1 /*
   2  * Copyright (c) 2003, 2015, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 #include "precompiled.hpp"
  26 #include "asm/macroAssembler.hpp"
  27 #include "interpreter/bytecodeHistogram.hpp"
  28 #include "interpreter/interpreter.hpp"
  29 #include "interpreter/interpreterGenerator.hpp"
  30 #include "interpreter/interpreterRuntime.hpp"
  31 #include "interpreter/interp_masm.hpp"
  32 #include "interpreter/templateTable.hpp"
  33 #include "oops/arrayOop.hpp"
  34 #include "oops/methodData.hpp"
  35 #include "oops/method.hpp"
  36 #include "oops/oop.inline.hpp"
  37 #include "prims/jvmtiExport.hpp"
  38 #include "prims/jvmtiThreadState.hpp"
  39 #include "runtime/arguments.hpp"
  40 #include "runtime/deoptimization.hpp"
  41 #include "runtime/frame.inline.hpp"
  42 #include "runtime/sharedRuntime.hpp"
  43 #include "runtime/stubRoutines.hpp"
  44 #include "runtime/synchronizer.hpp"
  45 #include "runtime/timer.hpp"
  46 #include "runtime/vframeArray.hpp"
  47 #include "utilities/debug.hpp"
  48 #include "utilities/macros.hpp"
  49 
  50 #define __ _masm->
  51 
  52 #ifndef CC_INTERP
  53 
  54 // Global Register Names
  55 static const Register rbcp     = LP64_ONLY(r13) NOT_LP64(rsi);
  56 static const Register rlocals  = LP64_ONLY(r14) NOT_LP64(rdi);
  57 
  58 const int method_offset = frame::interpreter_frame_method_offset * wordSize;
  59 const int bcp_offset    = frame::interpreter_frame_bcp_offset    * wordSize;
  60 const int locals_offset = frame::interpreter_frame_locals_offset * wordSize;
  61 
  62 //-----------------------------------------------------------------------------
  63 
  64 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() {
  65   address entry = __ pc();
  66 
  67 #ifdef ASSERT
  68   {
  69     Label L;
  70     __ lea(rax, Address(rbp,
  71                         frame::interpreter_frame_monitor_block_top_offset *
  72                         wordSize));
  73     __ cmpptr(rax, rsp); // rax = maximal rsp for current rbp (stack
  74                          // grows negative)
  75     __ jcc(Assembler::aboveEqual, L); // check if frame is complete
  76     __ stop ("interpreter frame not set up");
  77     __ bind(L);
  78   }
  79 #endif // ASSERT
  80   // Restore bcp under the assumption that the current frame is still
  81   // interpreted
  82   __ restore_bcp();
  83 
  84   // expression stack must be empty before entering the VM if an
  85   // exception happened
  86   __ empty_expression_stack();
  87   // throw exception
  88   __ call_VM(noreg,
  89              CAST_FROM_FN_PTR(address,
  90                               InterpreterRuntime::throw_StackOverflowError));
  91   return entry;
  92 }
  93 
  94 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler(
  95         const char* name) {
  96   address entry = __ pc();
  97   // expression stack must be empty before entering the VM if an
  98   // exception happened
  99   __ empty_expression_stack();
 100   // setup parameters
 101   // ??? convention: expect aberrant index in register ebx
 102   Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
 103   __ lea(rarg, ExternalAddress((address)name));
 104   __ call_VM(noreg,
 105              CAST_FROM_FN_PTR(address,
 106                               InterpreterRuntime::
 107                               throw_ArrayIndexOutOfBoundsException),
 108              rarg, rbx);
 109   return entry;
 110 }
 111 
 112 address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
 113   address entry = __ pc();
 114 
 115   // object is at TOS
 116   Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
 117   __ pop(rarg);
 118 
 119   // expression stack must be empty before entering the VM if an
 120   // exception happened
 121   __ empty_expression_stack();
 122 
 123   __ call_VM(noreg,
 124              CAST_FROM_FN_PTR(address,
 125                               InterpreterRuntime::
 126                               throw_ClassCastException),
 127              rarg);
 128   return entry;
 129 }
 130 
 131 address TemplateInterpreterGenerator::generate_exception_handler_common(
 132         const char* name, const char* message, bool pass_oop) {
 133   assert(!pass_oop || message == NULL, "either oop or message but not both");
 134   address entry = __ pc();
 135 
 136   Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
 137   Register rarg2 = NOT_LP64(rbx) LP64_ONLY(c_rarg2);
 138 
 139   if (pass_oop) {
 140     // object is at TOS
 141     __ pop(rarg2);
 142   }
 143   // expression stack must be empty before entering the VM if an
 144   // exception happened
 145   __ empty_expression_stack();
 146   // setup parameters
 147   __ lea(rarg, ExternalAddress((address)name));
 148   if (pass_oop) {
 149     __ call_VM(rax, CAST_FROM_FN_PTR(address,
 150                                      InterpreterRuntime::
 151                                      create_klass_exception),
 152                rarg, rarg2);
 153   } else {
 154     // kind of lame ExternalAddress can't take NULL because
 155     // external_word_Relocation will assert.
 156     if (message != NULL) {
 157       __ lea(rarg2, ExternalAddress((address)message));
 158     } else {
 159       __ movptr(rarg2, NULL_WORD);
 160     }
 161     __ call_VM(rax,
 162                CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception),
 163                rarg, rarg2);
 164   }
 165   // throw exception
 166   __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
 167   return entry;
 168 }
 169 
 170 
 171 address TemplateInterpreterGenerator::generate_continuation_for(TosState state) {
 172   address entry = __ pc();
 173   // NULL last_sp until next java call
 174   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
 175   __ dispatch_next(state);
 176   return entry;
 177 }
 178 
 179 
 180 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step, size_t index_size) {
 181   address entry = __ pc();
 182 
 183 #ifndef _LP64
 184 #ifdef COMPILER2
 185   // The FPU stack is clean if UseSSE >= 2 but must be cleaned in other cases
 186   if ((state == ftos && UseSSE < 1) || (state == dtos && UseSSE < 2)) {
 187     for (int i = 1; i < 8; i++) {
 188         __ ffree(i);
 189     }
 190   } else if (UseSSE < 2) {
 191     __ empty_FPU_stack();
 192   }
 193 #endif // COMPILER2
 194   if ((state == ftos && UseSSE < 1) || (state == dtos && UseSSE < 2)) {
 195     __ MacroAssembler::verify_FPU(1, "generate_return_entry_for compiled");
 196   } else {
 197     __ MacroAssembler::verify_FPU(0, "generate_return_entry_for compiled");
 198   }
 199 
 200   if (state == ftos) {
 201     __ MacroAssembler::verify_FPU(UseSSE >= 1 ? 0 : 1, "generate_return_entry_for in interpreter");
 202   } else if (state == dtos) {
 203     __ MacroAssembler::verify_FPU(UseSSE >= 2 ? 0 : 1, "generate_return_entry_for in interpreter");
 204   }
 205 #endif // _LP64
 206 
 207   // Restore stack bottom in case i2c adjusted stack
 208   __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
 209   // and NULL it as marker that esp is now tos until next java call
 210   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
 211 
 212   __ restore_bcp();
 213   __ restore_locals();
 214 
 215   if (state == atos) {
 216     Register mdp = rbx;
 217     Register tmp = rcx;
 218     __ profile_return_type(mdp, rax, tmp);
 219   }
 220 
 221   const Register cache = rbx;
 222   const Register index = rcx;
 223   __ get_cache_and_index_at_bcp(cache, index, 1, index_size);
 224 
 225   const Register flags = cache;
 226   __ movl(flags, Address(cache, index, Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()));
 227   __ andl(flags, ConstantPoolCacheEntry::parameter_size_mask);
 228   __ lea(rsp, Address(rsp, flags, Interpreter::stackElementScale()));
 229   __ dispatch_next(state, step);
 230 
 231   return entry;
 232 }
 233 
 234 
 235 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state, int step) {
 236   address entry = __ pc();
 237 
 238 #ifndef _LP64
 239   if (state == ftos) {
 240     __ MacroAssembler::verify_FPU(UseSSE >= 1 ? 0 : 1, "generate_deopt_entry_for in interpreter");
 241   } else if (state == dtos) {
 242     __ MacroAssembler::verify_FPU(UseSSE >= 2 ? 0 : 1, "generate_deopt_entry_for in interpreter");
 243   }
 244 #endif // _LP64
 245 
 246   // NULL last_sp until next java call
 247   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
 248   __ restore_bcp();
 249   __ restore_locals();
 250   const Register thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
 251   NOT_LP64(__ get_thread(thread));
 252 #if INCLUDE_JVMCI
 253   // Check if we need to take lock at entry of synchronized method.
 254   if (UseJVMCICompiler) {
 255     Label L;
 256     __ cmpb(Address(thread, JavaThread::pending_monitorenter_offset()), 0);
 257     __ jcc(Assembler::zero, L);
 258     // Clear flag.
 259     __ movb(Address(thread, JavaThread::pending_monitorenter_offset()), 0);
 260     // Satisfy calling convention for lock_method().
 261     __ get_method(rbx);
 262     // Take lock.
 263     lock_method();
 264     __ bind(L);
 265   }
 266 #endif
 267   // handle exceptions
 268   {
 269     Label L;
 270     __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t) NULL_WORD);
 271     __ jcc(Assembler::zero, L);
 272     __ call_VM(noreg,
 273                CAST_FROM_FN_PTR(address,
 274                                 InterpreterRuntime::throw_pending_exception));
 275     __ should_not_reach_here();
 276     __ bind(L);
 277   }
 278   __ dispatch_next(state, step);
 279   return entry;
 280 }
 281 
 282 address TemplateInterpreterGenerator::generate_result_handler_for(
 283         BasicType type) {
 284   address entry = __ pc();
 285   switch (type) {
 286   case T_BOOLEAN: __ c2bool(rax);            break;
 287 #ifndef _LP64
 288   case T_CHAR   : __ andptr(rax, 0xFFFF);    break;
 289 #else
 290   case T_CHAR   : __ movzwl(rax, rax);       break;
 291 #endif // _LP64
 292   case T_BYTE   : __ sign_extend_byte(rax);  break;
 293   case T_SHORT  : __ sign_extend_short(rax); break;
 294   case T_INT    : /* nothing to do */        break;
 295   case T_LONG   : /* nothing to do */        break;
 296   case T_VOID   : /* nothing to do */        break;
 297 #ifndef _LP64
 298   case T_DOUBLE :
 299   case T_FLOAT  :
 300     { const Register t = InterpreterRuntime::SignatureHandlerGenerator::temp();
 301       __ pop(t);                            // remove return address first
 302       // Must return a result for interpreter or compiler. In SSE
 303       // mode, results are returned in xmm0 and the FPU stack must
 304       // be empty.
 305       if (type == T_FLOAT && UseSSE >= 1) {
 306         // Load ST0
 307         __ fld_d(Address(rsp, 0));
 308         // Store as float and empty fpu stack
 309         __ fstp_s(Address(rsp, 0));
 310         // and reload
 311         __ movflt(xmm0, Address(rsp, 0));
 312       } else if (type == T_DOUBLE && UseSSE >= 2 ) {
 313         __ movdbl(xmm0, Address(rsp, 0));
 314       } else {
 315         // restore ST0
 316         __ fld_d(Address(rsp, 0));
 317       }
 318       // and pop the temp
 319       __ addptr(rsp, 2 * wordSize);
 320       __ push(t);                           // restore return address
 321     }
 322     break;
 323 #else
 324   case T_FLOAT  : /* nothing to do */        break;
 325   case T_DOUBLE : /* nothing to do */        break;
 326 #endif // _LP64
 327 
 328   case T_OBJECT :
 329     // retrieve result from frame
 330     __ movptr(rax, Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize));
 331     // and verify it
 332     __ verify_oop(rax);
 333     break;
 334   default       : ShouldNotReachHere();
 335   }
 336   __ ret(0);                                   // return from result handler
 337   return entry;
 338 }
 339 
 340 address TemplateInterpreterGenerator::generate_safept_entry_for(
 341         TosState state,
 342         address runtime_entry) {
 343   address entry = __ pc();
 344   __ push(state);
 345   __ call_VM(noreg, runtime_entry);
 346   __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
 347   return entry;
 348 }
 349 
 350 
 351 
 352 // Helpers for commoning out cases in the various type of method entries.
 353 //
 354 
 355 
 356 // increment invocation count & check for overflow
 357 //
 358 // Note: checking for negative value instead of overflow
 359 //       so we have a 'sticky' overflow test
 360 //
 361 // rbx: method
 362 // rcx: invocation counter
 363 //
 364 void InterpreterGenerator::generate_counter_incr(
 365         Label* overflow,
 366         Label* profile_method,
 367         Label* profile_method_continue) {
 368   Label done;
 369   // Note: In tiered we increment either counters in Method* or in MDO depending if we're profiling or not.
 370   if (TieredCompilation) {
 371     int increment = InvocationCounter::count_increment;
 372     Label no_mdo;
 373     if (ProfileInterpreter) {
 374       // Are we profiling?
 375       __ movptr(rax, Address(rbx, Method::method_data_offset()));
 376       __ testptr(rax, rax);
 377       __ jccb(Assembler::zero, no_mdo);
 378       // Increment counter in the MDO
 379       const Address mdo_invocation_counter(rax, in_bytes(MethodData::invocation_counter_offset()) +
 380                                                 in_bytes(InvocationCounter::counter_offset()));
 381       const Address mask(rax, in_bytes(MethodData::invoke_mask_offset()));
 382       __ increment_mask_and_jump(mdo_invocation_counter, increment, mask, rcx, false, Assembler::zero, overflow);
 383       __ jmp(done);
 384     }
 385     __ bind(no_mdo);
 386     // Increment counter in MethodCounters
 387     const Address invocation_counter(rax,
 388                   MethodCounters::invocation_counter_offset() +
 389                   InvocationCounter::counter_offset());
 390     __ get_method_counters(rbx, rax, done);
 391     const Address mask(rax, in_bytes(MethodCounters::invoke_mask_offset()));
 392     __ increment_mask_and_jump(invocation_counter, increment, mask, rcx,
 393                                false, Assembler::zero, overflow);
 394     __ bind(done);
 395   } else { // not TieredCompilation
 396     const Address backedge_counter(rax,
 397                   MethodCounters::backedge_counter_offset() +
 398                   InvocationCounter::counter_offset());
 399     const Address invocation_counter(rax,
 400                   MethodCounters::invocation_counter_offset() +
 401                   InvocationCounter::counter_offset());
 402 
 403     __ get_method_counters(rbx, rax, done);
 404 
 405     if (ProfileInterpreter) {
 406       __ incrementl(Address(rax,
 407               MethodCounters::interpreter_invocation_counter_offset()));
 408     }
 409     // Update standard invocation counters
 410     __ movl(rcx, invocation_counter);
 411     __ incrementl(rcx, InvocationCounter::count_increment);
 412     __ movl(invocation_counter, rcx); // save invocation count
 413 
 414     __ movl(rax, backedge_counter);   // load backedge counter
 415     __ andl(rax, InvocationCounter::count_mask_value); // mask out the status bits
 416 
 417     __ addl(rcx, rax);                // add both counters
 418 
 419     // profile_method is non-null only for interpreted method so
 420     // profile_method != NULL == !native_call
 421 
 422     if (ProfileInterpreter && profile_method != NULL) {
 423       // Test to see if we should create a method data oop
 424       __ movptr(rax, Address(rbx, Method::method_counters_offset()));
 425       __ cmp32(rcx, Address(rax, in_bytes(MethodCounters::interpreter_profile_limit_offset())));
 426       __ jcc(Assembler::less, *profile_method_continue);
 427 
 428       // if no method data exists, go to profile_method
 429       __ test_method_data_pointer(rax, *profile_method);
 430     }
 431 
 432     __ movptr(rax, Address(rbx, Method::method_counters_offset()));
 433     __ cmp32(rcx, Address(rax, in_bytes(MethodCounters::interpreter_invocation_limit_offset())));
 434     __ jcc(Assembler::aboveEqual, *overflow);
 435     __ bind(done);
 436   }
 437 }
 438 
 439 void InterpreterGenerator::generate_counter_overflow(Label* do_continue) {
 440 
 441   // Asm interpreter on entry
 442   // r14/rdi - locals
 443   // r13/rsi - bcp
 444   // rbx - method
 445   // rdx - cpool --- DOES NOT APPEAR TO BE TRUE
 446   // rbp - interpreter frame
 447 
 448   // On return (i.e. jump to entry_point) [ back to invocation of interpreter ]
 449   // Everything as it was on entry
 450   // rdx is not restored. Doesn't appear to really be set.
 451 
 452   // InterpreterRuntime::frequency_counter_overflow takes two
 453   // arguments, the first (thread) is passed by call_VM, the second
 454   // indicates if the counter overflow occurs at a backwards branch
 455   // (NULL bcp).  We pass zero for it.  The call returns the address
 456   // of the verified entry point for the method or NULL if the
 457   // compilation did not complete (either went background or bailed
 458   // out).
 459   Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
 460   __ movl(rarg, 0);
 461   __ call_VM(noreg,
 462              CAST_FROM_FN_PTR(address,
 463                               InterpreterRuntime::frequency_counter_overflow),
 464              rarg);
 465 
 466   __ movptr(rbx, Address(rbp, method_offset));   // restore Method*
 467   // Preserve invariant that r13/r14 contain bcp/locals of sender frame
 468   // and jump to the interpreted entry.
 469   __ jmp(*do_continue, relocInfo::none);
 470 }
 471 
 472 // See if we've got enough room on the stack for locals plus overhead.
 473 // The expression stack grows down incrementally, so the normal guard
 474 // page mechanism will work for that.
 475 //
 476 // NOTE: Since the additional locals are also always pushed (wasn't
 477 // obvious in generate_fixed_frame) so the guard should work for them
 478 // too.
 479 //
 480 // Args:
 481 //      rdx: number of additional locals this frame needs (what we must check)
 482 //      rbx: Method*
 483 //
 484 // Kills:
 485 //      rax
 486 void InterpreterGenerator::generate_stack_overflow_check(void) {
 487 
 488   // monitor entry size: see picture of stack in frame_x86.hpp
 489   const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
 490 
 491   // total overhead size: entry_size + (saved rbp through expr stack
 492   // bottom).  be sure to change this if you add/subtract anything
 493   // to/from the overhead area
 494   const int overhead_size =
 495     -(frame::interpreter_frame_initial_sp_offset * wordSize) + entry_size;
 496 
 497   const int page_size = os::vm_page_size();
 498 
 499   Label after_frame_check;
 500 
 501   // see if the frame is greater than one page in size. If so,
 502   // then we need to verify there is enough stack space remaining
 503   // for the additional locals.
 504   __ cmpl(rdx, (page_size - overhead_size) / Interpreter::stackElementSize);
 505   __ jcc(Assembler::belowEqual, after_frame_check);
 506 
 507   // compute rsp as if this were going to be the last frame on
 508   // the stack before the red zone
 509 
 510   Label after_frame_check_pop;
 511   const Register thread = NOT_LP64(rsi) LP64_ONLY(r15_thread);
 512 #ifndef _LP64
 513   __ push(thread);
 514   __ get_thread(thread);
 515 #endif
 516 
 517   const Address stack_base(thread, Thread::stack_base_offset());
 518   const Address stack_size(thread, Thread::stack_size_offset());
 519 
 520   // locals + overhead, in bytes
 521   __ mov(rax, rdx);
 522   __ shlptr(rax, Interpreter::logStackElementSize);  // 2 slots per parameter.
 523   __ addptr(rax, overhead_size);
 524 
 525 #ifdef ASSERT
 526   Label stack_base_okay, stack_size_okay;
 527   // verify that thread stack base is non-zero
 528   __ cmpptr(stack_base, (int32_t)NULL_WORD);
 529   __ jcc(Assembler::notEqual, stack_base_okay);
 530   __ stop("stack base is zero");
 531   __ bind(stack_base_okay);
 532   // verify that thread stack size is non-zero
 533   __ cmpptr(stack_size, 0);
 534   __ jcc(Assembler::notEqual, stack_size_okay);
 535   __ stop("stack size is zero");
 536   __ bind(stack_size_okay);
 537 #endif
 538 
 539   // Add stack base to locals and subtract stack size
 540   __ addptr(rax, stack_base);
 541   __ subptr(rax, stack_size);
 542 
 543   // Use the maximum number of pages we might bang.
 544   const int max_pages = StackShadowPages > (StackRedPages+StackYellowPages+StackReservedPages) ? StackShadowPages :
 545                         (StackRedPages+StackYellowPages+StackReservedPages);
 546 
 547   // add in the red and yellow zone sizes
 548   __ addptr(rax, max_pages * page_size);
 549 
 550   // check against the current stack bottom
 551   __ cmpptr(rsp, rax);
 552 
 553   __ jcc(Assembler::above, after_frame_check_pop);
 554   NOT_LP64(__ pop(rsi));  // get saved bcp
 555 
 556   // Restore sender's sp as SP. This is necessary if the sender's
 557   // frame is an extended compiled frame (see gen_c2i_adapter())
 558   // and safer anyway in case of JSR292 adaptations.
 559 
 560   __ pop(rax); // return address must be moved if SP is changed
 561   __ mov(rsp, rbcp);
 562   __ push(rax);
 563 
 564   // Note: the restored frame is not necessarily interpreted.
 565   // Use the shared runtime version of the StackOverflowError.
 566   assert(StubRoutines::throw_StackOverflowError_entry() != NULL, "stub not yet generated");
 567   __ jump(ExternalAddress(StubRoutines::throw_StackOverflowError_entry()));
 568   // all done with frame size check
 569   __ bind(after_frame_check_pop);
 570   NOT_LP64(__ pop(rsi));
 571 
 572   // all done with frame size check
 573   __ bind(after_frame_check);
 574 }
 575 
 576 // Allocate monitor and lock method (asm interpreter)
 577 //
 578 // Args:
 579 //      rbx: Method*
 580 //      r14/rdi: locals
 581 //
 582 // Kills:
 583 //      rax
 584 //      c_rarg0, c_rarg1, c_rarg2, c_rarg3, ...(param regs)
 585 //      rscratch1, rscratch2 (scratch regs)
 586 void TemplateInterpreterGenerator::lock_method() {
 587   // synchronize method
 588   const Address access_flags(rbx, Method::access_flags_offset());
 589   const Address monitor_block_top(
 590         rbp,
 591         frame::interpreter_frame_monitor_block_top_offset * wordSize);
 592   const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
 593 
 594 #ifdef ASSERT
 595   {
 596     Label L;
 597     __ movl(rax, access_flags);
 598     __ testl(rax, JVM_ACC_SYNCHRONIZED);
 599     __ jcc(Assembler::notZero, L);
 600     __ stop("method doesn't need synchronization");
 601     __ bind(L);
 602   }
 603 #endif // ASSERT
 604 
 605   // get synchronization object
 606   {
 607     const int mirror_offset = in_bytes(Klass::java_mirror_offset());
 608     Label done;
 609     __ movl(rax, access_flags);
 610     __ testl(rax, JVM_ACC_STATIC);
 611     // get receiver (assume this is frequent case)
 612     __ movptr(rax, Address(rlocals, Interpreter::local_offset_in_bytes(0)));
 613     __ jcc(Assembler::zero, done);
 614     __ movptr(rax, Address(rbx, Method::const_offset()));
 615     __ movptr(rax, Address(rax, ConstMethod::constants_offset()));
 616     __ movptr(rax, Address(rax,
 617                            ConstantPool::pool_holder_offset_in_bytes()));
 618     __ movptr(rax, Address(rax, mirror_offset));
 619 
 620 #ifdef ASSERT
 621     {
 622       Label L;
 623       __ testptr(rax, rax);
 624       __ jcc(Assembler::notZero, L);
 625       __ stop("synchronization object is NULL");
 626       __ bind(L);
 627     }
 628 #endif // ASSERT
 629 
 630     __ bind(done);
 631   }
 632 
 633   // add space for monitor & lock
 634   __ subptr(rsp, entry_size); // add space for a monitor entry
 635   __ movptr(monitor_block_top, rsp);  // set new monitor block top
 636   // store object
 637   __ movptr(Address(rsp, BasicObjectLock::obj_offset_in_bytes()), rax);
 638   const Register lockreg = NOT_LP64(rdx) LP64_ONLY(c_rarg1);
 639   __ movptr(lockreg, rsp); // object address
 640   __ lock_object(lockreg);
 641 }
 642 
 643 // Generate a fixed interpreter frame. This is identical setup for
 644 // interpreted methods and for native methods hence the shared code.
 645 //
 646 // Args:
 647 //      rax: return address
 648 //      rbx: Method*
 649 //      r14/rdi: pointer to locals
 650 //      r13/rsi: sender sp
 651 //      rdx: cp cache
 652 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
 653   // initialize fixed part of activation frame
 654   __ push(rax);        // save return address
 655   __ enter();          // save old & set new rbp
 656   __ push(rbcp);        // set sender sp
 657   __ push((int)NULL_WORD); // leave last_sp as null
 658   __ movptr(rbcp, Address(rbx, Method::const_offset()));      // get ConstMethod*
 659   __ lea(rbcp, Address(rbcp, ConstMethod::codes_offset())); // get codebase
 660   __ push(rbx);        // save Method*
 661   if (ProfileInterpreter) {
 662     Label method_data_continue;
 663     __ movptr(rdx, Address(rbx, in_bytes(Method::method_data_offset())));
 664     __ testptr(rdx, rdx);
 665     __ jcc(Assembler::zero, method_data_continue);
 666     __ addptr(rdx, in_bytes(MethodData::data_offset()));
 667     __ bind(method_data_continue);
 668     __ push(rdx);      // set the mdp (method data pointer)
 669   } else {
 670     __ push(0);
 671   }
 672 
 673   __ movptr(rdx, Address(rbx, Method::const_offset()));
 674   __ movptr(rdx, Address(rdx, ConstMethod::constants_offset()));
 675   __ movptr(rdx, Address(rdx, ConstantPool::cache_offset_in_bytes()));
 676   __ push(rdx); // set constant pool cache
 677   __ push(rlocals); // set locals pointer
 678   if (native_call) {
 679     __ push(0); // no bcp
 680   } else {
 681     __ push(rbcp); // set bcp
 682   }
 683   __ push(0); // reserve word for pointer to expression stack bottom
 684   __ movptr(Address(rsp, 0), rsp); // set expression stack bottom
 685 }
 686 
 687 // End of helpers
 688 
 689 // Method entry for java.lang.ref.Reference.get.
 690 address InterpreterGenerator::generate_Reference_get_entry(void) {
 691 #if INCLUDE_ALL_GCS
 692   // Code: _aload_0, _getfield, _areturn
 693   // parameter size = 1
 694   //
 695   // The code that gets generated by this routine is split into 2 parts:
 696   //    1. The "intrinsified" code for G1 (or any SATB based GC),
 697   //    2. The slow path - which is an expansion of the regular method entry.
 698   //
 699   // Notes:-
 700   // * In the G1 code we do not check whether we need to block for
 701   //   a safepoint. If G1 is enabled then we must execute the specialized
 702   //   code for Reference.get (except when the Reference object is null)
 703   //   so that we can log the value in the referent field with an SATB
 704   //   update buffer.
 705   //   If the code for the getfield template is modified so that the
 706   //   G1 pre-barrier code is executed when the current method is
 707   //   Reference.get() then going through the normal method entry
 708   //   will be fine.
 709   // * The G1 code can, however, check the receiver object (the instance
 710   //   of java.lang.Reference) and jump to the slow path if null. If the
 711   //   Reference object is null then we obviously cannot fetch the referent
 712   //   and so we don't need to call the G1 pre-barrier. Thus we can use the
 713   //   regular method entry code to generate the NPE.
 714   //
 715   // rbx: Method*
 716 
 717   // r13: senderSP must preserve for slow path, set SP to it on fast path
 718 
 719   address entry = __ pc();
 720 
 721   const int referent_offset = java_lang_ref_Reference::referent_offset;
 722   guarantee(referent_offset > 0, "referent offset not initialized");
 723 
 724   if (UseG1GC) {
 725     Label slow_path;
 726     // rbx: method
 727 
 728     // Check if local 0 != NULL
 729     // If the receiver is null then it is OK to jump to the slow path.
 730     __ movptr(rax, Address(rsp, wordSize));
 731 
 732     __ testptr(rax, rax);
 733     __ jcc(Assembler::zero, slow_path);
 734 
 735     // rax: local 0
 736     // rbx: method (but can be used as scratch now)
 737     // rdx: scratch
 738     // rdi: scratch
 739 
 740     // Preserve the sender sp in case the pre-barrier
 741     // calls the runtime
 742     NOT_LP64(__ push(rsi));
 743 
 744     // Generate the G1 pre-barrier code to log the value of
 745     // the referent field in an SATB buffer.
 746 
 747     // Load the value of the referent field.
 748     const Address field_address(rax, referent_offset);
 749     __ load_heap_oop(rax, field_address);
 750 
 751     const Register sender_sp = NOT_LP64(rsi) LP64_ONLY(r13);
 752     const Register thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
 753     NOT_LP64(__ get_thread(thread));
 754 
 755     // Generate the G1 pre-barrier code to log the value of
 756     // the referent field in an SATB buffer.
 757     __ g1_write_barrier_pre(noreg /* obj */,
 758                             rax /* pre_val */,
 759                             thread /* thread */,
 760                             rbx /* tmp */,
 761                             true /* tosca_live */,
 762                             true /* expand_call */);
 763 
 764     // _areturn
 765     NOT_LP64(__ pop(rsi));      // get sender sp
 766     __ pop(rdi);                // get return address
 767     __ mov(rsp, sender_sp);     // set sp to sender sp
 768     __ jmp(rdi);
 769     __ ret(0);
 770 
 771     // generate a vanilla interpreter entry as the slow path
 772     __ bind(slow_path);
 773     __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::zerolocals));
 774     return entry;
 775   }
 776 #endif // INCLUDE_ALL_GCS
 777 
 778   // If G1 is not enabled then attempt to go through the accessor entry point
 779   // Reference.get is an accessor
 780   return NULL;
 781 }
 782 
 783 // Interpreter stub for calling a native method. (asm interpreter)
 784 // This sets up a somewhat different looking stack for calling the
 785 // native method than the typical interpreter frame setup.
 786 address InterpreterGenerator::generate_native_entry(bool synchronized) {
 787   // determine code generation flags
 788   bool inc_counter  = UseCompiler || CountCompiledCalls || LogTouchedMethods;
 789 
 790   // rbx: Method*
 791   // rbcp: sender sp
 792 
 793   address entry_point = __ pc();
 794 
 795   const Address constMethod       (rbx, Method::const_offset());
 796   const Address access_flags      (rbx, Method::access_flags_offset());
 797   const Address size_of_parameters(rcx, ConstMethod::
 798                                         size_of_parameters_offset());
 799 
 800 
 801   // get parameter size (always needed)
 802   __ movptr(rcx, constMethod);
 803   __ load_unsigned_short(rcx, size_of_parameters);
 804 
 805   // native calls don't need the stack size check since they have no
 806   // expression stack and the arguments are already on the stack and
 807   // we only add a handful of words to the stack
 808 
 809   // rbx: Method*
 810   // rcx: size of parameters
 811   // rbcp: sender sp
 812   __ pop(rax);                                       // get return address
 813 
 814   // for natives the size of locals is zero
 815 
 816   // compute beginning of parameters
 817   __ lea(rlocals, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize));
 818 
 819   // add 2 zero-initialized slots for native calls
 820   // initialize result_handler slot
 821   __ push((int) NULL_WORD);
 822   // slot for oop temp
 823   // (static native method holder mirror/jni oop result)
 824   __ push((int) NULL_WORD);
 825 
 826   // initialize fixed part of activation frame
 827   generate_fixed_frame(true);
 828 
 829   // make sure method is native & not abstract
 830 #ifdef ASSERT
 831   __ movl(rax, access_flags);
 832   {
 833     Label L;
 834     __ testl(rax, JVM_ACC_NATIVE);
 835     __ jcc(Assembler::notZero, L);
 836     __ stop("tried to execute non-native method as native");
 837     __ bind(L);
 838   }
 839   {
 840     Label L;
 841     __ testl(rax, JVM_ACC_ABSTRACT);
 842     __ jcc(Assembler::zero, L);
 843     __ stop("tried to execute abstract method in interpreter");
 844     __ bind(L);
 845   }
 846 #endif
 847 
 848   // Since at this point in the method invocation the exception handler
 849   // would try to exit the monitor of synchronized methods which hasn't
 850   // been entered yet, we set the thread local variable
 851   // _do_not_unlock_if_synchronized to true. The remove_activation will
 852   // check this flag.
 853 
 854   const Register thread1 = NOT_LP64(rax) LP64_ONLY(r15_thread);
 855   NOT_LP64(__ get_thread(thread1));
 856   const Address do_not_unlock_if_synchronized(thread1,
 857         in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
 858   __ movbool(do_not_unlock_if_synchronized, true);
 859 
 860   // increment invocation count & check for overflow
 861   Label invocation_counter_overflow;
 862   if (inc_counter) {
 863     generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
 864   }
 865 
 866   Label continue_after_compile;
 867   __ bind(continue_after_compile);
 868 
 869   bang_stack_shadow_pages(true);
 870 
 871   // reset the _do_not_unlock_if_synchronized flag
 872   NOT_LP64(__ get_thread(thread1));
 873   __ movbool(do_not_unlock_if_synchronized, false);
 874 
 875   // check for synchronized methods
 876   // Must happen AFTER invocation_counter check and stack overflow check,
 877   // so method is not locked if overflows.
 878   if (synchronized) {
 879     lock_method();
 880   } else {
 881     // no synchronization necessary
 882 #ifdef ASSERT
 883     {
 884       Label L;
 885       __ movl(rax, access_flags);
 886       __ testl(rax, JVM_ACC_SYNCHRONIZED);
 887       __ jcc(Assembler::zero, L);
 888       __ stop("method needs synchronization");
 889       __ bind(L);
 890     }
 891 #endif
 892   }
 893 
 894   // start execution
 895 #ifdef ASSERT
 896   {
 897     Label L;
 898     const Address monitor_block_top(rbp,
 899                  frame::interpreter_frame_monitor_block_top_offset * wordSize);
 900     __ movptr(rax, monitor_block_top);
 901     __ cmpptr(rax, rsp);
 902     __ jcc(Assembler::equal, L);
 903     __ stop("broken stack frame setup in interpreter");
 904     __ bind(L);
 905   }
 906 #endif
 907 
 908   // jvmti support
 909   __ notify_method_entry();
 910 
 911   // work registers
 912   const Register method = rbx;
 913   const Register thread = NOT_LP64(rdi) LP64_ONLY(r15_thread);
 914   const Register t      = NOT_LP64(rcx) LP64_ONLY(r11);
 915 
 916   // allocate space for parameters
 917   __ get_method(method);
 918   __ movptr(t, Address(method, Method::const_offset()));
 919   __ load_unsigned_short(t, Address(t, ConstMethod::size_of_parameters_offset()));
 920 
 921 #ifndef _LP64
 922   __ shlptr(t, Interpreter::logStackElementSize);
 923   __ addptr(t, 2*wordSize);     // allocate two more slots for JNIEnv and possible mirror
 924   __ subptr(rsp, t);
 925   __ andptr(rsp, -(StackAlignmentInBytes)); // gcc needs 16 byte aligned stacks to do XMM intrinsics
 926 #else
 927   __ shll(t, Interpreter::logStackElementSize);
 928 
 929   __ subptr(rsp, t);
 930   __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
 931   __ andptr(rsp, -16); // must be 16 byte boundary (see amd64 ABI)
 932 #endif // _LP64
 933 
 934   // get signature handler
 935   {
 936     Label L;
 937     __ movptr(t, Address(method, Method::signature_handler_offset()));
 938     __ testptr(t, t);
 939     __ jcc(Assembler::notZero, L);
 940     __ call_VM(noreg,
 941                CAST_FROM_FN_PTR(address,
 942                                 InterpreterRuntime::prepare_native_call),
 943                method);
 944     __ get_method(method);
 945     __ movptr(t, Address(method, Method::signature_handler_offset()));
 946     __ bind(L);
 947   }
 948 
 949   // call signature handler
 950   assert(InterpreterRuntime::SignatureHandlerGenerator::from() == rlocals,
 951          "adjust this code");
 952   assert(InterpreterRuntime::SignatureHandlerGenerator::to() == rsp,
 953          "adjust this code");
 954   assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == NOT_LP64(t) LP64_ONLY(rscratch1),
 955          "adjust this code");
 956 
 957   // The generated handlers do not touch RBX (the method oop).
 958   // However, large signatures cannot be cached and are generated
 959   // each time here.  The slow-path generator can do a GC on return,
 960   // so we must reload it after the call.
 961   __ call(t);
 962   __ get_method(method);        // slow path can do a GC, reload RBX
 963 
 964 
 965   // result handler is in rax
 966   // set result handler
 967   __ movptr(Address(rbp,
 968                     (frame::interpreter_frame_result_handler_offset) * wordSize),
 969             rax);
 970 
 971   // pass mirror handle if static call
 972   {
 973     Label L;
 974     const int mirror_offset = in_bytes(Klass::java_mirror_offset());
 975     __ movl(t, Address(method, Method::access_flags_offset()));
 976     __ testl(t, JVM_ACC_STATIC);
 977     __ jcc(Assembler::zero, L);
 978     // get mirror
 979     __ movptr(t, Address(method, Method::const_offset()));
 980     __ movptr(t, Address(t, ConstMethod::constants_offset()));
 981     __ movptr(t, Address(t, ConstantPool::pool_holder_offset_in_bytes()));
 982     __ movptr(t, Address(t, mirror_offset));
 983     // copy mirror into activation frame
 984     __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize),
 985             t);
 986     // pass handle to mirror
 987 #ifndef _LP64
 988     __ lea(t, Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize));
 989     __ movptr(Address(rsp, wordSize), t);
 990 #else
 991     __ lea(c_rarg1,
 992            Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize));
 993 #endif // _LP64
 994     __ bind(L);
 995   }
 996 
 997   // get native function entry point
 998   {
 999     Label L;
1000     __ movptr(rax, Address(method, Method::native_function_offset()));
1001     ExternalAddress unsatisfied(SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
1002     __ cmpptr(rax, unsatisfied.addr());
1003     __ jcc(Assembler::notEqual, L);
1004     __ call_VM(noreg,
1005                CAST_FROM_FN_PTR(address,
1006                                 InterpreterRuntime::prepare_native_call),
1007                method);
1008     __ get_method(method);
1009     __ movptr(rax, Address(method, Method::native_function_offset()));
1010     __ bind(L);
1011   }
1012 
1013   // pass JNIEnv
1014 #ifndef _LP64
1015    __ get_thread(thread);
1016    __ lea(t, Address(thread, JavaThread::jni_environment_offset()));
1017    __ movptr(Address(rsp, 0), t);
1018 
1019    // set_last_Java_frame_before_call
1020    // It is enough that the pc()
1021    // points into the right code segment. It does not have to be the correct return pc.
1022    __ set_last_Java_frame(thread, noreg, rbp, __ pc());
1023 #else
1024    __ lea(c_rarg0, Address(r15_thread, JavaThread::jni_environment_offset()));
1025 
1026    // It is enough that the pc() points into the right code
1027    // segment. It does not have to be the correct return pc.
1028    __ set_last_Java_frame(rsp, rbp, (address) __ pc());
1029 #endif // _LP64
1030 
1031   // change thread state
1032 #ifdef ASSERT
1033   {
1034     Label L;
1035     __ movl(t, Address(thread, JavaThread::thread_state_offset()));
1036     __ cmpl(t, _thread_in_Java);
1037     __ jcc(Assembler::equal, L);
1038     __ stop("Wrong thread state in native stub");
1039     __ bind(L);
1040   }
1041 #endif
1042 
1043   // Change state to native
1044 
1045   __ movl(Address(thread, JavaThread::thread_state_offset()),
1046           _thread_in_native);
1047 
1048   // Call the native method.
1049   __ call(rax);
1050   // 32: result potentially in rdx:rax or ST0
1051   // 64: result potentially in rax or xmm0
1052 
1053   // Verify or restore cpu control state after JNI call
1054   __ restore_cpu_control_state_after_jni();
1055 
1056   // NOTE: The order of these pushes is known to frame::interpreter_frame_result
1057   // in order to extract the result of a method call. If the order of these
1058   // pushes change or anything else is added to the stack then the code in
1059   // interpreter_frame_result must also change.
1060 
1061 #ifndef _LP64
1062   // save potential result in ST(0) & rdx:rax
1063   // (if result handler is the T_FLOAT or T_DOUBLE handler, result must be in ST0 -
1064   // the check is necessary to avoid potential Intel FPU overflow problems by saving/restoring 'empty' FPU registers)
1065   // It is safe to do this push because state is _thread_in_native and return address will be found
1066   // via _last_native_pc and not via _last_jave_sp
1067 
1068   // NOTE: the order of theses push(es) is known to frame::interpreter_frame_result.
1069   // If the order changes or anything else is added to the stack the code in
1070   // interpreter_frame_result will have to be changed.
1071 
1072   { Label L;
1073     Label push_double;
1074     ExternalAddress float_handler(AbstractInterpreter::result_handler(T_FLOAT));
1075     ExternalAddress double_handler(AbstractInterpreter::result_handler(T_DOUBLE));
1076     __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize),
1077               float_handler.addr());
1078     __ jcc(Assembler::equal, push_double);
1079     __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize),
1080               double_handler.addr());
1081     __ jcc(Assembler::notEqual, L);
1082     __ bind(push_double);
1083     __ push_d(); // FP values are returned using the FPU, so push FPU contents (even if UseSSE > 0).
1084     __ bind(L);
1085   }
1086 #else
1087   __ push(dtos);
1088 #endif // _LP64
1089 
1090   __ push(ltos);
1091 
1092   // change thread state
1093   NOT_LP64(__ get_thread(thread));
1094   __ movl(Address(thread, JavaThread::thread_state_offset()),
1095           _thread_in_native_trans);
1096 
1097   if (os::is_MP()) {
1098     if (UseMembar) {
1099       // Force this write out before the read below
1100       __ membar(Assembler::Membar_mask_bits(
1101            Assembler::LoadLoad | Assembler::LoadStore |
1102            Assembler::StoreLoad | Assembler::StoreStore));
1103     } else {
1104       // Write serialization page so VM thread can do a pseudo remote membar.
1105       // We use the current thread pointer to calculate a thread specific
1106       // offset to write to within the page. This minimizes bus traffic
1107       // due to cache line collision.
1108       __ serialize_memory(thread, rcx);
1109     }
1110   }
1111 
1112 #ifndef _LP64
1113   if (AlwaysRestoreFPU) {
1114     //  Make sure the control word is correct.
1115     __ fldcw(ExternalAddress(StubRoutines::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     __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
1123              SafepointSynchronize::_not_synchronized);
1124 
1125     Label L;
1126     __ jcc(Assembler::notEqual, L);
1127     __ cmpl(Address(thread, JavaThread::suspend_flags_offset()), 0);
1128     __ jcc(Assembler::equal, Continue);
1129     __ bind(L);
1130 
1131     // Don't use call_VM as it will see a possible pending exception
1132     // and forward it and never return here preventing us from
1133     // clearing _last_native_pc down below.  Also can't use
1134     // call_VM_leaf either as it will check to see if r13 & r14 are
1135     // preserved and correspond to the bcp/locals pointers. So we do a
1136     // runtime call by hand.
1137     //
1138 #ifndef _LP64
1139     __ push(thread);
1140     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address,
1141                                             JavaThread::check_special_condition_for_native_trans)));
1142     __ increment(rsp, wordSize);
1143     __ get_thread(thread);
1144 #else
1145     __ mov(c_rarg0, r15_thread);
1146     __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1147     __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1148     __ andptr(rsp, -16); // align stack as required by ABI
1149     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans)));
1150     __ mov(rsp, r12); // restore sp
1151     __ reinit_heapbase();
1152 #endif // _LP64
1153     __ bind(Continue);
1154   }
1155 
1156   // change thread state
1157   __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_Java);
1158 
1159   // reset_last_Java_frame
1160   __ reset_last_Java_frame(thread, true, true);
1161 
1162   // reset handle block
1163   __ movptr(t, Address(thread, JavaThread::active_handles_offset()));
1164   __ movl(Address(t, JNIHandleBlock::top_offset_in_bytes()), (int32_t)NULL_WORD);
1165 
1166   // If result is an oop unbox and store it in frame where gc will see it
1167   // and result handler will pick it up
1168 
1169   {
1170     Label no_oop, store_result;
1171     __ lea(t, ExternalAddress(AbstractInterpreter::result_handler(T_OBJECT)));
1172     __ cmpptr(t, Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize));
1173     __ jcc(Assembler::notEqual, no_oop);
1174     // retrieve result
1175     __ pop(ltos);
1176     __ testptr(rax, rax);
1177     __ jcc(Assembler::zero, store_result);
1178     __ movptr(rax, Address(rax, 0));
1179     __ bind(store_result);
1180     __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize), rax);
1181     // keep stack depth as expected by pushing oop which will eventually be discarded
1182     __ push(ltos);
1183     __ bind(no_oop);
1184   }
1185 
1186 
1187   {
1188     Label no_reguard;
1189     __ cmpl(Address(thread, JavaThread::stack_guard_state_offset()),
1190             JavaThread::stack_guard_yellow_disabled);
1191     __ jcc(Assembler::notEqual, no_reguard);
1192 
1193     __ pusha(); // XXX only save smashed registers
1194 #ifndef _LP64
1195     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
1196     __ popa();
1197 #else
1198     __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1199     __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1200     __ andptr(rsp, -16); // align stack as required by ABI
1201     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
1202     __ mov(rsp, r12); // restore sp
1203     __ popa(); // XXX only restore smashed registers
1204     __ reinit_heapbase();
1205 #endif // _LP64
1206 
1207     __ bind(no_reguard);
1208   }
1209 
1210 
1211   // The method register is junk from after the thread_in_native transition
1212   // until here.  Also can't call_VM until the bcp has been
1213   // restored.  Need bcp for throwing exception below so get it now.
1214   __ get_method(method);
1215 
1216   // restore to have legal interpreter frame, i.e., bci == 0 <=> code_base()
1217   __ movptr(rbcp, Address(method, Method::const_offset()));   // get ConstMethod*
1218   __ lea(rbcp, Address(rbcp, ConstMethod::codes_offset()));    // get codebase
1219 
1220   // handle exceptions (exception handling will handle unlocking!)
1221   {
1222     Label L;
1223     __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t) NULL_WORD);
1224     __ jcc(Assembler::zero, L);
1225     // Note: At some point we may want to unify this with the code
1226     // used in call_VM_base(); i.e., we should use the
1227     // StubRoutines::forward_exception code. For now this doesn't work
1228     // here because the rsp is not correctly set at this point.
1229     __ MacroAssembler::call_VM(noreg,
1230                                CAST_FROM_FN_PTR(address,
1231                                InterpreterRuntime::throw_pending_exception));
1232     __ should_not_reach_here();
1233     __ bind(L);
1234   }
1235 
1236   // do unlocking if necessary
1237   {
1238     Label L;
1239     __ movl(t, Address(method, Method::access_flags_offset()));
1240     __ testl(t, JVM_ACC_SYNCHRONIZED);
1241     __ jcc(Assembler::zero, L);
1242     // the code below should be shared with interpreter macro
1243     // assembler implementation
1244     {
1245       Label unlock;
1246       // BasicObjectLock will be first in list, since this is a
1247       // synchronized method. However, need to check that the object
1248       // has not been unlocked by an explicit monitorexit bytecode.
1249       const Address monitor(rbp,
1250                             (intptr_t)(frame::interpreter_frame_initial_sp_offset *
1251                                        wordSize - (int)sizeof(BasicObjectLock)));
1252 
1253       const Register regmon = NOT_LP64(rdx) LP64_ONLY(c_rarg1);
1254 
1255       // monitor expect in c_rarg1 for slow unlock path
1256       __ lea(regmon, monitor); // address of first monitor
1257 
1258       __ movptr(t, Address(regmon, BasicObjectLock::obj_offset_in_bytes()));
1259       __ testptr(t, t);
1260       __ jcc(Assembler::notZero, unlock);
1261 
1262       // Entry already unlocked, need to throw exception
1263       __ MacroAssembler::call_VM(noreg,
1264                                  CAST_FROM_FN_PTR(address,
1265                    InterpreterRuntime::throw_illegal_monitor_state_exception));
1266       __ should_not_reach_here();
1267 
1268       __ bind(unlock);
1269       __ unlock_object(regmon);
1270     }
1271     __ bind(L);
1272   }
1273 
1274   // jvmti support
1275   // Note: This must happen _after_ handling/throwing any exceptions since
1276   //       the exception handler code notifies the runtime of method exits
1277   //       too. If this happens before, method entry/exit notifications are
1278   //       not properly paired (was bug - gri 11/22/99).
1279   __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI);
1280 
1281   // restore potential result in edx:eax, call result handler to
1282   // restore potential result in ST0 & handle result
1283 
1284   __ pop(ltos);
1285   LP64_ONLY( __ pop(dtos));
1286 
1287   __ movptr(t, Address(rbp,
1288                        (frame::interpreter_frame_result_handler_offset) * wordSize));
1289   __ call(t);
1290 
1291   // remove activation
1292   __ movptr(t, Address(rbp,
1293                        frame::interpreter_frame_sender_sp_offset *
1294                        wordSize)); // get sender sp
1295   __ leave();                                // remove frame anchor
1296   __ pop(rdi);                               // get return address
1297   __ mov(rsp, t);                            // set sp to sender sp
1298   __ jmp(rdi);
1299 
1300   if (inc_counter) {
1301     // Handle overflow of counter and compile method
1302     __ bind(invocation_counter_overflow);
1303     generate_counter_overflow(&continue_after_compile);
1304   }
1305 
1306   return entry_point;
1307 }
1308 
1309 //
1310 // Generic interpreted method entry to (asm) interpreter
1311 //
1312 address InterpreterGenerator::generate_normal_entry(bool synchronized) {
1313   // determine code generation flags
1314   bool inc_counter  = UseCompiler || CountCompiledCalls || LogTouchedMethods;
1315 
1316   // ebx: Method*
1317   // rbcp: sender sp
1318   address entry_point = __ pc();
1319 
1320   const Address constMethod(rbx, Method::const_offset());
1321   const Address access_flags(rbx, Method::access_flags_offset());
1322   const Address size_of_parameters(rdx,
1323                                    ConstMethod::size_of_parameters_offset());
1324   const Address size_of_locals(rdx, ConstMethod::size_of_locals_offset());
1325 
1326 
1327   // get parameter size (always needed)
1328   __ movptr(rdx, constMethod);
1329   __ load_unsigned_short(rcx, size_of_parameters);
1330 
1331   // rbx: Method*
1332   // rcx: size of parameters
1333   // rbcp: sender_sp (could differ from sp+wordSize if we were called via c2i )
1334 
1335   __ load_unsigned_short(rdx, size_of_locals); // get size of locals in words
1336   __ subl(rdx, rcx); // rdx = no. of additional locals
1337 
1338   // YYY
1339 //   __ incrementl(rdx);
1340 //   __ andl(rdx, -2);
1341 
1342   // see if we've got enough room on the stack for locals plus overhead.
1343   generate_stack_overflow_check();
1344 
1345   // get return address
1346   __ pop(rax);
1347 
1348   // compute beginning of parameters
1349   __ lea(rlocals, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize));
1350 
1351   // rdx - # of additional locals
1352   // allocate space for locals
1353   // explicitly initialize locals
1354   {
1355     Label exit, loop;
1356     __ testl(rdx, rdx);
1357     __ jcc(Assembler::lessEqual, exit); // do nothing if rdx <= 0
1358     __ bind(loop);
1359     __ push((int) NULL_WORD); // initialize local variables
1360     __ decrementl(rdx); // until everything initialized
1361     __ jcc(Assembler::greater, loop);
1362     __ bind(exit);
1363   }
1364 
1365   // initialize fixed part of activation frame
1366   generate_fixed_frame(false);
1367 
1368   // make sure method is not native & not abstract
1369 #ifdef ASSERT
1370   __ movl(rax, access_flags);
1371   {
1372     Label L;
1373     __ testl(rax, JVM_ACC_NATIVE);
1374     __ jcc(Assembler::zero, L);
1375     __ stop("tried to execute native method as non-native");
1376     __ bind(L);
1377   }
1378   {
1379     Label L;
1380     __ testl(rax, JVM_ACC_ABSTRACT);
1381     __ jcc(Assembler::zero, L);
1382     __ stop("tried to execute abstract method in interpreter");
1383     __ bind(L);
1384   }
1385 #endif
1386 
1387   // Since at this point in the method invocation the exception
1388   // handler would try to exit the monitor of synchronized methods
1389   // which hasn't been entered yet, we set the thread local variable
1390   // _do_not_unlock_if_synchronized to true. The remove_activation
1391   // will check this flag.
1392 
1393   const Register thread = NOT_LP64(rax) LP64_ONLY(r15_thread);
1394   NOT_LP64(__ get_thread(thread));
1395   const Address do_not_unlock_if_synchronized(thread,
1396         in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1397   __ movbool(do_not_unlock_if_synchronized, true);
1398 
1399   __ profile_parameters_type(rax, rcx, rdx);
1400   // increment invocation count & check for overflow
1401   Label invocation_counter_overflow;
1402   Label profile_method;
1403   Label profile_method_continue;
1404   if (inc_counter) {
1405     generate_counter_incr(&invocation_counter_overflow,
1406                           &profile_method,
1407                           &profile_method_continue);
1408     if (ProfileInterpreter) {
1409       __ bind(profile_method_continue);
1410     }
1411   }
1412 
1413   Label continue_after_compile;
1414   __ bind(continue_after_compile);
1415 
1416   // check for synchronized interpreted methods
1417   bang_stack_shadow_pages(false);
1418 
1419   // reset the _do_not_unlock_if_synchronized flag
1420   NOT_LP64(__ get_thread(thread));
1421   __ movbool(do_not_unlock_if_synchronized, false);
1422 
1423   // check for synchronized methods
1424   // Must happen AFTER invocation_counter check and stack overflow check,
1425   // so method is not locked if overflows.
1426   if (synchronized) {
1427     // Allocate monitor and lock method
1428     lock_method();
1429   } else {
1430     // no synchronization necessary
1431 #ifdef ASSERT
1432     {
1433       Label L;
1434       __ movl(rax, access_flags);
1435       __ testl(rax, JVM_ACC_SYNCHRONIZED);
1436       __ jcc(Assembler::zero, L);
1437       __ stop("method needs synchronization");
1438       __ bind(L);
1439     }
1440 #endif
1441   }
1442 
1443   // start execution
1444 #ifdef ASSERT
1445   {
1446     Label L;
1447      const Address monitor_block_top (rbp,
1448                  frame::interpreter_frame_monitor_block_top_offset * wordSize);
1449     __ movptr(rax, monitor_block_top);
1450     __ cmpptr(rax, rsp);
1451     __ jcc(Assembler::equal, L);
1452     __ stop("broken stack frame setup in interpreter");
1453     __ bind(L);
1454   }
1455 #endif
1456 
1457   // jvmti support
1458   __ notify_method_entry();
1459 
1460   __ dispatch_next(vtos);
1461 
1462   // invocation counter overflow
1463   if (inc_counter) {
1464     if (ProfileInterpreter) {
1465       // We have decided to profile this method in the interpreter
1466       __ bind(profile_method);
1467       __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1468       __ set_method_data_pointer_for_bcp();
1469       __ get_method(rbx);
1470       __ jmp(profile_method_continue);
1471     }
1472     // Handle overflow of counter and compile method
1473     __ bind(invocation_counter_overflow);
1474     generate_counter_overflow(&continue_after_compile);
1475   }
1476 
1477   return entry_point;
1478 }
1479 
1480 //-----------------------------------------------------------------------------
1481 // Exceptions
1482 
1483 void TemplateInterpreterGenerator::generate_throw_exception() {
1484   // Entry point in previous activation (i.e., if the caller was
1485   // interpreted)
1486   Interpreter::_rethrow_exception_entry = __ pc();
1487   // Restore sp to interpreter_frame_last_sp even though we are going
1488   // to empty the expression stack for the exception processing.
1489   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
1490   // rax: exception
1491   // rdx: return address/pc that threw exception
1492   __ restore_bcp();    // r13/rsi points to call/send
1493   __ restore_locals();
1494   LP64_ONLY(__ reinit_heapbase());  // restore r12 as heapbase.
1495   // Entry point for exceptions thrown within interpreter code
1496   Interpreter::_throw_exception_entry = __ pc();
1497   // expression stack is undefined here
1498   // rax: exception
1499   // r13/rsi: exception bcp
1500   __ verify_oop(rax);
1501   Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
1502   LP64_ONLY(__ mov(c_rarg1, rax));
1503 
1504   // expression stack must be empty before entering the VM in case of
1505   // an exception
1506   __ empty_expression_stack();
1507   // find exception handler address and preserve exception oop
1508   __ call_VM(rdx,
1509              CAST_FROM_FN_PTR(address,
1510                           InterpreterRuntime::exception_handler_for_exception),
1511              rarg);
1512   // rax: exception handler entry point
1513   // rdx: preserved exception oop
1514   // r13/rsi: bcp for exception handler
1515   __ push_ptr(rdx); // push exception which is now the only value on the stack
1516   __ jmp(rax); // jump to exception handler (may be _remove_activation_entry!)
1517 
1518   // If the exception is not handled in the current frame the frame is
1519   // removed and the exception is rethrown (i.e. exception
1520   // continuation is _rethrow_exception).
1521   //
1522   // Note: At this point the bci is still the bxi for the instruction
1523   // which caused the exception and the expression stack is
1524   // empty. Thus, for any VM calls at this point, GC will find a legal
1525   // oop map (with empty expression stack).
1526 
1527   // In current activation
1528   // tos: exception
1529   // esi: exception bcp
1530 
1531   //
1532   // JVMTI PopFrame support
1533   //
1534 
1535   Interpreter::_remove_activation_preserving_args_entry = __ pc();
1536   __ empty_expression_stack();
1537   // Set the popframe_processing bit in pending_popframe_condition
1538   // indicating that we are currently handling popframe, so that
1539   // call_VMs that may happen later do not trigger new popframe
1540   // handling cycles.
1541   const Register thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
1542   NOT_LP64(__ get_thread(thread));
1543   __ movl(rdx, Address(thread, JavaThread::popframe_condition_offset()));
1544   __ orl(rdx, JavaThread::popframe_processing_bit);
1545   __ movl(Address(thread, JavaThread::popframe_condition_offset()), rdx);
1546 
1547   {
1548     // Check to see whether we are returning to a deoptimized frame.
1549     // (The PopFrame call ensures that the caller of the popped frame is
1550     // either interpreted or compiled and deoptimizes it if compiled.)
1551     // In this case, we can't call dispatch_next() after the frame is
1552     // popped, but instead must save the incoming arguments and restore
1553     // them after deoptimization has occurred.
1554     //
1555     // Note that we don't compare the return PC against the
1556     // deoptimization blob's unpack entry because of the presence of
1557     // adapter frames in C2.
1558     Label caller_not_deoptimized;
1559     Register rarg = NOT_LP64(rdx) LP64_ONLY(c_rarg1);
1560     __ movptr(rarg, Address(rbp, frame::return_addr_offset * wordSize));
1561     __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1562                                InterpreterRuntime::interpreter_contains), rarg);
1563     __ testl(rax, rax);
1564     __ jcc(Assembler::notZero, caller_not_deoptimized);
1565 
1566     // Compute size of arguments for saving when returning to
1567     // deoptimized caller
1568     __ get_method(rax);
1569     __ movptr(rax, Address(rax, Method::const_offset()));
1570     __ load_unsigned_short(rax, Address(rax, in_bytes(ConstMethod::
1571                                                 size_of_parameters_offset())));
1572     __ shll(rax, Interpreter::logStackElementSize);
1573     __ restore_locals();
1574     __ subptr(rlocals, rax);
1575     __ addptr(rlocals, wordSize);
1576     // Save these arguments
1577     NOT_LP64(__ get_thread(thread));
1578     __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1579                                            Deoptimization::
1580                                            popframe_preserve_args),
1581                           thread, rax, rlocals);
1582 
1583     __ remove_activation(vtos, rdx,
1584                          /* throw_monitor_exception */ false,
1585                          /* install_monitor_exception */ false,
1586                          /* notify_jvmdi */ false);
1587 
1588     // Inform deoptimization that it is responsible for restoring
1589     // these arguments
1590     NOT_LP64(__ get_thread(thread));
1591     __ movl(Address(thread, JavaThread::popframe_condition_offset()),
1592             JavaThread::popframe_force_deopt_reexecution_bit);
1593 
1594     // Continue in deoptimization handler
1595     __ jmp(rdx);
1596 
1597     __ bind(caller_not_deoptimized);
1598   }
1599 
1600   __ remove_activation(vtos, rdx, /* rdx result (retaddr) is not used */
1601                        /* throw_monitor_exception */ false,
1602                        /* install_monitor_exception */ false,
1603                        /* notify_jvmdi */ false);
1604 
1605   // Finish with popframe handling
1606   // A previous I2C followed by a deoptimization might have moved the
1607   // outgoing arguments further up the stack. PopFrame expects the
1608   // mutations to those outgoing arguments to be preserved and other
1609   // constraints basically require this frame to look exactly as
1610   // though it had previously invoked an interpreted activation with
1611   // no space between the top of the expression stack (current
1612   // last_sp) and the top of stack. Rather than force deopt to
1613   // maintain this kind of invariant all the time we call a small
1614   // fixup routine to move the mutated arguments onto the top of our
1615   // expression stack if necessary.
1616 #ifndef _LP64
1617   __ mov(rax, rsp);
1618   __ movptr(rbx, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1619   __ get_thread(thread);
1620   // PC must point into interpreter here
1621   __ set_last_Java_frame(thread, noreg, rbp, __ pc());
1622   __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), thread, rax, rbx);
1623   __ get_thread(thread);
1624 #else
1625   __ mov(c_rarg1, rsp);
1626   __ movptr(c_rarg2, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1627   // PC must point into interpreter here
1628   __ set_last_Java_frame(noreg, rbp, __ pc());
1629   __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), r15_thread, c_rarg1, c_rarg2);
1630 #endif
1631   __ reset_last_Java_frame(thread, true, true);
1632 
1633   // Restore the last_sp and null it out
1634   __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1635   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
1636 
1637   __ restore_bcp();
1638   __ restore_locals();
1639   // The method data pointer was incremented already during
1640   // call profiling. We have to restore the mdp for the current bcp.
1641   if (ProfileInterpreter) {
1642     __ set_method_data_pointer_for_bcp();
1643   }
1644 
1645   // Clear the popframe condition flag
1646   NOT_LP64(__ get_thread(thread));
1647   __ movl(Address(thread, JavaThread::popframe_condition_offset()),
1648           JavaThread::popframe_inactive);
1649 
1650 #if INCLUDE_JVMTI
1651   {
1652     Label L_done;
1653     const Register local0 = rlocals;
1654 
1655     __ cmpb(Address(rbcp, 0), Bytecodes::_invokestatic);
1656     __ jcc(Assembler::notEqual, L_done);
1657 
1658     // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call.
1659     // Detect such a case in the InterpreterRuntime function and return the member name argument, or NULL.
1660 
1661     __ get_method(rdx);
1662     __ movptr(rax, Address(local0, 0));
1663     __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), rax, rdx, rbcp);
1664 
1665     __ testptr(rax, rax);
1666     __ jcc(Assembler::zero, L_done);
1667 
1668     __ movptr(Address(rbx, 0), rax);
1669     __ bind(L_done);
1670   }
1671 #endif // INCLUDE_JVMTI
1672 
1673   __ dispatch_next(vtos);
1674   // end of PopFrame support
1675 
1676   Interpreter::_remove_activation_entry = __ pc();
1677 
1678   // preserve exception over this code sequence
1679   __ pop_ptr(rax);
1680   NOT_LP64(__ get_thread(thread));
1681   __ movptr(Address(thread, JavaThread::vm_result_offset()), rax);
1682   // remove the activation (without doing throws on illegalMonitorExceptions)
1683   __ remove_activation(vtos, rdx, false, true, false);
1684   // restore exception
1685   NOT_LP64(__ get_thread(thread));
1686   __ get_vm_result(rax, thread);
1687 
1688   // In between activations - previous activation type unknown yet
1689   // compute continuation point - the continuation point expects the
1690   // following registers set up:
1691   //
1692   // rax: exception
1693   // rdx: return address/pc that threw exception
1694   // rsp: expression stack of caller
1695   // rbp: ebp of caller
1696   __ push(rax);                                  // save exception
1697   __ push(rdx);                                  // save return address
1698   __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1699                           SharedRuntime::exception_handler_for_return_address),
1700                         thread, rdx);
1701   __ mov(rbx, rax);                              // save exception handler
1702   __ pop(rdx);                                   // restore return address
1703   __ pop(rax);                                   // restore exception
1704   // Note that an "issuing PC" is actually the next PC after the call
1705   __ jmp(rbx);                                   // jump to exception
1706                                                  // handler of caller
1707 }
1708 
1709 
1710 //
1711 // JVMTI ForceEarlyReturn support
1712 //
1713 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
1714   address entry = __ pc();
1715 
1716   __ restore_bcp();
1717   __ restore_locals();
1718   __ empty_expression_stack();
1719   __ load_earlyret_value(state);  // 32 bits returns value in rdx, so don't reuse
1720 
1721   const Register thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
1722   NOT_LP64(__ get_thread(thread));
1723   __ movptr(rcx, Address(thread, JavaThread::jvmti_thread_state_offset()));
1724   Address cond_addr(rcx, JvmtiThreadState::earlyret_state_offset());
1725 
1726   // Clear the earlyret state
1727   __ movl(cond_addr, JvmtiThreadState::earlyret_inactive);
1728 
1729   __ remove_activation(state, rsi,
1730                        false, /* throw_monitor_exception */
1731                        false, /* install_monitor_exception */
1732                        true); /* notify_jvmdi */
1733   __ jmp(rsi);
1734 
1735   return entry;
1736 } // end of ForceEarlyReturn support
1737 
1738 
1739 //-----------------------------------------------------------------------------
1740 // Helper for vtos entry point generation
1741 
1742 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
1743                                                          address& bep,
1744                                                          address& cep,
1745                                                          address& sep,
1746                                                          address& aep,
1747                                                          address& iep,
1748                                                          address& lep,
1749                                                          address& fep,
1750                                                          address& dep,
1751                                                          address& vep) {
1752   assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
1753   Label L;
1754   aep = __ pc();  __ push_ptr();   __ jmp(L);
1755 #ifndef _LP64
1756   fep = __ pc(); __ push(ftos); __ jmp(L);
1757   dep = __ pc(); __ push(dtos); __ jmp(L);
1758 #else
1759   fep = __ pc();  __ push_f(xmm0); __ jmp(L);
1760   dep = __ pc();  __ push_d(xmm0); __ jmp(L);
1761 #endif // _LP64
1762   lep = __ pc();  __ push_l();     __ jmp(L);
1763   bep = cep = sep =
1764   iep = __ pc();  __ push_i();
1765   vep = __ pc();
1766   __ bind(L);
1767   generate_and_dispatch(t);
1768 }
1769 
1770 
1771 //-----------------------------------------------------------------------------
1772 // Generation of individual instructions
1773 
1774 // helpers for generate_and_dispatch
1775 
1776 
1777 InterpreterGenerator::InterpreterGenerator(StubQueue* code)
1778   : TemplateInterpreterGenerator(code) {
1779    generate_all(); // down here so it can be "virtual"
1780 }
1781 
1782 //-----------------------------------------------------------------------------
1783 
1784 // Non-product code
1785 #ifndef PRODUCT
1786 
1787 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
1788   address entry = __ pc();
1789 
1790 #ifndef _LP64
1791   // prepare expression stack
1792   __ pop(rcx);          // pop return address so expression stack is 'pure'
1793   __ push(state);       // save tosca
1794 
1795   // pass tosca registers as arguments & call tracer
1796   __ call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::trace_bytecode), rcx, rax, rdx);
1797   __ mov(rcx, rax);     // make sure return address is not destroyed by pop(state)
1798   __ pop(state);        // restore tosca
1799 
1800   // return
1801   __ jmp(rcx);
1802 #else
1803   __ push(state);
1804   __ push(c_rarg0);
1805   __ push(c_rarg1);
1806   __ push(c_rarg2);
1807   __ push(c_rarg3);
1808   __ mov(c_rarg2, rax);  // Pass itos
1809 #ifdef _WIN64
1810   __ movflt(xmm3, xmm0); // Pass ftos
1811 #endif
1812   __ call_VM(noreg,
1813              CAST_FROM_FN_PTR(address, SharedRuntime::trace_bytecode),
1814              c_rarg1, c_rarg2, c_rarg3);
1815   __ pop(c_rarg3);
1816   __ pop(c_rarg2);
1817   __ pop(c_rarg1);
1818   __ pop(c_rarg0);
1819   __ pop(state);
1820   __ ret(0);                                   // return from result handler
1821 #endif // _LP64
1822 
1823   return entry;
1824 }
1825 
1826 void TemplateInterpreterGenerator::count_bytecode() {
1827   __ incrementl(ExternalAddress((address) &BytecodeCounter::_counter_value));
1828 }
1829 
1830 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
1831   __ incrementl(ExternalAddress((address) &BytecodeHistogram::_counters[t->bytecode()]));
1832 }
1833 
1834 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
1835   __ mov32(rbx, ExternalAddress((address) &BytecodePairHistogram::_index));
1836   __ shrl(rbx, BytecodePairHistogram::log2_number_of_codes);
1837   __ orl(rbx,
1838          ((int) t->bytecode()) <<
1839          BytecodePairHistogram::log2_number_of_codes);
1840   __ mov32(ExternalAddress((address) &BytecodePairHistogram::_index), rbx);
1841   __ lea(rscratch1, ExternalAddress((address) BytecodePairHistogram::_counters));
1842   __ incrementl(Address(rscratch1, rbx, Address::times_4));
1843 }
1844 
1845 
1846 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
1847   // Call a little run-time stub to avoid blow-up for each bytecode.
1848   // The run-time runtime saves the right registers, depending on
1849   // the tosca in-state for the given template.
1850 
1851   assert(Interpreter::trace_code(t->tos_in()) != NULL,
1852          "entry must have been generated");
1853 #ifndef _LP64
1854   __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in())));
1855 #else
1856   __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1857   __ andptr(rsp, -16); // align stack as required by ABI
1858   __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in())));
1859   __ mov(rsp, r12); // restore sp
1860   __ reinit_heapbase();
1861 #endif // _LP64
1862 }
1863 
1864 
1865 void TemplateInterpreterGenerator::stop_interpreter_at() {
1866   Label L;
1867   __ cmp32(ExternalAddress((address) &BytecodeCounter::_counter_value),
1868            StopInterpreterAt);
1869   __ jcc(Assembler::notEqual, L);
1870   __ int3();
1871   __ bind(L);
1872 }
1873 #endif // !PRODUCT
1874 #endif // ! CC_INTERP