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 address TemplateInterpreterGenerator::generate_exception_handler_common(
 124         const char* name, const char* message, bool pass_oop) {
 125   assert(!pass_oop || message == NULL, "either oop or message but not both");
 126   address entry = __ pc();
 127   if (pass_oop) {
 128     // object is at TOS
 129     __ pop(c_rarg2);
 130   }
 131   // expression stack must be empty before entering the VM if an
 132   // exception happened
 133   __ empty_expression_stack();
 134   // setup parameters
 135   __ lea(c_rarg1, ExternalAddress((address)name));
 136   if (pass_oop) {
 137     __ call_VM(rax, CAST_FROM_FN_PTR(address,
 138                                      InterpreterRuntime::
 139                                      create_klass_exception),
 140                c_rarg1, c_rarg2);
 141   } else {
 142     // kind of lame ExternalAddress can't take NULL because
 143     // external_word_Relocation will assert.
 144     if (message != NULL) {
 145       __ lea(c_rarg2, ExternalAddress((address)message));
 146     } else {
 147       __ movptr(c_rarg2, NULL_WORD);
 148     }
 149     __ call_VM(rax,
 150                CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception),
 151                c_rarg1, c_rarg2);
 152   }
 153   // throw exception
 154   __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
 155   return entry;
 156 }
 157 
 158 
 159 address TemplateInterpreterGenerator::generate_continuation_for(TosState state) {
 160   address entry = __ pc();
 161   // NULL last_sp until next java call
 162   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
 163   __ dispatch_next(state);
 164   return entry;
 165 }
 166 
 167 
 168 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step) {
 169   address entry = __ pc();
 170 
 171   // Restore stack bottom in case i2c adjusted stack
 172   __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
 173   // and NULL it as marker that esp is now tos until next java call
 174   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
 175 
 176   __ restore_bcp();
 177   __ restore_locals();
 178 
 179   Label L_got_cache, L_giant_index;
 180   if (EnableInvokeDynamic) {
 181     __ cmpb(Address(r13, 0), Bytecodes::_invokedynamic);
 182     __ jcc(Assembler::equal, L_giant_index);
 183   }
 184   __ get_cache_and_index_at_bcp(rbx, rcx, 1, sizeof(u2));
 185   __ bind(L_got_cache);
 186   __ movl(rbx, Address(rbx, rcx,
 187                        Address::times_ptr,
 188                        in_bytes(constantPoolCacheOopDesc::base_offset()) +
 189                        3 * wordSize));
 190   __ andl(rbx, 0xFF);
 191   __ lea(rsp, Address(rsp, rbx, Address::times_8));
 192   __ dispatch_next(state, step);
 193 
 194   // out of the main line of code...
 195   if (EnableInvokeDynamic) {
 196     __ bind(L_giant_index);
 197     __ get_cache_and_index_at_bcp(rbx, rcx, 1, sizeof(u4));
 198     __ jmp(L_got_cache);
 199   }
 200 
 201   return entry;
 202 }
 203 
 204 
 205 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state,
 206                                                                int step) {
 207   address entry = __ pc();
 208   // NULL last_sp until next java call
 209   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
 210   __ restore_bcp();
 211   __ restore_locals();
 212   // handle exceptions
 213   {
 214     Label L;
 215     __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t) NULL_WORD);
 216     __ jcc(Assembler::zero, L);
 217     __ call_VM(noreg,
 218                CAST_FROM_FN_PTR(address,
 219                                 InterpreterRuntime::throw_pending_exception));
 220     __ should_not_reach_here();
 221     __ bind(L);
 222   }
 223   __ dispatch_next(state, step);
 224   return entry;
 225 }
 226 
 227 int AbstractInterpreter::BasicType_as_index(BasicType type) {
 228   int i = 0;
 229   switch (type) {
 230     case T_BOOLEAN: i = 0; break;
 231     case T_CHAR   : i = 1; break;
 232     case T_BYTE   : i = 2; break;
 233     case T_SHORT  : i = 3; break;
 234     case T_INT    : i = 4; break;
 235     case T_LONG   : i = 5; break;
 236     case T_VOID   : i = 6; break;
 237     case T_FLOAT  : i = 7; break;
 238     case T_DOUBLE : i = 8; break;
 239     case T_OBJECT : i = 9; break;
 240     case T_ARRAY  : i = 9; break;
 241     default       : ShouldNotReachHere();
 242   }
 243   assert(0 <= i && i < AbstractInterpreter::number_of_result_handlers,
 244          "index out of bounds");
 245   return i;
 246 }
 247 
 248 
 249 address TemplateInterpreterGenerator::generate_result_handler_for(
 250         BasicType type) {
 251   address entry = __ pc();
 252   switch (type) {
 253   case T_BOOLEAN: __ c2bool(rax);            break;
 254   case T_CHAR   : __ movzwl(rax, rax);       break;
 255   case T_BYTE   : __ sign_extend_byte(rax);  break;
 256   case T_SHORT  : __ sign_extend_short(rax); break;
 257   case T_INT    : /* nothing to do */        break;
 258   case T_LONG   : /* nothing to do */        break;
 259   case T_VOID   : /* nothing to do */        break;
 260   case T_FLOAT  : /* nothing to do */        break;
 261   case T_DOUBLE : /* nothing to do */        break;
 262   case T_OBJECT :
 263     // retrieve result from frame
 264     __ movptr(rax, Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize));
 265     // and verify it
 266     __ verify_oop(rax);
 267     break;
 268   default       : ShouldNotReachHere();
 269   }
 270   __ ret(0);                                   // return from result handler
 271   return entry;
 272 }
 273 
 274 address TemplateInterpreterGenerator::generate_safept_entry_for(
 275         TosState state,
 276         address runtime_entry) {
 277   address entry = __ pc();
 278   __ push(state);
 279   __ call_VM(noreg, runtime_entry);
 280   __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
 281   return entry;
 282 }
 283 
 284 
 285 
 286 // Helpers for commoning out cases in the various type of method entries.
 287 //
 288 
 289 
 290 // increment invocation count & check for overflow
 291 //
 292 // Note: checking for negative value instead of overflow
 293 //       so we have a 'sticky' overflow test
 294 //
 295 // rbx: method
 296 // ecx: invocation counter
 297 //
 298 void InterpreterGenerator::generate_counter_incr(
 299         Label* overflow,
 300         Label* profile_method,
 301         Label* profile_method_continue) {
 302   const Address invocation_counter(rbx, in_bytes(methodOopDesc::invocation_counter_offset()) +
 303                                         in_bytes(InvocationCounter::counter_offset()));
 304   // Note: In tiered we increment either counters in methodOop or in MDO depending if we're profiling or not.
 305   if (TieredCompilation) {
 306     int increment = InvocationCounter::count_increment;
 307     int mask = ((1 << Tier0InvokeNotifyFreqLog)  - 1) << InvocationCounter::count_shift;
 308     Label no_mdo, done;
 309     if (ProfileInterpreter) {
 310       // Are we profiling?
 311       __ movptr(rax, Address(rbx, methodOopDesc::method_data_offset()));
 312       __ testptr(rax, rax);
 313       __ jccb(Assembler::zero, no_mdo);
 314       // Increment counter in the MDO
 315       const Address mdo_invocation_counter(rax, in_bytes(methodDataOopDesc::invocation_counter_offset()) +
 316                                                 in_bytes(InvocationCounter::counter_offset()));
 317       __ increment_mask_and_jump(mdo_invocation_counter, increment, mask, rcx, false, Assembler::zero, overflow);
 318       __ jmpb(done);
 319     }
 320     __ bind(no_mdo);
 321     // Increment counter in methodOop (we don't need to load it, it's in ecx).
 322     __ increment_mask_and_jump(invocation_counter, increment, mask, rcx, true, Assembler::zero, overflow);
 323     __ bind(done);
 324   } else {
 325     const Address backedge_counter(rbx,
 326                                    methodOopDesc::backedge_counter_offset() +
 327                                    InvocationCounter::counter_offset());
 328 
 329     if (ProfileInterpreter) { // %%% Merge this into methodDataOop
 330       __ incrementl(Address(rbx,
 331                             methodOopDesc::interpreter_invocation_counter_offset()));
 332     }
 333     // Update standard invocation counters
 334     __ movl(rax, backedge_counter);   // load backedge counter
 335 
 336     __ incrementl(rcx, InvocationCounter::count_increment);
 337     __ andl(rax, InvocationCounter::count_mask_value); // mask out the status bits
 338 
 339     __ movl(invocation_counter, rcx); // save invocation count
 340     __ addl(rcx, rax);                // add both counters
 341 
 342     // profile_method is non-null only for interpreted method so
 343     // profile_method != NULL == !native_call
 344 
 345     if (ProfileInterpreter && profile_method != NULL) {
 346       // Test to see if we should create a method data oop
 347       __ cmp32(rcx, ExternalAddress((address)&InvocationCounter::InterpreterProfileLimit));
 348       __ jcc(Assembler::less, *profile_method_continue);
 349 
 350       // if no method data exists, go to profile_method
 351       __ test_method_data_pointer(rax, *profile_method);
 352     }
 353 
 354     __ cmp32(rcx, ExternalAddress((address)&InvocationCounter::InterpreterInvocationLimit));
 355     __ jcc(Assembler::aboveEqual, *overflow);
 356   }
 357 }
 358 
 359 void InterpreterGenerator::generate_counter_overflow(Label* do_continue) {
 360 
 361   // Asm interpreter on entry
 362   // r14 - locals
 363   // r13 - bcp
 364   // rbx - method
 365   // edx - cpool --- DOES NOT APPEAR TO BE TRUE
 366   // rbp - interpreter frame
 367 
 368   // On return (i.e. jump to entry_point) [ back to invocation of interpreter ]
 369   // Everything as it was on entry
 370   // rdx is not restored. Doesn't appear to really be set.
 371 
 372   const Address size_of_parameters(rbx,
 373                                    methodOopDesc::size_of_parameters_offset());
 374 
 375   // InterpreterRuntime::frequency_counter_overflow takes two
 376   // arguments, the first (thread) is passed by call_VM, the second
 377   // indicates if the counter overflow occurs at a backwards branch
 378   // (NULL bcp).  We pass zero for it.  The call returns the address
 379   // of the verified entry point for the method or NULL if the
 380   // compilation did not complete (either went background or bailed
 381   // out).
 382   __ movl(c_rarg1, 0);
 383   __ call_VM(noreg,
 384              CAST_FROM_FN_PTR(address,
 385                               InterpreterRuntime::frequency_counter_overflow),
 386              c_rarg1);
 387 
 388   __ movptr(rbx, Address(rbp, method_offset));   // restore methodOop
 389   // Preserve invariant that r13/r14 contain bcp/locals of sender frame
 390   // and jump to the interpreted entry.
 391   __ jmp(*do_continue, relocInfo::none);
 392 }
 393 
 394 // See if we've got enough room on the stack for locals plus overhead.
 395 // The expression stack grows down incrementally, so the normal guard
 396 // page mechanism will work for that.
 397 //
 398 // NOTE: Since the additional locals are also always pushed (wasn't
 399 // obvious in generate_method_entry) so the guard should work for them
 400 // too.
 401 //
 402 // Args:
 403 //      rdx: number of additional locals this frame needs (what we must check)
 404 //      rbx: methodOop
 405 //
 406 // Kills:
 407 //      rax
 408 void InterpreterGenerator::generate_stack_overflow_check(void) {
 409 
 410   // monitor entry size: see picture of stack set
 411   // (generate_method_entry) and frame_amd64.hpp
 412   const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
 413 
 414   // total overhead size: entry_size + (saved rbp through expr stack
 415   // bottom).  be sure to change this if you add/subtract anything
 416   // to/from the overhead area
 417   const int overhead_size =
 418     -(frame::interpreter_frame_initial_sp_offset * wordSize) + entry_size;
 419 
 420   const int page_size = os::vm_page_size();
 421 
 422   Label after_frame_check;
 423 
 424   // see if the frame is greater than one page in size. If so,
 425   // then we need to verify there is enough stack space remaining
 426   // for the additional locals.
 427   __ cmpl(rdx, (page_size - overhead_size) / Interpreter::stackElementSize);
 428   __ jcc(Assembler::belowEqual, after_frame_check);
 429 
 430   // compute rsp as if this were going to be the last frame on
 431   // the stack before the red zone
 432 
 433   const Address stack_base(r15_thread, Thread::stack_base_offset());
 434   const Address stack_size(r15_thread, Thread::stack_size_offset());
 435 
 436   // locals + overhead, in bytes
 437   __ mov(rax, rdx);
 438   __ shlptr(rax, Interpreter::logStackElementSize);  // 2 slots per parameter.
 439   __ addptr(rax, overhead_size);
 440 
 441 #ifdef ASSERT
 442   Label stack_base_okay, stack_size_okay;
 443   // verify that thread stack base is non-zero
 444   __ cmpptr(stack_base, (int32_t)NULL_WORD);
 445   __ jcc(Assembler::notEqual, stack_base_okay);
 446   __ stop("stack base is zero");
 447   __ bind(stack_base_okay);
 448   // verify that thread stack size is non-zero
 449   __ cmpptr(stack_size, 0);
 450   __ jcc(Assembler::notEqual, stack_size_okay);
 451   __ stop("stack size is zero");
 452   __ bind(stack_size_okay);
 453 #endif
 454 
 455   // Add stack base to locals and subtract stack size
 456   __ addptr(rax, stack_base);
 457   __ subptr(rax, stack_size);
 458 
 459   // Use the maximum number of pages we might bang.
 460   const int max_pages = StackShadowPages > (StackRedPages+StackYellowPages) ? StackShadowPages :
 461                                                                               (StackRedPages+StackYellowPages);
 462 
 463   // add in the red and yellow zone sizes
 464   __ addptr(rax, max_pages * page_size);
 465 
 466   // check against the current stack bottom
 467   __ cmpptr(rsp, rax);
 468   __ jcc(Assembler::above, after_frame_check);
 469 
 470   __ pop(rax); // get return address
 471   __ jump(ExternalAddress(Interpreter::throw_StackOverflowError_entry()));
 472 
 473   // all done with frame size check
 474   __ bind(after_frame_check);
 475 }
 476 
 477 // Allocate monitor and lock method (asm interpreter)
 478 //
 479 // Args:
 480 //      rbx: methodOop
 481 //      r14: locals
 482 //
 483 // Kills:
 484 //      rax
 485 //      c_rarg0, c_rarg1, c_rarg2, c_rarg3, ...(param regs)
 486 //      rscratch1, rscratch2 (scratch regs)
 487 void InterpreterGenerator::lock_method(void) {
 488   // synchronize method
 489   const Address access_flags(rbx, methodOopDesc::access_flags_offset());
 490   const Address monitor_block_top(
 491         rbp,
 492         frame::interpreter_frame_monitor_block_top_offset * wordSize);
 493   const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
 494 
 495 #ifdef ASSERT
 496   {
 497     Label L;
 498     __ movl(rax, access_flags);
 499     __ testl(rax, JVM_ACC_SYNCHRONIZED);
 500     __ jcc(Assembler::notZero, L);
 501     __ stop("method doesn't need synchronization");
 502     __ bind(L);
 503   }
 504 #endif // ASSERT
 505 
 506   // get synchronization object
 507   {
 508     const int mirror_offset = in_bytes(Klass::java_mirror_offset());
 509     Label done;
 510     __ movl(rax, access_flags);
 511     __ testl(rax, JVM_ACC_STATIC);
 512     // get receiver (assume this is frequent case)
 513     __ movptr(rax, Address(r14, Interpreter::local_offset_in_bytes(0)));
 514     __ jcc(Assembler::zero, done);
 515     __ movptr(rax, Address(rbx, methodOopDesc::constants_offset()));
 516     __ movptr(rax, Address(rax,
 517                            constantPoolOopDesc::pool_holder_offset_in_bytes()));
 518     __ movptr(rax, Address(rax, mirror_offset));
 519 
 520 #ifdef ASSERT
 521     {
 522       Label L;
 523       __ testptr(rax, rax);
 524       __ jcc(Assembler::notZero, L);
 525       __ stop("synchronization object is NULL");
 526       __ bind(L);
 527     }
 528 #endif // ASSERT
 529 
 530     __ bind(done);
 531   }
 532 
 533   // add space for monitor & lock
 534   __ subptr(rsp, entry_size); // add space for a monitor entry
 535   __ movptr(monitor_block_top, rsp);  // set new monitor block top
 536   // store object
 537   __ movptr(Address(rsp, BasicObjectLock::obj_offset_in_bytes()), rax);
 538   __ movptr(c_rarg1, rsp); // object address
 539   __ lock_object(c_rarg1);
 540 }
 541 
 542 // Generate a fixed interpreter frame. This is identical setup for
 543 // interpreted methods and for native methods hence the shared code.
 544 //
 545 // Args:
 546 //      rax: return address
 547 //      rbx: methodOop
 548 //      r14: pointer to locals
 549 //      r13: sender sp
 550 //      rdx: cp cache
 551 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
 552   // initialize fixed part of activation frame
 553   __ push(rax);        // save return address
 554   __ enter();          // save old & set new rbp
 555   __ push(r13);        // set sender sp
 556   __ push((int)NULL_WORD); // leave last_sp as null
 557   __ movptr(r13, Address(rbx, methodOopDesc::const_offset()));      // get constMethodOop
 558   __ lea(r13, Address(r13, constMethodOopDesc::codes_offset())); // get codebase
 559   __ push(rbx);        // save methodOop
 560   if (ProfileInterpreter) {
 561     Label method_data_continue;
 562     __ movptr(rdx, Address(rbx, in_bytes(methodOopDesc::method_data_offset())));
 563     __ testptr(rdx, rdx);
 564     __ jcc(Assembler::zero, method_data_continue);
 565     __ addptr(rdx, in_bytes(methodDataOopDesc::data_offset()));
 566     __ bind(method_data_continue);
 567     __ push(rdx);      // set the mdp (method data pointer)
 568   } else {
 569     __ push(0);
 570   }
 571 
 572   __ movptr(rdx, Address(rbx, methodOopDesc::constants_offset()));
 573   __ movptr(rdx, Address(rdx, constantPoolOopDesc::cache_offset_in_bytes()));
 574   __ push(rdx); // set constant pool cache
 575   __ push(r14); // set locals pointer
 576   if (native_call) {
 577     __ push(0); // no bcp
 578   } else {
 579     __ push(r13); // set bcp
 580   }
 581   __ push(0); // reserve word for pointer to expression stack bottom
 582   __ movptr(Address(rsp, 0), rsp); // set expression stack bottom
 583 }
 584 
 585 // End of helpers
 586 
 587 // Various method entries
 588 //------------------------------------------------------------------------------------------------------------------------
 589 //
 590 //
 591 
 592 // Call an accessor method (assuming it is resolved, otherwise drop
 593 // into vanilla (slow path) entry
 594 address InterpreterGenerator::generate_accessor_entry(void) {
 595   // rbx: methodOop
 596 
 597   // r13: senderSP must preserver for slow path, set SP to it on fast path
 598 
 599   address entry_point = __ pc();
 600   Label xreturn_path;
 601 
 602   // do fastpath for resolved accessor methods
 603   if (UseFastAccessorMethods) {
 604     // Code: _aload_0, _(i|a)getfield, _(i|a)return or any rewrites
 605     //       thereof; parameter size = 1
 606     // Note: We can only use this code if the getfield has been resolved
 607     //       and if we don't have a null-pointer exception => check for
 608     //       these conditions first and use slow path if necessary.
 609     Label slow_path;
 610     // If we need a safepoint check, generate full interpreter entry.
 611     __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
 612              SafepointSynchronize::_not_synchronized);
 613 
 614     __ jcc(Assembler::notEqual, slow_path);
 615     // rbx: method
 616     __ movptr(rax, Address(rsp, wordSize));
 617 
 618     // check if local 0 != NULL and read field
 619     __ testptr(rax, rax);
 620     __ jcc(Assembler::zero, slow_path);
 621 
 622     __ movptr(rdi, Address(rbx, methodOopDesc::constants_offset()));
 623     // read first instruction word and extract bytecode @ 1 and index @ 2
 624     __ movptr(rdx, Address(rbx, methodOopDesc::const_offset()));
 625     __ movl(rdx, Address(rdx, constMethodOopDesc::codes_offset()));
 626     // Shift codes right to get the index on the right.
 627     // The bytecode fetched looks like <index><0xb4><0x2a>
 628     __ shrl(rdx, 2 * BitsPerByte);
 629     __ shll(rdx, exact_log2(in_words(ConstantPoolCacheEntry::size())));
 630     __ movptr(rdi, Address(rdi, constantPoolOopDesc::cache_offset_in_bytes()));
 631 
 632     // rax: local 0
 633     // rbx: method
 634     // rdx: constant pool cache index
 635     // rdi: constant pool cache
 636 
 637     // check if getfield has been resolved and read constant pool cache entry
 638     // check the validity of the cache entry by testing whether _indices field
 639     // contains Bytecode::_getfield in b1 byte.
 640     assert(in_words(ConstantPoolCacheEntry::size()) == 4,
 641            "adjust shift below");
 642     __ movl(rcx,
 643             Address(rdi,
 644                     rdx,
 645                     Address::times_8,
 646                     constantPoolCacheOopDesc::base_offset() +
 647                     ConstantPoolCacheEntry::indices_offset()));
 648     __ shrl(rcx, 2 * BitsPerByte);
 649     __ andl(rcx, 0xFF);
 650     __ cmpl(rcx, Bytecodes::_getfield);
 651     __ jcc(Assembler::notEqual, slow_path);
 652 
 653     // Note: constant pool entry is not valid before bytecode is resolved
 654     __ movptr(rcx,
 655               Address(rdi,
 656                       rdx,
 657                       Address::times_8,
 658                       constantPoolCacheOopDesc::base_offset() +
 659                       ConstantPoolCacheEntry::f2_offset()));
 660     // edx: flags
 661     __ movl(rdx,
 662             Address(rdi,
 663                     rdx,
 664                     Address::times_8,
 665                     constantPoolCacheOopDesc::base_offset() +
 666                     ConstantPoolCacheEntry::flags_offset()));
 667 
 668     Label notObj, notInt, notByte, notShort;
 669     const Address field_address(rax, rcx, Address::times_1);
 670 
 671     // Need to differentiate between igetfield, agetfield, bgetfield etc.
 672     // because they are different sizes.
 673     // Use the type from the constant pool cache
 674     __ shrl(rdx, ConstantPoolCacheEntry::tosBits);
 675     // Make sure we don't need to mask edx for tosBits after the above shift
 676     ConstantPoolCacheEntry::verify_tosBits();
 677 
 678     __ cmpl(rdx, atos);
 679     __ jcc(Assembler::notEqual, notObj);
 680     // atos
 681     __ load_heap_oop(rax, field_address);
 682     __ jmp(xreturn_path);
 683 
 684     __ bind(notObj);
 685     __ cmpl(rdx, itos);
 686     __ jcc(Assembler::notEqual, notInt);
 687     // itos
 688     __ movl(rax, field_address);
 689     __ jmp(xreturn_path);
 690 
 691     __ bind(notInt);
 692     __ cmpl(rdx, btos);
 693     __ jcc(Assembler::notEqual, notByte);
 694     // btos
 695     __ load_signed_byte(rax, field_address);
 696     __ jmp(xreturn_path);
 697 
 698     __ bind(notByte);
 699     __ cmpl(rdx, stos);
 700     __ jcc(Assembler::notEqual, notShort);
 701     // stos
 702     __ load_signed_short(rax, field_address);
 703     __ jmp(xreturn_path);
 704 
 705     __ bind(notShort);
 706 #ifdef ASSERT
 707     Label okay;
 708     __ cmpl(rdx, ctos);
 709     __ jcc(Assembler::equal, okay);
 710     __ stop("what type is this?");
 711     __ bind(okay);
 712 #endif
 713     // ctos
 714     __ load_unsigned_short(rax, field_address);
 715 
 716     __ bind(xreturn_path);
 717 
 718     // _ireturn/_areturn
 719     __ pop(rdi);
 720     __ mov(rsp, r13);
 721     __ jmp(rdi);
 722     __ ret(0);
 723 
 724     // generate a vanilla interpreter entry as the slow path
 725     __ bind(slow_path);
 726     (void) generate_normal_entry(false);
 727   } else {
 728     (void) generate_normal_entry(false);
 729   }
 730 
 731   return entry_point;
 732 }
 733 
 734 // Method entry for java.lang.ref.Reference.get.
 735 address InterpreterGenerator::generate_Reference_get_entry(void) {
 736 #ifndef SERIALGC
 737   // Code: _aload_0, _getfield, _areturn
 738   // parameter size = 1
 739   //
 740   // The code that gets generated by this routine is split into 2 parts:
 741   //    1. The "intrinsified" code for G1 (or any SATB based GC),
 742   //    2. The slow path - which is an expansion of the regular method entry.
 743   //
 744   // Notes:-
 745   // * In the G1 code we do not check whether we need to block for
 746   //   a safepoint. If G1 is enabled then we must execute the specialized
 747   //   code for Reference.get (except when the Reference object is null)
 748   //   so that we can log the value in the referent field with an SATB
 749   //   update buffer.
 750   //   If the code for the getfield template is modified so that the
 751   //   G1 pre-barrier code is executed when the current method is
 752   //   Reference.get() then going through the normal method entry
 753   //   will be fine.
 754   // * The G1 code can, however, check the receiver object (the instance
 755   //   of java.lang.Reference) and jump to the slow path if null. If the
 756   //   Reference object is null then we obviously cannot fetch the referent
 757   //   and so we don't need to call the G1 pre-barrier. Thus we can use the
 758   //   regular method entry code to generate the NPE.
 759   //
 760   // This code is based on generate_accessor_enty.
 761   //
 762   // rbx: methodOop
 763 
 764   // r13: senderSP must preserve for slow path, set SP to it on fast path
 765 
 766   address entry = __ pc();
 767 
 768   const int referent_offset = java_lang_ref_Reference::referent_offset;
 769   guarantee(referent_offset > 0, "referent offset not initialized");
 770 
 771   if (UseG1GC) {
 772     Label slow_path;
 773     // rbx: method
 774 
 775     // Check if local 0 != NULL
 776     // If the receiver is null then it is OK to jump to the slow path.
 777     __ movptr(rax, Address(rsp, wordSize));
 778 
 779     __ testptr(rax, rax);
 780     __ jcc(Assembler::zero, slow_path);
 781 
 782     // rax: local 0
 783     // rbx: method (but can be used as scratch now)
 784     // rdx: scratch
 785     // rdi: scratch
 786 
 787     // Generate the G1 pre-barrier code to log the value of
 788     // the referent field in an SATB buffer.
 789 
 790     // Load the value of the referent field.
 791     const Address field_address(rax, referent_offset);
 792     __ load_heap_oop(rax, field_address);
 793 
 794     // Generate the G1 pre-barrier code to log the value of
 795     // the referent field in an SATB buffer.
 796     __ g1_write_barrier_pre(noreg /* obj */,
 797                             rax /* pre_val */,
 798                             r15_thread /* thread */,
 799                             rbx /* tmp */,
 800                             true /* tosca_live */,
 801                             true /* expand_call */);
 802 
 803     // _areturn
 804     __ pop(rdi);                // get return address
 805     __ mov(rsp, r13);           // set sp to sender sp
 806     __ jmp(rdi);
 807     __ ret(0);
 808 
 809     // generate a vanilla interpreter entry as the slow path
 810     __ bind(slow_path);
 811     (void) generate_normal_entry(false);
 812 
 813     return entry;
 814   }
 815 #endif // SERIALGC
 816 
 817   // If G1 is not enabled then attempt to go through the accessor entry point
 818   // Reference.get is an accessor
 819   return generate_accessor_entry();
 820 }
 821 
 822 
 823 // Interpreter stub for calling a native method. (asm interpreter)
 824 // This sets up a somewhat different looking stack for calling the
 825 // native method than the typical interpreter frame setup.
 826 address InterpreterGenerator::generate_native_entry(bool synchronized) {
 827   // determine code generation flags
 828   bool inc_counter  = UseCompiler || CountCompiledCalls;
 829 
 830   // rbx: methodOop
 831   // r13: sender sp
 832 
 833   address entry_point = __ pc();
 834 
 835   const Address size_of_parameters(rbx, methodOopDesc::
 836                                         size_of_parameters_offset());
 837   const Address invocation_counter(rbx, methodOopDesc::
 838                                         invocation_counter_offset() +
 839                                         InvocationCounter::counter_offset());
 840   const Address access_flags      (rbx, methodOopDesc::access_flags_offset());
 841 
 842   // get parameter size (always needed)
 843   __ load_unsigned_short(rcx, size_of_parameters);
 844 
 845   // native calls don't need the stack size check since they have no
 846   // expression stack and the arguments are already on the stack and
 847   // we only add a handful of words to the stack
 848 
 849   // rbx: methodOop
 850   // rcx: size of parameters
 851   // r13: sender sp
 852   __ pop(rax);                                       // get return address
 853 
 854   // for natives the size of locals is zero
 855 
 856   // compute beginning of parameters (r14)
 857   __ lea(r14, Address(rsp, rcx, Address::times_8, -wordSize));
 858 
 859   // add 2 zero-initialized slots for native calls
 860   // initialize result_handler slot
 861   __ push((int) NULL_WORD);
 862   // slot for oop temp
 863   // (static native method holder mirror/jni oop result)
 864   __ push((int) NULL_WORD);
 865 
 866   if (inc_counter) {
 867     __ movl(rcx, invocation_counter);  // (pre-)fetch invocation count
 868   }
 869 
 870   // initialize fixed part of activation frame
 871   generate_fixed_frame(true);
 872 
 873   // make sure method is native & not abstract
 874 #ifdef ASSERT
 875   __ movl(rax, access_flags);
 876   {
 877     Label L;
 878     __ testl(rax, JVM_ACC_NATIVE);
 879     __ jcc(Assembler::notZero, L);
 880     __ stop("tried to execute non-native method as native");
 881     __ bind(L);
 882   }
 883   {
 884     Label L;
 885     __ testl(rax, JVM_ACC_ABSTRACT);
 886     __ jcc(Assembler::zero, L);
 887     __ stop("tried to execute abstract method in interpreter");
 888     __ bind(L);
 889   }
 890 #endif
 891 
 892   // Since at this point in the method invocation the exception handler
 893   // would try to exit the monitor of synchronized methods which hasn't
 894   // been entered yet, we set the thread local variable
 895   // _do_not_unlock_if_synchronized to true. The remove_activation will
 896   // check this flag.
 897 
 898   const Address do_not_unlock_if_synchronized(r15_thread,
 899         in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
 900   __ movbool(do_not_unlock_if_synchronized, true);
 901 
 902   // increment invocation count & check for overflow
 903   Label invocation_counter_overflow;
 904   if (inc_counter) {
 905     generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
 906   }
 907 
 908   Label continue_after_compile;
 909   __ bind(continue_after_compile);
 910 
 911   bang_stack_shadow_pages(true);
 912 
 913   // reset the _do_not_unlock_if_synchronized flag
 914   __ movbool(do_not_unlock_if_synchronized, false);
 915 
 916   // check for synchronized methods
 917   // Must happen AFTER invocation_counter check and stack overflow check,
 918   // so method is not locked if overflows.
 919   if (synchronized) {
 920     lock_method();
 921   } else {
 922     // no synchronization necessary
 923 #ifdef ASSERT
 924     {
 925       Label L;
 926       __ movl(rax, access_flags);
 927       __ testl(rax, JVM_ACC_SYNCHRONIZED);
 928       __ jcc(Assembler::zero, L);
 929       __ stop("method needs synchronization");
 930       __ bind(L);
 931     }
 932 #endif
 933   }
 934 
 935   // start execution
 936 #ifdef ASSERT
 937   {
 938     Label L;
 939     const Address monitor_block_top(rbp,
 940                  frame::interpreter_frame_monitor_block_top_offset * wordSize);
 941     __ movptr(rax, monitor_block_top);
 942     __ cmpptr(rax, rsp);
 943     __ jcc(Assembler::equal, L);
 944     __ stop("broken stack frame setup in interpreter");
 945     __ bind(L);
 946   }
 947 #endif
 948 
 949   // jvmti support
 950   __ notify_method_entry();
 951 
 952   // work registers
 953   const Register method = rbx;
 954   const Register t      = r11;
 955 
 956   // allocate space for parameters
 957   __ get_method(method);
 958   __ verify_oop(method);
 959   __ load_unsigned_short(t,
 960                          Address(method,
 961                                  methodOopDesc::size_of_parameters_offset()));
 962   __ shll(t, Interpreter::logStackElementSize);
 963 
 964   __ subptr(rsp, t);
 965   __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
 966   __ andptr(rsp, -16); // must be 16 byte boundary (see amd64 ABI)
 967 
 968   // get signature handler
 969   {
 970     Label L;
 971     __ movptr(t, Address(method, methodOopDesc::signature_handler_offset()));
 972     __ testptr(t, t);
 973     __ jcc(Assembler::notZero, L);
 974     __ call_VM(noreg,
 975                CAST_FROM_FN_PTR(address,
 976                                 InterpreterRuntime::prepare_native_call),
 977                method);
 978     __ get_method(method);
 979     __ movptr(t, Address(method, methodOopDesc::signature_handler_offset()));
 980     __ bind(L);
 981   }
 982 
 983   // call signature handler
 984   assert(InterpreterRuntime::SignatureHandlerGenerator::from() == r14,
 985          "adjust this code");
 986   assert(InterpreterRuntime::SignatureHandlerGenerator::to() == rsp,
 987          "adjust this code");
 988   assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == rscratch1,
 989           "adjust this code");
 990 
 991   // The generated handlers do not touch RBX (the method oop).
 992   // However, large signatures cannot be cached and are generated
 993   // each time here.  The slow-path generator can do a GC on return,
 994   // so we must reload it after the call.
 995   __ call(t);
 996   __ get_method(method);        // slow path can do a GC, reload RBX
 997 
 998 
 999   // result handler is in rax
1000   // set result handler
1001   __ movptr(Address(rbp,
1002                     (frame::interpreter_frame_result_handler_offset) * wordSize),
1003             rax);
1004 
1005   // pass mirror handle if static call
1006   {
1007     Label L;
1008     const int mirror_offset = in_bytes(Klass::java_mirror_offset());
1009     __ movl(t, Address(method, methodOopDesc::access_flags_offset()));
1010     __ testl(t, JVM_ACC_STATIC);
1011     __ jcc(Assembler::zero, L);
1012     // get mirror
1013     __ movptr(t, Address(method, methodOopDesc::constants_offset()));
1014     __ movptr(t, Address(t, constantPoolOopDesc::pool_holder_offset_in_bytes()));
1015     __ movptr(t, Address(t, mirror_offset));
1016     // copy mirror into activation frame
1017     __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize),
1018             t);
1019     // pass handle to mirror
1020     __ lea(c_rarg1,
1021            Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize));
1022     __ bind(L);
1023   }
1024 
1025   // get native function entry point
1026   {
1027     Label L;
1028     __ movptr(rax, Address(method, methodOopDesc::native_function_offset()));
1029     ExternalAddress unsatisfied(SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
1030     __ movptr(rscratch2, unsatisfied.addr());
1031     __ cmpptr(rax, rscratch2);
1032     __ jcc(Assembler::notEqual, L);
1033     __ call_VM(noreg,
1034                CAST_FROM_FN_PTR(address,
1035                                 InterpreterRuntime::prepare_native_call),
1036                method);
1037     __ get_method(method);
1038     __ verify_oop(method);
1039     __ movptr(rax, Address(method, methodOopDesc::native_function_offset()));
1040     __ bind(L);
1041   }
1042 
1043   // pass JNIEnv
1044   __ lea(c_rarg0, Address(r15_thread, JavaThread::jni_environment_offset()));
1045 
1046   // It is enough that the pc() points into the right code
1047   // segment. It does not have to be the correct return pc.
1048   __ set_last_Java_frame(rsp, rbp, (address) __ pc());
1049 
1050   // change thread state
1051 #ifdef ASSERT
1052   {
1053     Label L;
1054     __ movl(t, Address(r15_thread, JavaThread::thread_state_offset()));
1055     __ cmpl(t, _thread_in_Java);
1056     __ jcc(Assembler::equal, L);
1057     __ stop("Wrong thread state in native stub");
1058     __ bind(L);
1059   }
1060 #endif
1061 
1062   // Change state to native
1063 
1064   __ movl(Address(r15_thread, JavaThread::thread_state_offset()),
1065           _thread_in_native);
1066 
1067   // Call the native method.
1068   __ call(rax);
1069   // result potentially in rax or xmm0
1070 
1071   // Depending on runtime options, either restore the MXCSR
1072   // register after returning from the JNI Call or verify that
1073   // it wasn't changed during -Xcheck:jni.
1074   if (RestoreMXCSROnJNICalls) {
1075     __ ldmxcsr(ExternalAddress(StubRoutines::x86::mxcsr_std()));
1076   }
1077   else if (CheckJNICalls) {
1078     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::x86::verify_mxcsr_entry())));
1079   }
1080 
1081   // NOTE: The order of these pushes is known to frame::interpreter_frame_result
1082   // in order to extract the result of a method call. If the order of these
1083   // pushes change or anything else is added to the stack then the code in
1084   // interpreter_frame_result must also change.
1085 
1086   __ push(dtos);
1087   __ push(ltos);
1088 
1089   // change thread state
1090   __ movl(Address(r15_thread, JavaThread::thread_state_offset()),
1091           _thread_in_native_trans);
1092 
1093   if (os::is_MP()) {
1094     if (UseMembar) {
1095       // Force this write out before the read below
1096       __ membar(Assembler::Membar_mask_bits(
1097            Assembler::LoadLoad | Assembler::LoadStore |
1098            Assembler::StoreLoad | Assembler::StoreStore));
1099     } else {
1100       // Write serialization page so VM thread can do a pseudo remote membar.
1101       // We use the current thread pointer to calculate a thread specific
1102       // offset to write to within the page. This minimizes bus traffic
1103       // due to cache line collision.
1104       __ serialize_memory(r15_thread, rscratch2);
1105     }
1106   }
1107 
1108   // check for safepoint operation in progress and/or pending suspend requests
1109   {
1110     Label Continue;
1111     __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
1112              SafepointSynchronize::_not_synchronized);
1113 
1114     Label L;
1115     __ jcc(Assembler::notEqual, L);
1116     __ cmpl(Address(r15_thread, JavaThread::suspend_flags_offset()), 0);
1117     __ jcc(Assembler::equal, Continue);
1118     __ bind(L);
1119 
1120     // Don't use call_VM as it will see a possible pending exception
1121     // and forward it and never return here preventing us from
1122     // clearing _last_native_pc down below.  Also can't use
1123     // call_VM_leaf either as it will check to see if r13 & r14 are
1124     // preserved and correspond to the bcp/locals pointers. So we do a
1125     // runtime call by hand.
1126     //
1127     __ mov(c_rarg0, r15_thread);
1128     __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1129     __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1130     __ andptr(rsp, -16); // align stack as required by ABI
1131     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans)));
1132     __ mov(rsp, r12); // restore sp
1133     __ reinit_heapbase();
1134     __ bind(Continue);
1135   }
1136 
1137   // change thread state
1138   __ movl(Address(r15_thread, JavaThread::thread_state_offset()), _thread_in_Java);
1139 
1140   // reset_last_Java_frame
1141   __ reset_last_Java_frame(true, true);
1142 
1143   // reset handle block
1144   __ movptr(t, Address(r15_thread, JavaThread::active_handles_offset()));
1145   __ movptr(Address(t, JNIHandleBlock::top_offset_in_bytes()), (int32_t)NULL_WORD);
1146 
1147   // If result is an oop unbox and store it in frame where gc will see it
1148   // and result handler will pick it up
1149 
1150   {
1151     Label no_oop, store_result;
1152     __ lea(t, ExternalAddress(AbstractInterpreter::result_handler(T_OBJECT)));
1153     __ cmpptr(t, Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize));
1154     __ jcc(Assembler::notEqual, no_oop);
1155     // retrieve result
1156     __ pop(ltos);
1157     __ testptr(rax, rax);
1158     __ jcc(Assembler::zero, store_result);
1159     __ movptr(rax, Address(rax, 0));
1160     __ bind(store_result);
1161     __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize), rax);
1162     // keep stack depth as expected by pushing oop which will eventually be discarde
1163     __ push(ltos);
1164     __ bind(no_oop);
1165   }
1166 
1167 
1168   {
1169     Label no_reguard;
1170     __ cmpl(Address(r15_thread, JavaThread::stack_guard_state_offset()),
1171             JavaThread::stack_guard_yellow_disabled);
1172     __ jcc(Assembler::notEqual, no_reguard);
1173 
1174     __ pusha(); // XXX only save smashed registers
1175     __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1176     __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1177     __ andptr(rsp, -16); // align stack as required by ABI
1178     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
1179     __ mov(rsp, r12); // restore sp
1180     __ popa(); // XXX only restore smashed registers
1181     __ reinit_heapbase();
1182 
1183     __ bind(no_reguard);
1184   }
1185 
1186 
1187   // The method register is junk from after the thread_in_native transition
1188   // until here.  Also can't call_VM until the bcp has been
1189   // restored.  Need bcp for throwing exception below so get it now.
1190   __ get_method(method);
1191   __ verify_oop(method);
1192 
1193   // restore r13 to have legal interpreter frame, i.e., bci == 0 <=>
1194   // r13 == code_base()
1195   __ movptr(r13, Address(method, methodOopDesc::const_offset()));   // get constMethodOop
1196   __ lea(r13, Address(r13, constMethodOopDesc::codes_offset()));    // get codebase
1197   // handle exceptions (exception handling will handle unlocking!)
1198   {
1199     Label L;
1200     __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t) NULL_WORD);
1201     __ jcc(Assembler::zero, L);
1202     // Note: At some point we may want to unify this with the code
1203     // used in call_VM_base(); i.e., we should use the
1204     // StubRoutines::forward_exception code. For now this doesn't work
1205     // here because the rsp is not correctly set at this point.
1206     __ MacroAssembler::call_VM(noreg,
1207                                CAST_FROM_FN_PTR(address,
1208                                InterpreterRuntime::throw_pending_exception));
1209     __ should_not_reach_here();
1210     __ bind(L);
1211   }
1212 
1213   // do unlocking if necessary
1214   {
1215     Label L;
1216     __ movl(t, Address(method, methodOopDesc::access_flags_offset()));
1217     __ testl(t, JVM_ACC_SYNCHRONIZED);
1218     __ jcc(Assembler::zero, L);
1219     // the code below should be shared with interpreter macro
1220     // assembler implementation
1221     {
1222       Label unlock;
1223       // BasicObjectLock will be first in list, since this is a
1224       // synchronized method. However, need to check that the object
1225       // has not been unlocked by an explicit monitorexit bytecode.
1226       const Address monitor(rbp,
1227                             (intptr_t)(frame::interpreter_frame_initial_sp_offset *
1228                                        wordSize - sizeof(BasicObjectLock)));
1229 
1230       // monitor expect in c_rarg1 for slow unlock path
1231       __ lea(c_rarg1, monitor); // address of first monitor
1232 
1233       __ movptr(t, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
1234       __ testptr(t, t);
1235       __ jcc(Assembler::notZero, unlock);
1236 
1237       // Entry already unlocked, need to throw exception
1238       __ MacroAssembler::call_VM(noreg,
1239                                  CAST_FROM_FN_PTR(address,
1240                    InterpreterRuntime::throw_illegal_monitor_state_exception));
1241       __ should_not_reach_here();
1242 
1243       __ bind(unlock);
1244       __ unlock_object(c_rarg1);
1245     }
1246     __ bind(L);
1247   }
1248 
1249   // jvmti support
1250   // Note: This must happen _after_ handling/throwing any exceptions since
1251   //       the exception handler code notifies the runtime of method exits
1252   //       too. If this happens before, method entry/exit notifications are
1253   //       not properly paired (was bug - gri 11/22/99).
1254   __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI);
1255 
1256   // restore potential result in edx:eax, call result handler to
1257   // restore potential result in ST0 & handle result
1258 
1259   __ pop(ltos);
1260   __ pop(dtos);
1261 
1262   __ movptr(t, Address(rbp,
1263                        (frame::interpreter_frame_result_handler_offset) * wordSize));
1264   __ call(t);
1265 
1266   // remove activation
1267   __ movptr(t, Address(rbp,
1268                        frame::interpreter_frame_sender_sp_offset *
1269                        wordSize)); // get sender sp
1270   __ leave();                                // remove frame anchor
1271   __ pop(rdi);                               // get return address
1272   __ mov(rsp, t);                            // set sp to sender sp
1273   __ jmp(rdi);
1274 
1275   if (inc_counter) {
1276     // Handle overflow of counter and compile method
1277     __ bind(invocation_counter_overflow);
1278     generate_counter_overflow(&continue_after_compile);
1279   }
1280 
1281   return entry_point;
1282 }
1283 
1284 //
1285 // Generic interpreted method entry to (asm) interpreter
1286 //
1287 address InterpreterGenerator::generate_normal_entry(bool synchronized) {
1288   // determine code generation flags
1289   bool inc_counter  = UseCompiler || CountCompiledCalls;
1290 
1291   // ebx: methodOop
1292   // r13: sender sp
1293   address entry_point = __ pc();
1294 
1295   const Address size_of_parameters(rbx,
1296                                    methodOopDesc::size_of_parameters_offset());
1297   const Address size_of_locals(rbx, methodOopDesc::size_of_locals_offset());
1298   const Address invocation_counter(rbx,
1299                                    methodOopDesc::invocation_counter_offset() +
1300                                    InvocationCounter::counter_offset());
1301   const Address access_flags(rbx, methodOopDesc::access_flags_offset());
1302 
1303   // get parameter size (always needed)
1304   __ load_unsigned_short(rcx, size_of_parameters);
1305 
1306   // rbx: methodOop
1307   // rcx: size of parameters
1308   // r13: sender_sp (could differ from sp+wordSize if we were called via c2i )
1309 
1310   __ load_unsigned_short(rdx, size_of_locals); // get size of locals in words
1311   __ subl(rdx, rcx); // rdx = no. of additional locals
1312 
1313   // YYY
1314 //   __ incrementl(rdx);
1315 //   __ andl(rdx, -2);
1316 
1317   // see if we've got enough room on the stack for locals plus overhead.
1318   generate_stack_overflow_check();
1319 
1320   // get return address
1321   __ pop(rax);
1322 
1323   // compute beginning of parameters (r14)
1324   __ lea(r14, Address(rsp, rcx, Address::times_8, -wordSize));
1325 
1326   // rdx - # of additional locals
1327   // allocate space for locals
1328   // explicitly initialize locals
1329   {
1330     Label exit, loop;
1331     __ testl(rdx, rdx);
1332     __ jcc(Assembler::lessEqual, exit); // do nothing if rdx <= 0
1333     __ bind(loop);
1334     __ push((int) NULL_WORD); // initialize local variables
1335     __ decrementl(rdx); // until everything initialized
1336     __ jcc(Assembler::greater, loop);
1337     __ bind(exit);
1338   }
1339 
1340   // (pre-)fetch invocation count
1341   if (inc_counter) {
1342     __ movl(rcx, invocation_counter);
1343   }
1344   // initialize fixed part of activation frame
1345   generate_fixed_frame(false);
1346 
1347   // make sure method is not native & not abstract
1348 #ifdef ASSERT
1349   __ movl(rax, access_flags);
1350   {
1351     Label L;
1352     __ testl(rax, JVM_ACC_NATIVE);
1353     __ jcc(Assembler::zero, L);
1354     __ stop("tried to execute native method as non-native");
1355     __ bind(L);
1356   }
1357   {
1358     Label L;
1359     __ testl(rax, JVM_ACC_ABSTRACT);
1360     __ jcc(Assembler::zero, L);
1361     __ stop("tried to execute abstract method in interpreter");
1362     __ bind(L);
1363   }
1364 #endif
1365 
1366   // Since at this point in the method invocation the exception
1367   // handler would try to exit the monitor of synchronized methods
1368   // which hasn't been entered yet, we set the thread local variable
1369   // _do_not_unlock_if_synchronized to true. The remove_activation
1370   // will check this flag.
1371 
1372   const Address do_not_unlock_if_synchronized(r15_thread,
1373         in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1374   __ movbool(do_not_unlock_if_synchronized, true);
1375 
1376   // increment invocation count & check for overflow
1377   Label invocation_counter_overflow;
1378   Label profile_method;
1379   Label profile_method_continue;
1380   if (inc_counter) {
1381     generate_counter_incr(&invocation_counter_overflow,
1382                           &profile_method,
1383                           &profile_method_continue);
1384     if (ProfileInterpreter) {
1385       __ bind(profile_method_continue);
1386     }
1387   }
1388 
1389   Label continue_after_compile;
1390   __ bind(continue_after_compile);
1391 
1392   // check for synchronized interpreted methods
1393   bang_stack_shadow_pages(false);
1394 
1395   // reset the _do_not_unlock_if_synchronized flag
1396   __ movbool(do_not_unlock_if_synchronized, false);
1397 
1398   // check for synchronized methods
1399   // Must happen AFTER invocation_counter check and stack overflow check,
1400   // so method is not locked if overflows.
1401   if (synchronized) {
1402     // Allocate monitor and lock method
1403     lock_method();
1404   } else {
1405     // no synchronization necessary
1406 #ifdef ASSERT
1407     {
1408       Label L;
1409       __ movl(rax, access_flags);
1410       __ testl(rax, JVM_ACC_SYNCHRONIZED);
1411       __ jcc(Assembler::zero, L);
1412       __ stop("method needs synchronization");
1413       __ bind(L);
1414     }
1415 #endif
1416   }
1417 
1418   // start execution
1419 #ifdef ASSERT
1420   {
1421     Label L;
1422      const Address monitor_block_top (rbp,
1423                  frame::interpreter_frame_monitor_block_top_offset * wordSize);
1424     __ movptr(rax, monitor_block_top);
1425     __ cmpptr(rax, rsp);
1426     __ jcc(Assembler::equal, L);
1427     __ stop("broken stack frame setup in interpreter");
1428     __ bind(L);
1429   }
1430 #endif
1431 
1432   // jvmti support
1433   __ notify_method_entry();
1434 
1435   __ dispatch_next(vtos);
1436 
1437   // invocation counter overflow
1438   if (inc_counter) {
1439     if (ProfileInterpreter) {
1440       // We have decided to profile this method in the interpreter
1441       __ bind(profile_method);
1442       __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1443       __ set_method_data_pointer_for_bcp();
1444       __ get_method(rbx);
1445       __ jmp(profile_method_continue);
1446     }
1447     // Handle overflow of counter and compile method
1448     __ bind(invocation_counter_overflow);
1449     generate_counter_overflow(&continue_after_compile);
1450   }
1451 
1452   return entry_point;
1453 }
1454 
1455 // Entry points
1456 //
1457 // Here we generate the various kind of entries into the interpreter.
1458 // The two main entry type are generic bytecode methods and native
1459 // call method.  These both come in synchronized and non-synchronized
1460 // versions but the frame layout they create is very similar. The
1461 // other method entry types are really just special purpose entries
1462 // that are really entry and interpretation all in one. These are for
1463 // trivial methods like accessor, empty, or special math methods.
1464 //
1465 // When control flow reaches any of the entry types for the interpreter
1466 // the following holds ->
1467 //
1468 // Arguments:
1469 //
1470 // rbx: methodOop
1471 //
1472 // Stack layout immediately at entry
1473 //
1474 // [ return address     ] <--- rsp
1475 // [ parameter n        ]
1476 //   ...
1477 // [ parameter 1        ]
1478 // [ expression stack   ] (caller's java expression stack)
1479 
1480 // Assuming that we don't go to one of the trivial specialized entries
1481 // the stack will look like below when we are ready to execute the
1482 // first bytecode (or call the native routine). The register usage
1483 // will be as the template based interpreter expects (see
1484 // interpreter_amd64.hpp).
1485 //
1486 // local variables follow incoming parameters immediately; i.e.
1487 // the return address is moved to the end of the locals).
1488 //
1489 // [ monitor entry      ] <--- rsp
1490 //   ...
1491 // [ monitor entry      ]
1492 // [ expr. stack bottom ]
1493 // [ saved r13          ]
1494 // [ current r14        ]
1495 // [ methodOop          ]
1496 // [ saved ebp          ] <--- rbp
1497 // [ return address     ]
1498 // [ local variable m   ]
1499 //   ...
1500 // [ local variable 1   ]
1501 // [ parameter n        ]
1502 //   ...
1503 // [ parameter 1        ] <--- r14
1504 
1505 address AbstractInterpreterGenerator::generate_method_entry(
1506                                         AbstractInterpreter::MethodKind kind) {
1507   // determine code generation flags
1508   bool synchronized = false;
1509   address entry_point = NULL;
1510 
1511   switch (kind) {
1512   case Interpreter::zerolocals             :                                                                             break;
1513   case Interpreter::zerolocals_synchronized: synchronized = true;                                                        break;
1514   case Interpreter::native                 : entry_point = ((InterpreterGenerator*) this)->generate_native_entry(false); break;
1515   case Interpreter::native_synchronized    : entry_point = ((InterpreterGenerator*) this)->generate_native_entry(true);  break;
1516   case Interpreter::empty                  : entry_point = ((InterpreterGenerator*) this)->generate_empty_entry();       break;
1517   case Interpreter::accessor               : entry_point = ((InterpreterGenerator*) this)->generate_accessor_entry();    break;
1518   case Interpreter::abstract               : entry_point = ((InterpreterGenerator*) this)->generate_abstract_entry();    break;
1519   case Interpreter::method_handle          : entry_point = ((InterpreterGenerator*) this)->generate_method_handle_entry();break;
1520 
1521   case Interpreter::java_lang_math_sin     : // fall thru
1522   case Interpreter::java_lang_math_cos     : // fall thru
1523   case Interpreter::java_lang_math_tan     : // fall thru
1524   case Interpreter::java_lang_math_abs     : // fall thru
1525   case Interpreter::java_lang_math_log     : // fall thru
1526   case Interpreter::java_lang_math_log10   : // fall thru
1527   case Interpreter::java_lang_math_sqrt    : entry_point = ((InterpreterGenerator*) this)->generate_math_entry(kind);    break;
1528   case Interpreter::java_lang_ref_reference_get
1529                                            : entry_point = ((InterpreterGenerator*)this)->generate_Reference_get_entry(); break;
1530   default                                  : ShouldNotReachHere();                                                       break;
1531   }
1532 
1533   if (entry_point) {
1534     return entry_point;
1535   }
1536 
1537   return ((InterpreterGenerator*) this)->
1538                                 generate_normal_entry(synchronized);
1539 }
1540 
1541 // These should never be compiled since the interpreter will prefer
1542 // the compiled version to the intrinsic version.
1543 bool AbstractInterpreter::can_be_compiled(methodHandle m) {
1544   switch (method_kind(m)) {
1545     case Interpreter::java_lang_math_sin     : // fall thru
1546     case Interpreter::java_lang_math_cos     : // fall thru
1547     case Interpreter::java_lang_math_tan     : // fall thru
1548     case Interpreter::java_lang_math_abs     : // fall thru
1549     case Interpreter::java_lang_math_log     : // fall thru
1550     case Interpreter::java_lang_math_log10   : // fall thru
1551     case Interpreter::java_lang_math_sqrt    :
1552       return false;
1553     default:
1554       return true;
1555   }
1556 }
1557 
1558 // How much stack a method activation needs in words.
1559 int AbstractInterpreter::size_top_interpreter_activation(methodOop method) {
1560   const int entry_size = frame::interpreter_frame_monitor_size();
1561 
1562   // total overhead size: entry_size + (saved rbp thru expr stack
1563   // bottom).  be sure to change this if you add/subtract anything
1564   // to/from the overhead area
1565   const int overhead_size =
1566     -(frame::interpreter_frame_initial_sp_offset) + entry_size;
1567 
1568   const int stub_code = frame::entry_frame_after_call_words;
1569   const int extra_stack = methodOopDesc::extra_stack_entries();
1570   const int method_stack = (method->max_locals() + method->max_stack() + extra_stack) *
1571                            Interpreter::stackElementWords;
1572   return (overhead_size + method_stack + stub_code);
1573 }
1574 
1575 int AbstractInterpreter::layout_activation(methodOop method,
1576                                            int tempcount,
1577                                            int popframe_extra_args,
1578                                            int moncount,
1579                                            int caller_actual_parameters,
1580                                            int callee_param_count,
1581                                            int callee_locals,
1582                                            frame* caller,
1583                                            frame* interpreter_frame,
1584                                            bool is_top_frame) {
1585   // Note: This calculation must exactly parallel the frame setup
1586   // in AbstractInterpreterGenerator::generate_method_entry.
1587   // If interpreter_frame!=NULL, set up the method, locals, and monitors.
1588   // The frame interpreter_frame, if not NULL, is guaranteed to be the
1589   // right size, as determined by a previous call to this method.
1590   // It is also guaranteed to be walkable even though it is in a skeletal state
1591 
1592   // fixed size of an interpreter frame:
1593   int max_locals = method->max_locals() * Interpreter::stackElementWords;
1594   int extra_locals = (method->max_locals() - method->size_of_parameters()) *
1595                      Interpreter::stackElementWords;
1596 
1597   int overhead = frame::sender_sp_offset -
1598                  frame::interpreter_frame_initial_sp_offset;
1599   // Our locals were accounted for by the caller (or last_frame_adjust
1600   // on the transistion) Since the callee parameters already account
1601   // for the callee's params we only need to account for the extra
1602   // locals.
1603   int size = overhead +
1604          (callee_locals - callee_param_count)*Interpreter::stackElementWords +
1605          moncount * frame::interpreter_frame_monitor_size() +
1606          tempcount* Interpreter::stackElementWords + popframe_extra_args;
1607   if (interpreter_frame != NULL) {
1608 #ifdef ASSERT
1609     if (!EnableInvokeDynamic)
1610       // @@@ FIXME: Should we correct interpreter_frame_sender_sp in the calling sequences?
1611       // Probably, since deoptimization doesn't work yet.
1612       assert(caller->unextended_sp() == interpreter_frame->interpreter_frame_sender_sp(), "Frame not properly walkable");
1613     assert(caller->sp() == interpreter_frame->sender_sp(), "Frame not properly walkable(2)");
1614 #endif
1615 
1616     interpreter_frame->interpreter_frame_set_method(method);
1617     // NOTE the difference in using sender_sp and
1618     // interpreter_frame_sender_sp interpreter_frame_sender_sp is
1619     // the original sp of the caller (the unextended_sp) and
1620     // sender_sp is fp+16 XXX
1621     intptr_t* locals = interpreter_frame->sender_sp() + max_locals - 1;
1622 
1623 #ifdef ASSERT
1624     if (caller->is_interpreted_frame()) {
1625       assert(locals < caller->fp() + frame::interpreter_frame_initial_sp_offset, "bad placement");
1626     }
1627 #endif
1628 
1629     interpreter_frame->interpreter_frame_set_locals(locals);
1630     BasicObjectLock* montop = interpreter_frame->interpreter_frame_monitor_begin();
1631     BasicObjectLock* monbot = montop - moncount;
1632     interpreter_frame->interpreter_frame_set_monitor_end(monbot);
1633 
1634     // Set last_sp
1635     intptr_t*  esp = (intptr_t*) monbot -
1636                      tempcount*Interpreter::stackElementWords -
1637                      popframe_extra_args;
1638     interpreter_frame->interpreter_frame_set_last_sp(esp);
1639 
1640     // All frames but the initial (oldest) interpreter frame we fill in have
1641     // a value for sender_sp that allows walking the stack but isn't
1642     // truly correct. Correct the value here.
1643     if (extra_locals != 0 &&
1644         interpreter_frame->sender_sp() ==
1645         interpreter_frame->interpreter_frame_sender_sp()) {
1646       interpreter_frame->set_interpreter_frame_sender_sp(caller->sp() +
1647                                                          extra_locals);
1648     }
1649     *interpreter_frame->interpreter_frame_cache_addr() =
1650       method->constants()->cache();
1651   }
1652   return size;
1653 }
1654 
1655 //-----------------------------------------------------------------------------
1656 // Exceptions
1657 
1658 void TemplateInterpreterGenerator::generate_throw_exception() {
1659   // Entry point in previous activation (i.e., if the caller was
1660   // interpreted)
1661   Interpreter::_rethrow_exception_entry = __ pc();
1662   // Restore sp to interpreter_frame_last_sp even though we are going
1663   // to empty the expression stack for the exception processing.
1664   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
1665   // rax: exception
1666   // rdx: return address/pc that threw exception
1667   __ restore_bcp();    // r13 points to call/send
1668   __ restore_locals();
1669   __ reinit_heapbase();  // restore r12 as heapbase.
1670   // Entry point for exceptions thrown within interpreter code
1671   Interpreter::_throw_exception_entry = __ pc();
1672   // expression stack is undefined here
1673   // rax: exception
1674   // r13: exception bcp
1675   __ verify_oop(rax);
1676   __ mov(c_rarg1, rax);
1677 
1678   // expression stack must be empty before entering the VM in case of
1679   // an exception
1680   __ empty_expression_stack();
1681   // find exception handler address and preserve exception oop
1682   __ call_VM(rdx,
1683              CAST_FROM_FN_PTR(address,
1684                           InterpreterRuntime::exception_handler_for_exception),
1685              c_rarg1);
1686   // rax: exception handler entry point
1687   // rdx: preserved exception oop
1688   // r13: bcp for exception handler
1689   __ push_ptr(rdx); // push exception which is now the only value on the stack
1690   __ jmp(rax); // jump to exception handler (may be _remove_activation_entry!)
1691 
1692   // If the exception is not handled in the current frame the frame is
1693   // removed and the exception is rethrown (i.e. exception
1694   // continuation is _rethrow_exception).
1695   //
1696   // Note: At this point the bci is still the bxi for the instruction
1697   // which caused the exception and the expression stack is
1698   // empty. Thus, for any VM calls at this point, GC will find a legal
1699   // oop map (with empty expression stack).
1700 
1701   // In current activation
1702   // tos: exception
1703   // esi: exception bcp
1704 
1705   //
1706   // JVMTI PopFrame support
1707   //
1708 
1709   Interpreter::_remove_activation_preserving_args_entry = __ pc();
1710   __ empty_expression_stack();
1711   // Set the popframe_processing bit in pending_popframe_condition
1712   // indicating that we are currently handling popframe, so that
1713   // call_VMs that may happen later do not trigger new popframe
1714   // handling cycles.
1715   __ movl(rdx, Address(r15_thread, JavaThread::popframe_condition_offset()));
1716   __ orl(rdx, JavaThread::popframe_processing_bit);
1717   __ movl(Address(r15_thread, JavaThread::popframe_condition_offset()), rdx);
1718 
1719   {
1720     // Check to see whether we are returning to a deoptimized frame.
1721     // (The PopFrame call ensures that the caller of the popped frame is
1722     // either interpreted or compiled and deoptimizes it if compiled.)
1723     // In this case, we can't call dispatch_next() after the frame is
1724     // popped, but instead must save the incoming arguments and restore
1725     // them after deoptimization has occurred.
1726     //
1727     // Note that we don't compare the return PC against the
1728     // deoptimization blob's unpack entry because of the presence of
1729     // adapter frames in C2.
1730     Label caller_not_deoptimized;
1731     __ movptr(c_rarg1, Address(rbp, frame::return_addr_offset * wordSize));
1732     __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1733                                InterpreterRuntime::interpreter_contains), c_rarg1);
1734     __ testl(rax, rax);
1735     __ jcc(Assembler::notZero, caller_not_deoptimized);
1736 
1737     // Compute size of arguments for saving when returning to
1738     // deoptimized caller
1739     __ get_method(rax);
1740     __ load_unsigned_short(rax, Address(rax, in_bytes(methodOopDesc::
1741                                                 size_of_parameters_offset())));
1742     __ shll(rax, Interpreter::logStackElementSize);
1743     __ restore_locals(); // XXX do we need this?
1744     __ subptr(r14, rax);
1745     __ addptr(r14, wordSize);
1746     // Save these arguments
1747     __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1748                                            Deoptimization::
1749                                            popframe_preserve_args),
1750                           r15_thread, rax, r14);
1751 
1752     __ remove_activation(vtos, rdx,
1753                          /* throw_monitor_exception */ false,
1754                          /* install_monitor_exception */ false,
1755                          /* notify_jvmdi */ false);
1756 
1757     // Inform deoptimization that it is responsible for restoring
1758     // these arguments
1759     __ movl(Address(r15_thread, JavaThread::popframe_condition_offset()),
1760             JavaThread::popframe_force_deopt_reexecution_bit);
1761 
1762     // Continue in deoptimization handler
1763     __ jmp(rdx);
1764 
1765     __ bind(caller_not_deoptimized);
1766   }
1767 
1768   __ remove_activation(vtos, rdx, /* rdx result (retaddr) is not used */
1769                        /* throw_monitor_exception */ false,
1770                        /* install_monitor_exception */ false,
1771                        /* notify_jvmdi */ false);
1772 
1773   // Finish with popframe handling
1774   // A previous I2C followed by a deoptimization might have moved the
1775   // outgoing arguments further up the stack. PopFrame expects the
1776   // mutations to those outgoing arguments to be preserved and other
1777   // constraints basically require this frame to look exactly as
1778   // though it had previously invoked an interpreted activation with
1779   // no space between the top of the expression stack (current
1780   // last_sp) and the top of stack. Rather than force deopt to
1781   // maintain this kind of invariant all the time we call a small
1782   // fixup routine to move the mutated arguments onto the top of our
1783   // expression stack if necessary.
1784   __ mov(c_rarg1, rsp);
1785   __ movptr(c_rarg2, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1786   // PC must point into interpreter here
1787   __ set_last_Java_frame(noreg, rbp, __ pc());
1788   __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), r15_thread, c_rarg1, c_rarg2);
1789   __ reset_last_Java_frame(true, true);
1790   // Restore the last_sp and null it out
1791   __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1792   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
1793 
1794   __ restore_bcp();  // XXX do we need this?
1795   __ restore_locals(); // XXX do we need this?
1796   // The method data pointer was incremented already during
1797   // call profiling. We have to restore the mdp for the current bcp.
1798   if (ProfileInterpreter) {
1799     __ set_method_data_pointer_for_bcp();
1800   }
1801 
1802   // Clear the popframe condition flag
1803   __ movl(Address(r15_thread, JavaThread::popframe_condition_offset()),
1804           JavaThread::popframe_inactive);
1805 
1806   __ dispatch_next(vtos);
1807   // end of PopFrame support
1808 
1809   Interpreter::_remove_activation_entry = __ pc();
1810 
1811   // preserve exception over this code sequence
1812   __ pop_ptr(rax);
1813   __ movptr(Address(r15_thread, JavaThread::vm_result_offset()), rax);
1814   // remove the activation (without doing throws on illegalMonitorExceptions)
1815   __ remove_activation(vtos, rdx, false, true, false);
1816   // restore exception
1817   __ movptr(rax, Address(r15_thread, JavaThread::vm_result_offset()));
1818   __ movptr(Address(r15_thread, JavaThread::vm_result_offset()), (int32_t)NULL_WORD);
1819   __ verify_oop(rax);
1820 
1821   // In between activations - previous activation type unknown yet
1822   // compute continuation point - the continuation point expects the
1823   // following registers set up:
1824   //
1825   // rax: exception
1826   // rdx: return address/pc that threw exception
1827   // rsp: expression stack of caller
1828   // rbp: ebp of caller
1829   __ push(rax);                                  // save exception
1830   __ push(rdx);                                  // save return address
1831   __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1832                           SharedRuntime::exception_handler_for_return_address),
1833                         r15_thread, rdx);
1834   __ mov(rbx, rax);                              // save exception handler
1835   __ pop(rdx);                                   // restore return address
1836   __ pop(rax);                                   // restore exception
1837   // Note that an "issuing PC" is actually the next PC after the call
1838   __ jmp(rbx);                                   // jump to exception
1839                                                  // handler of caller
1840 }
1841 
1842 
1843 //
1844 // JVMTI ForceEarlyReturn support
1845 //
1846 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
1847   address entry = __ pc();
1848 
1849   __ restore_bcp();
1850   __ restore_locals();
1851   __ empty_expression_stack();
1852   __ load_earlyret_value(state);
1853 
1854   __ movptr(rdx, Address(r15_thread, JavaThread::jvmti_thread_state_offset()));
1855   Address cond_addr(rdx, JvmtiThreadState::earlyret_state_offset());
1856 
1857   // Clear the earlyret state
1858   __ movl(cond_addr, JvmtiThreadState::earlyret_inactive);
1859 
1860   __ remove_activation(state, rsi,
1861                        false, /* throw_monitor_exception */
1862                        false, /* install_monitor_exception */
1863                        true); /* notify_jvmdi */
1864   __ jmp(rsi);
1865 
1866   return entry;
1867 } // end of ForceEarlyReturn support
1868 
1869 
1870 //-----------------------------------------------------------------------------
1871 // Helper for vtos entry point generation
1872 
1873 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
1874                                                          address& bep,
1875                                                          address& cep,
1876                                                          address& sep,
1877                                                          address& aep,
1878                                                          address& iep,
1879                                                          address& lep,
1880                                                          address& fep,
1881                                                          address& dep,
1882                                                          address& vep) {
1883   assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
1884   Label L;
1885   aep = __ pc();  __ push_ptr();  __ jmp(L);
1886   fep = __ pc();  __ push_f();    __ jmp(L);
1887   dep = __ pc();  __ push_d();    __ jmp(L);
1888   lep = __ pc();  __ push_l();    __ jmp(L);
1889   bep = cep = sep =
1890   iep = __ pc();  __ push_i();
1891   vep = __ pc();
1892   __ bind(L);
1893   generate_and_dispatch(t);
1894 }
1895 
1896 
1897 //-----------------------------------------------------------------------------
1898 // Generation of individual instructions
1899 
1900 // helpers for generate_and_dispatch
1901 
1902 
1903 InterpreterGenerator::InterpreterGenerator(StubQueue* code)
1904   : TemplateInterpreterGenerator(code) {
1905    generate_all(); // down here so it can be "virtual"
1906 }
1907 
1908 //-----------------------------------------------------------------------------
1909 
1910 // Non-product code
1911 #ifndef PRODUCT
1912 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
1913   address entry = __ pc();
1914 
1915   __ push(state);
1916   __ push(c_rarg0);
1917   __ push(c_rarg1);
1918   __ push(c_rarg2);
1919   __ push(c_rarg3);
1920   __ mov(c_rarg2, rax);  // Pass itos
1921 #ifdef _WIN64
1922   __ movflt(xmm3, xmm0); // Pass ftos
1923 #endif
1924   __ call_VM(noreg,
1925              CAST_FROM_FN_PTR(address, SharedRuntime::trace_bytecode),
1926              c_rarg1, c_rarg2, c_rarg3);
1927   __ pop(c_rarg3);
1928   __ pop(c_rarg2);
1929   __ pop(c_rarg1);
1930   __ pop(c_rarg0);
1931   __ pop(state);
1932   __ ret(0);                                   // return from result handler
1933 
1934   return entry;
1935 }
1936 
1937 void TemplateInterpreterGenerator::count_bytecode() {
1938   __ incrementl(ExternalAddress((address) &BytecodeCounter::_counter_value));
1939 }
1940 
1941 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
1942   __ incrementl(ExternalAddress((address) &BytecodeHistogram::_counters[t->bytecode()]));
1943 }
1944 
1945 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
1946   __ mov32(rbx, ExternalAddress((address) &BytecodePairHistogram::_index));
1947   __ shrl(rbx, BytecodePairHistogram::log2_number_of_codes);
1948   __ orl(rbx,
1949          ((int) t->bytecode()) <<
1950          BytecodePairHistogram::log2_number_of_codes);
1951   __ mov32(ExternalAddress((address) &BytecodePairHistogram::_index), rbx);
1952   __ lea(rscratch1, ExternalAddress((address) BytecodePairHistogram::_counters));
1953   __ incrementl(Address(rscratch1, rbx, Address::times_4));
1954 }
1955 
1956 
1957 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
1958   // Call a little run-time stub to avoid blow-up for each bytecode.
1959   // The run-time runtime saves the right registers, depending on
1960   // the tosca in-state for the given template.
1961 
1962   assert(Interpreter::trace_code(t->tos_in()) != NULL,
1963          "entry must have been generated");
1964   __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1965   __ andptr(rsp, -16); // align stack as required by ABI
1966   __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in())));
1967   __ mov(rsp, r12); // restore sp
1968   __ reinit_heapbase();
1969 }
1970 
1971 
1972 void TemplateInterpreterGenerator::stop_interpreter_at() {
1973   Label L;
1974   __ cmp32(ExternalAddress((address) &BytecodeCounter::_counter_value),
1975            StopInterpreterAt);
1976   __ jcc(Assembler::notEqual, L);
1977   __ int3();
1978   __ bind(L);
1979 }
1980 #endif // !PRODUCT
1981 #endif // ! CC_INTERP