1 /*
   2  * Copyright (c) 2003, 2018, 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 "gc/shared/barrierSetAssembler.hpp"
  28 #include "interpreter/bytecodeHistogram.hpp"
  29 #include "interpreter/interp_masm.hpp"
  30 #include "interpreter/interpreter.hpp"
  31 #include "interpreter/interpreterRuntime.hpp"
  32 #include "interpreter/templateInterpreterGenerator.hpp"
  33 #include "interpreter/templateTable.hpp"
  34 #include "oops/arrayOop.hpp"
  35 #include "oops/methodData.hpp"
  36 #include "oops/method.hpp"
  37 #include "oops/oop.inline.hpp"
  38 #include "prims/jvmtiExport.hpp"
  39 #include "prims/jvmtiThreadState.hpp"
  40 #include "runtime/arguments.hpp"
  41 #include "runtime/deoptimization.hpp"
  42 #include "runtime/frame.inline.hpp"
  43 #include "runtime/sharedRuntime.hpp"
  44 #include "runtime/stubRoutines.hpp"
  45 #include "runtime/synchronizer.hpp"
  46 #include "runtime/timer.hpp"
  47 #include "runtime/vframeArray.hpp"
  48 #include "utilities/debug.hpp"
  49 #include "utilities/macros.hpp"
  50 
  51 #define __ _masm->
  52 
  53 // Size of interpreter code.  Increase if too small.  Interpreter will
  54 // fail with a guarantee ("not enough space for interpreter generation");
  55 // if too small.
  56 // Run with +PrintInterpreter to get the VM to print out the size.
  57 // Max size with JVMTI
  58 #ifdef AMD64
  59 int TemplateInterpreter::InterpreterCodeSize = JVMCI_ONLY(268) NOT_JVMCI(256) * 1024;
  60 #else
  61 int TemplateInterpreter::InterpreterCodeSize = 224 * 1024;
  62 #endif // AMD64
  63 
  64 // Global Register Names
  65 static const Register rbcp     = LP64_ONLY(r13) NOT_LP64(rsi);
  66 static const Register rlocals  = LP64_ONLY(r14) NOT_LP64(rdi);
  67 
  68 const int method_offset = frame::interpreter_frame_method_offset * wordSize;
  69 const int bcp_offset    = frame::interpreter_frame_bcp_offset    * wordSize;
  70 const int locals_offset = frame::interpreter_frame_locals_offset * wordSize;
  71 
  72 
  73 //-----------------------------------------------------------------------------
  74 
  75 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() {
  76   address entry = __ pc();
  77 
  78 #ifdef ASSERT
  79   {
  80     Label L;
  81     __ lea(rax, Address(rbp,
  82                         frame::interpreter_frame_monitor_block_top_offset *
  83                         wordSize));
  84     __ cmpptr(rax, rsp); // rax = maximal rsp for current rbp (stack
  85                          // grows negative)
  86     __ jcc(Assembler::aboveEqual, L); // check if frame is complete
  87     __ stop ("interpreter frame not set up");
  88     __ bind(L);
  89   }
  90 #endif // ASSERT
  91   // Restore bcp under the assumption that the current frame is still
  92   // interpreted
  93   __ restore_bcp();
  94 
  95   // expression stack must be empty before entering the VM if an
  96   // exception happened
  97   __ empty_expression_stack();
  98   // throw exception
  99   __ call_VM(noreg,
 100              CAST_FROM_FN_PTR(address,
 101                               InterpreterRuntime::throw_StackOverflowError));
 102   return entry;
 103 }
 104 
 105 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler(
 106         const char* name) {
 107   address entry = __ pc();
 108   // expression stack must be empty before entering the VM if an
 109   // exception happened
 110   __ empty_expression_stack();
 111   // setup parameters
 112   // ??? convention: expect aberrant index in register ebx
 113   Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
 114   __ lea(rarg, ExternalAddress((address)name));
 115   __ call_VM(noreg,
 116              CAST_FROM_FN_PTR(address,
 117                               InterpreterRuntime::
 118                               throw_ArrayIndexOutOfBoundsException),
 119              rarg, rbx);
 120   return entry;
 121 }
 122 
 123 address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
 124   address entry = __ pc();
 125 
 126   // object is at TOS
 127   Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
 128   __ pop(rarg);
 129 
 130   // expression stack must be empty before entering the VM if an
 131   // exception happened
 132   __ empty_expression_stack();
 133 
 134   __ call_VM(noreg,
 135              CAST_FROM_FN_PTR(address,
 136                               InterpreterRuntime::
 137                               throw_ClassCastException),
 138              rarg);
 139   return entry;
 140 }
 141 
 142 address TemplateInterpreterGenerator::generate_exception_handler_common(
 143         const char* name, const char* message, bool pass_oop) {
 144   assert(!pass_oop || message == NULL, "either oop or message but not both");
 145   address entry = __ pc();
 146 
 147   Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
 148   Register rarg2 = NOT_LP64(rbx) LP64_ONLY(c_rarg2);
 149 
 150   if (pass_oop) {
 151     // object is at TOS
 152     __ pop(rarg2);
 153   }
 154   // expression stack must be empty before entering the VM if an
 155   // exception happened
 156   __ empty_expression_stack();
 157   // setup parameters
 158   __ lea(rarg, ExternalAddress((address)name));
 159   if (pass_oop) {
 160     __ call_VM(rax, CAST_FROM_FN_PTR(address,
 161                                      InterpreterRuntime::
 162                                      create_klass_exception),
 163                rarg, rarg2);
 164   } else {
 165     __ lea(rarg2, ExternalAddress((address)message));
 166     __ call_VM(rax,
 167                CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception),
 168                rarg, rarg2);
 169   }
 170   // throw exception
 171   __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
 172   return entry;
 173 }
 174 
 175 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step, size_t index_size) {
 176   address entry = __ pc();
 177 
 178 #ifndef _LP64
 179 #ifdef COMPILER2
 180   // The FPU stack is clean if UseSSE >= 2 but must be cleaned in other cases
 181   if ((state == ftos && UseSSE < 1) || (state == dtos && UseSSE < 2)) {
 182     for (int i = 1; i < 8; i++) {
 183         __ ffree(i);
 184     }
 185   } else if (UseSSE < 2) {
 186     __ empty_FPU_stack();
 187   }
 188 #endif // COMPILER2
 189   if ((state == ftos && UseSSE < 1) || (state == dtos && UseSSE < 2)) {
 190     __ MacroAssembler::verify_FPU(1, "generate_return_entry_for compiled");
 191   } else {
 192     __ MacroAssembler::verify_FPU(0, "generate_return_entry_for compiled");
 193   }
 194 
 195   if (state == ftos) {
 196     __ MacroAssembler::verify_FPU(UseSSE >= 1 ? 0 : 1, "generate_return_entry_for in interpreter");
 197   } else if (state == dtos) {
 198     __ MacroAssembler::verify_FPU(UseSSE >= 2 ? 0 : 1, "generate_return_entry_for in interpreter");
 199   }
 200 #endif // _LP64
 201 
 202   // Restore stack bottom in case i2c adjusted stack
 203   __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
 204   // and NULL it as marker that esp is now tos until next java call
 205   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
 206 
 207   __ restore_bcp();
 208   __ restore_locals();
 209 
 210   if (state == atos) {
 211     Register mdp = rbx;
 212     Register tmp = rcx;
 213     __ profile_return_type(mdp, rax, tmp);
 214   }
 215 
 216   const Register cache = rbx;
 217   const Register index = rcx;
 218   __ get_cache_and_index_at_bcp(cache, index, 1, index_size);
 219 
 220   const Register flags = cache;
 221   __ movl(flags, Address(cache, index, Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()));
 222   __ andl(flags, ConstantPoolCacheEntry::parameter_size_mask);
 223   __ lea(rsp, Address(rsp, flags, Interpreter::stackElementScale()));
 224 
 225    const Register java_thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
 226    if (JvmtiExport::can_pop_frame()) {
 227      NOT_LP64(__ get_thread(java_thread));
 228      __ check_and_handle_popframe(java_thread);
 229    }
 230    if (JvmtiExport::can_force_early_return()) {
 231      NOT_LP64(__ get_thread(java_thread));
 232      __ check_and_handle_earlyret(java_thread);
 233    }
 234 
 235   __ dispatch_next(state, step);
 236 
 237   return entry;
 238 }
 239 
 240 
 241 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state, int step, address continuation) {
 242   address entry = __ pc();
 243 
 244 #ifndef _LP64
 245   if (state == ftos) {
 246     __ MacroAssembler::verify_FPU(UseSSE >= 1 ? 0 : 1, "generate_deopt_entry_for in interpreter");
 247   } else if (state == dtos) {
 248     __ MacroAssembler::verify_FPU(UseSSE >= 2 ? 0 : 1, "generate_deopt_entry_for in interpreter");
 249   }
 250 #endif // _LP64
 251 
 252   // NULL last_sp until next java call
 253   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
 254   __ restore_bcp();
 255   __ restore_locals();
 256   const Register thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
 257   NOT_LP64(__ get_thread(thread));
 258 #if INCLUDE_JVMCI
 259   // Check if we need to take lock at entry of synchronized method.  This can
 260   // only occur on method entry so emit it only for vtos with step 0.
 261   if ((EnableJVMCI || UseAOT) && state == vtos && step == 0) {
 262     Label L;
 263     __ cmpb(Address(thread, JavaThread::pending_monitorenter_offset()), 0);
 264     __ jcc(Assembler::zero, L);
 265     // Clear flag.
 266     __ movb(Address(thread, JavaThread::pending_monitorenter_offset()), 0);
 267     // Satisfy calling convention for lock_method().
 268     __ get_method(rbx);
 269     // Take lock.
 270     lock_method();
 271     __ bind(L);
 272   } else {
 273 #ifdef ASSERT
 274     if (EnableJVMCI) {
 275       Label L;
 276       __ cmpb(Address(r15_thread, JavaThread::pending_monitorenter_offset()), 0);
 277       __ jccb(Assembler::zero, L);
 278       __ stop("unexpected pending monitor in deopt entry");
 279       __ bind(L);
 280     }
 281 #endif
 282   }
 283 #endif
 284   // handle exceptions
 285   {
 286     Label L;
 287     __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t) NULL_WORD);
 288     __ jcc(Assembler::zero, L);
 289     __ call_VM(noreg,
 290                CAST_FROM_FN_PTR(address,
 291                                 InterpreterRuntime::throw_pending_exception));
 292     __ should_not_reach_here();
 293     __ bind(L);
 294   }
 295   if (continuation == NULL) {
 296     __ dispatch_next(state, step);
 297   } else {
 298     __ jump_to_entry(continuation);
 299   }
 300   return entry;
 301 }
 302 
 303 address TemplateInterpreterGenerator::generate_result_handler_for(
 304         BasicType type) {
 305   address entry = __ pc();
 306   switch (type) {
 307   case T_BOOLEAN: __ c2bool(rax);            break;
 308 #ifndef _LP64
 309   case T_CHAR   : __ andptr(rax, 0xFFFF);    break;
 310 #else
 311   case T_CHAR   : __ movzwl(rax, rax);       break;
 312 #endif // _LP64
 313   case T_BYTE   : __ sign_extend_byte(rax);  break;
 314   case T_SHORT  : __ sign_extend_short(rax); break;
 315   case T_INT    : /* nothing to do */        break;
 316   case T_LONG   : /* nothing to do */        break;
 317   case T_VOID   : /* nothing to do */        break;
 318 #ifndef _LP64
 319   case T_DOUBLE :
 320   case T_FLOAT  :
 321     { const Register t = InterpreterRuntime::SignatureHandlerGenerator::temp();
 322       __ pop(t);                            // remove return address first
 323       // Must return a result for interpreter or compiler. In SSE
 324       // mode, results are returned in xmm0 and the FPU stack must
 325       // be empty.
 326       if (type == T_FLOAT && UseSSE >= 1) {
 327         // Load ST0
 328         __ fld_d(Address(rsp, 0));
 329         // Store as float and empty fpu stack
 330         __ fstp_s(Address(rsp, 0));
 331         // and reload
 332         __ movflt(xmm0, Address(rsp, 0));
 333       } else if (type == T_DOUBLE && UseSSE >= 2 ) {
 334         __ movdbl(xmm0, Address(rsp, 0));
 335       } else {
 336         // restore ST0
 337         __ fld_d(Address(rsp, 0));
 338       }
 339       // and pop the temp
 340       __ addptr(rsp, 2 * wordSize);
 341       __ push(t);                           // restore return address
 342     }
 343     break;
 344 #else
 345   case T_FLOAT  : /* nothing to do */        break;
 346   case T_DOUBLE : /* nothing to do */        break;
 347 #endif // _LP64
 348 
 349   case T_OBJECT :
 350     // retrieve result from frame
 351     __ movptr(rax, Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize));
 352     // and verify it
 353     __ verify_oop(rax);
 354     break;
 355   default       : ShouldNotReachHere();
 356   }
 357   __ ret(0);                                   // return from result handler
 358   return entry;
 359 }
 360 
 361 address TemplateInterpreterGenerator::generate_safept_entry_for(
 362         TosState state,
 363         address runtime_entry) {
 364   address entry = __ pc();
 365   __ push(state);
 366   __ call_VM(noreg, runtime_entry);
 367   __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
 368   return entry;
 369 }
 370 
 371 
 372 
 373 // Helpers for commoning out cases in the various type of method entries.
 374 //
 375 
 376 
 377 // increment invocation count & check for overflow
 378 //
 379 // Note: checking for negative value instead of overflow
 380 //       so we have a 'sticky' overflow test
 381 //
 382 // rbx: method
 383 // rcx: invocation counter
 384 //
 385 void TemplateInterpreterGenerator::generate_counter_incr(
 386         Label* overflow,
 387         Label* profile_method,
 388         Label* profile_method_continue) {
 389   Label done;
 390   // Note: In tiered we increment either counters in Method* or in MDO depending if we're profiling or not.
 391   if (TieredCompilation) {
 392     int increment = InvocationCounter::count_increment;
 393     Label no_mdo;
 394     if (ProfileInterpreter) {
 395       // Are we profiling?
 396       __ movptr(rax, Address(rbx, Method::method_data_offset()));
 397       __ testptr(rax, rax);
 398       __ jccb(Assembler::zero, no_mdo);
 399       // Increment counter in the MDO
 400       const Address mdo_invocation_counter(rax, in_bytes(MethodData::invocation_counter_offset()) +
 401                                                 in_bytes(InvocationCounter::counter_offset()));
 402       const Address mask(rax, in_bytes(MethodData::invoke_mask_offset()));
 403       __ increment_mask_and_jump(mdo_invocation_counter, increment, mask, rcx, false, Assembler::zero, overflow);
 404       __ jmp(done);
 405     }
 406     __ bind(no_mdo);
 407     // Increment counter in MethodCounters
 408     const Address invocation_counter(rax,
 409                   MethodCounters::invocation_counter_offset() +
 410                   InvocationCounter::counter_offset());
 411     __ get_method_counters(rbx, rax, done);
 412     const Address mask(rax, in_bytes(MethodCounters::invoke_mask_offset()));
 413     __ increment_mask_and_jump(invocation_counter, increment, mask, rcx,
 414                                false, Assembler::zero, overflow);
 415     __ bind(done);
 416   } else { // not TieredCompilation
 417     const Address backedge_counter(rax,
 418                   MethodCounters::backedge_counter_offset() +
 419                   InvocationCounter::counter_offset());
 420     const Address invocation_counter(rax,
 421                   MethodCounters::invocation_counter_offset() +
 422                   InvocationCounter::counter_offset());
 423 
 424     __ get_method_counters(rbx, rax, done);
 425 
 426     if (ProfileInterpreter) {
 427       __ incrementl(Address(rax,
 428               MethodCounters::interpreter_invocation_counter_offset()));
 429     }
 430     // Update standard invocation counters
 431     __ movl(rcx, invocation_counter);
 432     __ incrementl(rcx, InvocationCounter::count_increment);
 433     __ movl(invocation_counter, rcx); // save invocation count
 434 
 435     __ movl(rax, backedge_counter);   // load backedge counter
 436     __ andl(rax, InvocationCounter::count_mask_value); // mask out the status bits
 437 
 438     __ addl(rcx, rax);                // add both counters
 439 
 440     // profile_method is non-null only for interpreted method so
 441     // profile_method != NULL == !native_call
 442 
 443     if (ProfileInterpreter && profile_method != NULL) {
 444       // Test to see if we should create a method data oop
 445       __ movptr(rax, Address(rbx, Method::method_counters_offset()));
 446       __ cmp32(rcx, Address(rax, in_bytes(MethodCounters::interpreter_profile_limit_offset())));
 447       __ jcc(Assembler::less, *profile_method_continue);
 448 
 449       // if no method data exists, go to profile_method
 450       __ test_method_data_pointer(rax, *profile_method);
 451     }
 452 
 453     __ movptr(rax, Address(rbx, Method::method_counters_offset()));
 454     __ cmp32(rcx, Address(rax, in_bytes(MethodCounters::interpreter_invocation_limit_offset())));
 455     __ jcc(Assembler::aboveEqual, *overflow);
 456     __ bind(done);
 457   }
 458 }
 459 
 460 void TemplateInterpreterGenerator::generate_counter_overflow(Label& do_continue) {
 461 
 462   // Asm interpreter on entry
 463   // r14/rdi - locals
 464   // r13/rsi - bcp
 465   // rbx - method
 466   // rdx - cpool --- DOES NOT APPEAR TO BE TRUE
 467   // rbp - interpreter frame
 468 
 469   // On return (i.e. jump to entry_point) [ back to invocation of interpreter ]
 470   // Everything as it was on entry
 471   // rdx is not restored. Doesn't appear to really be set.
 472 
 473   // InterpreterRuntime::frequency_counter_overflow takes two
 474   // arguments, the first (thread) is passed by call_VM, the second
 475   // indicates if the counter overflow occurs at a backwards branch
 476   // (NULL bcp).  We pass zero for it.  The call returns the address
 477   // of the verified entry point for the method or NULL if the
 478   // compilation did not complete (either went background or bailed
 479   // out).
 480   Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
 481   __ movl(rarg, 0);
 482   __ call_VM(noreg,
 483              CAST_FROM_FN_PTR(address,
 484                               InterpreterRuntime::frequency_counter_overflow),
 485              rarg);
 486 
 487   __ movptr(rbx, Address(rbp, method_offset));   // restore Method*
 488   // Preserve invariant that r13/r14 contain bcp/locals of sender frame
 489   // and jump to the interpreted entry.
 490   __ jmp(do_continue, relocInfo::none);
 491 }
 492 
 493 // See if we've got enough room on the stack for locals plus overhead below
 494 // JavaThread::stack_overflow_limit(). If not, throw a StackOverflowError
 495 // without going through the signal handler, i.e., reserved and yellow zones
 496 // will not be made usable. The shadow zone must suffice to handle the
 497 // overflow.
 498 // The expression stack grows down incrementally, so the normal guard
 499 // page mechanism will work for that.
 500 //
 501 // NOTE: Since the additional locals are also always pushed (wasn't
 502 // obvious in generate_fixed_frame) so the guard should work for them
 503 // too.
 504 //
 505 // Args:
 506 //      rdx: number of additional locals this frame needs (what we must check)
 507 //      rbx: Method*
 508 //
 509 // Kills:
 510 //      rax
 511 void TemplateInterpreterGenerator::generate_stack_overflow_check(void) {
 512 
 513   // monitor entry size: see picture of stack in frame_x86.hpp
 514   const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
 515 
 516   // total overhead size: entry_size + (saved rbp through expr stack
 517   // bottom).  be sure to change this if you add/subtract anything
 518   // to/from the overhead area
 519   const int overhead_size =
 520     -(frame::interpreter_frame_initial_sp_offset * wordSize) + entry_size;
 521 
 522   const int page_size = os::vm_page_size();
 523 
 524   Label after_frame_check;
 525 
 526   // see if the frame is greater than one page in size. If so,
 527   // then we need to verify there is enough stack space remaining
 528   // for the additional locals.
 529   __ cmpl(rdx, (page_size - overhead_size) / Interpreter::stackElementSize);
 530   __ jcc(Assembler::belowEqual, after_frame_check);
 531 
 532   // compute rsp as if this were going to be the last frame on
 533   // the stack before the red zone
 534 
 535   Label after_frame_check_pop;
 536   const Register thread = NOT_LP64(rsi) LP64_ONLY(r15_thread);
 537 #ifndef _LP64
 538   __ push(thread);
 539   __ get_thread(thread);
 540 #endif
 541 
 542   const Address stack_limit(thread, JavaThread::stack_overflow_limit_offset());
 543 
 544   // locals + overhead, in bytes
 545   __ mov(rax, rdx);
 546   __ shlptr(rax, Interpreter::logStackElementSize); // Convert parameter count to bytes.
 547   __ addptr(rax, overhead_size);
 548 
 549 #ifdef ASSERT
 550   Label limit_okay;
 551   // Verify that thread stack overflow limit is non-zero.
 552   __ cmpptr(stack_limit, (int32_t)NULL_WORD);
 553   __ jcc(Assembler::notEqual, limit_okay);
 554   __ stop("stack overflow limit is zero");
 555   __ bind(limit_okay);
 556 #endif
 557 
 558   // Add locals/frame size to stack limit.
 559   __ addptr(rax, stack_limit);
 560 
 561   // Check against the current stack bottom.
 562   __ cmpptr(rsp, rax);
 563 
 564   __ jcc(Assembler::above, after_frame_check_pop);
 565   NOT_LP64(__ pop(rsi));  // get saved bcp
 566 
 567   // Restore sender's sp as SP. This is necessary if the sender's
 568   // frame is an extended compiled frame (see gen_c2i_adapter())
 569   // and safer anyway in case of JSR292 adaptations.
 570 
 571   __ pop(rax); // return address must be moved if SP is changed
 572   __ mov(rsp, rbcp);
 573   __ push(rax);
 574 
 575   // Note: the restored frame is not necessarily interpreted.
 576   // Use the shared runtime version of the StackOverflowError.
 577   assert(StubRoutines::throw_StackOverflowError_entry() != NULL, "stub not yet generated");
 578   __ jump(ExternalAddress(StubRoutines::throw_StackOverflowError_entry()));
 579   // all done with frame size check
 580   __ bind(after_frame_check_pop);
 581   NOT_LP64(__ pop(rsi));
 582 
 583   // all done with frame size check
 584   __ bind(after_frame_check);
 585 }
 586 
 587 // Allocate monitor and lock method (asm interpreter)
 588 //
 589 // Args:
 590 //      rbx: Method*
 591 //      r14/rdi: locals
 592 //
 593 // Kills:
 594 //      rax
 595 //      c_rarg0, c_rarg1, c_rarg2, c_rarg3, ...(param regs)
 596 //      rscratch1, rscratch2 (scratch regs)
 597 void TemplateInterpreterGenerator::lock_method() {
 598   // synchronize method
 599   const Address access_flags(rbx, Method::access_flags_offset());
 600   const Address monitor_block_top(
 601         rbp,
 602         frame::interpreter_frame_monitor_block_top_offset * wordSize);
 603   const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
 604 
 605 #ifdef ASSERT
 606   {
 607     Label L;
 608     __ movl(rax, access_flags);
 609     __ testl(rax, JVM_ACC_SYNCHRONIZED);
 610     __ jcc(Assembler::notZero, L);
 611     __ stop("method doesn't need synchronization");
 612     __ bind(L);
 613   }
 614 #endif // ASSERT
 615 
 616   // get synchronization object
 617   {
 618     Label done;
 619     __ movl(rax, access_flags);
 620     __ testl(rax, JVM_ACC_STATIC);
 621     // get receiver (assume this is frequent case)
 622     __ movptr(rax, Address(rlocals, Interpreter::local_offset_in_bytes(0)));
 623     __ jcc(Assembler::zero, done);
 624     __ load_mirror(rax, rbx);
 625 
 626 #ifdef ASSERT
 627     {
 628       Label L;
 629       __ testptr(rax, rax);
 630       __ jcc(Assembler::notZero, L);
 631       __ stop("synchronization object is NULL");
 632       __ bind(L);
 633     }
 634 #endif // ASSERT
 635 
 636     __ bind(done);
 637   }
 638 
 639   // add space for monitor & lock
 640   __ subptr(rsp, entry_size); // add space for a monitor entry
 641   __ movptr(monitor_block_top, rsp);  // set new monitor block top
 642   // store object
 643   __ movptr(Address(rsp, BasicObjectLock::obj_offset_in_bytes()), rax);
 644   const Register lockreg = NOT_LP64(rdx) LP64_ONLY(c_rarg1);
 645   __ movptr(lockreg, rsp); // object address
 646   __ lock_object(lockreg);
 647 }
 648 
 649 // Generate a fixed interpreter frame. This is identical setup for
 650 // interpreted methods and for native methods hence the shared code.
 651 //
 652 // Args:
 653 //      rax: return address
 654 //      rbx: Method*
 655 //      r14/rdi: pointer to locals
 656 //      r13/rsi: sender sp
 657 //      rdx: cp cache
 658 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
 659   // initialize fixed part of activation frame
 660   __ push(rax);        // save return address
 661   __ enter();          // save old & set new rbp
 662   __ push(rbcp);        // set sender sp
 663   __ push((int)NULL_WORD); // leave last_sp as null
 664   __ movptr(rbcp, Address(rbx, Method::const_offset()));      // get ConstMethod*
 665   __ lea(rbcp, Address(rbcp, ConstMethod::codes_offset())); // get codebase
 666   __ push(rbx);        // save Method*
 667   // Get mirror and store it in the frame as GC root for this Method*
 668   __ load_mirror(rdx, rbx);
 669   __ push(rdx);
 670   if (ProfileInterpreter) {
 671     Label method_data_continue;
 672     __ movptr(rdx, Address(rbx, in_bytes(Method::method_data_offset())));
 673     __ testptr(rdx, rdx);
 674     __ jcc(Assembler::zero, method_data_continue);
 675     __ addptr(rdx, in_bytes(MethodData::data_offset()));
 676     __ bind(method_data_continue);
 677     __ push(rdx);      // set the mdp (method data pointer)
 678   } else {
 679     __ push(0);
 680   }
 681 
 682   __ movptr(rdx, Address(rbx, Method::const_offset()));
 683   __ movptr(rdx, Address(rdx, ConstMethod::constants_offset()));
 684   __ movptr(rdx, Address(rdx, ConstantPool::cache_offset_in_bytes()));
 685   __ push(rdx); // set constant pool cache
 686   __ push(rlocals); // set locals pointer
 687   if (native_call) {
 688     __ push(0); // no bcp
 689   } else {
 690     __ push(rbcp); // set bcp
 691   }
 692   __ push(0); // reserve word for pointer to expression stack bottom
 693   __ movptr(Address(rsp, 0), rsp); // set expression stack bottom
 694 }
 695 
 696 // End of helpers
 697 
 698 // Method entry for java.lang.ref.Reference.get.
 699 address TemplateInterpreterGenerator::generate_Reference_get_entry(void) {
 700   // Code: _aload_0, _getfield, _areturn
 701   // parameter size = 1
 702   //
 703   // The code that gets generated by this routine is split into 2 parts:
 704   //    1. The "intrinsified" code performing an ON_WEAK_OOP_REF load,
 705   //    2. The slow path - which is an expansion of the regular method entry.
 706   //
 707   // Notes:-
 708   // * An intrinsic is always executed, where an ON_WEAK_OOP_REF load is performed.
 709   // * We may jump to the slow path iff the receiver is null. If the
 710   //   Reference object is null then we no longer perform an ON_WEAK_OOP_REF load
 711   //   Thus we can use the regular method entry code to generate the NPE.
 712   //
 713   // rbx: Method*
 714 
 715   // r13: senderSP must preserve for slow path, set SP to it on fast path
 716 
 717   address entry = __ pc();
 718 
 719   const int referent_offset = java_lang_ref_Reference::referent_offset;
 720   guarantee(referent_offset > 0, "referent offset not initialized");
 721 
 722   Label slow_path;
 723   // rbx: method
 724 
 725   // Check if local 0 != NULL
 726   // If the receiver is null then it is OK to jump to the slow path.
 727   __ movptr(rax, Address(rsp, wordSize));
 728 
 729   __ testptr(rax, rax);
 730   __ jcc(Assembler::zero, slow_path);
 731 
 732   // rax: local 0
 733   // rbx: method (but can be used as scratch now)
 734   // rdx: scratch
 735   // rdi: scratch
 736 
 737   // Preserve the sender sp in case the load barrier
 738   // calls the runtime
 739   NOT_LP64(__ push(rsi));
 740 
 741   // Load the value of the referent field.
 742   const Address field_address(rax, referent_offset);
 743   __ load_heap_oop(rax, field_address, /*tmp1*/ rbx, /*tmp_thread*/ rdx, ON_WEAK_OOP_REF);
 744 
 745   // _areturn
 746   const Register sender_sp = NOT_LP64(rsi) LP64_ONLY(r13);
 747   NOT_LP64(__ pop(rsi));      // get sender sp
 748   __ pop(rdi);                // get return address
 749   __ mov(rsp, sender_sp);     // set sp to sender sp
 750   __ jmp(rdi);
 751   __ ret(0);
 752 
 753   // generate a vanilla interpreter entry as the slow path
 754   __ bind(slow_path);
 755   __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::zerolocals));
 756   return entry;
 757 }
 758 
 759 void TemplateInterpreterGenerator::bang_stack_shadow_pages(bool native_call) {
 760   // Quick & dirty stack overflow checking: bang the stack & handle trap.
 761   // Note that we do the banging after the frame is setup, since the exception
 762   // handling code expects to find a valid interpreter frame on the stack.
 763   // Doing the banging earlier fails if the caller frame is not an interpreter
 764   // frame.
 765   // (Also, the exception throwing code expects to unlock any synchronized
 766   // method receiever, so do the banging after locking the receiver.)
 767 
 768   // Bang each page in the shadow zone. We can't assume it's been done for
 769   // an interpreter frame with greater than a page of locals, so each page
 770   // needs to be checked.  Only true for non-native.
 771   if (UseStackBanging) {
 772     const int page_size = os::vm_page_size();
 773     const int n_shadow_pages = ((int)JavaThread::stack_shadow_zone_size()) / page_size;
 774     const int start_page = native_call ? n_shadow_pages : 1;
 775     for (int pages = start_page; pages <= n_shadow_pages; pages++) {
 776       __ bang_stack_with_offset(pages*page_size);
 777     }
 778   }
 779 }
 780 
 781 // Interpreter stub for calling a native method. (asm interpreter)
 782 // This sets up a somewhat different looking stack for calling the
 783 // native method than the typical interpreter frame setup.
 784 address TemplateInterpreterGenerator::generate_native_entry(bool synchronized) {
 785   // determine code generation flags
 786   bool inc_counter  = UseCompiler || CountCompiledCalls || LogTouchedMethods;
 787 
 788   // rbx: Method*
 789   // rbcp: sender sp
 790 
 791   address entry_point = __ pc();
 792 
 793   const Address constMethod       (rbx, Method::const_offset());
 794   const Address access_flags      (rbx, Method::access_flags_offset());
 795   const Address size_of_parameters(rcx, ConstMethod::
 796                                         size_of_parameters_offset());
 797 
 798 
 799   // get parameter size (always needed)
 800   __ movptr(rcx, constMethod);
 801   __ load_unsigned_short(rcx, size_of_parameters);
 802 
 803   // native calls don't need the stack size check since they have no
 804   // expression stack and the arguments are already on the stack and
 805   // we only add a handful of words to the stack
 806 
 807   // rbx: Method*
 808   // rcx: size of parameters
 809   // rbcp: sender sp
 810   __ pop(rax);                                       // get return address
 811 
 812   // for natives the size of locals is zero
 813 
 814   // compute beginning of parameters
 815   __ lea(rlocals, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize));
 816 
 817   // add 2 zero-initialized slots for native calls
 818   // initialize result_handler slot
 819   __ push((int) NULL_WORD);
 820   // slot for oop temp
 821   // (static native method holder mirror/jni oop result)
 822   __ push((int) NULL_WORD);
 823 
 824   // initialize fixed part of activation frame
 825   generate_fixed_frame(true);
 826 
 827   // make sure method is native & not abstract
 828 #ifdef ASSERT
 829   __ movl(rax, access_flags);
 830   {
 831     Label L;
 832     __ testl(rax, JVM_ACC_NATIVE);
 833     __ jcc(Assembler::notZero, L);
 834     __ stop("tried to execute non-native method as native");
 835     __ bind(L);
 836   }
 837   {
 838     Label L;
 839     __ testl(rax, JVM_ACC_ABSTRACT);
 840     __ jcc(Assembler::zero, L);
 841     __ stop("tried to execute abstract method in interpreter");
 842     __ bind(L);
 843   }
 844 #endif
 845 
 846   // Since at this point in the method invocation the exception handler
 847   // would try to exit the monitor of synchronized methods which hasn't
 848   // been entered yet, we set the thread local variable
 849   // _do_not_unlock_if_synchronized to true. The remove_activation will
 850   // check this flag.
 851 
 852   const Register thread1 = NOT_LP64(rax) LP64_ONLY(r15_thread);
 853   NOT_LP64(__ get_thread(thread1));
 854   const Address do_not_unlock_if_synchronized(thread1,
 855         in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
 856   __ movbool(do_not_unlock_if_synchronized, true);
 857 
 858   // increment invocation count & check for overflow
 859   Label invocation_counter_overflow;
 860   if (inc_counter) {
 861     generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
 862   }
 863 
 864   Label continue_after_compile;
 865   __ bind(continue_after_compile);
 866 
 867   bang_stack_shadow_pages(true);
 868 
 869   // reset the _do_not_unlock_if_synchronized flag
 870   NOT_LP64(__ get_thread(thread1));
 871   __ movbool(do_not_unlock_if_synchronized, false);
 872 
 873   // check for synchronized methods
 874   // Must happen AFTER invocation_counter check and stack overflow check,
 875   // so method is not locked if overflows.
 876   if (synchronized) {
 877     lock_method();
 878   } else {
 879     // no synchronization necessary
 880 #ifdef ASSERT
 881     {
 882       Label L;
 883       __ movl(rax, access_flags);
 884       __ testl(rax, JVM_ACC_SYNCHRONIZED);
 885       __ jcc(Assembler::zero, L);
 886       __ stop("method needs synchronization");
 887       __ bind(L);
 888     }
 889 #endif
 890   }
 891 
 892   // start execution
 893 #ifdef ASSERT
 894   {
 895     Label L;
 896     const Address monitor_block_top(rbp,
 897                  frame::interpreter_frame_monitor_block_top_offset * wordSize);
 898     __ movptr(rax, monitor_block_top);
 899     __ cmpptr(rax, rsp);
 900     __ jcc(Assembler::equal, L);
 901     __ stop("broken stack frame setup in interpreter");
 902     __ bind(L);
 903   }
 904 #endif
 905 
 906   // jvmti support
 907   __ notify_method_entry();
 908 
 909   // work registers
 910   const Register method = rbx;
 911   const Register thread = NOT_LP64(rdi) LP64_ONLY(r15_thread);
 912   const Register t      = NOT_LP64(rcx) LP64_ONLY(r11);
 913 
 914   // allocate space for parameters
 915   __ get_method(method);
 916   __ movptr(t, Address(method, Method::const_offset()));
 917   __ load_unsigned_short(t, Address(t, ConstMethod::size_of_parameters_offset()));
 918 
 919 #ifndef _LP64
 920   __ shlptr(t, Interpreter::logStackElementSize); // 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 oop).
 956   // However, large signatures cannot be cached and are generated
 957   // each time here.  The slow-path generator can do a GC on return,
 958   // so we must reload it after the call.
 959   __ call(t);
 960   __ get_method(method);        // slow path can do a GC, reload RBX
 961 
 962 
 963   // result handler is in rax
 964   // set result handler
 965   __ movptr(Address(rbp,
 966                     (frame::interpreter_frame_result_handler_offset) * wordSize),
 967             rax);
 968 
 969   // pass mirror handle if static call
 970   {
 971     Label L;
 972     __ 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);
 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());
 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());
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());
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();
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 theses 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());
1072     __ jcc(Assembler::equal, push_double);
1073     __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize),
1074               double_handler.addr());
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   if (os::is_MP()) {
1092     if (UseMembar) {
1093       // Force this write out before the read below
1094       __ membar(Assembler::Membar_mask_bits(
1095            Assembler::LoadLoad | Assembler::LoadStore |
1096            Assembler::StoreLoad | Assembler::StoreStore));
1097     } else {
1098       // Write serialization page so VM thread can do a pseudo remote membar.
1099       // We use the current thread pointer to calculate a thread specific
1100       // offset to write to within the page. This minimizes bus traffic
1101       // due to cache line collision.
1102       __ serialize_memory(thread, rcx);
1103     }
1104   }
1105 
1106 #ifndef _LP64
1107   if (AlwaysRestoreFPU) {
1108     //  Make sure the control word is correct.
1109     __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
1110   }
1111 #endif // _LP64
1112 
1113   // check for safepoint operation in progress and/or pending suspend requests
1114   {
1115     Label Continue;
1116     Label slow_path;
1117 
1118 #ifndef _LP64
1119     __ safepoint_poll(slow_path, thread, noreg);
1120 #else
1121     __ safepoint_poll(slow_path, r15_thread, rscratch1);
1122 #endif
1123 
1124     __ cmpl(Address(thread, JavaThread::suspend_flags_offset()), 0);
1125     __ jcc(Assembler::equal, Continue);
1126     __ bind(slow_path);
1127 
1128     // Don't use call_VM as it will see a possible pending exception
1129     // and forward it and never return here preventing us from
1130     // clearing _last_native_pc down below.  Also can't use
1131     // call_VM_leaf either as it will check to see if r13 & r14 are
1132     // preserved and correspond to the bcp/locals pointers. So we do a
1133     // runtime call by hand.
1134     //
1135 #ifndef _LP64
1136     __ push(thread);
1137     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address,
1138                                             JavaThread::check_special_condition_for_native_trans)));
1139     __ increment(rsp, wordSize);
1140     __ get_thread(thread);
1141 #else
1142     __ mov(c_rarg0, r15_thread);
1143     __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1144     __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1145     __ andptr(rsp, -16); // align stack as required by ABI
1146     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans)));
1147     __ mov(rsp, r12); // restore sp
1148     __ reinit_heapbase();
1149 #endif // _LP64
1150     __ bind(Continue);
1151   }
1152 
1153   // change thread state
1154   __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_Java);
1155 
1156   // reset_last_Java_frame
1157   __ reset_last_Java_frame(thread, true);
1158 
1159   if (CheckJNICalls) {
1160     // clear_pending_jni_exception_check
1161     __ movptr(Address(thread, JavaThread::pending_jni_exception_check_fn_offset()), NULL_WORD);
1162   }
1163 
1164   // reset handle block
1165   __ movptr(t, Address(thread, JavaThread::active_handles_offset()));
1166   __ movl(Address(t, JNIHandleBlock::top_offset_in_bytes()), (int32_t)NULL_WORD);
1167 
1168   // If result is an oop unbox and store it in frame where gc will see it
1169   // and result handler will pick it up
1170 
1171   {
1172     Label no_oop, not_weak, store_result;
1173     __ lea(t, ExternalAddress(AbstractInterpreter::result_handler(T_OBJECT)));
1174     __ cmpptr(t, Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize));
1175     __ jcc(Assembler::notEqual, no_oop);
1176     // retrieve result
1177     __ pop(ltos);
1178     // Unbox oop result, e.g. JNIHandles::resolve value.
1179     __ resolve_jobject(rax /* value */,
1180                        thread /* thread */,
1181                        t /* tmp */);
1182     __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize), rax);
1183     // keep stack depth as expected by pushing oop which will eventually be discarded
1184     __ push(ltos);
1185     __ bind(no_oop);
1186   }
1187 
1188 
1189   {
1190     Label no_reguard;
1191     __ cmpl(Address(thread, JavaThread::stack_guard_state_offset()),
1192             JavaThread::stack_guard_yellow_reserved_disabled);
1193     __ jcc(Assembler::notEqual, no_reguard);
1194 
1195     __ pusha(); // XXX only save smashed registers
1196 #ifndef _LP64
1197     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
1198     __ popa();
1199 #else
1200     __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1201     __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1202     __ andptr(rsp, -16); // align stack as required by ABI
1203     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
1204     __ mov(rsp, r12); // restore sp
1205     __ popa(); // XXX only restore smashed registers
1206     __ reinit_heapbase();
1207 #endif // _LP64
1208 
1209     __ bind(no_reguard);
1210   }
1211 
1212 
1213   // The method register is junk from after the thread_in_native transition
1214   // until here.  Also can't call_VM until the bcp has been
1215   // restored.  Need bcp for throwing exception below so get it now.
1216   __ get_method(method);
1217 
1218   // restore to have legal interpreter frame, i.e., bci == 0 <=> code_base()
1219   __ movptr(rbcp, Address(method, Method::const_offset()));   // get ConstMethod*
1220   __ lea(rbcp, Address(rbcp, ConstMethod::codes_offset()));    // get codebase
1221 
1222   // handle exceptions (exception handling will handle unlocking!)
1223   {
1224     Label L;
1225     __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t) NULL_WORD);
1226     __ jcc(Assembler::zero, L);
1227     // Note: At some point we may want to unify this with the code
1228     // used in call_VM_base(); i.e., we should use the
1229     // StubRoutines::forward_exception code. For now this doesn't work
1230     // here because the rsp is not correctly set at this point.
1231     __ MacroAssembler::call_VM(noreg,
1232                                CAST_FROM_FN_PTR(address,
1233                                InterpreterRuntime::throw_pending_exception));
1234     __ should_not_reach_here();
1235     __ bind(L);
1236   }
1237 
1238   // do unlocking if necessary
1239   {
1240     Label L;
1241     __ movl(t, Address(method, Method::access_flags_offset()));
1242     __ testl(t, JVM_ACC_SYNCHRONIZED);
1243     __ jcc(Assembler::zero, L);
1244     // the code below should be shared with interpreter macro
1245     // assembler implementation
1246     {
1247       Label unlock;
1248       // BasicObjectLock will be first in list, since this is a
1249       // synchronized method. However, need to check that the object
1250       // has not been unlocked by an explicit monitorexit bytecode.
1251       const Address monitor(rbp,
1252                             (intptr_t)(frame::interpreter_frame_initial_sp_offset *
1253                                        wordSize - (int)sizeof(BasicObjectLock)));
1254 
1255       const Register regmon = NOT_LP64(rdx) LP64_ONLY(c_rarg1);
1256 
1257       // monitor expect in c_rarg1 for slow unlock path
1258       __ lea(regmon, monitor); // address of first monitor
1259 
1260       __ movptr(t, Address(regmon, BasicObjectLock::obj_offset_in_bytes()));
1261       __ testptr(t, t);
1262       __ jcc(Assembler::notZero, unlock);
1263 
1264       // Entry already unlocked, need to throw exception
1265       __ MacroAssembler::call_VM(noreg,
1266                                  CAST_FROM_FN_PTR(address,
1267                    InterpreterRuntime::throw_illegal_monitor_state_exception));
1268       __ should_not_reach_here();
1269 
1270       __ bind(unlock);
1271       __ unlock_object(regmon);
1272     }
1273     __ bind(L);
1274   }
1275 
1276   // jvmti support
1277   // Note: This must happen _after_ handling/throwing any exceptions since
1278   //       the exception handler code notifies the runtime of method exits
1279   //       too. If this happens before, method entry/exit notifications are
1280   //       not properly paired (was bug - gri 11/22/99).
1281   __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI);
1282 
1283   // restore potential result in edx:eax, call result handler to
1284   // restore potential result in ST0 & handle result
1285 
1286   __ pop(ltos);
1287   LP64_ONLY( __ pop(dtos));
1288 
1289   __ movptr(t, Address(rbp,
1290                        (frame::interpreter_frame_result_handler_offset) * wordSize));
1291   __ call(t);
1292 
1293   // remove activation
1294   __ movptr(t, Address(rbp,
1295                        frame::interpreter_frame_sender_sp_offset *
1296                        wordSize)); // get sender sp
1297   __ leave();                                // remove frame anchor
1298   __ pop(rdi);                               // get return address
1299   __ mov(rsp, t);                            // set sp to sender sp
1300   __ jmp(rdi);
1301 
1302   if (inc_counter) {
1303     // Handle overflow of counter and compile method
1304     __ bind(invocation_counter_overflow);
1305     generate_counter_overflow(continue_after_compile);
1306   }
1307 
1308   return entry_point;
1309 }
1310 
1311 // Abstract method entry
1312 // Attempt to execute abstract method. Throw exception
1313 address TemplateInterpreterGenerator::generate_abstract_entry(void) {
1314 
1315   address entry_point = __ pc();
1316 
1317   // abstract method entry
1318 
1319   //  pop return address, reset last_sp to NULL
1320   __ empty_expression_stack();
1321   __ restore_bcp();      // rsi must be correct for exception handler   (was destroyed)
1322   __ restore_locals();   // make sure locals pointer is correct as well (was destroyed)
1323 
1324   // throw exception
1325   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodErrorWithMethod), rbx);
1326   // the call_VM checks for exception, so we should never return here.
1327   __ should_not_reach_here();
1328 
1329   return entry_point;
1330 }
1331 
1332 //
1333 // Generic interpreted method entry to (asm) interpreter
1334 //
1335 address TemplateInterpreterGenerator::generate_normal_entry(bool synchronized) {
1336   // determine code generation flags
1337   bool inc_counter  = UseCompiler || CountCompiledCalls || LogTouchedMethods;
1338 
1339   // ebx: Method*
1340   // rbcp: sender sp
1341   address entry_point = __ pc();
1342 
1343   const Address constMethod(rbx, Method::const_offset());
1344   const Address access_flags(rbx, Method::access_flags_offset());
1345   const Address size_of_parameters(rdx,
1346                                    ConstMethod::size_of_parameters_offset());
1347   const Address size_of_locals(rdx, ConstMethod::size_of_locals_offset());
1348 
1349 
1350   // get parameter size (always needed)
1351   __ movptr(rdx, constMethod);
1352   __ load_unsigned_short(rcx, size_of_parameters);
1353 
1354   // rbx: Method*
1355   // rcx: size of parameters
1356   // rbcp: sender_sp (could differ from sp+wordSize if we were called via c2i )
1357 
1358   __ load_unsigned_short(rdx, size_of_locals); // get size of locals in words
1359   __ subl(rdx, rcx); // rdx = no. of additional locals
1360 
1361   // YYY
1362 //   __ incrementl(rdx);
1363 //   __ andl(rdx, -2);
1364 
1365   // see if we've got enough room on the stack for locals plus overhead.
1366   generate_stack_overflow_check();
1367 
1368   // get return address
1369   __ pop(rax);
1370 
1371   // compute beginning of parameters
1372   __ lea(rlocals, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize));
1373 
1374   // rdx - # of additional locals
1375   // allocate space for locals
1376   // explicitly initialize locals
1377   {
1378     Label exit, loop;
1379     __ testl(rdx, rdx);
1380     __ jcc(Assembler::lessEqual, exit); // do nothing if rdx <= 0
1381     __ bind(loop);
1382     __ push((int) NULL_WORD); // initialize local variables
1383     __ decrementl(rdx); // until everything initialized
1384     __ jcc(Assembler::greater, loop);
1385     __ bind(exit);
1386   }
1387 
1388   // initialize fixed part of activation frame
1389   generate_fixed_frame(false);
1390 
1391   // make sure method is not native & not abstract
1392 #ifdef ASSERT
1393   __ movl(rax, access_flags);
1394   {
1395     Label L;
1396     __ testl(rax, JVM_ACC_NATIVE);
1397     __ jcc(Assembler::zero, L);
1398     __ stop("tried to execute native method as non-native");
1399     __ bind(L);
1400   }
1401   {
1402     Label L;
1403     __ testl(rax, JVM_ACC_ABSTRACT);
1404     __ jcc(Assembler::zero, L);
1405     __ stop("tried to execute abstract method in interpreter");
1406     __ bind(L);
1407   }
1408 #endif
1409 
1410   // Since at this point in the method invocation the exception
1411   // handler would try to exit the monitor of synchronized methods
1412   // which hasn't been entered yet, we set the thread local variable
1413   // _do_not_unlock_if_synchronized to true. The remove_activation
1414   // will check this flag.
1415 
1416   const Register thread = NOT_LP64(rax) LP64_ONLY(r15_thread);
1417   NOT_LP64(__ get_thread(thread));
1418   const Address do_not_unlock_if_synchronized(thread,
1419         in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1420   __ movbool(do_not_unlock_if_synchronized, true);
1421 
1422   __ profile_parameters_type(rax, rcx, rdx);
1423   // increment invocation count & check for overflow
1424   Label invocation_counter_overflow;
1425   Label profile_method;
1426   Label profile_method_continue;
1427   if (inc_counter) {
1428     generate_counter_incr(&invocation_counter_overflow,
1429                           &profile_method,
1430                           &profile_method_continue);
1431     if (ProfileInterpreter) {
1432       __ bind(profile_method_continue);
1433     }
1434   }
1435 
1436   Label continue_after_compile;
1437   __ bind(continue_after_compile);
1438 
1439   // check for synchronized interpreted methods
1440   bang_stack_shadow_pages(false);
1441 
1442   // reset the _do_not_unlock_if_synchronized flag
1443   NOT_LP64(__ get_thread(thread));
1444   __ movbool(do_not_unlock_if_synchronized, false);
1445 
1446   // check for synchronized methods
1447   // Must happen AFTER invocation_counter check and stack overflow check,
1448   // so method is not locked if overflows.
1449   if (synchronized) {
1450     // Allocate monitor and lock method
1451     lock_method();
1452   } else {
1453     // no synchronization necessary
1454 #ifdef ASSERT
1455     {
1456       Label L;
1457       __ movl(rax, access_flags);
1458       __ testl(rax, JVM_ACC_SYNCHRONIZED);
1459       __ jcc(Assembler::zero, L);
1460       __ stop("method needs synchronization");
1461       __ bind(L);
1462     }
1463 #endif
1464   }
1465 
1466   // start execution
1467 #ifdef ASSERT
1468   {
1469     Label L;
1470      const Address monitor_block_top (rbp,
1471                  frame::interpreter_frame_monitor_block_top_offset * wordSize);
1472     __ movptr(rax, monitor_block_top);
1473     __ cmpptr(rax, rsp);
1474     __ jcc(Assembler::equal, L);
1475     __ stop("broken stack frame setup in interpreter");
1476     __ bind(L);
1477   }
1478 #endif
1479 
1480   // jvmti support
1481   __ notify_method_entry();
1482 
1483   __ dispatch_next(vtos);
1484 
1485   // invocation counter overflow
1486   if (inc_counter) {
1487     if (ProfileInterpreter) {
1488       // We have decided to profile this method in the interpreter
1489       __ bind(profile_method);
1490       __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1491       __ set_method_data_pointer_for_bcp();
1492       __ get_method(rbx);
1493       __ jmp(profile_method_continue);
1494     }
1495     // Handle overflow of counter and compile method
1496     __ bind(invocation_counter_overflow);
1497     generate_counter_overflow(continue_after_compile);
1498   }
1499 
1500   return entry_point;
1501 }
1502 
1503 //-----------------------------------------------------------------------------
1504 // Exceptions
1505 
1506 void TemplateInterpreterGenerator::generate_throw_exception() {
1507   // Entry point in previous activation (i.e., if the caller was
1508   // interpreted)
1509   Interpreter::_rethrow_exception_entry = __ pc();
1510   // Restore sp to interpreter_frame_last_sp even though we are going
1511   // to empty the expression stack for the exception processing.
1512   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
1513   // rax: exception
1514   // rdx: return address/pc that threw exception
1515   __ restore_bcp();    // r13/rsi points to call/send
1516   __ restore_locals();
1517   LP64_ONLY(__ reinit_heapbase());  // restore r12 as heapbase.
1518   // Entry point for exceptions thrown within interpreter code
1519   Interpreter::_throw_exception_entry = __ pc();
1520   // expression stack is undefined here
1521   // rax: exception
1522   // r13/rsi: exception bcp
1523   __ verify_oop(rax);
1524   Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
1525   LP64_ONLY(__ mov(c_rarg1, rax));
1526 
1527   // expression stack must be empty before entering the VM in case of
1528   // an exception
1529   __ empty_expression_stack();
1530   // find exception handler address and preserve exception oop
1531   __ call_VM(rdx,
1532              CAST_FROM_FN_PTR(address,
1533                           InterpreterRuntime::exception_handler_for_exception),
1534              rarg);
1535   // rax: exception handler entry point
1536   // rdx: preserved exception oop
1537   // r13/rsi: bcp for exception handler
1538   __ push_ptr(rdx); // push exception which is now the only value on the stack
1539   __ jmp(rax); // jump to exception handler (may be _remove_activation_entry!)
1540 
1541   // If the exception is not handled in the current frame the frame is
1542   // removed and the exception is rethrown (i.e. exception
1543   // continuation is _rethrow_exception).
1544   //
1545   // Note: At this point the bci is still the bxi for the instruction
1546   // which caused the exception and the expression stack is
1547   // empty. Thus, for any VM calls at this point, GC will find a legal
1548   // oop map (with empty expression stack).
1549 
1550   // In current activation
1551   // tos: exception
1552   // esi: exception bcp
1553 
1554   //
1555   // JVMTI PopFrame support
1556   //
1557 
1558   Interpreter::_remove_activation_preserving_args_entry = __ pc();
1559   __ empty_expression_stack();
1560   // Set the popframe_processing bit in pending_popframe_condition
1561   // indicating that we are currently handling popframe, so that
1562   // call_VMs that may happen later do not trigger new popframe
1563   // handling cycles.
1564   const Register thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
1565   NOT_LP64(__ get_thread(thread));
1566   __ movl(rdx, Address(thread, JavaThread::popframe_condition_offset()));
1567   __ orl(rdx, JavaThread::popframe_processing_bit);
1568   __ movl(Address(thread, JavaThread::popframe_condition_offset()), rdx);
1569 
1570   {
1571     // Check to see whether we are returning to a deoptimized frame.
1572     // (The PopFrame call ensures that the caller of the popped frame is
1573     // either interpreted or compiled and deoptimizes it if compiled.)
1574     // In this case, we can't call dispatch_next() after the frame is
1575     // popped, but instead must save the incoming arguments and restore
1576     // them after deoptimization has occurred.
1577     //
1578     // Note that we don't compare the return PC against the
1579     // deoptimization blob's unpack entry because of the presence of
1580     // adapter frames in C2.
1581     Label caller_not_deoptimized;
1582     Register rarg = NOT_LP64(rdx) LP64_ONLY(c_rarg1);
1583     __ movptr(rarg, Address(rbp, frame::return_addr_offset * wordSize));
1584     __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1585                                InterpreterRuntime::interpreter_contains), rarg);
1586     __ testl(rax, rax);
1587     __ jcc(Assembler::notZero, caller_not_deoptimized);
1588 
1589     // Compute size of arguments for saving when returning to
1590     // deoptimized caller
1591     __ get_method(rax);
1592     __ movptr(rax, Address(rax, Method::const_offset()));
1593     __ load_unsigned_short(rax, Address(rax, in_bytes(ConstMethod::
1594                                                 size_of_parameters_offset())));
1595     __ shll(rax, Interpreter::logStackElementSize);
1596     __ restore_locals();
1597     __ subptr(rlocals, rax);
1598     __ addptr(rlocals, wordSize);
1599     // Save these arguments
1600     NOT_LP64(__ get_thread(thread));
1601     __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1602                                            Deoptimization::
1603                                            popframe_preserve_args),
1604                           thread, rax, rlocals);
1605 
1606     __ remove_activation(vtos, rdx,
1607                          /* throw_monitor_exception */ false,
1608                          /* install_monitor_exception */ false,
1609                          /* notify_jvmdi */ false);
1610 
1611     // Inform deoptimization that it is responsible for restoring
1612     // these arguments
1613     NOT_LP64(__ get_thread(thread));
1614     __ movl(Address(thread, JavaThread::popframe_condition_offset()),
1615             JavaThread::popframe_force_deopt_reexecution_bit);
1616 
1617     // Continue in deoptimization handler
1618     __ jmp(rdx);
1619 
1620     __ bind(caller_not_deoptimized);
1621   }
1622 
1623   __ remove_activation(vtos, rdx, /* rdx result (retaddr) is not used */
1624                        /* throw_monitor_exception */ false,
1625                        /* install_monitor_exception */ false,
1626                        /* notify_jvmdi */ false);
1627 
1628   // Finish with popframe handling
1629   // A previous I2C followed by a deoptimization might have moved the
1630   // outgoing arguments further up the stack. PopFrame expects the
1631   // mutations to those outgoing arguments to be preserved and other
1632   // constraints basically require this frame to look exactly as
1633   // though it had previously invoked an interpreted activation with
1634   // no space between the top of the expression stack (current
1635   // last_sp) and the top of stack. Rather than force deopt to
1636   // maintain this kind of invariant all the time we call a small
1637   // fixup routine to move the mutated arguments onto the top of our
1638   // expression stack if necessary.
1639 #ifndef _LP64
1640   __ mov(rax, rsp);
1641   __ movptr(rbx, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1642   __ get_thread(thread);
1643   // PC must point into interpreter here
1644   __ set_last_Java_frame(thread, noreg, rbp, __ pc());
1645   __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), thread, rax, rbx);
1646   __ get_thread(thread);
1647 #else
1648   __ mov(c_rarg1, rsp);
1649   __ movptr(c_rarg2, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1650   // PC must point into interpreter here
1651   __ set_last_Java_frame(noreg, rbp, __ pc());
1652   __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), r15_thread, c_rarg1, c_rarg2);
1653 #endif
1654   __ reset_last_Java_frame(thread, true);
1655 
1656   // Restore the last_sp and null it out
1657   __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1658   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
1659 
1660   __ restore_bcp();
1661   __ restore_locals();
1662   // The method data pointer was incremented already during
1663   // call profiling. We have to restore the mdp for the current bcp.
1664   if (ProfileInterpreter) {
1665     __ set_method_data_pointer_for_bcp();
1666   }
1667 
1668   // Clear the popframe condition flag
1669   NOT_LP64(__ get_thread(thread));
1670   __ movl(Address(thread, JavaThread::popframe_condition_offset()),
1671           JavaThread::popframe_inactive);
1672 
1673 #if INCLUDE_JVMTI
1674   {
1675     Label L_done;
1676     const Register local0 = rlocals;
1677 
1678     __ cmpb(Address(rbcp, 0), Bytecodes::_invokestatic);
1679     __ jcc(Assembler::notEqual, L_done);
1680 
1681     // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call.
1682     // Detect such a case in the InterpreterRuntime function and return the member name argument, or NULL.
1683 
1684     __ get_method(rdx);
1685     __ movptr(rax, Address(local0, 0));
1686     __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), rax, rdx, rbcp);
1687 
1688     __ testptr(rax, rax);
1689     __ jcc(Assembler::zero, L_done);
1690 
1691     __ movptr(Address(rbx, 0), rax);
1692     __ bind(L_done);
1693   }
1694 #endif // INCLUDE_JVMTI
1695 
1696   __ dispatch_next(vtos);
1697   // end of PopFrame support
1698 
1699   Interpreter::_remove_activation_entry = __ pc();
1700 
1701   // preserve exception over this code sequence
1702   __ pop_ptr(rax);
1703   NOT_LP64(__ get_thread(thread));
1704   __ movptr(Address(thread, JavaThread::vm_result_offset()), rax);
1705   // remove the activation (without doing throws on illegalMonitorExceptions)
1706   __ remove_activation(vtos, rdx, false, true, false);
1707   // restore exception
1708   NOT_LP64(__ get_thread(thread));
1709   __ get_vm_result(rax, thread);
1710 
1711   // In between activations - previous activation type unknown yet
1712   // compute continuation point - the continuation point expects the
1713   // following registers set up:
1714   //
1715   // rax: exception
1716   // rdx: return address/pc that threw exception
1717   // rsp: expression stack of caller
1718   // rbp: ebp of caller
1719   __ push(rax);                                  // save exception
1720   __ push(rdx);                                  // save return address
1721   __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1722                           SharedRuntime::exception_handler_for_return_address),
1723                         thread, rdx);
1724   __ mov(rbx, rax);                              // save exception handler
1725   __ pop(rdx);                                   // restore return address
1726   __ pop(rax);                                   // restore exception
1727   // Note that an "issuing PC" is actually the next PC after the call
1728   __ jmp(rbx);                                   // jump to exception
1729                                                  // handler of caller
1730 }
1731 
1732 
1733 //
1734 // JVMTI ForceEarlyReturn support
1735 //
1736 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
1737   address entry = __ pc();
1738 
1739   __ restore_bcp();
1740   __ restore_locals();
1741   __ empty_expression_stack();
1742   __ load_earlyret_value(state);  // 32 bits returns value in rdx, so don't reuse
1743 
1744   const Register thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
1745   NOT_LP64(__ get_thread(thread));
1746   __ movptr(rcx, Address(thread, JavaThread::jvmti_thread_state_offset()));
1747   Address cond_addr(rcx, JvmtiThreadState::earlyret_state_offset());
1748 
1749   // Clear the earlyret state
1750   __ movl(cond_addr, JvmtiThreadState::earlyret_inactive);
1751 
1752   __ remove_activation(state, rsi,
1753                        false, /* throw_monitor_exception */
1754                        false, /* install_monitor_exception */
1755                        true); /* notify_jvmdi */
1756   __ jmp(rsi);
1757 
1758   return entry;
1759 } // end of ForceEarlyReturn support
1760 
1761 
1762 //-----------------------------------------------------------------------------
1763 // Helper for vtos entry point generation
1764 
1765 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
1766                                                          address& bep,
1767                                                          address& cep,
1768                                                          address& sep,
1769                                                          address& aep,
1770                                                          address& iep,
1771                                                          address& lep,
1772                                                          address& fep,
1773                                                          address& dep,
1774                                                          address& vep) {
1775   assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
1776   Label L;
1777   aep = __ pc();  __ push_ptr();   __ jmp(L);
1778 #ifndef _LP64
1779   fep = __ pc(); __ push(ftos); __ jmp(L);
1780   dep = __ pc(); __ push(dtos); __ jmp(L);
1781 #else
1782   fep = __ pc();  __ push_f(xmm0); __ jmp(L);
1783   dep = __ pc();  __ push_d(xmm0); __ jmp(L);
1784 #endif // _LP64
1785   lep = __ pc();  __ push_l();     __ jmp(L);
1786   bep = cep = sep =
1787   iep = __ pc();  __ push_i();
1788   vep = __ pc();
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));
1839 }
1840 
1841 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
1842   __ incrementl(ExternalAddress((address) &BytecodeHistogram::_counters[t->bytecode()]));
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);
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()) != NULL,
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   __ jcc(Assembler::notEqual, L);
1881   __ int3();
1882   __ bind(L);
1883 }
1884 #endif // !PRODUCT