1 /*
   2  * Copyright (c) 2003, 2010, 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/assembler.hpp"
  27 #include "interpreter/bytecodeHistogram.hpp"
  28 #include "interpreter/interpreter.hpp"
  29 #include "interpreter/interpreterGenerator.hpp"
  30 #include "interpreter/interpreterRuntime.hpp"
  31 #include "interpreter/templateTable.hpp"
  32 #include "oops/arrayOop.hpp"
  33 #include "oops/methodDataOop.hpp"
  34 #include "oops/methodOop.hpp"
  35 #include "oops/oop.inline.hpp"
  36 #include "prims/jvmtiExport.hpp"
  37 #include "prims/jvmtiThreadState.hpp"
  38 #include "runtime/arguments.hpp"
  39 #include "runtime/deoptimization.hpp"
  40 #include "runtime/frame.inline.hpp"
  41 #include "runtime/sharedRuntime.hpp"
  42 #include "runtime/stubRoutines.hpp"
  43 #include "runtime/synchronizer.hpp"
  44 #include "runtime/timer.hpp"
  45 #include "runtime/vframeArray.hpp"
  46 #include "utilities/debug.hpp"
  47 
  48 #define __ _masm->
  49 
  50 #ifndef CC_INTERP
  51 
  52 const int method_offset = frame::interpreter_frame_method_offset * wordSize;
  53 const int bci_offset    = frame::interpreter_frame_bcx_offset    * wordSize;
  54 const int locals_offset = frame::interpreter_frame_locals_offset * wordSize;
  55 
  56 //-----------------------------------------------------------------------------
  57 
  58 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() {
  59   address entry = __ pc();
  60 
  61 #ifdef ASSERT
  62   {
  63     Label L;
  64     __ lea(rax, Address(rbp,
  65                         frame::interpreter_frame_monitor_block_top_offset *
  66                         wordSize));
  67     __ cmpptr(rax, rsp); // rax = maximal rsp for current rbp (stack
  68                          // grows negative)
  69     __ jcc(Assembler::aboveEqual, L); // check if frame is complete
  70     __ stop ("interpreter frame not set up");
  71     __ bind(L);
  72   }
  73 #endif // ASSERT
  74   // Restore bcp under the assumption that the current frame is still
  75   // interpreted
  76   __ restore_bcp();
  77 
  78   // expression stack must be empty before entering the VM if an
  79   // exception happened
  80   __ empty_expression_stack();
  81   // throw exception
  82   __ call_VM(noreg,
  83              CAST_FROM_FN_PTR(address,
  84                               InterpreterRuntime::throw_StackOverflowError));
  85   return entry;
  86 }
  87 
  88 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler(
  89         const char* name) {
  90   address entry = __ pc();
  91   // expression stack must be empty before entering the VM if an
  92   // exception happened
  93   __ empty_expression_stack();
  94   // setup parameters
  95   // ??? convention: expect aberrant index in register ebx
  96   __ lea(c_rarg1, ExternalAddress((address)name));
  97   __ call_VM(noreg,
  98              CAST_FROM_FN_PTR(address,
  99                               InterpreterRuntime::
 100                               throw_ArrayIndexOutOfBoundsException),
 101              c_rarg1, rbx);
 102   return entry;
 103 }
 104 
 105 address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
 106   address entry = __ pc();
 107 
 108   // object is at TOS
 109   __ pop(c_rarg1);
 110 
 111   // expression stack must be empty before entering the VM if an
 112   // exception happened
 113   __ empty_expression_stack();
 114 
 115   __ call_VM(noreg,
 116              CAST_FROM_FN_PTR(address,
 117                               InterpreterRuntime::
 118                               throw_ClassCastException),
 119              c_rarg1);
 120   return entry;
 121 }
 122 
 123 // Arguments are: required type at TOS+8, failing object (or NULL) at TOS+4.
 124 address TemplateInterpreterGenerator::generate_WrongMethodType_handler() {
 125   address entry = __ pc();
 126 
 127   __ pop(c_rarg2);              // failing object is at TOS
 128   __ pop(c_rarg1);              // required type is at TOS+8
 129 
 130   __ verify_oop(c_rarg1);
 131   __ verify_oop(c_rarg2);
 132 
 133   // Various method handle types use interpreter registers as temps.
 134   __ restore_bcp();
 135   __ restore_locals();
 136 
 137   // Expression stack must be empty before entering the VM for an exception.
 138   __ empty_expression_stack();
 139 
 140   __ call_VM(noreg,
 141              CAST_FROM_FN_PTR(address,
 142                               InterpreterRuntime::throw_WrongMethodTypeException),
 143              // pass required type, failing object (or NULL)
 144              c_rarg1, c_rarg2);
 145   return entry;
 146 }
 147 
 148 address TemplateInterpreterGenerator::generate_exception_handler_common(
 149         const char* name, const char* message, bool pass_oop) {
 150   assert(!pass_oop || message == NULL, "either oop or message but not both");
 151   address entry = __ pc();
 152   if (pass_oop) {
 153     // object is at TOS
 154     __ pop(c_rarg2);
 155   }
 156   // expression stack must be empty before entering the VM if an
 157   // exception happened
 158   __ empty_expression_stack();
 159   // setup parameters
 160   __ lea(c_rarg1, ExternalAddress((address)name));
 161   if (pass_oop) {
 162     __ call_VM(rax, CAST_FROM_FN_PTR(address,
 163                                      InterpreterRuntime::
 164                                      create_klass_exception),
 165                c_rarg1, c_rarg2);
 166   } else {
 167     // kind of lame ExternalAddress can't take NULL because
 168     // external_word_Relocation will assert.
 169     if (message != NULL) {
 170       __ lea(c_rarg2, ExternalAddress((address)message));
 171     } else {
 172       __ movptr(c_rarg2, NULL_WORD);
 173     }
 174     __ call_VM(rax,
 175                CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception),
 176                c_rarg1, c_rarg2);
 177   }
 178   // throw exception
 179   __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
 180   return entry;
 181 }
 182 
 183 
 184 address TemplateInterpreterGenerator::generate_continuation_for(TosState state) {
 185   address entry = __ pc();
 186   // NULL last_sp until next java call
 187   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
 188   __ dispatch_next(state);
 189   return entry;
 190 }
 191 
 192 
 193 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state,
 194                                                                 int step) {
 195 
 196   // amd64 doesn't need to do anything special about compiled returns
 197   // to the interpreter so the code that exists on x86 to place a sentinel
 198   // here and the specialized cleanup code is not needed here.
 199 
 200   address entry = __ pc();
 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   Label L_got_cache, L_giant_index;
 211   if (EnableInvokeDynamic) {
 212     __ cmpb(Address(r13, 0), Bytecodes::_invokedynamic);
 213     __ jcc(Assembler::equal, L_giant_index);
 214   }
 215   __ get_cache_and_index_at_bcp(rbx, rcx, 1, sizeof(u2));
 216   __ bind(L_got_cache);
 217   __ movl(rbx, Address(rbx, rcx,
 218                        Address::times_ptr,
 219                        in_bytes(constantPoolCacheOopDesc::base_offset()) +
 220                        3 * wordSize));
 221   __ andl(rbx, 0xFF);
 222   __ lea(rsp, Address(rsp, rbx, Address::times_8));
 223   __ dispatch_next(state, step);
 224 
 225   // out of the main line of code...
 226   if (EnableInvokeDynamic) {
 227     __ bind(L_giant_index);
 228     __ get_cache_and_index_at_bcp(rbx, rcx, 1, sizeof(u4));
 229     __ jmp(L_got_cache);
 230   }
 231 
 232   return entry;
 233 }
 234 
 235 
 236 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state,
 237                                                                int step) {
 238   address entry = __ pc();
 239   // NULL last_sp until next java call
 240   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
 241   __ restore_bcp();
 242   __ restore_locals();
 243   // handle exceptions
 244   {
 245     Label L;
 246     __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t) NULL_WORD);
 247     __ jcc(Assembler::zero, L);
 248     __ call_VM(noreg,
 249                CAST_FROM_FN_PTR(address,
 250                                 InterpreterRuntime::throw_pending_exception));
 251     __ should_not_reach_here();
 252     __ bind(L);
 253   }
 254   __ dispatch_next(state, step);
 255   return entry;
 256 }
 257 
 258 int AbstractInterpreter::BasicType_as_index(BasicType type) {
 259   int i = 0;
 260   switch (type) {
 261     case T_BOOLEAN: i = 0; break;
 262     case T_CHAR   : i = 1; break;
 263     case T_BYTE   : i = 2; break;
 264     case T_SHORT  : i = 3; break;
 265     case T_INT    : i = 4; break;
 266     case T_LONG   : i = 5; break;
 267     case T_VOID   : i = 6; break;
 268     case T_FLOAT  : i = 7; break;
 269     case T_DOUBLE : i = 8; break;
 270     case T_OBJECT : i = 9; break;
 271     case T_ARRAY  : i = 9; break;
 272     default       : ShouldNotReachHere();
 273   }
 274   assert(0 <= i && i < AbstractInterpreter::number_of_result_handlers,
 275          "index out of bounds");
 276   return i;
 277 }
 278 
 279 
 280 address TemplateInterpreterGenerator::generate_result_handler_for(
 281         BasicType type) {
 282   address entry = __ pc();
 283   switch (type) {
 284   case T_BOOLEAN: __ c2bool(rax);            break;
 285   case T_CHAR   : __ movzwl(rax, rax);       break;
 286   case T_BYTE   : __ sign_extend_byte(rax);  break;
 287   case T_SHORT  : __ sign_extend_short(rax); break;
 288   case T_INT    : /* nothing to do */        break;
 289   case T_LONG   : /* nothing to do */        break;
 290   case T_VOID   : /* nothing to do */        break;
 291   case T_FLOAT  : /* nothing to do */        break;
 292   case T_DOUBLE : /* nothing to do */        break;
 293   case T_OBJECT :
 294     // retrieve result from frame
 295     __ movptr(rax, Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize));
 296     // and verify it
 297     __ verify_oop(rax);
 298     break;
 299   default       : ShouldNotReachHere();
 300   }
 301   __ ret(0);                                   // return from result handler
 302   return entry;
 303 }
 304 
 305 address TemplateInterpreterGenerator::generate_safept_entry_for(
 306         TosState state,
 307         address runtime_entry) {
 308   address entry = __ pc();
 309   __ push(state);
 310   __ call_VM(noreg, runtime_entry);
 311   __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
 312   return entry;
 313 }
 314 
 315 
 316 
 317 // Helpers for commoning out cases in the various type of method entries.
 318 //
 319 
 320 
 321 // increment invocation count & check for overflow
 322 //
 323 // Note: checking for negative value instead of overflow
 324 //       so we have a 'sticky' overflow test
 325 //
 326 // rbx: method
 327 // ecx: invocation counter
 328 //
 329 void InterpreterGenerator::generate_counter_incr(
 330         Label* overflow,
 331         Label* profile_method,
 332         Label* profile_method_continue) {
 333   const Address invocation_counter(rbx, in_bytes(methodOopDesc::invocation_counter_offset()) +
 334                                         in_bytes(InvocationCounter::counter_offset()));
 335   // Note: In tiered we increment either counters in methodOop or in MDO depending if we're profiling or not.
 336   if (TieredCompilation) {
 337     int increment = InvocationCounter::count_increment;
 338     int mask = ((1 << Tier0InvokeNotifyFreqLog)  - 1) << InvocationCounter::count_shift;
 339     Label no_mdo, done;
 340     if (ProfileInterpreter) {
 341       // Are we profiling?
 342       __ movptr(rax, Address(rbx, methodOopDesc::method_data_offset()));
 343       __ testptr(rax, rax);
 344       __ jccb(Assembler::zero, no_mdo);
 345       // Increment counter in the MDO
 346       const Address mdo_invocation_counter(rax, in_bytes(methodDataOopDesc::invocation_counter_offset()) +
 347                                                 in_bytes(InvocationCounter::counter_offset()));
 348       __ increment_mask_and_jump(mdo_invocation_counter, increment, mask, rcx, false, Assembler::zero, overflow);
 349       __ jmpb(done);
 350     }
 351     __ bind(no_mdo);
 352     // Increment counter in methodOop (we don't need to load it, it's in ecx).
 353     __ increment_mask_and_jump(invocation_counter, increment, mask, rcx, true, Assembler::zero, overflow);
 354     __ bind(done);
 355   } else {
 356     const Address backedge_counter(rbx,
 357                                    methodOopDesc::backedge_counter_offset() +
 358                                    InvocationCounter::counter_offset());
 359 
 360     if (ProfileInterpreter) { // %%% Merge this into methodDataOop
 361       __ incrementl(Address(rbx,
 362                             methodOopDesc::interpreter_invocation_counter_offset()));
 363     }
 364     // Update standard invocation counters
 365     __ movl(rax, backedge_counter);   // load backedge counter
 366 
 367     __ incrementl(rcx, InvocationCounter::count_increment);
 368     __ andl(rax, InvocationCounter::count_mask_value); // mask out the status bits
 369 
 370     __ movl(invocation_counter, rcx); // save invocation count
 371     __ addl(rcx, rax);                // add both counters
 372 
 373     // profile_method is non-null only for interpreted method so
 374     // profile_method != NULL == !native_call
 375 
 376     if (ProfileInterpreter && profile_method != NULL) {
 377       // Test to see if we should create a method data oop
 378       __ cmp32(rcx, ExternalAddress((address)&InvocationCounter::InterpreterProfileLimit));
 379       __ jcc(Assembler::less, *profile_method_continue);
 380 
 381       // if no method data exists, go to profile_method
 382       __ test_method_data_pointer(rax, *profile_method);
 383     }
 384 
 385     __ cmp32(rcx, ExternalAddress((address)&InvocationCounter::InterpreterInvocationLimit));
 386     __ jcc(Assembler::aboveEqual, *overflow);
 387   }
 388 }
 389 
 390 void InterpreterGenerator::generate_counter_overflow(Label* do_continue) {
 391 
 392   // Asm interpreter on entry
 393   // r14 - locals
 394   // r13 - bcp
 395   // rbx - method
 396   // edx - cpool --- DOES NOT APPEAR TO BE TRUE
 397   // rbp - interpreter frame
 398 
 399   // On return (i.e. jump to entry_point) [ back to invocation of interpreter ]
 400   // Everything as it was on entry
 401   // rdx is not restored. Doesn't appear to really be set.
 402 
 403   const Address size_of_parameters(rbx,
 404                                    methodOopDesc::size_of_parameters_offset());
 405 
 406   // InterpreterRuntime::frequency_counter_overflow takes two
 407   // arguments, the first (thread) is passed by call_VM, the second
 408   // indicates if the counter overflow occurs at a backwards branch
 409   // (NULL bcp).  We pass zero for it.  The call returns the address
 410   // of the verified entry point for the method or NULL if the
 411   // compilation did not complete (either went background or bailed
 412   // out).
 413   __ movl(c_rarg1, 0);
 414   __ call_VM(noreg,
 415              CAST_FROM_FN_PTR(address,
 416                               InterpreterRuntime::frequency_counter_overflow),
 417              c_rarg1);
 418 
 419   __ movptr(rbx, Address(rbp, method_offset));   // restore methodOop
 420   // Preserve invariant that r13/r14 contain bcp/locals of sender frame
 421   // and jump to the interpreted entry.
 422   __ jmp(*do_continue, relocInfo::none);
 423 }
 424 
 425 // See if we've got enough room on the stack for locals plus overhead.
 426 // The expression stack grows down incrementally, so the normal guard
 427 // page mechanism will work for that.
 428 //
 429 // NOTE: Since the additional locals are also always pushed (wasn't
 430 // obvious in generate_method_entry) so the guard should work for them
 431 // too.
 432 //
 433 // Args:
 434 //      rdx: number of additional locals this frame needs (what we must check)
 435 //      rbx: methodOop
 436 //
 437 // Kills:
 438 //      rax
 439 void InterpreterGenerator::generate_stack_overflow_check(void) {
 440 
 441   // monitor entry size: see picture of stack set
 442   // (generate_method_entry) and frame_amd64.hpp
 443   const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
 444 
 445   // total overhead size: entry_size + (saved rbp through expr stack
 446   // bottom).  be sure to change this if you add/subtract anything
 447   // to/from the overhead area
 448   const int overhead_size =
 449     -(frame::interpreter_frame_initial_sp_offset * wordSize) + entry_size;
 450 
 451   const int page_size = os::vm_page_size();
 452 
 453   Label after_frame_check;
 454 
 455   // see if the frame is greater than one page in size. If so,
 456   // then we need to verify there is enough stack space remaining
 457   // for the additional locals.
 458   __ cmpl(rdx, (page_size - overhead_size) / Interpreter::stackElementSize);
 459   __ jcc(Assembler::belowEqual, after_frame_check);
 460 
 461   // compute rsp as if this were going to be the last frame on
 462   // the stack before the red zone
 463 
 464   const Address stack_base(r15_thread, Thread::stack_base_offset());
 465   const Address stack_size(r15_thread, Thread::stack_size_offset());
 466 
 467   // locals + overhead, in bytes
 468   __ mov(rax, rdx);
 469   __ shlptr(rax, Interpreter::logStackElementSize);  // 2 slots per parameter.
 470   __ addptr(rax, overhead_size);
 471 
 472 #ifdef ASSERT
 473   Label stack_base_okay, stack_size_okay;
 474   // verify that thread stack base is non-zero
 475   __ cmpptr(stack_base, (int32_t)NULL_WORD);
 476   __ jcc(Assembler::notEqual, stack_base_okay);
 477   __ stop("stack base is zero");
 478   __ bind(stack_base_okay);
 479   // verify that thread stack size is non-zero
 480   __ cmpptr(stack_size, 0);
 481   __ jcc(Assembler::notEqual, stack_size_okay);
 482   __ stop("stack size is zero");
 483   __ bind(stack_size_okay);
 484 #endif
 485 
 486   // Add stack base to locals and subtract stack size
 487   __ addptr(rax, stack_base);
 488   __ subptr(rax, stack_size);
 489 
 490   // Use the maximum number of pages we might bang.
 491   const int max_pages = StackShadowPages > (StackRedPages+StackYellowPages) ? StackShadowPages :
 492                                                                               (StackRedPages+StackYellowPages);
 493 
 494   // add in the red and yellow zone sizes
 495   __ addptr(rax, max_pages * page_size);
 496 
 497   // check against the current stack bottom
 498   __ cmpptr(rsp, rax);
 499   __ jcc(Assembler::above, after_frame_check);
 500 
 501   __ pop(rax); // get return address
 502   __ jump(ExternalAddress(Interpreter::throw_StackOverflowError_entry()));
 503 
 504   // all done with frame size check
 505   __ bind(after_frame_check);
 506 }
 507 
 508 // Allocate monitor and lock method (asm interpreter)
 509 //
 510 // Args:
 511 //      rbx: methodOop
 512 //      r14: locals
 513 //
 514 // Kills:
 515 //      rax
 516 //      c_rarg0, c_rarg1, c_rarg2, c_rarg3, ...(param regs)
 517 //      rscratch1, rscratch2 (scratch regs)
 518 void InterpreterGenerator::lock_method(void) {
 519   // synchronize method
 520   const Address access_flags(rbx, methodOopDesc::access_flags_offset());
 521   const Address monitor_block_top(
 522         rbp,
 523         frame::interpreter_frame_monitor_block_top_offset * wordSize);
 524   const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
 525 
 526 #ifdef ASSERT
 527   {
 528     Label L;
 529     __ movl(rax, access_flags);
 530     __ testl(rax, JVM_ACC_SYNCHRONIZED);
 531     __ jcc(Assembler::notZero, L);
 532     __ stop("method doesn't need synchronization");
 533     __ bind(L);
 534   }
 535 #endif // ASSERT
 536 
 537   // get synchronization object
 538   {
 539     const int mirror_offset = klassOopDesc::klass_part_offset_in_bytes() +
 540                               Klass::java_mirror_offset_in_bytes();
 541     Label done;
 542     __ movl(rax, access_flags);
 543     __ testl(rax, JVM_ACC_STATIC);
 544     // get receiver (assume this is frequent case)
 545     __ movptr(rax, Address(r14, Interpreter::local_offset_in_bytes(0)));
 546     __ jcc(Assembler::zero, done);
 547     __ movptr(rax, Address(rbx, methodOopDesc::constants_offset()));
 548     __ movptr(rax, Address(rax,
 549                            constantPoolOopDesc::pool_holder_offset_in_bytes()));
 550     __ movptr(rax, Address(rax, mirror_offset));
 551 
 552 #ifdef ASSERT
 553     {
 554       Label L;
 555       __ testptr(rax, rax);
 556       __ jcc(Assembler::notZero, L);
 557       __ stop("synchronization object is NULL");
 558       __ bind(L);
 559     }
 560 #endif // ASSERT
 561 
 562     __ bind(done);
 563   }
 564 
 565   // add space for monitor & lock
 566   __ subptr(rsp, entry_size); // add space for a monitor entry
 567   __ movptr(monitor_block_top, rsp);  // set new monitor block top
 568   // store object
 569   __ movptr(Address(rsp, BasicObjectLock::obj_offset_in_bytes()), rax);
 570   __ movptr(c_rarg1, rsp); // object address
 571   __ lock_object(c_rarg1);
 572 }
 573 
 574 // Generate a fixed interpreter frame. This is identical setup for
 575 // interpreted methods and for native methods hence the shared code.
 576 //
 577 // Args:
 578 //      rax: return address
 579 //      rbx: methodOop
 580 //      r14: pointer to locals
 581 //      r13: sender sp
 582 //      rdx: cp cache
 583 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
 584   // initialize fixed part of activation frame
 585   __ push(rax);        // save return address
 586   __ enter();          // save old & set new rbp
 587   __ push(r13);        // set sender sp
 588   __ push((int)NULL_WORD); // leave last_sp as null
 589   __ movptr(r13, Address(rbx, methodOopDesc::const_offset()));      // get constMethodOop
 590   __ lea(r13, Address(r13, constMethodOopDesc::codes_offset())); // get codebase
 591   __ push(rbx);        // save methodOop
 592   if (ProfileInterpreter) {
 593     Label method_data_continue;
 594     __ movptr(rdx, Address(rbx, in_bytes(methodOopDesc::method_data_offset())));
 595     __ testptr(rdx, rdx);
 596     __ jcc(Assembler::zero, method_data_continue);
 597     __ addptr(rdx, in_bytes(methodDataOopDesc::data_offset()));
 598     __ bind(method_data_continue);
 599     __ push(rdx);      // set the mdp (method data pointer)
 600   } else {
 601     __ push(0);
 602   }
 603 
 604   __ movptr(rdx, Address(rbx, methodOopDesc::constants_offset()));
 605   __ movptr(rdx, Address(rdx, constantPoolOopDesc::cache_offset_in_bytes()));
 606   __ push(rdx); // set constant pool cache
 607   __ push(r14); // set locals pointer
 608   if (native_call) {
 609     __ push(0); // no bcp
 610   } else {
 611     __ push(r13); // set bcp
 612   }
 613   __ push(0); // reserve word for pointer to expression stack bottom
 614   __ movptr(Address(rsp, 0), rsp); // set expression stack bottom
 615 }
 616 
 617 // End of helpers
 618 
 619 // Various method entries
 620 //------------------------------------------------------------------------------------------------------------------------
 621 //
 622 //
 623 
 624 // Call an accessor method (assuming it is resolved, otherwise drop
 625 // into vanilla (slow path) entry
 626 address InterpreterGenerator::generate_accessor_entry(void) {
 627   // rbx: methodOop
 628 
 629   // r13: senderSP must preserver for slow path, set SP to it on fast path
 630 
 631   address entry_point = __ pc();
 632   Label xreturn_path;
 633 
 634   // do fastpath for resolved accessor methods
 635   if (UseFastAccessorMethods) {
 636     // Code: _aload_0, _(i|a)getfield, _(i|a)return or any rewrites
 637     //       thereof; parameter size = 1
 638     // Note: We can only use this code if the getfield has been resolved
 639     //       and if we don't have a null-pointer exception => check for
 640     //       these conditions first and use slow path if necessary.
 641     Label slow_path;
 642     // If we need a safepoint check, generate full interpreter entry.
 643     __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
 644              SafepointSynchronize::_not_synchronized);
 645 
 646     __ jcc(Assembler::notEqual, slow_path);
 647     // rbx: method
 648     __ movptr(rax, Address(rsp, wordSize));
 649 
 650     // check if local 0 != NULL and read field
 651     __ testptr(rax, rax);
 652     __ jcc(Assembler::zero, slow_path);
 653 
 654     __ movptr(rdi, Address(rbx, methodOopDesc::constants_offset()));
 655     // read first instruction word and extract bytecode @ 1 and index @ 2
 656     __ movptr(rdx, Address(rbx, methodOopDesc::const_offset()));
 657     __ movl(rdx, Address(rdx, constMethodOopDesc::codes_offset()));
 658     // Shift codes right to get the index on the right.
 659     // The bytecode fetched looks like <index><0xb4><0x2a>
 660     __ shrl(rdx, 2 * BitsPerByte);
 661     __ shll(rdx, exact_log2(in_words(ConstantPoolCacheEntry::size())));
 662     __ movptr(rdi, Address(rdi, constantPoolOopDesc::cache_offset_in_bytes()));
 663 
 664     // rax: local 0
 665     // rbx: method
 666     // rdx: constant pool cache index
 667     // rdi: constant pool cache
 668 
 669     // check if getfield has been resolved and read constant pool cache entry
 670     // check the validity of the cache entry by testing whether _indices field
 671     // contains Bytecode::_getfield in b1 byte.
 672     assert(in_words(ConstantPoolCacheEntry::size()) == 4,
 673            "adjust shift below");
 674     __ movl(rcx,
 675             Address(rdi,
 676                     rdx,
 677                     Address::times_8,
 678                     constantPoolCacheOopDesc::base_offset() +
 679                     ConstantPoolCacheEntry::indices_offset()));
 680     __ shrl(rcx, 2 * BitsPerByte);
 681     __ andl(rcx, 0xFF);
 682     __ cmpl(rcx, Bytecodes::_getfield);
 683     __ jcc(Assembler::notEqual, slow_path);
 684 
 685     // Note: constant pool entry is not valid before bytecode is resolved
 686     __ movptr(rcx,
 687               Address(rdi,
 688                       rdx,
 689                       Address::times_8,
 690                       constantPoolCacheOopDesc::base_offset() +
 691                       ConstantPoolCacheEntry::f2_offset()));
 692     // edx: flags
 693     __ movl(rdx,
 694             Address(rdi,
 695                     rdx,
 696                     Address::times_8,
 697                     constantPoolCacheOopDesc::base_offset() +
 698                     ConstantPoolCacheEntry::flags_offset()));
 699 
 700     Label notObj, notInt, notByte, notShort;
 701     const Address field_address(rax, rcx, Address::times_1);
 702 
 703     // Need to differentiate between igetfield, agetfield, bgetfield etc.
 704     // because they are different sizes.
 705     // Use the type from the constant pool cache
 706     __ shrl(rdx, ConstantPoolCacheEntry::tosBits);
 707     // Make sure we don't need to mask edx for tosBits after the above shift
 708     ConstantPoolCacheEntry::verify_tosBits();
 709 
 710     __ cmpl(rdx, atos);
 711     __ jcc(Assembler::notEqual, notObj);
 712     // atos
 713     __ load_heap_oop(rax, field_address);
 714     __ jmp(xreturn_path);
 715 
 716     __ bind(notObj);
 717     __ cmpl(rdx, itos);
 718     __ jcc(Assembler::notEqual, notInt);
 719     // itos
 720     __ movl(rax, field_address);
 721     __ jmp(xreturn_path);
 722 
 723     __ bind(notInt);
 724     __ cmpl(rdx, btos);
 725     __ jcc(Assembler::notEqual, notByte);
 726     // btos
 727     __ load_signed_byte(rax, field_address);
 728     __ jmp(xreturn_path);
 729 
 730     __ bind(notByte);
 731     __ cmpl(rdx, stos);
 732     __ jcc(Assembler::notEqual, notShort);
 733     // stos
 734     __ load_signed_short(rax, field_address);
 735     __ jmp(xreturn_path);
 736 
 737     __ bind(notShort);
 738 #ifdef ASSERT
 739     Label okay;
 740     __ cmpl(rdx, ctos);
 741     __ jcc(Assembler::equal, okay);
 742     __ stop("what type is this?");
 743     __ bind(okay);
 744 #endif
 745     // ctos
 746     __ load_unsigned_short(rax, field_address);
 747 
 748     __ bind(xreturn_path);
 749 
 750     // _ireturn/_areturn
 751     __ pop(rdi);
 752     __ mov(rsp, r13);
 753     __ jmp(rdi);
 754     __ ret(0);
 755 
 756     // generate a vanilla interpreter entry as the slow path
 757     __ bind(slow_path);
 758     (void) generate_normal_entry(false);
 759   } else {
 760     (void) generate_normal_entry(false);
 761   }
 762 
 763   return entry_point;
 764 }
 765 
 766 // Interpreter stub for calling a native method. (asm interpreter)
 767 // This sets up a somewhat different looking stack for calling the
 768 // native method than the typical interpreter frame setup.
 769 address InterpreterGenerator::generate_native_entry(bool synchronized) {
 770   // determine code generation flags
 771   bool inc_counter  = UseCompiler || CountCompiledCalls;
 772 
 773   // rbx: methodOop
 774   // r13: sender sp
 775 
 776   address entry_point = __ pc();
 777 
 778   const Address size_of_parameters(rbx, methodOopDesc::
 779                                         size_of_parameters_offset());
 780   const Address invocation_counter(rbx, methodOopDesc::
 781                                         invocation_counter_offset() +
 782                                         InvocationCounter::counter_offset());
 783   const Address access_flags      (rbx, methodOopDesc::access_flags_offset());
 784 
 785   // get parameter size (always needed)
 786   __ load_unsigned_short(rcx, size_of_parameters);
 787 
 788   // native calls don't need the stack size check since they have no
 789   // expression stack and the arguments are already on the stack and
 790   // we only add a handful of words to the stack
 791 
 792   // rbx: methodOop
 793   // rcx: size of parameters
 794   // r13: sender sp
 795   __ pop(rax);                                       // get return address
 796 
 797   // for natives the size of locals is zero
 798 
 799   // compute beginning of parameters (r14)
 800   __ lea(r14, Address(rsp, rcx, Address::times_8, -wordSize));
 801 
 802   // add 2 zero-initialized slots for native calls
 803   // initialize result_handler slot
 804   __ push((int) NULL_WORD);
 805   // slot for oop temp
 806   // (static native method holder mirror/jni oop result)
 807   __ push((int) NULL_WORD);
 808 
 809   if (inc_counter) {
 810     __ movl(rcx, invocation_counter);  // (pre-)fetch invocation count
 811   }
 812 
 813   // initialize fixed part of activation frame
 814   generate_fixed_frame(true);
 815 
 816   // make sure method is native & not abstract
 817 #ifdef ASSERT
 818   __ movl(rax, access_flags);
 819   {
 820     Label L;
 821     __ testl(rax, JVM_ACC_NATIVE);
 822     __ jcc(Assembler::notZero, L);
 823     __ stop("tried to execute non-native method as native");
 824     __ bind(L);
 825   }
 826   {
 827     Label L;
 828     __ testl(rax, JVM_ACC_ABSTRACT);
 829     __ jcc(Assembler::zero, L);
 830     __ stop("tried to execute abstract method in interpreter");
 831     __ bind(L);
 832   }
 833 #endif
 834 
 835   // Since at this point in the method invocation the exception handler
 836   // would try to exit the monitor of synchronized methods which hasn't
 837   // been entered yet, we set the thread local variable
 838   // _do_not_unlock_if_synchronized to true. The remove_activation will
 839   // check this flag.
 840 
 841   const Address do_not_unlock_if_synchronized(r15_thread,
 842         in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
 843   __ movbool(do_not_unlock_if_synchronized, true);
 844 
 845   // increment invocation count & check for overflow
 846   Label invocation_counter_overflow;
 847   if (inc_counter) {
 848     generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
 849   }
 850 
 851   Label continue_after_compile;
 852   __ bind(continue_after_compile);
 853 
 854   bang_stack_shadow_pages(true);
 855 
 856   // reset the _do_not_unlock_if_synchronized flag
 857   __ movbool(do_not_unlock_if_synchronized, false);
 858 
 859   // check for synchronized methods
 860   // Must happen AFTER invocation_counter check and stack overflow check,
 861   // so method is not locked if overflows.
 862   if (synchronized) {
 863     lock_method();
 864   } else {
 865     // no synchronization necessary
 866 #ifdef ASSERT
 867     {
 868       Label L;
 869       __ movl(rax, access_flags);
 870       __ testl(rax, JVM_ACC_SYNCHRONIZED);
 871       __ jcc(Assembler::zero, L);
 872       __ stop("method needs synchronization");
 873       __ bind(L);
 874     }
 875 #endif
 876   }
 877 
 878   // start execution
 879 #ifdef ASSERT
 880   {
 881     Label L;
 882     const Address monitor_block_top(rbp,
 883                  frame::interpreter_frame_monitor_block_top_offset * wordSize);
 884     __ movptr(rax, monitor_block_top);
 885     __ cmpptr(rax, rsp);
 886     __ jcc(Assembler::equal, L);
 887     __ stop("broken stack frame setup in interpreter");
 888     __ bind(L);
 889   }
 890 #endif
 891 
 892   // jvmti support
 893   __ notify_method_entry();
 894 
 895   // work registers
 896   const Register method = rbx;
 897   const Register t      = r11;
 898 
 899   // allocate space for parameters
 900   __ get_method(method);
 901   __ verify_oop(method);
 902   __ load_unsigned_short(t,
 903                          Address(method,
 904                                  methodOopDesc::size_of_parameters_offset()));
 905   __ shll(t, Interpreter::logStackElementSize);
 906 
 907   __ subptr(rsp, t);
 908   __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
 909   __ andptr(rsp, -16); // must be 16 byte boundary (see amd64 ABI)
 910 
 911   // get signature handler
 912   {
 913     Label L;
 914     __ movptr(t, Address(method, methodOopDesc::signature_handler_offset()));
 915     __ testptr(t, t);
 916     __ jcc(Assembler::notZero, L);
 917     __ call_VM(noreg,
 918                CAST_FROM_FN_PTR(address,
 919                                 InterpreterRuntime::prepare_native_call),
 920                method);
 921     __ get_method(method);
 922     __ movptr(t, Address(method, methodOopDesc::signature_handler_offset()));
 923     __ bind(L);
 924   }
 925 
 926   // call signature handler
 927   assert(InterpreterRuntime::SignatureHandlerGenerator::from() == r14,
 928          "adjust this code");
 929   assert(InterpreterRuntime::SignatureHandlerGenerator::to() == rsp,
 930          "adjust this code");
 931   assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == rscratch1,
 932           "adjust this code");
 933 
 934   // The generated handlers do not touch RBX (the method oop).
 935   // However, large signatures cannot be cached and are generated
 936   // each time here.  The slow-path generator can do a GC on return,
 937   // so we must reload it after the call.
 938   __ call(t);
 939   __ get_method(method);        // slow path can do a GC, reload RBX
 940 
 941 
 942   // result handler is in rax
 943   // set result handler
 944   __ movptr(Address(rbp,
 945                     (frame::interpreter_frame_result_handler_offset) * wordSize),
 946             rax);
 947 
 948   // pass mirror handle if static call
 949   {
 950     Label L;
 951     const int mirror_offset = klassOopDesc::klass_part_offset_in_bytes() +
 952                               Klass::java_mirror_offset_in_bytes();
 953     __ movl(t, Address(method, methodOopDesc::access_flags_offset()));
 954     __ testl(t, JVM_ACC_STATIC);
 955     __ jcc(Assembler::zero, L);
 956     // get mirror
 957     __ movptr(t, Address(method, methodOopDesc::constants_offset()));
 958     __ movptr(t, Address(t, constantPoolOopDesc::pool_holder_offset_in_bytes()));
 959     __ movptr(t, Address(t, mirror_offset));
 960     // copy mirror into activation frame
 961     __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize),
 962             t);
 963     // pass handle to mirror
 964     __ lea(c_rarg1,
 965            Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize));
 966     __ bind(L);
 967   }
 968 
 969   // get native function entry point
 970   {
 971     Label L;
 972     __ movptr(rax, Address(method, methodOopDesc::native_function_offset()));
 973     ExternalAddress unsatisfied(SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
 974     __ movptr(rscratch2, unsatisfied.addr());
 975     __ cmpptr(rax, rscratch2);
 976     __ jcc(Assembler::notEqual, L);
 977     __ call_VM(noreg,
 978                CAST_FROM_FN_PTR(address,
 979                                 InterpreterRuntime::prepare_native_call),
 980                method);
 981     __ get_method(method);
 982     __ verify_oop(method);
 983     __ movptr(rax, Address(method, methodOopDesc::native_function_offset()));
 984     __ bind(L);
 985   }
 986 
 987   // pass JNIEnv
 988   __ lea(c_rarg0, Address(r15_thread, JavaThread::jni_environment_offset()));
 989 
 990   // It is enough that the pc() points into the right code
 991   // segment. It does not have to be the correct return pc.
 992   __ set_last_Java_frame(rsp, rbp, (address) __ pc());
 993 
 994   // change thread state
 995 #ifdef ASSERT
 996   {
 997     Label L;
 998     __ movl(t, Address(r15_thread, JavaThread::thread_state_offset()));
 999     __ cmpl(t, _thread_in_Java);
1000     __ jcc(Assembler::equal, L);
1001     __ stop("Wrong thread state in native stub");
1002     __ bind(L);
1003   }
1004 #endif
1005 
1006   // Change state to native
1007 
1008   __ movl(Address(r15_thread, JavaThread::thread_state_offset()),
1009           _thread_in_native);
1010 
1011   // Call the native method.
1012   __ call(rax);
1013   // result potentially in rax or xmm0
1014 
1015   // Depending on runtime options, either restore the MXCSR
1016   // register after returning from the JNI Call or verify that
1017   // it wasn't changed during -Xcheck:jni.
1018   if (RestoreMXCSROnJNICalls) {
1019     __ ldmxcsr(ExternalAddress(StubRoutines::x86::mxcsr_std()));
1020   }
1021   else if (CheckJNICalls) {
1022     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::x86::verify_mxcsr_entry())));
1023   }
1024 
1025   // NOTE: The order of these pushes is known to frame::interpreter_frame_result
1026   // in order to extract the result of a method call. If the order of these
1027   // pushes change or anything else is added to the stack then the code in
1028   // interpreter_frame_result must also change.
1029 
1030   __ push(dtos);
1031   __ push(ltos);
1032 
1033   // change thread state
1034   __ movl(Address(r15_thread, JavaThread::thread_state_offset()),
1035           _thread_in_native_trans);
1036 
1037   if (os::is_MP()) {
1038     if (UseMembar) {
1039       // Force this write out before the read below
1040       __ membar(Assembler::Membar_mask_bits(
1041            Assembler::LoadLoad | Assembler::LoadStore |
1042            Assembler::StoreLoad | Assembler::StoreStore));
1043     } else {
1044       // Write serialization page so VM thread can do a pseudo remote membar.
1045       // We use the current thread pointer to calculate a thread specific
1046       // offset to write to within the page. This minimizes bus traffic
1047       // due to cache line collision.
1048       __ serialize_memory(r15_thread, rscratch2);
1049     }
1050   }
1051 
1052   // check for safepoint operation in progress and/or pending suspend requests
1053   {
1054     Label Continue;
1055     __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
1056              SafepointSynchronize::_not_synchronized);
1057 
1058     Label L;
1059     __ jcc(Assembler::notEqual, L);
1060     __ cmpl(Address(r15_thread, JavaThread::suspend_flags_offset()), 0);
1061     __ jcc(Assembler::equal, Continue);
1062     __ bind(L);
1063 
1064     // Don't use call_VM as it will see a possible pending exception
1065     // and forward it and never return here preventing us from
1066     // clearing _last_native_pc down below.  Also can't use
1067     // call_VM_leaf either as it will check to see if r13 & r14 are
1068     // preserved and correspond to the bcp/locals pointers. So we do a
1069     // runtime call by hand.
1070     //
1071     __ mov(c_rarg0, r15_thread);
1072     __ mov(r12, rsp); // remember sp
1073     __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1074     __ andptr(rsp, -16); // align stack as required by ABI
1075     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans)));
1076     __ mov(rsp, r12); // restore sp
1077     __ reinit_heapbase();
1078     __ bind(Continue);
1079   }
1080 
1081   // change thread state
1082   __ movl(Address(r15_thread, JavaThread::thread_state_offset()), _thread_in_Java);
1083 
1084   // reset_last_Java_frame
1085   __ reset_last_Java_frame(true, true);
1086 
1087   // reset handle block
1088   __ movptr(t, Address(r15_thread, JavaThread::active_handles_offset()));
1089   __ movptr(Address(t, JNIHandleBlock::top_offset_in_bytes()), (int32_t)NULL_WORD);
1090 
1091   // If result is an oop unbox and store it in frame where gc will see it
1092   // and result handler will pick it up
1093 
1094   {
1095     Label no_oop, store_result;
1096     __ lea(t, ExternalAddress(AbstractInterpreter::result_handler(T_OBJECT)));
1097     __ cmpptr(t, Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize));
1098     __ jcc(Assembler::notEqual, no_oop);
1099     // retrieve result
1100     __ pop(ltos);
1101     __ testptr(rax, rax);
1102     __ jcc(Assembler::zero, store_result);
1103     __ movptr(rax, Address(rax, 0));
1104     __ bind(store_result);
1105     __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize), rax);
1106     // keep stack depth as expected by pushing oop which will eventually be discarde
1107     __ push(ltos);
1108     __ bind(no_oop);
1109   }
1110 
1111 
1112   {
1113     Label no_reguard;
1114     __ cmpl(Address(r15_thread, JavaThread::stack_guard_state_offset()),
1115             JavaThread::stack_guard_yellow_disabled);
1116     __ jcc(Assembler::notEqual, no_reguard);
1117 
1118     __ pusha(); // XXX only save smashed registers
1119     __ mov(r12, rsp); // remember sp
1120     __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1121     __ andptr(rsp, -16); // align stack as required by ABI
1122     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
1123     __ mov(rsp, r12); // restore sp
1124     __ popa(); // XXX only restore smashed registers
1125     __ reinit_heapbase();
1126 
1127     __ bind(no_reguard);
1128   }
1129 
1130 
1131   // The method register is junk from after the thread_in_native transition
1132   // until here.  Also can't call_VM until the bcp has been
1133   // restored.  Need bcp for throwing exception below so get it now.
1134   __ get_method(method);
1135   __ verify_oop(method);
1136 
1137   // restore r13 to have legal interpreter frame, i.e., bci == 0 <=>
1138   // r13 == code_base()
1139   __ movptr(r13, Address(method, methodOopDesc::const_offset()));   // get constMethodOop
1140   __ lea(r13, Address(r13, constMethodOopDesc::codes_offset()));    // get codebase
1141   // handle exceptions (exception handling will handle unlocking!)
1142   {
1143     Label L;
1144     __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t) NULL_WORD);
1145     __ jcc(Assembler::zero, L);
1146     // Note: At some point we may want to unify this with the code
1147     // used in call_VM_base(); i.e., we should use the
1148     // StubRoutines::forward_exception code. For now this doesn't work
1149     // here because the rsp is not correctly set at this point.
1150     __ MacroAssembler::call_VM(noreg,
1151                                CAST_FROM_FN_PTR(address,
1152                                InterpreterRuntime::throw_pending_exception));
1153     __ should_not_reach_here();
1154     __ bind(L);
1155   }
1156 
1157   // do unlocking if necessary
1158   {
1159     Label L;
1160     __ movl(t, Address(method, methodOopDesc::access_flags_offset()));
1161     __ testl(t, JVM_ACC_SYNCHRONIZED);
1162     __ jcc(Assembler::zero, L);
1163     // the code below should be shared with interpreter macro
1164     // assembler implementation
1165     {
1166       Label unlock;
1167       // BasicObjectLock will be first in list, since this is a
1168       // synchronized method. However, need to check that the object
1169       // has not been unlocked by an explicit monitorexit bytecode.
1170       const Address monitor(rbp,
1171                             (intptr_t)(frame::interpreter_frame_initial_sp_offset *
1172                                        wordSize - sizeof(BasicObjectLock)));
1173 
1174       // monitor expect in c_rarg1 for slow unlock path
1175       __ lea(c_rarg1, monitor); // address of first monitor
1176 
1177       __ movptr(t, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
1178       __ testptr(t, t);
1179       __ jcc(Assembler::notZero, unlock);
1180 
1181       // Entry already unlocked, need to throw exception
1182       __ MacroAssembler::call_VM(noreg,
1183                                  CAST_FROM_FN_PTR(address,
1184                    InterpreterRuntime::throw_illegal_monitor_state_exception));
1185       __ should_not_reach_here();
1186 
1187       __ bind(unlock);
1188       __ unlock_object(c_rarg1);
1189     }
1190     __ bind(L);
1191   }
1192 
1193   // jvmti support
1194   // Note: This must happen _after_ handling/throwing any exceptions since
1195   //       the exception handler code notifies the runtime of method exits
1196   //       too. If this happens before, method entry/exit notifications are
1197   //       not properly paired (was bug - gri 11/22/99).
1198   __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI);
1199 
1200   // restore potential result in edx:eax, call result handler to
1201   // restore potential result in ST0 & handle result
1202 
1203   __ pop(ltos);
1204   __ pop(dtos);
1205 
1206   __ movptr(t, Address(rbp,
1207                        (frame::interpreter_frame_result_handler_offset) * wordSize));
1208   __ call(t);
1209 
1210   // remove activation
1211   __ movptr(t, Address(rbp,
1212                        frame::interpreter_frame_sender_sp_offset *
1213                        wordSize)); // get sender sp
1214   __ leave();                                // remove frame anchor
1215   __ pop(rdi);                               // get return address
1216   __ mov(rsp, t);                            // set sp to sender sp
1217   __ jmp(rdi);
1218 
1219   if (inc_counter) {
1220     // Handle overflow of counter and compile method
1221     __ bind(invocation_counter_overflow);
1222     generate_counter_overflow(&continue_after_compile);
1223   }
1224 
1225   return entry_point;
1226 }
1227 
1228 //
1229 // Generic interpreted method entry to (asm) interpreter
1230 //
1231 address InterpreterGenerator::generate_normal_entry(bool synchronized) {
1232   // determine code generation flags
1233   bool inc_counter  = UseCompiler || CountCompiledCalls;
1234 
1235   // ebx: methodOop
1236   // r13: sender sp
1237   address entry_point = __ pc();
1238 
1239   const Address size_of_parameters(rbx,
1240                                    methodOopDesc::size_of_parameters_offset());
1241   const Address size_of_locals(rbx, methodOopDesc::size_of_locals_offset());
1242   const Address invocation_counter(rbx,
1243                                    methodOopDesc::invocation_counter_offset() +
1244                                    InvocationCounter::counter_offset());
1245   const Address access_flags(rbx, methodOopDesc::access_flags_offset());
1246 
1247   // get parameter size (always needed)
1248   __ load_unsigned_short(rcx, size_of_parameters);
1249 
1250   // rbx: methodOop
1251   // rcx: size of parameters
1252   // r13: sender_sp (could differ from sp+wordSize if we were called via c2i )
1253 
1254   __ load_unsigned_short(rdx, size_of_locals); // get size of locals in words
1255   __ subl(rdx, rcx); // rdx = no. of additional locals
1256 
1257   // YYY
1258 //   __ incrementl(rdx);
1259 //   __ andl(rdx, -2);
1260 
1261   // see if we've got enough room on the stack for locals plus overhead.
1262   generate_stack_overflow_check();
1263 
1264   // get return address
1265   __ pop(rax);
1266 
1267   // compute beginning of parameters (r14)
1268   __ lea(r14, Address(rsp, rcx, Address::times_8, -wordSize));
1269 
1270   // rdx - # of additional locals
1271   // allocate space for locals
1272   // explicitly initialize locals
1273   {
1274     Label exit, loop;
1275     __ testl(rdx, rdx);
1276     __ jcc(Assembler::lessEqual, exit); // do nothing if rdx <= 0
1277     __ bind(loop);
1278     __ push((int) NULL_WORD); // initialize local variables
1279     __ decrementl(rdx); // until everything initialized
1280     __ jcc(Assembler::greater, loop);
1281     __ bind(exit);
1282   }
1283 
1284   // (pre-)fetch invocation count
1285   if (inc_counter) {
1286     __ movl(rcx, invocation_counter);
1287   }
1288   // initialize fixed part of activation frame
1289   generate_fixed_frame(false);
1290 
1291   // make sure method is not native & not abstract
1292 #ifdef ASSERT
1293   __ movl(rax, access_flags);
1294   {
1295     Label L;
1296     __ testl(rax, JVM_ACC_NATIVE);
1297     __ jcc(Assembler::zero, L);
1298     __ stop("tried to execute native method as non-native");
1299     __ bind(L);
1300   }
1301   {
1302     Label L;
1303     __ testl(rax, JVM_ACC_ABSTRACT);
1304     __ jcc(Assembler::zero, L);
1305     __ stop("tried to execute abstract method in interpreter");
1306     __ bind(L);
1307   }
1308 #endif
1309 
1310   // Since at this point in the method invocation the exception
1311   // handler would try to exit the monitor of synchronized methods
1312   // which hasn't been entered yet, we set the thread local variable
1313   // _do_not_unlock_if_synchronized to true. The remove_activation
1314   // will check this flag.
1315 
1316   const Address do_not_unlock_if_synchronized(r15_thread,
1317         in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1318   __ movbool(do_not_unlock_if_synchronized, true);
1319 
1320   // increment invocation count & check for overflow
1321   Label invocation_counter_overflow;
1322   Label profile_method;
1323   Label profile_method_continue;
1324   if (inc_counter) {
1325     generate_counter_incr(&invocation_counter_overflow,
1326                           &profile_method,
1327                           &profile_method_continue);
1328     if (ProfileInterpreter) {
1329       __ bind(profile_method_continue);
1330     }
1331   }
1332 
1333   Label continue_after_compile;
1334   __ bind(continue_after_compile);
1335 
1336   // check for synchronized interpreted methods
1337   bang_stack_shadow_pages(false);
1338 
1339   // reset the _do_not_unlock_if_synchronized flag
1340   __ movbool(do_not_unlock_if_synchronized, false);
1341 
1342   // check for synchronized methods
1343   // Must happen AFTER invocation_counter check and stack overflow check,
1344   // so method is not locked if overflows.
1345   if (synchronized) {
1346     // Allocate monitor and lock method
1347     lock_method();
1348   } else {
1349     // no synchronization necessary
1350 #ifdef ASSERT
1351     {
1352       Label L;
1353       __ movl(rax, access_flags);
1354       __ testl(rax, JVM_ACC_SYNCHRONIZED);
1355       __ jcc(Assembler::zero, L);
1356       __ stop("method needs synchronization");
1357       __ bind(L);
1358     }
1359 #endif
1360   }
1361 
1362   // start execution
1363 #ifdef ASSERT
1364   {
1365     Label L;
1366      const Address monitor_block_top (rbp,
1367                  frame::interpreter_frame_monitor_block_top_offset * wordSize);
1368     __ movptr(rax, monitor_block_top);
1369     __ cmpptr(rax, rsp);
1370     __ jcc(Assembler::equal, L);
1371     __ stop("broken stack frame setup in interpreter");
1372     __ bind(L);
1373   }
1374 #endif
1375 
1376   // jvmti support
1377   __ notify_method_entry();
1378 
1379   __ dispatch_next(vtos);
1380 
1381   // invocation counter overflow
1382   if (inc_counter) {
1383     if (ProfileInterpreter) {
1384       // We have decided to profile this method in the interpreter
1385       __ bind(profile_method);
1386 
1387       __ call_VM(noreg,
1388                  CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method),
1389                  r13, true);
1390 
1391       __ movptr(rbx, Address(rbp, method_offset)); // restore methodOop
1392       __ movptr(rax, Address(rbx,
1393                              in_bytes(methodOopDesc::method_data_offset())));
1394       __ movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize),
1395                 rax);
1396       __ test_method_data_pointer(rax, profile_method_continue);
1397       __ addptr(rax, in_bytes(methodDataOopDesc::data_offset()));
1398       __ movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize),
1399               rax);
1400       __ jmp(profile_method_continue);
1401     }
1402     // Handle overflow of counter and compile method
1403     __ bind(invocation_counter_overflow);
1404     generate_counter_overflow(&continue_after_compile);
1405   }
1406 
1407   return entry_point;
1408 }
1409 
1410 // Entry points
1411 //
1412 // Here we generate the various kind of entries into the interpreter.
1413 // The two main entry type are generic bytecode methods and native
1414 // call method.  These both come in synchronized and non-synchronized
1415 // versions but the frame layout they create is very similar. The
1416 // other method entry types are really just special purpose entries
1417 // that are really entry and interpretation all in one. These are for
1418 // trivial methods like accessor, empty, or special math methods.
1419 //
1420 // When control flow reaches any of the entry types for the interpreter
1421 // the following holds ->
1422 //
1423 // Arguments:
1424 //
1425 // rbx: methodOop
1426 //
1427 // Stack layout immediately at entry
1428 //
1429 // [ return address     ] <--- rsp
1430 // [ parameter n        ]
1431 //   ...
1432 // [ parameter 1        ]
1433 // [ expression stack   ] (caller's java expression stack)
1434 
1435 // Assuming that we don't go to one of the trivial specialized entries
1436 // the stack will look like below when we are ready to execute the
1437 // first bytecode (or call the native routine). The register usage
1438 // will be as the template based interpreter expects (see
1439 // interpreter_amd64.hpp).
1440 //
1441 // local variables follow incoming parameters immediately; i.e.
1442 // the return address is moved to the end of the locals).
1443 //
1444 // [ monitor entry      ] <--- rsp
1445 //   ...
1446 // [ monitor entry      ]
1447 // [ expr. stack bottom ]
1448 // [ saved r13          ]
1449 // [ current r14        ]
1450 // [ methodOop          ]
1451 // [ saved ebp          ] <--- rbp
1452 // [ return address     ]
1453 // [ local variable m   ]
1454 //   ...
1455 // [ local variable 1   ]
1456 // [ parameter n        ]
1457 //   ...
1458 // [ parameter 1        ] <--- r14
1459 
1460 address AbstractInterpreterGenerator::generate_method_entry(
1461                                         AbstractInterpreter::MethodKind kind) {
1462   // determine code generation flags
1463   bool synchronized = false;
1464   address entry_point = NULL;
1465 
1466   switch (kind) {
1467   case Interpreter::zerolocals             :                                                                             break;
1468   case Interpreter::zerolocals_synchronized: synchronized = true;                                                        break;
1469   case Interpreter::native                 : entry_point = ((InterpreterGenerator*) this)->generate_native_entry(false); break;
1470   case Interpreter::native_synchronized    : entry_point = ((InterpreterGenerator*) this)->generate_native_entry(true);  break;
1471   case Interpreter::empty                  : entry_point = ((InterpreterGenerator*) this)->generate_empty_entry();       break;
1472   case Interpreter::accessor               : entry_point = ((InterpreterGenerator*) this)->generate_accessor_entry();    break;
1473   case Interpreter::abstract               : entry_point = ((InterpreterGenerator*) this)->generate_abstract_entry();    break;
1474   case Interpreter::method_handle          : entry_point = ((InterpreterGenerator*) this)->generate_method_handle_entry();break;
1475 
1476   case Interpreter::java_lang_math_sin     : // fall thru
1477   case Interpreter::java_lang_math_cos     : // fall thru
1478   case Interpreter::java_lang_math_tan     : // fall thru
1479   case Interpreter::java_lang_math_abs     : // fall thru
1480   case Interpreter::java_lang_math_log     : // fall thru
1481   case Interpreter::java_lang_math_log10   : // fall thru
1482   case Interpreter::java_lang_math_sqrt    : entry_point = ((InterpreterGenerator*) this)->generate_math_entry(kind);    break;
1483   default                                  : ShouldNotReachHere();                                                       break;
1484   }
1485 
1486   if (entry_point) {
1487     return entry_point;
1488   }
1489 
1490   return ((InterpreterGenerator*) this)->
1491                                 generate_normal_entry(synchronized);
1492 }
1493 
1494 // These should never be compiled since the interpreter will prefer
1495 // the compiled version to the intrinsic version.
1496 bool AbstractInterpreter::can_be_compiled(methodHandle m) {
1497   switch (method_kind(m)) {
1498     case Interpreter::java_lang_math_sin     : // fall thru
1499     case Interpreter::java_lang_math_cos     : // fall thru
1500     case Interpreter::java_lang_math_tan     : // fall thru
1501     case Interpreter::java_lang_math_abs     : // fall thru
1502     case Interpreter::java_lang_math_log     : // fall thru
1503     case Interpreter::java_lang_math_log10   : // fall thru
1504     case Interpreter::java_lang_math_sqrt    :
1505       return false;
1506     default:
1507       return true;
1508   }
1509 }
1510 
1511 // How much stack a method activation needs in words.
1512 int AbstractInterpreter::size_top_interpreter_activation(methodOop method) {
1513   const int entry_size = frame::interpreter_frame_monitor_size();
1514 
1515   // total overhead size: entry_size + (saved rbp thru expr stack
1516   // bottom).  be sure to change this if you add/subtract anything
1517   // to/from the overhead area
1518   const int overhead_size =
1519     -(frame::interpreter_frame_initial_sp_offset) + entry_size;
1520 
1521   const int stub_code = frame::entry_frame_after_call_words;
1522   const int extra_stack = methodOopDesc::extra_stack_entries();
1523   const int method_stack = (method->max_locals() + method->max_stack() + extra_stack) *
1524                            Interpreter::stackElementWords;
1525   return (overhead_size + method_stack + stub_code);
1526 }
1527 
1528 int AbstractInterpreter::layout_activation(methodOop method,
1529                                            int tempcount,
1530                                            int popframe_extra_args,
1531                                            int moncount,
1532                                            int callee_param_count,
1533                                            int callee_locals,
1534                                            frame* caller,
1535                                            frame* interpreter_frame,
1536                                            bool is_top_frame) {
1537   // Note: This calculation must exactly parallel the frame setup
1538   // in AbstractInterpreterGenerator::generate_method_entry.
1539   // If interpreter_frame!=NULL, set up the method, locals, and monitors.
1540   // The frame interpreter_frame, if not NULL, is guaranteed to be the
1541   // right size, as determined by a previous call to this method.
1542   // It is also guaranteed to be walkable even though it is in a skeletal state
1543 
1544   // fixed size of an interpreter frame:
1545   int max_locals = method->max_locals() * Interpreter::stackElementWords;
1546   int extra_locals = (method->max_locals() - method->size_of_parameters()) *
1547                      Interpreter::stackElementWords;
1548 
1549   int overhead = frame::sender_sp_offset -
1550                  frame::interpreter_frame_initial_sp_offset;
1551   // Our locals were accounted for by the caller (or last_frame_adjust
1552   // on the transistion) Since the callee parameters already account
1553   // for the callee's params we only need to account for the extra
1554   // locals.
1555   int size = overhead +
1556          (callee_locals - callee_param_count)*Interpreter::stackElementWords +
1557          moncount * frame::interpreter_frame_monitor_size() +
1558          tempcount* Interpreter::stackElementWords + popframe_extra_args;
1559   if (interpreter_frame != NULL) {
1560 #ifdef ASSERT
1561     if (!EnableMethodHandles)
1562       // @@@ FIXME: Should we correct interpreter_frame_sender_sp in the calling sequences?
1563       // Probably, since deoptimization doesn't work yet.
1564       assert(caller->unextended_sp() == interpreter_frame->interpreter_frame_sender_sp(), "Frame not properly walkable");
1565     assert(caller->sp() == interpreter_frame->sender_sp(), "Frame not properly walkable(2)");
1566 #endif
1567 
1568     interpreter_frame->interpreter_frame_set_method(method);
1569     // NOTE the difference in using sender_sp and
1570     // interpreter_frame_sender_sp interpreter_frame_sender_sp is
1571     // the original sp of the caller (the unextended_sp) and
1572     // sender_sp is fp+16 XXX
1573     intptr_t* locals = interpreter_frame->sender_sp() + max_locals - 1;
1574 
1575     interpreter_frame->interpreter_frame_set_locals(locals);
1576     BasicObjectLock* montop = interpreter_frame->interpreter_frame_monitor_begin();
1577     BasicObjectLock* monbot = montop - moncount;
1578     interpreter_frame->interpreter_frame_set_monitor_end(monbot);
1579 
1580     // Set last_sp
1581     intptr_t*  esp = (intptr_t*) monbot -
1582                      tempcount*Interpreter::stackElementWords -
1583                      popframe_extra_args;
1584     interpreter_frame->interpreter_frame_set_last_sp(esp);
1585 
1586     // All frames but the initial (oldest) interpreter frame we fill in have
1587     // a value for sender_sp that allows walking the stack but isn't
1588     // truly correct. Correct the value here.
1589     if (extra_locals != 0 &&
1590         interpreter_frame->sender_sp() ==
1591         interpreter_frame->interpreter_frame_sender_sp()) {
1592       interpreter_frame->set_interpreter_frame_sender_sp(caller->sp() +
1593                                                          extra_locals);
1594     }
1595     *interpreter_frame->interpreter_frame_cache_addr() =
1596       method->constants()->cache();
1597   }
1598   return size;
1599 }
1600 
1601 //-----------------------------------------------------------------------------
1602 // Exceptions
1603 
1604 void TemplateInterpreterGenerator::generate_throw_exception() {
1605   // Entry point in previous activation (i.e., if the caller was
1606   // interpreted)
1607   Interpreter::_rethrow_exception_entry = __ pc();
1608   // Restore sp to interpreter_frame_last_sp even though we are going
1609   // to empty the expression stack for the exception processing.
1610   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
1611   // rax: exception
1612   // rdx: return address/pc that threw exception
1613   __ restore_bcp();    // r13 points to call/send
1614   __ restore_locals();
1615   __ reinit_heapbase();  // restore r12 as heapbase.
1616   // Entry point for exceptions thrown within interpreter code
1617   Interpreter::_throw_exception_entry = __ pc();
1618   // expression stack is undefined here
1619   // rax: exception
1620   // r13: exception bcp
1621   __ verify_oop(rax);
1622   __ mov(c_rarg1, rax);
1623 
1624   // expression stack must be empty before entering the VM in case of
1625   // an exception
1626   __ empty_expression_stack();
1627   // find exception handler address and preserve exception oop
1628   __ call_VM(rdx,
1629              CAST_FROM_FN_PTR(address,
1630                           InterpreterRuntime::exception_handler_for_exception),
1631              c_rarg1);
1632   // rax: exception handler entry point
1633   // rdx: preserved exception oop
1634   // r13: bcp for exception handler
1635   __ push_ptr(rdx); // push exception which is now the only value on the stack
1636   __ jmp(rax); // jump to exception handler (may be _remove_activation_entry!)
1637 
1638   // If the exception is not handled in the current frame the frame is
1639   // removed and the exception is rethrown (i.e. exception
1640   // continuation is _rethrow_exception).
1641   //
1642   // Note: At this point the bci is still the bxi for the instruction
1643   // which caused the exception and the expression stack is
1644   // empty. Thus, for any VM calls at this point, GC will find a legal
1645   // oop map (with empty expression stack).
1646 
1647   // In current activation
1648   // tos: exception
1649   // esi: exception bcp
1650 
1651   //
1652   // JVMTI PopFrame support
1653   //
1654 
1655   Interpreter::_remove_activation_preserving_args_entry = __ pc();
1656   __ empty_expression_stack();
1657   // Set the popframe_processing bit in pending_popframe_condition
1658   // indicating that we are currently handling popframe, so that
1659   // call_VMs that may happen later do not trigger new popframe
1660   // handling cycles.
1661   __ movl(rdx, Address(r15_thread, JavaThread::popframe_condition_offset()));
1662   __ orl(rdx, JavaThread::popframe_processing_bit);
1663   __ movl(Address(r15_thread, JavaThread::popframe_condition_offset()), rdx);
1664 
1665   {
1666     // Check to see whether we are returning to a deoptimized frame.
1667     // (The PopFrame call ensures that the caller of the popped frame is
1668     // either interpreted or compiled and deoptimizes it if compiled.)
1669     // In this case, we can't call dispatch_next() after the frame is
1670     // popped, but instead must save the incoming arguments and restore
1671     // them after deoptimization has occurred.
1672     //
1673     // Note that we don't compare the return PC against the
1674     // deoptimization blob's unpack entry because of the presence of
1675     // adapter frames in C2.
1676     Label caller_not_deoptimized;
1677     __ movptr(c_rarg1, Address(rbp, frame::return_addr_offset * wordSize));
1678     __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1679                                InterpreterRuntime::interpreter_contains), c_rarg1);
1680     __ testl(rax, rax);
1681     __ jcc(Assembler::notZero, caller_not_deoptimized);
1682 
1683     // Compute size of arguments for saving when returning to
1684     // deoptimized caller
1685     __ get_method(rax);
1686     __ load_unsigned_short(rax, Address(rax, in_bytes(methodOopDesc::
1687                                                 size_of_parameters_offset())));
1688     __ shll(rax, Interpreter::logStackElementSize);
1689     __ restore_locals(); // XXX do we need this?
1690     __ subptr(r14, rax);
1691     __ addptr(r14, wordSize);
1692     // Save these arguments
1693     __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1694                                            Deoptimization::
1695                                            popframe_preserve_args),
1696                           r15_thread, rax, r14);
1697 
1698     __ remove_activation(vtos, rdx,
1699                          /* throw_monitor_exception */ false,
1700                          /* install_monitor_exception */ false,
1701                          /* notify_jvmdi */ false);
1702 
1703     // Inform deoptimization that it is responsible for restoring
1704     // these arguments
1705     __ movl(Address(r15_thread, JavaThread::popframe_condition_offset()),
1706             JavaThread::popframe_force_deopt_reexecution_bit);
1707 
1708     // Continue in deoptimization handler
1709     __ jmp(rdx);
1710 
1711     __ bind(caller_not_deoptimized);
1712   }
1713 
1714   __ remove_activation(vtos, rdx, /* rdx result (retaddr) is not used */
1715                        /* throw_monitor_exception */ false,
1716                        /* install_monitor_exception */ false,
1717                        /* notify_jvmdi */ false);
1718 
1719   // Finish with popframe handling
1720   // A previous I2C followed by a deoptimization might have moved the
1721   // outgoing arguments further up the stack. PopFrame expects the
1722   // mutations to those outgoing arguments to be preserved and other
1723   // constraints basically require this frame to look exactly as
1724   // though it had previously invoked an interpreted activation with
1725   // no space between the top of the expression stack (current
1726   // last_sp) and the top of stack. Rather than force deopt to
1727   // maintain this kind of invariant all the time we call a small
1728   // fixup routine to move the mutated arguments onto the top of our
1729   // expression stack if necessary.
1730   __ mov(c_rarg1, rsp);
1731   __ movptr(c_rarg2, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1732   // PC must point into interpreter here
1733   __ set_last_Java_frame(noreg, rbp, __ pc());
1734   __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), r15_thread, c_rarg1, c_rarg2);
1735   __ reset_last_Java_frame(true, true);
1736   // Restore the last_sp and null it out
1737   __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1738   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
1739 
1740   __ restore_bcp();  // XXX do we need this?
1741   __ restore_locals(); // XXX do we need this?
1742   // The method data pointer was incremented already during
1743   // call profiling. We have to restore the mdp for the current bcp.
1744   if (ProfileInterpreter) {
1745     __ set_method_data_pointer_for_bcp();
1746   }
1747 
1748   // Clear the popframe condition flag
1749   __ movl(Address(r15_thread, JavaThread::popframe_condition_offset()),
1750           JavaThread::popframe_inactive);
1751 
1752   __ dispatch_next(vtos);
1753   // end of PopFrame support
1754 
1755   Interpreter::_remove_activation_entry = __ pc();
1756 
1757   // preserve exception over this code sequence
1758   __ pop_ptr(rax);
1759   __ movptr(Address(r15_thread, JavaThread::vm_result_offset()), rax);
1760   // remove the activation (without doing throws on illegalMonitorExceptions)
1761   __ remove_activation(vtos, rdx, false, true, false);
1762   // restore exception
1763   __ movptr(rax, Address(r15_thread, JavaThread::vm_result_offset()));
1764   __ movptr(Address(r15_thread, JavaThread::vm_result_offset()), (int32_t)NULL_WORD);
1765   __ verify_oop(rax);
1766 
1767   // In between activations - previous activation type unknown yet
1768   // compute continuation point - the continuation point expects the
1769   // following registers set up:
1770   //
1771   // rax: exception
1772   // rdx: return address/pc that threw exception
1773   // rsp: expression stack of caller
1774   // rbp: ebp of caller
1775   __ push(rax);                                  // save exception
1776   __ push(rdx);                                  // save return address
1777   __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1778                           SharedRuntime::exception_handler_for_return_address),
1779                         r15_thread, rdx);
1780   __ mov(rbx, rax);                              // save exception handler
1781   __ pop(rdx);                                   // restore return address
1782   __ pop(rax);                                   // restore exception
1783   // Note that an "issuing PC" is actually the next PC after the call
1784   __ jmp(rbx);                                   // jump to exception
1785                                                  // handler of caller
1786 }
1787 
1788 
1789 //
1790 // JVMTI ForceEarlyReturn support
1791 //
1792 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
1793   address entry = __ pc();
1794 
1795   __ restore_bcp();
1796   __ restore_locals();
1797   __ empty_expression_stack();
1798   __ load_earlyret_value(state);
1799 
1800   __ movptr(rdx, Address(r15_thread, JavaThread::jvmti_thread_state_offset()));
1801   Address cond_addr(rdx, JvmtiThreadState::earlyret_state_offset());
1802 
1803   // Clear the earlyret state
1804   __ movl(cond_addr, JvmtiThreadState::earlyret_inactive);
1805 
1806   __ remove_activation(state, rsi,
1807                        false, /* throw_monitor_exception */
1808                        false, /* install_monitor_exception */
1809                        true); /* notify_jvmdi */
1810   __ jmp(rsi);
1811 
1812   return entry;
1813 } // end of ForceEarlyReturn support
1814 
1815 
1816 //-----------------------------------------------------------------------------
1817 // Helper for vtos entry point generation
1818 
1819 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
1820                                                          address& bep,
1821                                                          address& cep,
1822                                                          address& sep,
1823                                                          address& aep,
1824                                                          address& iep,
1825                                                          address& lep,
1826                                                          address& fep,
1827                                                          address& dep,
1828                                                          address& vep) {
1829   assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
1830   Label L;
1831   aep = __ pc();  __ push_ptr();  __ jmp(L);
1832   fep = __ pc();  __ push_f();    __ jmp(L);
1833   dep = __ pc();  __ push_d();    __ jmp(L);
1834   lep = __ pc();  __ push_l();    __ jmp(L);
1835   bep = cep = sep =
1836   iep = __ pc();  __ push_i();
1837   vep = __ pc();
1838   __ bind(L);
1839   generate_and_dispatch(t);
1840 }
1841 
1842 
1843 //-----------------------------------------------------------------------------
1844 // Generation of individual instructions
1845 
1846 // helpers for generate_and_dispatch
1847 
1848 
1849 InterpreterGenerator::InterpreterGenerator(StubQueue* code)
1850   : TemplateInterpreterGenerator(code) {
1851    generate_all(); // down here so it can be "virtual"
1852 }
1853 
1854 //-----------------------------------------------------------------------------
1855 
1856 // Non-product code
1857 #ifndef PRODUCT
1858 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
1859   address entry = __ pc();
1860 
1861   __ push(state);
1862   __ push(c_rarg0);
1863   __ push(c_rarg1);
1864   __ push(c_rarg2);
1865   __ push(c_rarg3);
1866   __ mov(c_rarg2, rax);  // Pass itos
1867 #ifdef _WIN64
1868   __ movflt(xmm3, xmm0); // Pass ftos
1869 #endif
1870   __ call_VM(noreg,
1871              CAST_FROM_FN_PTR(address, SharedRuntime::trace_bytecode),
1872              c_rarg1, c_rarg2, c_rarg3);
1873   __ pop(c_rarg3);
1874   __ pop(c_rarg2);
1875   __ pop(c_rarg1);
1876   __ pop(c_rarg0);
1877   __ pop(state);
1878   __ ret(0);                                   // return from result handler
1879 
1880   return entry;
1881 }
1882 
1883 void TemplateInterpreterGenerator::count_bytecode() {
1884   __ incrementl(ExternalAddress((address) &BytecodeCounter::_counter_value));
1885 }
1886 
1887 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
1888   __ incrementl(ExternalAddress((address) &BytecodeHistogram::_counters[t->bytecode()]));
1889 }
1890 
1891 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
1892   __ mov32(rbx, ExternalAddress((address) &BytecodePairHistogram::_index));
1893   __ shrl(rbx, BytecodePairHistogram::log2_number_of_codes);
1894   __ orl(rbx,
1895          ((int) t->bytecode()) <<
1896          BytecodePairHistogram::log2_number_of_codes);
1897   __ mov32(ExternalAddress((address) &BytecodePairHistogram::_index), rbx);
1898   __ lea(rscratch1, ExternalAddress((address) BytecodePairHistogram::_counters));
1899   __ incrementl(Address(rscratch1, rbx, Address::times_4));
1900 }
1901 
1902 
1903 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
1904   // Call a little run-time stub to avoid blow-up for each bytecode.
1905   // The run-time runtime saves the right registers, depending on
1906   // the tosca in-state for the given template.
1907 
1908   assert(Interpreter::trace_code(t->tos_in()) != NULL,
1909          "entry must have been generated");
1910   __ mov(r12, rsp); // remember sp
1911   __ andptr(rsp, -16); // align stack as required by ABI
1912   __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in())));
1913   __ mov(rsp, r12); // restore sp
1914   __ reinit_heapbase();
1915 }
1916 
1917 
1918 void TemplateInterpreterGenerator::stop_interpreter_at() {
1919   Label L;
1920   __ cmp32(ExternalAddress((address) &BytecodeCounter::_counter_value),
1921            StopInterpreterAt);
1922   __ jcc(Assembler::notEqual, L);
1923   __ int3();
1924   __ bind(L);
1925 }
1926 #endif // !PRODUCT
1927 #endif // ! CC_INTERP