1 /*
   2  * Copyright (c) 2003, 2011, 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, int step) {
 194   address entry = __ pc();
 195 
 196   // Restore stack bottom in case i2c adjusted stack
 197   __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
 198   // and NULL it as marker that esp is now tos until next java call
 199   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
 200 
 201   __ restore_bcp();
 202   __ restore_locals();
 203 
 204   Label L_got_cache, L_giant_index;
 205   if (EnableInvokeDynamic) {
 206     __ cmpb(Address(r13, 0), Bytecodes::_invokedynamic);
 207     __ jcc(Assembler::equal, L_giant_index);
 208   }
 209   __ get_cache_and_index_at_bcp(rbx, rcx, 1, sizeof(u2));
 210   __ bind(L_got_cache);
 211   __ movl(rbx, Address(rbx, rcx,
 212                        Address::times_ptr,
 213                        in_bytes(constantPoolCacheOopDesc::base_offset()) +
 214                        3 * wordSize));
 215   __ andl(rbx, 0xFF);
 216   __ lea(rsp, Address(rsp, rbx, Address::times_8));
 217   __ dispatch_next(state, step);
 218 
 219   // out of the main line of code...
 220   if (EnableInvokeDynamic) {
 221     __ bind(L_giant_index);
 222     __ get_cache_and_index_at_bcp(rbx, rcx, 1, sizeof(u4));
 223     __ jmp(L_got_cache);
 224   }
 225 
 226   return entry;
 227 }
 228 
 229 
 230 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state,
 231                                                                int step) {
 232   address entry = __ pc();
 233   // NULL last_sp until next java call
 234   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
 235   __ restore_bcp();
 236   __ restore_locals();
 237   // handle exceptions
 238   {
 239     Label L;
 240     __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t) NULL_WORD);
 241     __ jcc(Assembler::zero, L);
 242     __ call_VM(noreg,
 243                CAST_FROM_FN_PTR(address,
 244                                 InterpreterRuntime::throw_pending_exception));
 245     __ should_not_reach_here();
 246     __ bind(L);
 247   }
 248   __ dispatch_next(state, step);
 249   return entry;
 250 }
 251 
 252 int AbstractInterpreter::BasicType_as_index(BasicType type) {
 253   int i = 0;
 254   switch (type) {
 255     case T_BOOLEAN: i = 0; break;
 256     case T_CHAR   : i = 1; break;
 257     case T_BYTE   : i = 2; break;
 258     case T_SHORT  : i = 3; break;
 259     case T_INT    : i = 4; break;
 260     case T_LONG   : i = 5; break;
 261     case T_VOID   : i = 6; break;
 262     case T_FLOAT  : i = 7; break;
 263     case T_DOUBLE : i = 8; break;
 264     case T_OBJECT : i = 9; break;
 265     case T_ARRAY  : i = 9; break;
 266     default       : ShouldNotReachHere();
 267   }
 268   assert(0 <= i && i < AbstractInterpreter::number_of_result_handlers,
 269          "index out of bounds");
 270   return i;
 271 }
 272 
 273 
 274 address TemplateInterpreterGenerator::generate_result_handler_for(
 275         BasicType type) {
 276   address entry = __ pc();
 277   switch (type) {
 278   case T_BOOLEAN: __ c2bool(rax);            break;
 279   case T_CHAR   : __ movzwl(rax, rax);       break;
 280   case T_BYTE   : __ sign_extend_byte(rax);  break;
 281   case T_SHORT  : __ sign_extend_short(rax); break;
 282   case T_INT    : /* nothing to do */        break;
 283   case T_LONG   : /* nothing to do */        break;
 284   case T_VOID   : /* nothing to do */        break;
 285   case T_FLOAT  : /* nothing to do */        break;
 286   case T_DOUBLE : /* nothing to do */        break;
 287   case T_OBJECT :
 288     // retrieve result from frame
 289     __ movptr(rax, Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize));
 290     // and verify it
 291     __ verify_oop(rax);
 292     break;
 293   default       : ShouldNotReachHere();
 294   }
 295   __ ret(0);                                   // return from result handler
 296   return entry;
 297 }
 298 
 299 address TemplateInterpreterGenerator::generate_safept_entry_for(
 300         TosState state,
 301         address runtime_entry) {
 302   address entry = __ pc();
 303   __ push(state);
 304   __ call_VM(noreg, runtime_entry);
 305   __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
 306   return entry;
 307 }
 308 
 309 
 310 
 311 // Helpers for commoning out cases in the various type of method entries.
 312 //
 313 
 314 
 315 // increment invocation count & check for overflow
 316 //
 317 // Note: checking for negative value instead of overflow
 318 //       so we have a 'sticky' overflow test
 319 //
 320 // rbx: method
 321 // ecx: invocation counter
 322 //
 323 void InterpreterGenerator::generate_counter_incr(
 324         Label* overflow,
 325         Label* profile_method,
 326         Label* profile_method_continue) {
 327   const Address invocation_counter(rbx, in_bytes(methodOopDesc::invocation_counter_offset()) +
 328                                         in_bytes(InvocationCounter::counter_offset()));
 329   // Note: In tiered we increment either counters in methodOop or in MDO depending if we're profiling or not.
 330   if (TieredCompilation) {
 331     int increment = InvocationCounter::count_increment;
 332     int mask = ((1 << Tier0InvokeNotifyFreqLog)  - 1) << InvocationCounter::count_shift;
 333     Label no_mdo, done;
 334     if (ProfileInterpreter) {
 335       // Are we profiling?
 336       __ movptr(rax, Address(rbx, methodOopDesc::method_data_offset()));
 337       __ testptr(rax, rax);
 338       __ jccb(Assembler::zero, no_mdo);
 339       // Increment counter in the MDO
 340       const Address mdo_invocation_counter(rax, in_bytes(methodDataOopDesc::invocation_counter_offset()) +
 341                                                 in_bytes(InvocationCounter::counter_offset()));
 342       __ increment_mask_and_jump(mdo_invocation_counter, increment, mask, rcx, false, Assembler::zero, overflow);
 343       __ jmpb(done);
 344     }
 345     __ bind(no_mdo);
 346     // Increment counter in methodOop (we don't need to load it, it's in ecx).
 347     __ increment_mask_and_jump(invocation_counter, increment, mask, rcx, true, Assembler::zero, overflow);
 348     __ bind(done);
 349   } else {
 350     const Address backedge_counter(rbx,
 351                                    methodOopDesc::backedge_counter_offset() +
 352                                    InvocationCounter::counter_offset());
 353 
 354     if (ProfileInterpreter) { // %%% Merge this into methodDataOop
 355       __ incrementl(Address(rbx,
 356                             methodOopDesc::interpreter_invocation_counter_offset()));
 357     }
 358     // Update standard invocation counters
 359     __ movl(rax, backedge_counter);   // load backedge counter
 360 
 361     __ incrementl(rcx, InvocationCounter::count_increment);
 362     __ andl(rax, InvocationCounter::count_mask_value); // mask out the status bits
 363 
 364     __ movl(invocation_counter, rcx); // save invocation count
 365     __ addl(rcx, rax);                // add both counters
 366 
 367     // profile_method is non-null only for interpreted method so
 368     // profile_method != NULL == !native_call
 369 
 370     if (ProfileInterpreter && profile_method != NULL) {
 371       // Test to see if we should create a method data oop
 372       __ cmp32(rcx, ExternalAddress((address)&InvocationCounter::InterpreterProfileLimit));
 373       __ jcc(Assembler::less, *profile_method_continue);
 374 
 375       // if no method data exists, go to profile_method
 376       __ test_method_data_pointer(rax, *profile_method);
 377     }
 378 
 379     __ cmp32(rcx, ExternalAddress((address)&InvocationCounter::InterpreterInvocationLimit));
 380     __ jcc(Assembler::aboveEqual, *overflow);
 381   }
 382 }
 383 
 384 void InterpreterGenerator::generate_counter_overflow(Label* do_continue) {
 385 
 386   // Asm interpreter on entry
 387   // r14 - locals
 388   // r13 - bcp
 389   // rbx - method
 390   // edx - cpool --- DOES NOT APPEAR TO BE TRUE
 391   // rbp - interpreter frame
 392 
 393   // On return (i.e. jump to entry_point) [ back to invocation of interpreter ]
 394   // Everything as it was on entry
 395   // rdx is not restored. Doesn't appear to really be set.
 396 
 397   const Address size_of_parameters(rbx,
 398                                    methodOopDesc::size_of_parameters_offset());
 399 
 400   // InterpreterRuntime::frequency_counter_overflow takes two
 401   // arguments, the first (thread) is passed by call_VM, the second
 402   // indicates if the counter overflow occurs at a backwards branch
 403   // (NULL bcp).  We pass zero for it.  The call returns the address
 404   // of the verified entry point for the method or NULL if the
 405   // compilation did not complete (either went background or bailed
 406   // out).
 407   __ movl(c_rarg1, 0);
 408   __ call_VM(noreg,
 409              CAST_FROM_FN_PTR(address,
 410                               InterpreterRuntime::frequency_counter_overflow),
 411              c_rarg1);
 412 
 413   __ movptr(rbx, Address(rbp, method_offset));   // restore methodOop
 414   // Preserve invariant that r13/r14 contain bcp/locals of sender frame
 415   // and jump to the interpreted entry.
 416   __ jmp(*do_continue, relocInfo::none);
 417 }
 418 
 419 // See if we've got enough room on the stack for locals plus overhead.
 420 // The expression stack grows down incrementally, so the normal guard
 421 // page mechanism will work for that.
 422 //
 423 // NOTE: Since the additional locals are also always pushed (wasn't
 424 // obvious in generate_method_entry) so the guard should work for them
 425 // too.
 426 //
 427 // Args:
 428 //      rdx: number of additional locals this frame needs (what we must check)
 429 //      rbx: methodOop
 430 //
 431 // Kills:
 432 //      rax
 433 void InterpreterGenerator::generate_stack_overflow_check(void) {
 434 
 435   // monitor entry size: see picture of stack set
 436   // (generate_method_entry) and frame_amd64.hpp
 437   const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
 438 
 439   // total overhead size: entry_size + (saved rbp through expr stack
 440   // bottom).  be sure to change this if you add/subtract anything
 441   // to/from the overhead area
 442   const int overhead_size =
 443     -(frame::interpreter_frame_initial_sp_offset * wordSize) + entry_size;
 444 
 445   const int page_size = os::vm_page_size();
 446 
 447   Label after_frame_check;
 448 
 449   // see if the frame is greater than one page in size. If so,
 450   // then we need to verify there is enough stack space remaining
 451   // for the additional locals.
 452   __ cmpl(rdx, (page_size - overhead_size) / Interpreter::stackElementSize);
 453   __ jcc(Assembler::belowEqual, after_frame_check);
 454 
 455   // compute rsp as if this were going to be the last frame on
 456   // the stack before the red zone
 457 
 458   const Address stack_base(r15_thread, Thread::stack_base_offset());
 459   const Address stack_size(r15_thread, Thread::stack_size_offset());
 460 
 461   // locals + overhead, in bytes
 462   __ mov(rax, rdx);
 463   __ shlptr(rax, Interpreter::logStackElementSize);  // 2 slots per parameter.
 464   __ addptr(rax, overhead_size);
 465 
 466 #ifdef ASSERT
 467   Label stack_base_okay, stack_size_okay;
 468   // verify that thread stack base is non-zero
 469   __ cmpptr(stack_base, (int32_t)NULL_WORD);
 470   __ jcc(Assembler::notEqual, stack_base_okay);
 471   __ stop("stack base is zero");
 472   __ bind(stack_base_okay);
 473   // verify that thread stack size is non-zero
 474   __ cmpptr(stack_size, 0);
 475   __ jcc(Assembler::notEqual, stack_size_okay);
 476   __ stop("stack size is zero");
 477   __ bind(stack_size_okay);
 478 #endif
 479 
 480   // Add stack base to locals and subtract stack size
 481   __ addptr(rax, stack_base);
 482   __ subptr(rax, stack_size);
 483 
 484   // Use the maximum number of pages we might bang.
 485   const int max_pages = StackShadowPages > (StackRedPages+StackYellowPages) ? StackShadowPages :
 486                                                                               (StackRedPages+StackYellowPages);
 487 
 488   // add in the red and yellow zone sizes
 489   __ addptr(rax, max_pages * page_size);
 490 
 491   // check against the current stack bottom
 492   __ cmpptr(rsp, rax);
 493   __ jcc(Assembler::above, after_frame_check);
 494 
 495   __ pop(rax); // get return address
 496   __ jump(ExternalAddress(Interpreter::throw_StackOverflowError_entry()));
 497 
 498   // all done with frame size check
 499   __ bind(after_frame_check);
 500 }
 501 
 502 // Allocate monitor and lock method (asm interpreter)
 503 //
 504 // Args:
 505 //      rbx: methodOop
 506 //      r14: locals
 507 //
 508 // Kills:
 509 //      rax
 510 //      c_rarg0, c_rarg1, c_rarg2, c_rarg3, ...(param regs)
 511 //      rscratch1, rscratch2 (scratch regs)
 512 void InterpreterGenerator::lock_method(void) {
 513   // synchronize method
 514   const Address access_flags(rbx, methodOopDesc::access_flags_offset());
 515   const Address monitor_block_top(
 516         rbp,
 517         frame::interpreter_frame_monitor_block_top_offset * wordSize);
 518   const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
 519 
 520 #ifdef ASSERT
 521   {
 522     Label L;
 523     __ movl(rax, access_flags);
 524     __ testl(rax, JVM_ACC_SYNCHRONIZED);
 525     __ jcc(Assembler::notZero, L);
 526     __ stop("method doesn't need synchronization");
 527     __ bind(L);
 528   }
 529 #endif // ASSERT
 530 
 531   // get synchronization object
 532   {
 533     const int mirror_offset = klassOopDesc::klass_part_offset_in_bytes() +
 534                               Klass::java_mirror_offset_in_bytes();
 535     Label done;
 536     __ movl(rax, access_flags);
 537     __ testl(rax, JVM_ACC_STATIC);
 538     // get receiver (assume this is frequent case)
 539     __ movptr(rax, Address(r14, Interpreter::local_offset_in_bytes(0)));
 540     __ jcc(Assembler::zero, done);
 541     __ movptr(rax, Address(rbx, methodOopDesc::constants_offset()));
 542     __ movptr(rax, Address(rax,
 543                            constantPoolOopDesc::pool_holder_offset_in_bytes()));
 544     __ movptr(rax, Address(rax, mirror_offset));
 545 
 546 #ifdef ASSERT
 547     {
 548       Label L;
 549       __ testptr(rax, rax);
 550       __ jcc(Assembler::notZero, L);
 551       __ stop("synchronization object is NULL");
 552       __ bind(L);
 553     }
 554 #endif // ASSERT
 555 
 556     __ bind(done);
 557   }
 558 
 559   // add space for monitor & lock
 560   __ subptr(rsp, entry_size); // add space for a monitor entry
 561   __ movptr(monitor_block_top, rsp);  // set new monitor block top
 562   // store object
 563   __ movptr(Address(rsp, BasicObjectLock::obj_offset_in_bytes()), rax);
 564   __ movptr(c_rarg1, rsp); // object address
 565   __ lock_object(c_rarg1);
 566 }
 567 
 568 // Generate a fixed interpreter frame. This is identical setup for
 569 // interpreted methods and for native methods hence the shared code.
 570 //
 571 // Args:
 572 //      rax: return address
 573 //      rbx: methodOop
 574 //      r14: pointer to locals
 575 //      r13: sender sp
 576 //      rdx: cp cache
 577 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
 578   // initialize fixed part of activation frame
 579   __ push(rax);        // save return address
 580   __ enter();          // save old & set new rbp
 581   __ push(r13);        // set sender sp
 582   __ push((int)NULL_WORD); // leave last_sp as null
 583   __ movptr(r13, Address(rbx, methodOopDesc::const_offset()));      // get constMethodOop
 584   __ lea(r13, Address(r13, constMethodOopDesc::codes_offset())); // get codebase
 585   __ push(rbx);        // save methodOop
 586   if (ProfileInterpreter) {
 587     Label method_data_continue;
 588     __ movptr(rdx, Address(rbx, in_bytes(methodOopDesc::method_data_offset())));
 589     __ testptr(rdx, rdx);
 590     __ jcc(Assembler::zero, method_data_continue);
 591     __ addptr(rdx, in_bytes(methodDataOopDesc::data_offset()));
 592     __ bind(method_data_continue);
 593     __ push(rdx);      // set the mdp (method data pointer)
 594   } else {
 595     __ push(0);
 596   }
 597 
 598   __ movptr(rdx, Address(rbx, methodOopDesc::constants_offset()));
 599   __ movptr(rdx, Address(rdx, constantPoolOopDesc::cache_offset_in_bytes()));
 600   __ push(rdx); // set constant pool cache
 601   __ push(r14); // set locals pointer
 602   if (native_call) {
 603     __ push(0); // no bcp
 604   } else {
 605     __ push(r13); // set bcp
 606   }
 607   __ push(0); // reserve word for pointer to expression stack bottom
 608   __ movptr(Address(rsp, 0), rsp); // set expression stack bottom
 609 }
 610 
 611 // End of helpers
 612 
 613 // Various method entries
 614 //------------------------------------------------------------------------------------------------------------------------
 615 //
 616 //
 617 
 618 // Call an accessor method (assuming it is resolved, otherwise drop
 619 // into vanilla (slow path) entry
 620 address InterpreterGenerator::generate_accessor_entry(void) {
 621   // rbx: methodOop
 622 
 623   // r13: senderSP must preserver for slow path, set SP to it on fast path
 624 
 625   address entry_point = __ pc();
 626   Label xreturn_path;
 627 
 628   // do fastpath for resolved accessor methods
 629   if (UseFastAccessorMethods) {
 630     // Code: _aload_0, _(i|a)getfield, _(i|a)return or any rewrites
 631     //       thereof; parameter size = 1
 632     // Note: We can only use this code if the getfield has been resolved
 633     //       and if we don't have a null-pointer exception => check for
 634     //       these conditions first and use slow path if necessary.
 635     Label slow_path;
 636     // If we need a safepoint check, generate full interpreter entry.
 637     __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
 638              SafepointSynchronize::_not_synchronized);
 639 
 640     __ jcc(Assembler::notEqual, slow_path);
 641     // rbx: method
 642     __ movptr(rax, Address(rsp, wordSize));
 643 
 644     // check if local 0 != NULL and read field
 645     __ testptr(rax, rax);
 646     __ jcc(Assembler::zero, slow_path);
 647 
 648     __ movptr(rdi, Address(rbx, methodOopDesc::constants_offset()));
 649     // read first instruction word and extract bytecode @ 1 and index @ 2
 650     __ movptr(rdx, Address(rbx, methodOopDesc::const_offset()));
 651     __ movl(rdx, Address(rdx, constMethodOopDesc::codes_offset()));
 652     // Shift codes right to get the index on the right.
 653     // The bytecode fetched looks like <index><0xb4><0x2a>
 654     __ shrl(rdx, 2 * BitsPerByte);
 655     __ shll(rdx, exact_log2(in_words(ConstantPoolCacheEntry::size())));
 656     __ movptr(rdi, Address(rdi, constantPoolOopDesc::cache_offset_in_bytes()));
 657 
 658     // rax: local 0
 659     // rbx: method
 660     // rdx: constant pool cache index
 661     // rdi: constant pool cache
 662 
 663     // check if getfield has been resolved and read constant pool cache entry
 664     // check the validity of the cache entry by testing whether _indices field
 665     // contains Bytecode::_getfield in b1 byte.
 666     assert(in_words(ConstantPoolCacheEntry::size()) == 4,
 667            "adjust shift below");
 668     __ movl(rcx,
 669             Address(rdi,
 670                     rdx,
 671                     Address::times_8,
 672                     constantPoolCacheOopDesc::base_offset() +
 673                     ConstantPoolCacheEntry::indices_offset()));
 674     __ shrl(rcx, 2 * BitsPerByte);
 675     __ andl(rcx, 0xFF);
 676     __ cmpl(rcx, Bytecodes::_getfield);
 677     __ jcc(Assembler::notEqual, slow_path);
 678 
 679     // Note: constant pool entry is not valid before bytecode is resolved
 680     __ movptr(rcx,
 681               Address(rdi,
 682                       rdx,
 683                       Address::times_8,
 684                       constantPoolCacheOopDesc::base_offset() +
 685                       ConstantPoolCacheEntry::f2_offset()));
 686     // edx: flags
 687     __ movl(rdx,
 688             Address(rdi,
 689                     rdx,
 690                     Address::times_8,
 691                     constantPoolCacheOopDesc::base_offset() +
 692                     ConstantPoolCacheEntry::flags_offset()));
 693 
 694     Label notObj, notInt, notByte, notShort;
 695     const Address field_address(rax, rcx, Address::times_1);
 696 
 697     // Need to differentiate between igetfield, agetfield, bgetfield etc.
 698     // because they are different sizes.
 699     // Use the type from the constant pool cache
 700     __ shrl(rdx, ConstantPoolCacheEntry::tosBits);
 701     // Make sure we don't need to mask edx for tosBits after the above shift
 702     ConstantPoolCacheEntry::verify_tosBits();
 703 
 704     __ cmpl(rdx, atos);
 705     __ jcc(Assembler::notEqual, notObj);
 706     // atos
 707     __ load_heap_oop(rax, field_address);
 708     __ jmp(xreturn_path);
 709 
 710     __ bind(notObj);
 711     __ cmpl(rdx, itos);
 712     __ jcc(Assembler::notEqual, notInt);
 713     // itos
 714     __ movl(rax, field_address);
 715     __ jmp(xreturn_path);
 716 
 717     __ bind(notInt);
 718     __ cmpl(rdx, btos);
 719     __ jcc(Assembler::notEqual, notByte);
 720     // btos
 721     __ load_signed_byte(rax, field_address);
 722     __ jmp(xreturn_path);
 723 
 724     __ bind(notByte);
 725     __ cmpl(rdx, stos);
 726     __ jcc(Assembler::notEqual, notShort);
 727     // stos
 728     __ load_signed_short(rax, field_address);
 729     __ jmp(xreturn_path);
 730 
 731     __ bind(notShort);
 732 #ifdef ASSERT
 733     Label okay;
 734     __ cmpl(rdx, ctos);
 735     __ jcc(Assembler::equal, okay);
 736     __ stop("what type is this?");
 737     __ bind(okay);
 738 #endif
 739     // ctos
 740     __ load_unsigned_short(rax, field_address);
 741 
 742     __ bind(xreturn_path);
 743 
 744     // _ireturn/_areturn
 745     __ pop(rdi);
 746     __ mov(rsp, r13);
 747     __ jmp(rdi);
 748     __ ret(0);
 749 
 750     // generate a vanilla interpreter entry as the slow path
 751     __ bind(slow_path);
 752     (void) generate_normal_entry(false);
 753   } else {
 754     (void) generate_normal_entry(false);
 755   }
 756 
 757   return entry_point;
 758 }
 759 
 760 // Interpreter stub for calling a native method. (asm interpreter)
 761 // This sets up a somewhat different looking stack for calling the
 762 // native method than the typical interpreter frame setup.
 763 address InterpreterGenerator::generate_native_entry(bool synchronized) {
 764   // determine code generation flags
 765   bool inc_counter  = UseCompiler || CountCompiledCalls;
 766 
 767   // rbx: methodOop
 768   // r13: sender sp
 769 
 770   address entry_point = __ pc();
 771 
 772   const Address size_of_parameters(rbx, methodOopDesc::
 773                                         size_of_parameters_offset());
 774   const Address invocation_counter(rbx, methodOopDesc::
 775                                         invocation_counter_offset() +
 776                                         InvocationCounter::counter_offset());
 777   const Address access_flags      (rbx, methodOopDesc::access_flags_offset());
 778 
 779   // get parameter size (always needed)
 780   __ load_unsigned_short(rcx, size_of_parameters);
 781 
 782   // native calls don't need the stack size check since they have no
 783   // expression stack and the arguments are already on the stack and
 784   // we only add a handful of words to the stack
 785 
 786   // rbx: methodOop
 787   // rcx: size of parameters
 788   // r13: sender sp
 789   __ pop(rax);                                       // get return address
 790 
 791   // for natives the size of locals is zero
 792 
 793   // compute beginning of parameters (r14)
 794   __ lea(r14, Address(rsp, rcx, Address::times_8, -wordSize));
 795 
 796   // add 2 zero-initialized slots for native calls
 797   // initialize result_handler slot
 798   __ push((int) NULL_WORD);
 799   // slot for oop temp
 800   // (static native method holder mirror/jni oop result)
 801   __ push((int) NULL_WORD);
 802 
 803   if (inc_counter) {
 804     __ movl(rcx, invocation_counter);  // (pre-)fetch invocation count
 805   }
 806 
 807   // initialize fixed part of activation frame
 808   generate_fixed_frame(true);
 809 
 810   // make sure method is native & not abstract
 811 #ifdef ASSERT
 812   __ movl(rax, access_flags);
 813   {
 814     Label L;
 815     __ testl(rax, JVM_ACC_NATIVE);
 816     __ jcc(Assembler::notZero, L);
 817     __ stop("tried to execute non-native method as native");
 818     __ bind(L);
 819   }
 820   {
 821     Label L;
 822     __ testl(rax, JVM_ACC_ABSTRACT);
 823     __ jcc(Assembler::zero, L);
 824     __ stop("tried to execute abstract method in interpreter");
 825     __ bind(L);
 826   }
 827 #endif
 828 
 829   // Since at this point in the method invocation the exception handler
 830   // would try to exit the monitor of synchronized methods which hasn't
 831   // been entered yet, we set the thread local variable
 832   // _do_not_unlock_if_synchronized to true. The remove_activation will
 833   // check this flag.
 834 
 835   const Address do_not_unlock_if_synchronized(r15_thread,
 836         in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
 837   __ movbool(do_not_unlock_if_synchronized, true);
 838 
 839   // increment invocation count & check for overflow
 840   Label invocation_counter_overflow;
 841   if (inc_counter) {
 842     generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
 843   }
 844 
 845   Label continue_after_compile;
 846   __ bind(continue_after_compile);
 847 
 848   bang_stack_shadow_pages(true);
 849 
 850   // reset the _do_not_unlock_if_synchronized flag
 851   __ movbool(do_not_unlock_if_synchronized, false);
 852 
 853   // check for synchronized methods
 854   // Must happen AFTER invocation_counter check and stack overflow check,
 855   // so method is not locked if overflows.
 856   if (synchronized) {
 857     lock_method();
 858   } else {
 859     // no synchronization necessary
 860 #ifdef ASSERT
 861     {
 862       Label L;
 863       __ movl(rax, access_flags);
 864       __ testl(rax, JVM_ACC_SYNCHRONIZED);
 865       __ jcc(Assembler::zero, L);
 866       __ stop("method needs synchronization");
 867       __ bind(L);
 868     }
 869 #endif
 870   }
 871 
 872   // start execution
 873 #ifdef ASSERT
 874   {
 875     Label L;
 876     const Address monitor_block_top(rbp,
 877                  frame::interpreter_frame_monitor_block_top_offset * wordSize);
 878     __ movptr(rax, monitor_block_top);
 879     __ cmpptr(rax, rsp);
 880     __ jcc(Assembler::equal, L);
 881     __ stop("broken stack frame setup in interpreter");
 882     __ bind(L);
 883   }
 884 #endif
 885 
 886   // jvmti support
 887   __ notify_method_entry();
 888 
 889   // work registers
 890   const Register method = rbx;
 891   const Register t      = r11;
 892 
 893   // allocate space for parameters
 894   __ get_method(method);
 895   __ verify_oop(method);
 896   __ load_unsigned_short(t,
 897                          Address(method,
 898                                  methodOopDesc::size_of_parameters_offset()));
 899   __ shll(t, Interpreter::logStackElementSize);
 900 
 901   __ subptr(rsp, t);
 902   __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
 903   __ andptr(rsp, -16); // must be 16 byte boundary (see amd64 ABI)
 904 
 905   // get signature handler
 906   {
 907     Label L;
 908     __ movptr(t, Address(method, methodOopDesc::signature_handler_offset()));
 909     __ testptr(t, t);
 910     __ jcc(Assembler::notZero, L);
 911     __ call_VM(noreg,
 912                CAST_FROM_FN_PTR(address,
 913                                 InterpreterRuntime::prepare_native_call),
 914                method);
 915     __ get_method(method);
 916     __ movptr(t, Address(method, methodOopDesc::signature_handler_offset()));
 917     __ bind(L);
 918   }
 919 
 920   // call signature handler
 921   assert(InterpreterRuntime::SignatureHandlerGenerator::from() == r14,
 922          "adjust this code");
 923   assert(InterpreterRuntime::SignatureHandlerGenerator::to() == rsp,
 924          "adjust this code");
 925   assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == rscratch1,
 926           "adjust this code");
 927 
 928   // The generated handlers do not touch RBX (the method oop).
 929   // However, large signatures cannot be cached and are generated
 930   // each time here.  The slow-path generator can do a GC on return,
 931   // so we must reload it after the call.
 932   __ call(t);
 933   __ get_method(method);        // slow path can do a GC, reload RBX
 934 
 935 
 936   // result handler is in rax
 937   // set result handler
 938   __ movptr(Address(rbp,
 939                     (frame::interpreter_frame_result_handler_offset) * wordSize),
 940             rax);
 941 
 942   // pass mirror handle if static call
 943   {
 944     Label L;
 945     const int mirror_offset = klassOopDesc::klass_part_offset_in_bytes() +
 946                               Klass::java_mirror_offset_in_bytes();
 947     __ movl(t, Address(method, methodOopDesc::access_flags_offset()));
 948     __ testl(t, JVM_ACC_STATIC);
 949     __ jcc(Assembler::zero, L);
 950     // get mirror
 951     __ movptr(t, Address(method, methodOopDesc::constants_offset()));
 952     __ movptr(t, Address(t, constantPoolOopDesc::pool_holder_offset_in_bytes()));
 953     __ movptr(t, Address(t, mirror_offset));
 954     // copy mirror into activation frame
 955     __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize),
 956             t);
 957     // pass handle to mirror
 958     __ lea(c_rarg1,
 959            Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize));
 960     __ bind(L);
 961   }
 962 
 963   // get native function entry point
 964   {
 965     Label L;
 966     __ movptr(rax, Address(method, methodOopDesc::native_function_offset()));
 967     ExternalAddress unsatisfied(SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
 968     __ movptr(rscratch2, unsatisfied.addr());
 969     __ cmpptr(rax, rscratch2);
 970     __ jcc(Assembler::notEqual, L);
 971     __ call_VM(noreg,
 972                CAST_FROM_FN_PTR(address,
 973                                 InterpreterRuntime::prepare_native_call),
 974                method);
 975     __ get_method(method);
 976     __ verify_oop(method);
 977     __ movptr(rax, Address(method, methodOopDesc::native_function_offset()));
 978     __ bind(L);
 979   }
 980 
 981   // pass JNIEnv
 982   __ lea(c_rarg0, Address(r15_thread, JavaThread::jni_environment_offset()));
 983 
 984   // It is enough that the pc() points into the right code
 985   // segment. It does not have to be the correct return pc.
 986   __ set_last_Java_frame(rsp, rbp, (address) __ pc());
 987 
 988   // change thread state
 989 #ifdef ASSERT
 990   {
 991     Label L;
 992     __ movl(t, Address(r15_thread, JavaThread::thread_state_offset()));
 993     __ cmpl(t, _thread_in_Java);
 994     __ jcc(Assembler::equal, L);
 995     __ stop("Wrong thread state in native stub");
 996     __ bind(L);
 997   }
 998 #endif
 999 
1000   // Change state to native
1001 
1002   __ movl(Address(r15_thread, JavaThread::thread_state_offset()),
1003           _thread_in_native);
1004 
1005   // Call the native method.
1006   __ call(rax);
1007   // result potentially in rax or xmm0
1008 
1009   // Depending on runtime options, either restore the MXCSR
1010   // register after returning from the JNI Call or verify that
1011   // it wasn't changed during -Xcheck:jni.
1012   if (RestoreMXCSROnJNICalls) {
1013     __ ldmxcsr(ExternalAddress(StubRoutines::x86::mxcsr_std()));
1014   }
1015   else if (CheckJNICalls) {
1016     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::x86::verify_mxcsr_entry())));
1017   }
1018 
1019   // NOTE: The order of these pushes is known to frame::interpreter_frame_result
1020   // in order to extract the result of a method call. If the order of these
1021   // pushes change or anything else is added to the stack then the code in
1022   // interpreter_frame_result must also change.
1023 
1024   __ push(dtos);
1025   __ push(ltos);
1026 
1027   // change thread state
1028   __ movl(Address(r15_thread, JavaThread::thread_state_offset()),
1029           _thread_in_native_trans);
1030 
1031   if (os::is_MP()) {
1032     if (UseMembar) {
1033       // Force this write out before the read below
1034       __ membar(Assembler::Membar_mask_bits(
1035            Assembler::LoadLoad | Assembler::LoadStore |
1036            Assembler::StoreLoad | Assembler::StoreStore));
1037     } else {
1038       // Write serialization page so VM thread can do a pseudo remote membar.
1039       // We use the current thread pointer to calculate a thread specific
1040       // offset to write to within the page. This minimizes bus traffic
1041       // due to cache line collision.
1042       __ serialize_memory(r15_thread, rscratch2);
1043     }
1044   }
1045 
1046   // check for safepoint operation in progress and/or pending suspend requests
1047   {
1048     Label Continue;
1049     __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
1050              SafepointSynchronize::_not_synchronized);
1051 
1052     Label L;
1053     __ jcc(Assembler::notEqual, L);
1054     __ cmpl(Address(r15_thread, JavaThread::suspend_flags_offset()), 0);
1055     __ jcc(Assembler::equal, Continue);
1056     __ bind(L);
1057 
1058     // Don't use call_VM as it will see a possible pending exception
1059     // and forward it and never return here preventing us from
1060     // clearing _last_native_pc down below.  Also can't use
1061     // call_VM_leaf either as it will check to see if r13 & r14 are
1062     // preserved and correspond to the bcp/locals pointers. So we do a
1063     // runtime call by hand.
1064     //
1065     __ mov(c_rarg0, r15_thread);
1066     __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1067     __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1068     __ andptr(rsp, -16); // align stack as required by ABI
1069     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans)));
1070     __ mov(rsp, r12); // restore sp
1071     __ reinit_heapbase();
1072     __ bind(Continue);
1073   }
1074 
1075   // change thread state
1076   __ movl(Address(r15_thread, JavaThread::thread_state_offset()), _thread_in_Java);
1077 
1078   // reset_last_Java_frame
1079   __ reset_last_Java_frame(true, true);
1080 
1081   // reset handle block
1082   __ movptr(t, Address(r15_thread, JavaThread::active_handles_offset()));
1083   __ movptr(Address(t, JNIHandleBlock::top_offset_in_bytes()), (int32_t)NULL_WORD);
1084 
1085   // If result is an oop unbox and store it in frame where gc will see it
1086   // and result handler will pick it up
1087 
1088   {
1089     Label no_oop, store_result;
1090     __ lea(t, ExternalAddress(AbstractInterpreter::result_handler(T_OBJECT)));
1091     __ cmpptr(t, Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize));
1092     __ jcc(Assembler::notEqual, no_oop);
1093     // retrieve result
1094     __ pop(ltos);
1095     __ testptr(rax, rax);
1096     __ jcc(Assembler::zero, store_result);
1097     __ movptr(rax, Address(rax, 0));
1098     __ bind(store_result);
1099     __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize), rax);
1100     // keep stack depth as expected by pushing oop which will eventually be discarde
1101     __ push(ltos);
1102     __ bind(no_oop);
1103   }
1104 
1105 
1106   {
1107     Label no_reguard;
1108     __ cmpl(Address(r15_thread, JavaThread::stack_guard_state_offset()),
1109             JavaThread::stack_guard_yellow_disabled);
1110     __ jcc(Assembler::notEqual, no_reguard);
1111 
1112     __ pusha(); // XXX only save smashed registers
1113     __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1114     __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1115     __ andptr(rsp, -16); // align stack as required by ABI
1116     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
1117     __ mov(rsp, r12); // restore sp
1118     __ popa(); // XXX only restore smashed registers
1119     __ reinit_heapbase();
1120 
1121     __ bind(no_reguard);
1122   }
1123 
1124 
1125   // The method register is junk from after the thread_in_native transition
1126   // until here.  Also can't call_VM until the bcp has been
1127   // restored.  Need bcp for throwing exception below so get it now.
1128   __ get_method(method);
1129   __ verify_oop(method);
1130 
1131   // restore r13 to have legal interpreter frame, i.e., bci == 0 <=>
1132   // r13 == code_base()
1133   __ movptr(r13, Address(method, methodOopDesc::const_offset()));   // get constMethodOop
1134   __ lea(r13, Address(r13, constMethodOopDesc::codes_offset()));    // get codebase
1135   // handle exceptions (exception handling will handle unlocking!)
1136   {
1137     Label L;
1138     __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t) NULL_WORD);
1139     __ jcc(Assembler::zero, L);
1140     // Note: At some point we may want to unify this with the code
1141     // used in call_VM_base(); i.e., we should use the
1142     // StubRoutines::forward_exception code. For now this doesn't work
1143     // here because the rsp is not correctly set at this point.
1144     __ MacroAssembler::call_VM(noreg,
1145                                CAST_FROM_FN_PTR(address,
1146                                InterpreterRuntime::throw_pending_exception));
1147     __ should_not_reach_here();
1148     __ bind(L);
1149   }
1150 
1151   // do unlocking if necessary
1152   {
1153     Label L;
1154     __ movl(t, Address(method, methodOopDesc::access_flags_offset()));
1155     __ testl(t, JVM_ACC_SYNCHRONIZED);
1156     __ jcc(Assembler::zero, L);
1157     // the code below should be shared with interpreter macro
1158     // assembler implementation
1159     {
1160       Label unlock;
1161       // BasicObjectLock will be first in list, since this is a
1162       // synchronized method. However, need to check that the object
1163       // has not been unlocked by an explicit monitorexit bytecode.
1164       const Address monitor(rbp,
1165                             (intptr_t)(frame::interpreter_frame_initial_sp_offset *
1166                                        wordSize - sizeof(BasicObjectLock)));
1167 
1168       // monitor expect in c_rarg1 for slow unlock path
1169       __ lea(c_rarg1, monitor); // address of first monitor
1170 
1171       __ movptr(t, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
1172       __ testptr(t, t);
1173       __ jcc(Assembler::notZero, unlock);
1174 
1175       // Entry already unlocked, need to throw exception
1176       __ MacroAssembler::call_VM(noreg,
1177                                  CAST_FROM_FN_PTR(address,
1178                    InterpreterRuntime::throw_illegal_monitor_state_exception));
1179       __ should_not_reach_here();
1180 
1181       __ bind(unlock);
1182       __ unlock_object(c_rarg1);
1183     }
1184     __ bind(L);
1185   }
1186 
1187   // jvmti support
1188   // Note: This must happen _after_ handling/throwing any exceptions since
1189   //       the exception handler code notifies the runtime of method exits
1190   //       too. If this happens before, method entry/exit notifications are
1191   //       not properly paired (was bug - gri 11/22/99).
1192   __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI);
1193 
1194   // restore potential result in edx:eax, call result handler to
1195   // restore potential result in ST0 & handle result
1196 
1197   __ pop(ltos);
1198   __ pop(dtos);
1199 
1200   __ movptr(t, Address(rbp,
1201                        (frame::interpreter_frame_result_handler_offset) * wordSize));
1202   __ call(t);
1203 
1204   // remove activation
1205   __ movptr(t, Address(rbp,
1206                        frame::interpreter_frame_sender_sp_offset *
1207                        wordSize)); // get sender sp
1208   __ leave();                                // remove frame anchor
1209   __ pop(rdi);                               // get return address
1210   __ mov(rsp, t);                            // set sp to sender sp
1211   __ jmp(rdi);
1212 
1213   if (inc_counter) {
1214     // Handle overflow of counter and compile method
1215     __ bind(invocation_counter_overflow);
1216     generate_counter_overflow(&continue_after_compile);
1217   }
1218 
1219   return entry_point;
1220 }
1221 
1222 //
1223 // Generic interpreted method entry to (asm) interpreter
1224 //
1225 address InterpreterGenerator::generate_normal_entry(bool synchronized) {
1226   // determine code generation flags
1227   bool inc_counter  = UseCompiler || CountCompiledCalls;
1228 
1229   // ebx: methodOop
1230   // r13: sender sp
1231   address entry_point = __ pc();
1232 
1233   const Address size_of_parameters(rbx,
1234                                    methodOopDesc::size_of_parameters_offset());
1235   const Address size_of_locals(rbx, methodOopDesc::size_of_locals_offset());
1236   const Address invocation_counter(rbx,
1237                                    methodOopDesc::invocation_counter_offset() +
1238                                    InvocationCounter::counter_offset());
1239   const Address access_flags(rbx, methodOopDesc::access_flags_offset());
1240 
1241   // get parameter size (always needed)
1242   __ load_unsigned_short(rcx, size_of_parameters);
1243 
1244   // rbx: methodOop
1245   // rcx: size of parameters
1246   // r13: sender_sp (could differ from sp+wordSize if we were called via c2i )
1247 
1248   __ load_unsigned_short(rdx, size_of_locals); // get size of locals in words
1249   __ subl(rdx, rcx); // rdx = no. of additional locals
1250 
1251   // YYY
1252 //   __ incrementl(rdx);
1253 //   __ andl(rdx, -2);
1254 
1255   // see if we've got enough room on the stack for locals plus overhead.
1256   generate_stack_overflow_check();
1257 
1258   // get return address
1259   __ pop(rax);
1260 
1261   // compute beginning of parameters (r14)
1262   __ lea(r14, Address(rsp, rcx, Address::times_8, -wordSize));
1263 
1264   // rdx - # of additional locals
1265   // allocate space for locals
1266   // explicitly initialize locals
1267   {
1268     Label exit, loop;
1269     __ testl(rdx, rdx);
1270     __ jcc(Assembler::lessEqual, exit); // do nothing if rdx <= 0
1271     __ bind(loop);
1272     __ push((int) NULL_WORD); // initialize local variables
1273     __ decrementl(rdx); // until everything initialized
1274     __ jcc(Assembler::greater, loop);
1275     __ bind(exit);
1276   }
1277 
1278   // (pre-)fetch invocation count
1279   if (inc_counter) {
1280     __ movl(rcx, invocation_counter);
1281   }
1282   // initialize fixed part of activation frame
1283   generate_fixed_frame(false);
1284 
1285   // make sure method is not native & not abstract
1286 #ifdef ASSERT
1287   __ movl(rax, access_flags);
1288   {
1289     Label L;
1290     __ testl(rax, JVM_ACC_NATIVE);
1291     __ jcc(Assembler::zero, L);
1292     __ stop("tried to execute native method as non-native");
1293     __ bind(L);
1294   }
1295   {
1296     Label L;
1297     __ testl(rax, JVM_ACC_ABSTRACT);
1298     __ jcc(Assembler::zero, L);
1299     __ stop("tried to execute abstract method in interpreter");
1300     __ bind(L);
1301   }
1302 #endif
1303 
1304   // Since at this point in the method invocation the exception
1305   // handler would try to exit the monitor of synchronized methods
1306   // which hasn't been entered yet, we set the thread local variable
1307   // _do_not_unlock_if_synchronized to true. The remove_activation
1308   // will check this flag.
1309 
1310   const Address do_not_unlock_if_synchronized(r15_thread,
1311         in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1312   __ movbool(do_not_unlock_if_synchronized, true);
1313 
1314   // increment invocation count & check for overflow
1315   Label invocation_counter_overflow;
1316   Label profile_method;
1317   Label profile_method_continue;
1318   if (inc_counter) {
1319     generate_counter_incr(&invocation_counter_overflow,
1320                           &profile_method,
1321                           &profile_method_continue);
1322     if (ProfileInterpreter) {
1323       __ bind(profile_method_continue);
1324     }
1325   }
1326 
1327   Label continue_after_compile;
1328   __ bind(continue_after_compile);
1329 
1330   // check for synchronized interpreted methods
1331   bang_stack_shadow_pages(false);
1332 
1333   // reset the _do_not_unlock_if_synchronized flag
1334   __ movbool(do_not_unlock_if_synchronized, false);
1335 
1336   // check for synchronized methods
1337   // Must happen AFTER invocation_counter check and stack overflow check,
1338   // so method is not locked if overflows.
1339   if (synchronized) {
1340     // Allocate monitor and lock method
1341     lock_method();
1342   } else {
1343     // no synchronization necessary
1344 #ifdef ASSERT
1345     {
1346       Label L;
1347       __ movl(rax, access_flags);
1348       __ testl(rax, JVM_ACC_SYNCHRONIZED);
1349       __ jcc(Assembler::zero, L);
1350       __ stop("method needs synchronization");
1351       __ bind(L);
1352     }
1353 #endif
1354   }
1355 
1356   // start execution
1357 #ifdef ASSERT
1358   {
1359     Label L;
1360      const Address monitor_block_top (rbp,
1361                  frame::interpreter_frame_monitor_block_top_offset * wordSize);
1362     __ movptr(rax, monitor_block_top);
1363     __ cmpptr(rax, rsp);
1364     __ jcc(Assembler::equal, L);
1365     __ stop("broken stack frame setup in interpreter");
1366     __ bind(L);
1367   }
1368 #endif
1369 
1370   // jvmti support
1371   __ notify_method_entry();
1372 
1373   __ dispatch_next(vtos);
1374 
1375   // invocation counter overflow
1376   if (inc_counter) {
1377     if (ProfileInterpreter) {
1378       // We have decided to profile this method in the interpreter
1379       __ bind(profile_method);
1380       __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1381       __ set_method_data_pointer_for_bcp();
1382       __ get_method(rbx);
1383       __ jmp(profile_method_continue);
1384     }
1385     // Handle overflow of counter and compile method
1386     __ bind(invocation_counter_overflow);
1387     generate_counter_overflow(&continue_after_compile);
1388   }
1389 
1390   return entry_point;
1391 }
1392 
1393 // Entry points
1394 //
1395 // Here we generate the various kind of entries into the interpreter.
1396 // The two main entry type are generic bytecode methods and native
1397 // call method.  These both come in synchronized and non-synchronized
1398 // versions but the frame layout they create is very similar. The
1399 // other method entry types are really just special purpose entries
1400 // that are really entry and interpretation all in one. These are for
1401 // trivial methods like accessor, empty, or special math methods.
1402 //
1403 // When control flow reaches any of the entry types for the interpreter
1404 // the following holds ->
1405 //
1406 // Arguments:
1407 //
1408 // rbx: methodOop
1409 //
1410 // Stack layout immediately at entry
1411 //
1412 // [ return address     ] <--- rsp
1413 // [ parameter n        ]
1414 //   ...
1415 // [ parameter 1        ]
1416 // [ expression stack   ] (caller's java expression stack)
1417 
1418 // Assuming that we don't go to one of the trivial specialized entries
1419 // the stack will look like below when we are ready to execute the
1420 // first bytecode (or call the native routine). The register usage
1421 // will be as the template based interpreter expects (see
1422 // interpreter_amd64.hpp).
1423 //
1424 // local variables follow incoming parameters immediately; i.e.
1425 // the return address is moved to the end of the locals).
1426 //
1427 // [ monitor entry      ] <--- rsp
1428 //   ...
1429 // [ monitor entry      ]
1430 // [ expr. stack bottom ]
1431 // [ saved r13          ]
1432 // [ current r14        ]
1433 // [ methodOop          ]
1434 // [ saved ebp          ] <--- rbp
1435 // [ return address     ]
1436 // [ local variable m   ]
1437 //   ...
1438 // [ local variable 1   ]
1439 // [ parameter n        ]
1440 //   ...
1441 // [ parameter 1        ] <--- r14
1442 
1443 address AbstractInterpreterGenerator::generate_method_entry(
1444                                         AbstractInterpreter::MethodKind kind) {
1445   // determine code generation flags
1446   bool synchronized = false;
1447   address entry_point = NULL;
1448 
1449   switch (kind) {
1450   case Interpreter::zerolocals             :                                                                             break;
1451   case Interpreter::zerolocals_synchronized: synchronized = true;                                                        break;
1452   case Interpreter::native                 : entry_point = ((InterpreterGenerator*) this)->generate_native_entry(false); break;
1453   case Interpreter::native_synchronized    : entry_point = ((InterpreterGenerator*) this)->generate_native_entry(true);  break;
1454   case Interpreter::empty                  : entry_point = ((InterpreterGenerator*) this)->generate_empty_entry();       break;
1455   case Interpreter::accessor               : entry_point = ((InterpreterGenerator*) this)->generate_accessor_entry();    break;
1456   case Interpreter::abstract               : entry_point = ((InterpreterGenerator*) this)->generate_abstract_entry();    break;
1457   case Interpreter::method_handle          : entry_point = ((InterpreterGenerator*) this)->generate_method_handle_entry();break;
1458 
1459   case Interpreter::java_lang_math_sin     : // fall thru
1460   case Interpreter::java_lang_math_cos     : // fall thru
1461   case Interpreter::java_lang_math_tan     : // fall thru
1462   case Interpreter::java_lang_math_abs     : // fall thru
1463   case Interpreter::java_lang_math_log     : // fall thru
1464   case Interpreter::java_lang_math_log10   : // fall thru
1465   case Interpreter::java_lang_math_sqrt    : entry_point = ((InterpreterGenerator*) this)->generate_math_entry(kind);    break;
1466   default                                  : ShouldNotReachHere();                                                       break;
1467   }
1468 
1469   if (entry_point) {
1470     return entry_point;
1471   }
1472 
1473   return ((InterpreterGenerator*) this)->
1474                                 generate_normal_entry(synchronized);
1475 }
1476 
1477 // These should never be compiled since the interpreter will prefer
1478 // the compiled version to the intrinsic version.
1479 bool AbstractInterpreter::can_be_compiled(methodHandle m) {
1480   switch (method_kind(m)) {
1481     case Interpreter::java_lang_math_sin     : // fall thru
1482     case Interpreter::java_lang_math_cos     : // fall thru
1483     case Interpreter::java_lang_math_tan     : // fall thru
1484     case Interpreter::java_lang_math_abs     : // fall thru
1485     case Interpreter::java_lang_math_log     : // fall thru
1486     case Interpreter::java_lang_math_log10   : // fall thru
1487     case Interpreter::java_lang_math_sqrt    :
1488       return false;
1489     default:
1490       return true;
1491   }
1492 }
1493 
1494 // How much stack a method activation needs in words.
1495 int AbstractInterpreter::size_top_interpreter_activation(methodOop method) {
1496   const int entry_size = frame::interpreter_frame_monitor_size();
1497 
1498   // total overhead size: entry_size + (saved rbp thru expr stack
1499   // bottom).  be sure to change this if you add/subtract anything
1500   // to/from the overhead area
1501   const int overhead_size =
1502     -(frame::interpreter_frame_initial_sp_offset) + entry_size;
1503 
1504   const int stub_code = frame::entry_frame_after_call_words;
1505   const int extra_stack = methodOopDesc::extra_stack_entries();
1506   const int method_stack = (method->max_locals() + method->max_stack() + extra_stack) *
1507                            Interpreter::stackElementWords;
1508   return (overhead_size + method_stack + stub_code);
1509 }
1510 
1511 int AbstractInterpreter::layout_activation(methodOop method,
1512                                            int tempcount,
1513                                            int popframe_extra_args,
1514                                            int moncount,
1515                                            int callee_param_count,
1516                                            int callee_locals,
1517                                            frame* caller,
1518                                            frame* interpreter_frame,
1519                                            bool is_top_frame) {
1520   // Note: This calculation must exactly parallel the frame setup
1521   // in AbstractInterpreterGenerator::generate_method_entry.
1522   // If interpreter_frame!=NULL, set up the method, locals, and monitors.
1523   // The frame interpreter_frame, if not NULL, is guaranteed to be the
1524   // right size, as determined by a previous call to this method.
1525   // It is also guaranteed to be walkable even though it is in a skeletal state
1526 
1527   // fixed size of an interpreter frame:
1528   int max_locals = method->max_locals() * Interpreter::stackElementWords;
1529   int extra_locals = (method->max_locals() - method->size_of_parameters()) *
1530                      Interpreter::stackElementWords;
1531 
1532   int overhead = frame::sender_sp_offset -
1533                  frame::interpreter_frame_initial_sp_offset;
1534   // Our locals were accounted for by the caller (or last_frame_adjust
1535   // on the transistion) Since the callee parameters already account
1536   // for the callee's params we only need to account for the extra
1537   // locals.
1538   int size = overhead +
1539          (callee_locals - callee_param_count)*Interpreter::stackElementWords +
1540          moncount * frame::interpreter_frame_monitor_size() +
1541          tempcount* Interpreter::stackElementWords + popframe_extra_args;
1542   if (interpreter_frame != NULL) {
1543 #ifdef ASSERT
1544     if (!EnableInvokeDynamic)
1545       // @@@ FIXME: Should we correct interpreter_frame_sender_sp in the calling sequences?
1546       // Probably, since deoptimization doesn't work yet.
1547       assert(caller->unextended_sp() == interpreter_frame->interpreter_frame_sender_sp(), "Frame not properly walkable");
1548     assert(caller->sp() == interpreter_frame->sender_sp(), "Frame not properly walkable(2)");
1549 #endif
1550 
1551     interpreter_frame->interpreter_frame_set_method(method);
1552     // NOTE the difference in using sender_sp and
1553     // interpreter_frame_sender_sp interpreter_frame_sender_sp is
1554     // the original sp of the caller (the unextended_sp) and
1555     // sender_sp is fp+16 XXX
1556     intptr_t* locals = interpreter_frame->sender_sp() + max_locals - 1;
1557 
1558     interpreter_frame->interpreter_frame_set_locals(locals);
1559     BasicObjectLock* montop = interpreter_frame->interpreter_frame_monitor_begin();
1560     BasicObjectLock* monbot = montop - moncount;
1561     interpreter_frame->interpreter_frame_set_monitor_end(monbot);
1562 
1563     // Set last_sp
1564     intptr_t*  esp = (intptr_t*) monbot -
1565                      tempcount*Interpreter::stackElementWords -
1566                      popframe_extra_args;
1567     interpreter_frame->interpreter_frame_set_last_sp(esp);
1568 
1569     // All frames but the initial (oldest) interpreter frame we fill in have
1570     // a value for sender_sp that allows walking the stack but isn't
1571     // truly correct. Correct the value here.
1572     if (extra_locals != 0 &&
1573         interpreter_frame->sender_sp() ==
1574         interpreter_frame->interpreter_frame_sender_sp()) {
1575       interpreter_frame->set_interpreter_frame_sender_sp(caller->sp() +
1576                                                          extra_locals);
1577     }
1578     *interpreter_frame->interpreter_frame_cache_addr() =
1579       method->constants()->cache();
1580   }
1581   return size;
1582 }
1583 
1584 //-----------------------------------------------------------------------------
1585 // Exceptions
1586 
1587 void TemplateInterpreterGenerator::generate_throw_exception() {
1588   // Entry point in previous activation (i.e., if the caller was
1589   // interpreted)
1590   Interpreter::_rethrow_exception_entry = __ pc();
1591   // Restore sp to interpreter_frame_last_sp even though we are going
1592   // to empty the expression stack for the exception processing.
1593   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
1594   // rax: exception
1595   // rdx: return address/pc that threw exception
1596   __ restore_bcp();    // r13 points to call/send
1597   __ restore_locals();
1598   __ reinit_heapbase();  // restore r12 as heapbase.
1599   // Entry point for exceptions thrown within interpreter code
1600   Interpreter::_throw_exception_entry = __ pc();
1601   // expression stack is undefined here
1602   // rax: exception
1603   // r13: exception bcp
1604   __ verify_oop(rax);
1605   __ mov(c_rarg1, rax);
1606 
1607   // expression stack must be empty before entering the VM in case of
1608   // an exception
1609   __ empty_expression_stack();
1610   // find exception handler address and preserve exception oop
1611   __ call_VM(rdx,
1612              CAST_FROM_FN_PTR(address,
1613                           InterpreterRuntime::exception_handler_for_exception),
1614              c_rarg1);
1615   // rax: exception handler entry point
1616   // rdx: preserved exception oop
1617   // r13: bcp for exception handler
1618   __ push_ptr(rdx); // push exception which is now the only value on the stack
1619   __ jmp(rax); // jump to exception handler (may be _remove_activation_entry!)
1620 
1621   // If the exception is not handled in the current frame the frame is
1622   // removed and the exception is rethrown (i.e. exception
1623   // continuation is _rethrow_exception).
1624   //
1625   // Note: At this point the bci is still the bxi for the instruction
1626   // which caused the exception and the expression stack is
1627   // empty. Thus, for any VM calls at this point, GC will find a legal
1628   // oop map (with empty expression stack).
1629 
1630   // In current activation
1631   // tos: exception
1632   // esi: exception bcp
1633 
1634   //
1635   // JVMTI PopFrame support
1636   //
1637 
1638   Interpreter::_remove_activation_preserving_args_entry = __ pc();
1639   __ empty_expression_stack();
1640   // Set the popframe_processing bit in pending_popframe_condition
1641   // indicating that we are currently handling popframe, so that
1642   // call_VMs that may happen later do not trigger new popframe
1643   // handling cycles.
1644   __ movl(rdx, Address(r15_thread, JavaThread::popframe_condition_offset()));
1645   __ orl(rdx, JavaThread::popframe_processing_bit);
1646   __ movl(Address(r15_thread, JavaThread::popframe_condition_offset()), rdx);
1647 
1648   {
1649     // Check to see whether we are returning to a deoptimized frame.
1650     // (The PopFrame call ensures that the caller of the popped frame is
1651     // either interpreted or compiled and deoptimizes it if compiled.)
1652     // In this case, we can't call dispatch_next() after the frame is
1653     // popped, but instead must save the incoming arguments and restore
1654     // them after deoptimization has occurred.
1655     //
1656     // Note that we don't compare the return PC against the
1657     // deoptimization blob's unpack entry because of the presence of
1658     // adapter frames in C2.
1659     Label caller_not_deoptimized;
1660     __ movptr(c_rarg1, Address(rbp, frame::return_addr_offset * wordSize));
1661     __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1662                                InterpreterRuntime::interpreter_contains), c_rarg1);
1663     __ testl(rax, rax);
1664     __ jcc(Assembler::notZero, caller_not_deoptimized);
1665 
1666     // Compute size of arguments for saving when returning to
1667     // deoptimized caller
1668     __ get_method(rax);
1669     __ load_unsigned_short(rax, Address(rax, in_bytes(methodOopDesc::
1670                                                 size_of_parameters_offset())));
1671     __ shll(rax, Interpreter::logStackElementSize);
1672     __ restore_locals(); // XXX do we need this?
1673     __ subptr(r14, rax);
1674     __ addptr(r14, wordSize);
1675     // Save these arguments
1676     __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1677                                            Deoptimization::
1678                                            popframe_preserve_args),
1679                           r15_thread, rax, r14);
1680 
1681     __ remove_activation(vtos, rdx,
1682                          /* throw_monitor_exception */ false,
1683                          /* install_monitor_exception */ false,
1684                          /* notify_jvmdi */ false);
1685 
1686     // Inform deoptimization that it is responsible for restoring
1687     // these arguments
1688     __ movl(Address(r15_thread, JavaThread::popframe_condition_offset()),
1689             JavaThread::popframe_force_deopt_reexecution_bit);
1690 
1691     // Continue in deoptimization handler
1692     __ jmp(rdx);
1693 
1694     __ bind(caller_not_deoptimized);
1695   }
1696 
1697   __ remove_activation(vtos, rdx, /* rdx result (retaddr) is not used */
1698                        /* throw_monitor_exception */ false,
1699                        /* install_monitor_exception */ false,
1700                        /* notify_jvmdi */ false);
1701 
1702   // Finish with popframe handling
1703   // A previous I2C followed by a deoptimization might have moved the
1704   // outgoing arguments further up the stack. PopFrame expects the
1705   // mutations to those outgoing arguments to be preserved and other
1706   // constraints basically require this frame to look exactly as
1707   // though it had previously invoked an interpreted activation with
1708   // no space between the top of the expression stack (current
1709   // last_sp) and the top of stack. Rather than force deopt to
1710   // maintain this kind of invariant all the time we call a small
1711   // fixup routine to move the mutated arguments onto the top of our
1712   // expression stack if necessary.
1713   __ mov(c_rarg1, rsp);
1714   __ movptr(c_rarg2, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1715   // PC must point into interpreter here
1716   __ set_last_Java_frame(noreg, rbp, __ pc());
1717   __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), r15_thread, c_rarg1, c_rarg2);
1718   __ reset_last_Java_frame(true, true);
1719   // Restore the last_sp and null it out
1720   __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1721   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
1722 
1723   __ restore_bcp();  // XXX do we need this?
1724   __ restore_locals(); // XXX do we need this?
1725   // The method data pointer was incremented already during
1726   // call profiling. We have to restore the mdp for the current bcp.
1727   if (ProfileInterpreter) {
1728     __ set_method_data_pointer_for_bcp();
1729   }
1730 
1731   // Clear the popframe condition flag
1732   __ movl(Address(r15_thread, JavaThread::popframe_condition_offset()),
1733           JavaThread::popframe_inactive);
1734 
1735   __ dispatch_next(vtos);
1736   // end of PopFrame support
1737 
1738   Interpreter::_remove_activation_entry = __ pc();
1739 
1740   // preserve exception over this code sequence
1741   __ pop_ptr(rax);
1742   __ movptr(Address(r15_thread, JavaThread::vm_result_offset()), rax);
1743   // remove the activation (without doing throws on illegalMonitorExceptions)
1744   __ remove_activation(vtos, rdx, false, true, false);
1745   // restore exception
1746   __ movptr(rax, Address(r15_thread, JavaThread::vm_result_offset()));
1747   __ movptr(Address(r15_thread, JavaThread::vm_result_offset()), (int32_t)NULL_WORD);
1748   __ verify_oop(rax);
1749 
1750   // In between activations - previous activation type unknown yet
1751   // compute continuation point - the continuation point expects the
1752   // following registers set up:
1753   //
1754   // rax: exception
1755   // rdx: return address/pc that threw exception
1756   // rsp: expression stack of caller
1757   // rbp: ebp of caller
1758   __ push(rax);                                  // save exception
1759   __ push(rdx);                                  // save return address
1760   __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1761                           SharedRuntime::exception_handler_for_return_address),
1762                         r15_thread, rdx);
1763   __ mov(rbx, rax);                              // save exception handler
1764   __ pop(rdx);                                   // restore return address
1765   __ pop(rax);                                   // restore exception
1766   // Note that an "issuing PC" is actually the next PC after the call
1767   __ jmp(rbx);                                   // jump to exception
1768                                                  // handler of caller
1769 }
1770 
1771 
1772 //
1773 // JVMTI ForceEarlyReturn support
1774 //
1775 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
1776   address entry = __ pc();
1777 
1778   __ restore_bcp();
1779   __ restore_locals();
1780   __ empty_expression_stack();
1781   __ load_earlyret_value(state);
1782 
1783   __ movptr(rdx, Address(r15_thread, JavaThread::jvmti_thread_state_offset()));
1784   Address cond_addr(rdx, JvmtiThreadState::earlyret_state_offset());
1785 
1786   // Clear the earlyret state
1787   __ movl(cond_addr, JvmtiThreadState::earlyret_inactive);
1788 
1789   __ remove_activation(state, rsi,
1790                        false, /* throw_monitor_exception */
1791                        false, /* install_monitor_exception */
1792                        true); /* notify_jvmdi */
1793   __ jmp(rsi);
1794 
1795   return entry;
1796 } // end of ForceEarlyReturn support
1797 
1798 
1799 //-----------------------------------------------------------------------------
1800 // Helper for vtos entry point generation
1801 
1802 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
1803                                                          address& bep,
1804                                                          address& cep,
1805                                                          address& sep,
1806                                                          address& aep,
1807                                                          address& iep,
1808                                                          address& lep,
1809                                                          address& fep,
1810                                                          address& dep,
1811                                                          address& vep) {
1812   assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
1813   Label L;
1814   aep = __ pc();  __ push_ptr();  __ jmp(L);
1815   fep = __ pc();  __ push_f();    __ jmp(L);
1816   dep = __ pc();  __ push_d();    __ jmp(L);
1817   lep = __ pc();  __ push_l();    __ jmp(L);
1818   bep = cep = sep =
1819   iep = __ pc();  __ push_i();
1820   vep = __ pc();
1821   __ bind(L);
1822   generate_and_dispatch(t);
1823 }
1824 
1825 
1826 //-----------------------------------------------------------------------------
1827 // Generation of individual instructions
1828 
1829 // helpers for generate_and_dispatch
1830 
1831 
1832 InterpreterGenerator::InterpreterGenerator(StubQueue* code)
1833   : TemplateInterpreterGenerator(code) {
1834    generate_all(); // down here so it can be "virtual"
1835 }
1836 
1837 //-----------------------------------------------------------------------------
1838 
1839 // Non-product code
1840 #ifndef PRODUCT
1841 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
1842   address entry = __ pc();
1843 
1844   __ push(state);
1845   __ push(c_rarg0);
1846   __ push(c_rarg1);
1847   __ push(c_rarg2);
1848   __ push(c_rarg3);
1849   __ mov(c_rarg2, rax);  // Pass itos
1850 #ifdef _WIN64
1851   __ movflt(xmm3, xmm0); // Pass ftos
1852 #endif
1853   __ call_VM(noreg,
1854              CAST_FROM_FN_PTR(address, SharedRuntime::trace_bytecode),
1855              c_rarg1, c_rarg2, c_rarg3);
1856   __ pop(c_rarg3);
1857   __ pop(c_rarg2);
1858   __ pop(c_rarg1);
1859   __ pop(c_rarg0);
1860   __ pop(state);
1861   __ ret(0);                                   // return from result handler
1862 
1863   return entry;
1864 }
1865 
1866 void TemplateInterpreterGenerator::count_bytecode() {
1867   __ incrementl(ExternalAddress((address) &BytecodeCounter::_counter_value));
1868 }
1869 
1870 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
1871   __ incrementl(ExternalAddress((address) &BytecodeHistogram::_counters[t->bytecode()]));
1872 }
1873 
1874 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
1875   __ mov32(rbx, ExternalAddress((address) &BytecodePairHistogram::_index));
1876   __ shrl(rbx, BytecodePairHistogram::log2_number_of_codes);
1877   __ orl(rbx,
1878          ((int) t->bytecode()) <<
1879          BytecodePairHistogram::log2_number_of_codes);
1880   __ mov32(ExternalAddress((address) &BytecodePairHistogram::_index), rbx);
1881   __ lea(rscratch1, ExternalAddress((address) BytecodePairHistogram::_counters));
1882   __ incrementl(Address(rscratch1, rbx, Address::times_4));
1883 }
1884 
1885 
1886 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
1887   // Call a little run-time stub to avoid blow-up for each bytecode.
1888   // The run-time runtime saves the right registers, depending on
1889   // the tosca in-state for the given template.
1890 
1891   assert(Interpreter::trace_code(t->tos_in()) != NULL,
1892          "entry must have been generated");
1893   __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1894   __ andptr(rsp, -16); // align stack as required by ABI
1895   __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in())));
1896   __ mov(rsp, r12); // restore sp
1897   __ reinit_heapbase();
1898 }
1899 
1900 
1901 void TemplateInterpreterGenerator::stop_interpreter_at() {
1902   Label L;
1903   __ cmp32(ExternalAddress((address) &BytecodeCounter::_counter_value),
1904            StopInterpreterAt);
1905   __ jcc(Assembler::notEqual, L);
1906   __ int3();
1907   __ bind(L);
1908 }
1909 #endif // !PRODUCT
1910 #endif // ! CC_INTERP