
1067   // pushes change or anything else is added to the stack then the code in
1068   // interpreter_frame_result must also change.
1069 
1070 #ifndef _LP64
1071   // save potential result in ST(0) & rdx:rax
1072   // (if result handler is the T_FLOAT or T_DOUBLE handler, result must be in ST0 -
1073   // the check is necessary to avoid potential Intel FPU overflow problems by saving/restoring 'empty' FPU registers)
1074   // It is safe to do this push because state is _thread_in_native and return address will be found
1075   // via _last_native_pc and not via _last_jave_sp
1076 
1077   // NOTE: the order of these push(es) is known to frame::interpreter_frame_result.
1078   // If the order changes or anything else is added to the stack the code in
1079   // interpreter_frame_result will have to be changed.
1080 
1081   { Label L;
1082     Label push_double;
1083     ExternalAddress float_handler(AbstractInterpreter::result_handler(T_FLOAT));
1084     ExternalAddress double_handler(AbstractInterpreter::result_handler(T_DOUBLE));
1085     __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize),
1086               float_handler.addr(), noreg);
1087     __ jcc(Assembler::equal, push_double);
1088     __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize),
1089               double_handler.addr(), noreg);
1090     __ jcc(Assembler::notEqual, L);
1091     __ bind(push_double);
1092     __ push_d(); // FP values are returned using the FPU, so push FPU contents (even if UseSSE > 0).
1093     __ bind(L);
1094   }
1095 #else
1096   __ push(dtos);
1097 #endif // _LP64
1098 
1099   __ push(ltos);
1100 
1101   // change thread state
1102   NOT_LP64(__ get_thread(thread));
1103   __ movl(Address(thread, JavaThread::thread_state_offset()),
1104           _thread_in_native_trans);
1105 
1106   // Force this write out before the read below
1107   if (!UseSystemMemoryBarrier) {
1108     __ membar(Assembler::Membar_mask_bits(
1109                 Assembler::LoadLoad | Assembler::LoadStore |
1110                 Assembler::StoreLoad | Assembler::StoreStore));
1111   }
1112 #ifndef _LP64
1113   if (AlwaysRestoreFPU) {
1114     //  Make sure the control word is correct.
1115     __ fldcw(ExternalAddress(StubRoutines::x86::addr_fpu_cntrl_wrd_std()));
1116   }
1117 #endif // _LP64
1118 
1119   // check for safepoint operation in progress and/or pending suspend requests
1120   {
1121     Label Continue;
1122     Label slow_path;
1123 
1124     __ safepoint_poll(slow_path, thread, true /* at_return */, false /* in_nmethod */);
1125 
1126     __ cmpl(Address(thread, JavaThread::suspend_flags_offset()), 0);
1127     __ jcc(Assembler::equal, Continue);
1128     __ bind(slow_path);
1129 
1130     // Don't use call_VM as it will see a possible pending exception
1131     // and forward it and never return here preventing us from
1132     // clearing _last_native_pc down below.  Also can't use
1133     // call_VM_leaf either as it will check to see if r13 & r14 are
1134     // preserved and correspond to the bcp/locals pointers. So we do a
1135     // runtime call by hand.
1136     //
1137 #ifndef _LP64
1138     __ push(thread);
1139     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address,
1140                                             JavaThread::check_special_condition_for_native_trans)));
1141     __ increment(rsp, wordSize);
1142     __ get_thread(thread);
1143 #else
1144     __ mov(c_rarg0, r15_thread);
1145     __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1146     __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1147     __ andptr(rsp, -16); // align stack as required by ABI
1148     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans)));
1149     __ mov(rsp, r12); // restore sp
1150     __ reinit_heapbase();
1151 #endif // _LP64
1152     __ bind(Continue);
1153   }
1154 
1155   // change thread state
1156   __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_Java);
1157 
1158   // reset_last_Java_frame
1159   __ reset_last_Java_frame(thread, true);
1160 
1161   if (CheckJNICalls) {
1162     // clear_pending_jni_exception_check
1163     __ movptr(Address(thread, JavaThread::pending_jni_exception_check_fn_offset()), NULL_WORD);
1164   }
1165 
1166   // reset handle block
1167   __ movptr(t, Address(thread, JavaThread::active_handles_offset()));
1168   __ movl(Address(t, JNIHandleBlock::top_offset()), NULL_WORD);
1169 
1170   // If result is an oop unbox and store it in frame where gc will see it
1171   // and result handler will pick it up
1172 
1173   {
1174     Label no_oop;
1175     __ lea(t, ExternalAddress(AbstractInterpreter::result_handler(T_OBJECT)));
1176     __ cmpptr(t, Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize));
1177     __ jcc(Assembler::notEqual, no_oop);
1178     // retrieve result
1179     __ pop(ltos);
1180     // Unbox oop result, e.g. JNIHandles::resolve value.
1181     __ resolve_jobject(rax /* value */,
1182                        thread /* thread */,
1183                        t /* tmp */);
1184     __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize), rax);
1185     // keep stack depth as expected by pushing oop which will eventually be discarded
1186     __ push(ltos);
1187     __ bind(no_oop);
1188   }
1189 
1190 
1191   {
1192     Label no_reguard;
1193     __ cmpl(Address(thread, JavaThread::stack_guard_state_offset()),
1194             StackOverflow::stack_guard_yellow_reserved_disabled);
1195     __ jcc(Assembler::notEqual, no_reguard);
1196 
1197     __ pusha(); // XXX only save smashed registers
1198 #ifndef _LP64
1199     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
1200     __ popa();
1201 #else
1202     __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1203     __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1204     __ andptr(rsp, -16); // align stack as required by ABI
1205     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
1206     __ mov(rsp, r12); // restore sp
1207     __ popa(); // XXX only restore smashed registers
1208     __ reinit_heapbase();
1209 #endif // _LP64
1210 
1211     __ bind(no_reguard);
1212   }
1213 
1214 
1215   // The method register is junk from after the thread_in_native transition
1216   // until here.  Also can't call_VM until the bcp has been
1217   // restored.  Need bcp for throwing exception below so get it now.
1218   __ get_method(method);
1219 
1220   // restore to have legal interpreter frame, i.e., bci == 0 <=> code_base()
1221   __ movptr(rbcp, Address(method, Method::const_offset()));   // get ConstMethod*
1222   __ lea(rbcp, Address(rbcp, ConstMethod::codes_offset()));    // get codebase
1223 
1224   // handle exceptions (exception handling will handle unlocking!)
1225   {
1226     Label L;
1227     __ cmpptr(Address(thread, Thread::pending_exception_offset()), NULL_WORD);
1228     __ jcc(Assembler::zero, L);
1229     // Note: At some point we may want to unify this with the code
1230     // used in call_VM_base(); i.e., we should use the
1231     // StubRoutines::forward_exception code. For now this doesn't work
1232     // here because the rsp is not correctly set at this point.
1233     __ MacroAssembler::call_VM(noreg,
1234                                CAST_FROM_FN_PTR(address,
1235                                InterpreterRuntime::throw_pending_exception));
1236     __ should_not_reach_here();
1237     __ bind(L);
1238   }
1239 
1240   // do unlocking if necessary
1241   {
1242     Label L;
1243     __ movl(t, Address(method, Method::access_flags_offset()));
1244     __ testl(t, JVM_ACC_SYNCHRONIZED);
1245     __ jcc(Assembler::zero, L);
1246     // the code below should be shared with interpreter macro
1247     // assembler implementation
1248     {
1249       Label unlock;
1250       // BasicObjectLock will be first in list, since this is a
1251       // synchronized method. However, need to check that the object
1252       // has not been unlocked by an explicit monitorexit bytecode.
1253       const Address monitor(rbp,
1254                             (intptr_t)(frame::interpreter_frame_initial_sp_offset *
1255                                        wordSize - (int)sizeof(BasicObjectLock)));
1256 
1257       const Register regmon = NOT_LP64(rdx) LP64_ONLY(c_rarg1);
1258 
1259       // monitor expect in c_rarg1 for slow unlock path
1260       __ lea(regmon, monitor); // address of first monitor
1261 
1262       __ movptr(t, Address(regmon, BasicObjectLock::obj_offset()));
1263       __ testptr(t, t);
1264       __ jcc(Assembler::notZero, unlock);
1265 
1266       // Entry already unlocked, need to throw exception
1267       __ MacroAssembler::call_VM(noreg,
1268                                  CAST_FROM_FN_PTR(address,
1269                    InterpreterRuntime::throw_illegal_monitor_state_exception));
1270       __ should_not_reach_here();
1271 
1272       __ bind(unlock);
1273       __ unlock_object(regmon);
1274     }
1275     __ bind(L);
1276   }
1277 
1278   // jvmti support
1279   // Note: This must happen _after_ handling/throwing any exceptions since
1280   //       the exception handler code notifies the runtime of method exits
1281   //       too. If this happens before, method entry/exit notifications are
1282   //       not properly paired (was bug - gri 11/22/99).
1283   __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI);
1284 
1285   // restore potential result in edx:eax, call result handler to
1286   // restore potential result in ST0 & handle result
1287 
1288   __ pop(ltos);
1289   LP64_ONLY( __ pop(dtos));
1290 
1291   __ movptr(t, Address(rbp,
1292                        (frame::interpreter_frame_result_handler_offset) * wordSize));
1293   __ call(t);
1294 
1295   // remove activation
1296   __ movptr(t, Address(rbp,
1297                        frame::interpreter_frame_sender_sp_offset *
1298                        wordSize)); // get sender sp
1299   __ leave();                                // remove frame anchor
1300   __ pop(rdi);                               // get return address
1301   __ mov(rsp, t);                            // set sp to sender sp
1302   __ jmp(rdi);
1303 
1304   if (inc_counter) {
1305     // Handle overflow of counter and compile method
1306     __ bind(invocation_counter_overflow);
1307     generate_counter_overflow(continue_after_compile);
1308   }
1309 
1310   return entry_point;
1311 }
1312 
1313 // Abstract method entry
1314 // Attempt to execute abstract method. Throw exception
1315 address TemplateInterpreterGenerator::generate_abstract_entry(void) {
1316 
1317   address entry_point = __ pc();
1318 
1319   // abstract method entry
1320 
1321   //  pop return address, reset last_sp to null
1322   __ empty_expression_stack();
1323   __ restore_bcp();      // rsi must be correct for exception handler   (was destroyed)
1324   __ restore_locals();   // make sure locals pointer is correct as well (was destroyed)
1325 
1326   // throw exception
1327   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodErrorWithMethod), rbx);
1328   // the call_VM checks for exception, so we should never return here.
1329   __ should_not_reach_here();
1330 
1331   return entry_point;
1332 }
1333 
1334 //
1335 // Generic interpreted method entry to (asm) interpreter
1336 //
1337 address TemplateInterpreterGenerator::generate_normal_entry(bool synchronized) {
1338   // determine code generation flags
1339   bool inc_counter  = UseCompiler || CountCompiledCalls;
1340 
1341   // ebx: Method*
1342   // rbcp: sender sp (set in InterpreterMacroAssembler::prepare_to_jump_from_interpreted / generate_call_stub)
1343   address entry_point = __ pc();
1344 
1345   const Address constMethod(rbx, Method::const_offset());
1346   const Address access_flags(rbx, Method::access_flags_offset());
1347   const Address size_of_parameters(rdx,
1348                                    ConstMethod::size_of_parameters_offset());
1349   const Address size_of_locals(rdx, ConstMethod::size_of_locals_offset());
1350 
1351 
1352   // get parameter size (always needed)
1353   __ movptr(rdx, constMethod);
1354   __ load_unsigned_short(rcx, size_of_parameters);
1355 
1356   // rbx: Method*
1357   // rcx: size of parameters
1358   // rbcp: sender_sp (could differ from sp+wordSize if we were called via c2i )
1359 
1360   __ load_unsigned_short(rdx, size_of_locals); // get size of locals in words
1361   __ subl(rdx, rcx); // rdx = no. of additional locals
1362 
1363   // YYY
1364 //   __ incrementl(rdx);
1365 //   __ andl(rdx, -2);
1366 
1367   // see if we've got enough room on the stack for locals plus overhead.
1368   generate_stack_overflow_check();
1369 
1370   // get return address
1371   __ pop(rax);
1372 
1373   // compute beginning of parameters
1374   __ lea(rlocals, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize));
1375 
1376   // rdx - # of additional locals
1377   // allocate space for locals
1378   // explicitly initialize locals
1379   {
1380     Label exit, loop;
1381     __ testl(rdx, rdx);
1382     __ jcc(Assembler::lessEqual, exit); // do nothing if rdx <= 0
1383     __ bind(loop);
1384     __ push(NULL_WORD); // initialize local variables
1385     __ decrementl(rdx); // until everything initialized
1386     __ jcc(Assembler::greater, loop);
1387     __ bind(exit);
1388   }
1389 
1390   // initialize fixed part of activation frame
1391   generate_fixed_frame(false);
1392 
1393   // make sure method is not native & not abstract
1394 #ifdef ASSERT
1395   __ movl(rax, access_flags);
1396   {
1397     Label L;
1398     __ testl(rax, JVM_ACC_NATIVE);
1399     __ jcc(Assembler::zero, L);
1400     __ stop("tried to execute native method as non-native");
1401     __ bind(L);
1402   }
1403   {
1404     Label L;
1405     __ testl(rax, JVM_ACC_ABSTRACT);
1406     __ jcc(Assembler::zero, L);
1407     __ stop("tried to execute abstract method in interpreter");
1408     __ bind(L);
1409   }
1410 #endif
1411 
1412   // Since at this point in the method invocation the exception
1413   // handler would try to exit the monitor of synchronized methods
1414   // which hasn't been entered yet, we set the thread local variable
1415   // _do_not_unlock_if_synchronized to true. The remove_activation
1416   // will check this flag.
1417 
1418   const Register thread = NOT_LP64(rax) LP64_ONLY(r15_thread);
1419   NOT_LP64(__ get_thread(thread));
1420   const Address do_not_unlock_if_synchronized(thread,
1421         in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1422   __ movbool(do_not_unlock_if_synchronized, true);
1423 
1424   __ profile_parameters_type(rax, rcx, rdx);
1425   // increment invocation count & check for overflow
1426   Label invocation_counter_overflow;
1427   if (inc_counter) {
1428     generate_counter_incr(&invocation_counter_overflow);
1429   }
1430 
1431   Label continue_after_compile;
1432   __ bind(continue_after_compile);
1433 
1434   // check for synchronized interpreted methods
1435   bang_stack_shadow_pages(false);
1436 
1437   // reset the _do_not_unlock_if_synchronized flag
1438   NOT_LP64(__ get_thread(thread));
1439   __ movbool(do_not_unlock_if_synchronized, false);
1440 
1441   // check for synchronized methods
1442   // Must happen AFTER invocation_counter check and stack overflow check,
1443   // so method is not locked if overflows.
1444   if (synchronized) {
1445     // Allocate monitor and lock method
1446     lock_method();
1447   } else {
1448     // no synchronization necessary
1449 #ifdef ASSERT
1450     {
1451       Label L;
1452       __ movl(rax, access_flags);
1453       __ testl(rax, JVM_ACC_SYNCHRONIZED);
1454       __ jcc(Assembler::zero, L);
1455       __ stop("method needs synchronization");
1456       __ bind(L);
1457     }
1458 #endif
1459   }
1460 
1461   // start execution
1462 #ifdef ASSERT
1463   {
1464     Label L;
1465      const Address monitor_block_top (rbp,
1466                  frame::interpreter_frame_monitor_block_top_offset * wordSize);
1467     __ movptr(rax, monitor_block_top);
1468     __ lea(rax, Address(rbp, rax, Address::times_ptr));
1469     __ cmpptr(rax, rsp);
1470     __ jcc(Assembler::equal, L);
1471     __ stop("broken stack frame setup in interpreter 6");
1472     __ bind(L);
1473   }
1474 #endif
1475 
1476   // jvmti support
1477   __ notify_method_entry();
1478 
1479   __ dispatch_next(vtos);
1480 
1481   // invocation counter overflow
1482   if (inc_counter) {
1483     // Handle overflow of counter and compile method
1484     __ bind(invocation_counter_overflow);
1485     generate_counter_overflow(continue_after_compile);
1486   }
1487 
1488   return entry_point;
1489 }
1490 
1491 //-----------------------------------------------------------------------------
1492 // Exceptions
1493 
1494 void TemplateInterpreterGenerator::generate_throw_exception() {
1495   // Entry point in previous activation (i.e., if the caller was
1496   // interpreted)
1497   Interpreter::_rethrow_exception_entry = __ pc();
1498   // Restore sp to interpreter_frame_last_sp even though we are going
1499   // to empty the expression stack for the exception processing.
1500   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
1501   // rax: exception
1502   // rdx: return address/pc that threw exception
1503   __ restore_bcp();    // r13/rsi points to call/send
1504   __ restore_locals();
1505   LP64_ONLY(__ reinit_heapbase());  // restore r12 as heapbase.
1506   // Entry point for exceptions thrown within interpreter code
1507   Interpreter::_throw_exception_entry = __ pc();
1508   // expression stack is undefined here
1509   // rax: exception
1510   // r13/rsi: exception bcp
1511   __ verify_oop(rax);
1512   Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
1513   LP64_ONLY(__ mov(c_rarg1, rax));
1514 
1515   // expression stack must be empty before entering the VM in case of
1516   // an exception
1517   __ empty_expression_stack();
1518   // find exception handler address and preserve exception oop
1519   __ call_VM(rdx,
1520              CAST_FROM_FN_PTR(address,
1521                           InterpreterRuntime::exception_handler_for_exception),
1522              rarg);
1523   // rax: exception handler entry point
1524   // rdx: preserved exception oop
1525   // r13/rsi: bcp for exception handler
1526   __ push_ptr(rdx); // push exception which is now the only value on the stack
1527   __ jmp(rax); // jump to exception handler (may be _remove_activation_entry!)
1528 
1529   // If the exception is not handled in the current frame the frame is
1530   // removed and the exception is rethrown (i.e. exception
1531   // continuation is _rethrow_exception).
1532   //
1533   // Note: At this point the bci is still the bxi for the instruction
1534   // which caused the exception and the expression stack is
1535   // empty. Thus, for any VM calls at this point, GC will find a legal
1536   // oop map (with empty expression stack).
1537 
1538   // In current activation
1539   // tos: exception
1540   // esi: exception bcp
1541 
1542   //
1543   // JVMTI PopFrame support
1544   //
1545 
1546   Interpreter::_remove_activation_preserving_args_entry = __ pc();
1547   __ empty_expression_stack();
1548   // Set the popframe_processing bit in pending_popframe_condition
1549   // indicating that we are currently handling popframe, so that
1550   // call_VMs that may happen later do not trigger new popframe
1551   // handling cycles.
1552   const Register thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
1553   NOT_LP64(__ get_thread(thread));
1554   __ movl(rdx, Address(thread, JavaThread::popframe_condition_offset()));
1555   __ orl(rdx, JavaThread::popframe_processing_bit);
1556   __ movl(Address(thread, JavaThread::popframe_condition_offset()), rdx);
1557 
1558   {
1559     // Check to see whether we are returning to a deoptimized frame.
1560     // (The PopFrame call ensures that the caller of the popped frame is
1561     // either interpreted or compiled and deoptimizes it if compiled.)
1562     // In this case, we can't call dispatch_next() after the frame is
1563     // popped, but instead must save the incoming arguments and restore
1564     // them after deoptimization has occurred.
1565     //
1566     // Note that we don't compare the return PC against the
1567     // deoptimization blob's unpack entry because of the presence of
1568     // adapter frames in C2.
1569     Label caller_not_deoptimized;
1570     Register rarg = NOT_LP64(rdx) LP64_ONLY(c_rarg1);
1571     __ movptr(rarg, Address(rbp, frame::return_addr_offset * wordSize));
1572     __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1573                                InterpreterRuntime::interpreter_contains), rarg);
1574     __ testl(rax, rax);
1575     __ jcc(Assembler::notZero, caller_not_deoptimized);
1576 
1577     // Compute size of arguments for saving when returning to
1578     // deoptimized caller
1579     __ get_method(rax);
1580     __ movptr(rax, Address(rax, Method::const_offset()));
1581     __ load_unsigned_short(rax, Address(rax, in_bytes(ConstMethod::
1582                                                 size_of_parameters_offset())));
1583     __ shll(rax, Interpreter::logStackElementSize);
1584     __ restore_locals();
1585     __ subptr(rlocals, rax);
1586     __ addptr(rlocals, wordSize);
1587     // Save these arguments
1588     NOT_LP64(__ get_thread(thread));
1589     __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1590                                            Deoptimization::
1591                                            popframe_preserve_args),
1592                           thread, rax, rlocals);
1593 
1594     __ remove_activation(vtos, rdx,
1595                          /* throw_monitor_exception */ false,
1596                          /* install_monitor_exception */ false,
1597                          /* notify_jvmdi */ false);
1598 
1599     // Inform deoptimization that it is responsible for restoring
1600     // these arguments
1601     NOT_LP64(__ get_thread(thread));
1602     __ movl(Address(thread, JavaThread::popframe_condition_offset()),
1603             JavaThread::popframe_force_deopt_reexecution_bit);
1604 
1605     // Continue in deoptimization handler
1606     __ jmp(rdx);
1607 
1608     __ bind(caller_not_deoptimized);
1609   }
1610 
1611   __ remove_activation(vtos, rdx, /* rdx result (retaddr) is not used */
1612                        /* throw_monitor_exception */ false,
1613                        /* install_monitor_exception */ false,
1614                        /* notify_jvmdi */ false);
1615 
1616   // Finish with popframe handling
1617   // A previous I2C followed by a deoptimization might have moved the
1618   // outgoing arguments further up the stack. PopFrame expects the
1619   // mutations to those outgoing arguments to be preserved and other
1620   // constraints basically require this frame to look exactly as
1621   // though it had previously invoked an interpreted activation with
1622   // no space between the top of the expression stack (current
1623   // last_sp) and the top of stack. Rather than force deopt to
1624   // maintain this kind of invariant all the time we call a small
1625   // fixup routine to move the mutated arguments onto the top of our
1626   // expression stack if necessary.
1627 #ifndef _LP64
1628   __ mov(rax, rsp);
1629   __ movptr(rbx, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1630   __ lea(rbx, Address(rbp, rbx, Address::times_ptr));
1631   __ get_thread(thread);
1632   // PC must point into interpreter here
1633   __ set_last_Java_frame(thread, noreg, rbp, __ pc(), noreg);
1634   __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), thread, rax, rbx);
1635   __ get_thread(thread);
1636 #else
1637   __ mov(c_rarg1, rsp);
1638   __ movptr(c_rarg2, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1639   __ lea(c_rarg2, Address(rbp, c_rarg2, Address::times_ptr));
1640   // PC must point into interpreter here
1641   __ set_last_Java_frame(noreg, rbp, __ pc(), rscratch1);
1642   __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), r15_thread, c_rarg1, c_rarg2);
1643 #endif
1644   __ reset_last_Java_frame(thread, true);
1645 
1646   // Restore the last_sp and null it out
1647   __ movptr(rcx, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1648   __ lea(rsp, Address(rbp, rcx, Address::times_ptr));
1649   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
1650 
1651   __ restore_bcp();
1652   __ restore_locals();
1653   // The method data pointer was incremented already during
1654   // call profiling. We have to restore the mdp for the current bcp.
1655   if (ProfileInterpreter) {
1656     __ set_method_data_pointer_for_bcp();
1657   }
1658 
1659   // Clear the popframe condition flag
1660   NOT_LP64(__ get_thread(thread));
1661   __ movl(Address(thread, JavaThread::popframe_condition_offset()),
1662           JavaThread::popframe_inactive);
1663 
1664 #if INCLUDE_JVMTI
1665   {
1666     Label L_done;
1667     const Register local0 = rlocals;
1668 
1669     __ cmpb(Address(rbcp, 0), Bytecodes::_invokestatic);
1670     __ jcc(Assembler::notEqual, L_done);
1671 
1672     // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call.
1673     // Detect such a case in the InterpreterRuntime function and return the member name argument, or null.
1674 
1675     __ get_method(rdx);
1676     __ movptr(rax, Address(local0, 0));
1677     __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), rax, rdx, rbcp);
1678 
1679     __ testptr(rax, rax);
1680     __ jcc(Assembler::zero, L_done);
1681 
1682     __ movptr(Address(rbx, 0), rax);
1683     __ bind(L_done);
1684   }
1685 #endif // INCLUDE_JVMTI
1686 
1687   __ dispatch_next(vtos);
1688   // end of PopFrame support
1689 
1690   Interpreter::_remove_activation_entry = __ pc();
1691 
1692   // preserve exception over this code sequence
1693   __ pop_ptr(rax);
1694   NOT_LP64(__ get_thread(thread));
1695   __ movptr(Address(thread, JavaThread::vm_result_offset()), rax);
1696   // remove the activation (without doing throws on illegalMonitorExceptions)
1697   __ remove_activation(vtos, rdx, false, true, false);
1698   // restore exception
1699   NOT_LP64(__ get_thread(thread));
1700   __ get_vm_result(rax, thread);
1701 
1702   // In between activations - previous activation type unknown yet
1703   // compute continuation point - the continuation point expects the
1704   // following registers set up:
1705   //
1706   // rax: exception
1707   // rdx: return address/pc that threw exception
1708   // rsp: expression stack of caller
1709   // rbp: ebp of caller
1710   __ push(rax);                                  // save exception
1711   __ push(rdx);                                  // save return address
1712   __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1713                           SharedRuntime::exception_handler_for_return_address),
1714                         thread, rdx);
1715   __ mov(rbx, rax);                              // save exception handler
1716   __ pop(rdx);                                   // restore return address
1717   __ pop(rax);                                   // restore exception
1718   // Note that an "issuing PC" is actually the next PC after the call
1719   __ jmp(rbx);                                   // jump to exception
1720                                                  // handler of caller
1721 }
1722 
1723 
1724 //
1725 // JVMTI ForceEarlyReturn support
1726 //
1727 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
1728   address entry = __ pc();
1729 
1730   __ restore_bcp();
1731   __ restore_locals();
1732   __ empty_expression_stack();
1733   __ load_earlyret_value(state);  // 32 bits returns value in rdx, so don't reuse
1734 
1735   const Register thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
1736   NOT_LP64(__ get_thread(thread));
1737   __ movptr(rcx, Address(thread, JavaThread::jvmti_thread_state_offset()));
1738   Address cond_addr(rcx, JvmtiThreadState::earlyret_state_offset());
1739 
1740   // Clear the earlyret state
1741   __ movl(cond_addr, JvmtiThreadState::earlyret_inactive);
1742 
1743   __ remove_activation(state, rsi,
1744                        false, /* throw_monitor_exception */
1745                        false, /* install_monitor_exception */
1746                        true); /* notify_jvmdi */
1747   __ jmp(rsi);
1748 
1749   return entry;
1750 } // end of ForceEarlyReturn support
1751 
1752 
1753 //-----------------------------------------------------------------------------
1754 // Helper for vtos entry point generation
1755 
1756 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
1757                                                          address& bep,
1758                                                          address& cep,
1759                                                          address& sep,
1760                                                          address& aep,
1761                                                          address& iep,
1762                                                          address& lep,
1763                                                          address& fep,
1764                                                          address& dep,
1765                                                          address& vep) {
1766   assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
1767   Label L;
1768 #ifndef _LP64
1769   fep = __ pc();     // ftos entry point
1770       __ push(ftos);
1771       __ jmpb(L);
1772   dep = __ pc();     // dtos entry point
1773       __ push(dtos);
1774       __ jmpb(L);
1775 #else
1776   fep = __ pc();     // ftos entry point
1777       __ push_f(xmm0);
1778       __ jmpb(L);
1779   dep = __ pc();     // dtos entry point
1780       __ push_d(xmm0);
1781       __ jmpb(L);
1782 #endif // _LP64
1783   lep = __ pc();     // ltos entry point
1784       __ push_l();
1785       __ jmpb(L);
1786   aep = bep = cep = sep = iep = __ pc();      // [abcsi]tos entry point
1787       __ push_i_or_ptr();
1788   vep = __ pc();    // vtos entry point
1789   __ bind(L);
1790   generate_and_dispatch(t);
1791 }
1792 
1793 //-----------------------------------------------------------------------------
1794 
1795 // Non-product code
1796 #ifndef PRODUCT
1797 
1798 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
1799   address entry = __ pc();
1800 
1801 #ifndef _LP64
1802   // prepare expression stack
1803   __ pop(rcx);          // pop return address so expression stack is 'pure'
1804   __ push(state);       // save tosca
1805 
1806   // pass tosca registers as arguments & call tracer
1807   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::trace_bytecode), rcx, rax, rdx);
1808   __ mov(rcx, rax);     // make sure return address is not destroyed by pop(state)
1809   __ pop(state);        // restore tosca
1810 
1811   // return
1812   __ jmp(rcx);
1813 #else
1814   __ push(state);
1815   __ push(c_rarg0);
1816   __ push(c_rarg1);
1817   __ push(c_rarg2);
1818   __ push(c_rarg3);
1819   __ mov(c_rarg2, rax);  // Pass itos
1820 #ifdef _WIN64
1821   __ movflt(xmm3, xmm0); // Pass ftos
1822 #endif
1823   __ call_VM(noreg,
1824              CAST_FROM_FN_PTR(address, InterpreterRuntime::trace_bytecode),
1825              c_rarg1, c_rarg2, c_rarg3);
1826   __ pop(c_rarg3);
1827   __ pop(c_rarg2);
1828   __ pop(c_rarg1);
1829   __ pop(c_rarg0);
1830   __ pop(state);
1831   __ ret(0);                                   // return from result handler
1832 #endif // _LP64
1833 
1834   return entry;
1835 }
1836 
1837 void TemplateInterpreterGenerator::count_bytecode() {
1838   __ incrementl(ExternalAddress((address) &BytecodeCounter::_counter_value), rscratch1);
1839 }
1840 
1841 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
1842   __ incrementl(ExternalAddress((address) &BytecodeHistogram::_counters[t->bytecode()]), rscratch1);
1843 }
1844 
1845 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
1846   __ mov32(rbx, ExternalAddress((address) &BytecodePairHistogram::_index));
1847   __ shrl(rbx, BytecodePairHistogram::log2_number_of_codes);
1848   __ orl(rbx,
1849          ((int) t->bytecode()) <<
1850          BytecodePairHistogram::log2_number_of_codes);
1851   __ mov32(ExternalAddress((address) &BytecodePairHistogram::_index), rbx, rscratch1);
1852   __ lea(rscratch1, ExternalAddress((address) BytecodePairHistogram::_counters));
1853   __ incrementl(Address(rscratch1, rbx, Address::times_4));
1854 }
1855 
1856 
1857 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
1858   // Call a little run-time stub to avoid blow-up for each bytecode.
1859   // The run-time runtime saves the right registers, depending on
1860   // the tosca in-state for the given template.
1861 
1862   assert(Interpreter::trace_code(t->tos_in()) != nullptr,
1863          "entry must have been generated");
1864 #ifndef _LP64
1865   __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in())));
1866 #else
1867   __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1868   __ andptr(rsp, -16); // align stack as required by ABI
1869   __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in())));
1870   __ mov(rsp, r12); // restore sp
1871   __ reinit_heapbase();
1872 #endif // _LP64
1873 }
1874 
1875 
1876 void TemplateInterpreterGenerator::stop_interpreter_at() {
1877   Label L;
1878   __ cmp32(ExternalAddress((address) &BytecodeCounter::_counter_value),
1879            StopInterpreterAt,
1880            rscratch1);
1881   __ jcc(Assembler::notEqual, L);
1882   __ int3();
1883   __ bind(L);
1884 }
1885 #endif // !PRODUCT