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   // Restore sender's sp as SP. This is necessary if the sender's
 471   // frame is an extended compiled frame (see gen_c2i_adapter())
 472   // and safer anyway in case of JSR292 adaptations.
 473 
 474   __ pop(rax); // return address must be moved if SP is changed
 475   __ mov(rsp, r13);
 476   __ push(rax);
 477 
 478   // Note: the restored frame is not necessarily interpreted.
 479   // Use the shared runtime version of the StackOverflowError.
 480   assert(StubRoutines::throw_StackOverflowError_entry() != NULL, "stub not yet generated");
 481   __ jump(ExternalAddress(StubRoutines::throw_StackOverflowError_entry()));
 482 
 483   // all done with frame size check
 484   __ bind(after_frame_check);
 485 }
 486 
 487 // Allocate monitor and lock method (asm interpreter)
 488 //
 489 // Args:
 490 //      rbx: methodOop
 491 //      r14: locals
 492 //
 493 // Kills:
 494 //      rax
 495 //      c_rarg0, c_rarg1, c_rarg2, c_rarg3, ...(param regs)
 496 //      rscratch1, rscratch2 (scratch regs)
 497 void InterpreterGenerator::lock_method(void) {
 498   // synchronize method
 499   const Address access_flags(rbx, methodOopDesc::access_flags_offset());
 500   const Address monitor_block_top(
 501         rbp,
 502         frame::interpreter_frame_monitor_block_top_offset * wordSize);
 503   const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
 504 
 505 #ifdef ASSERT
 506   {
 507     Label L;
 508     __ movl(rax, access_flags);
 509     __ testl(rax, JVM_ACC_SYNCHRONIZED);
 510     __ jcc(Assembler::notZero, L);
 511     __ stop("method doesn't need synchronization");
 512     __ bind(L);
 513   }
 514 #endif // ASSERT
 515 
 516   // get synchronization object
 517   {
 518     const int mirror_offset = in_bytes(Klass::java_mirror_offset());
 519     Label done;
 520     __ movl(rax, access_flags);
 521     __ testl(rax, JVM_ACC_STATIC);
 522     // get receiver (assume this is frequent case)
 523     __ movptr(rax, Address(r14, Interpreter::local_offset_in_bytes(0)));
 524     __ jcc(Assembler::zero, done);
 525     __ movptr(rax, Address(rbx, methodOopDesc::constants_offset()));
 526     __ movptr(rax, Address(rax,
 527                            constantPoolOopDesc::pool_holder_offset_in_bytes()));
 528     __ movptr(rax, Address(rax, mirror_offset));
 529 
 530 #ifdef ASSERT
 531     {
 532       Label L;
 533       __ testptr(rax, rax);
 534       __ jcc(Assembler::notZero, L);
 535       __ stop("synchronization object is NULL");
 536       __ bind(L);
 537     }
 538 #endif // ASSERT
 539 
 540     __ bind(done);
 541   }
 542 
 543   // add space for monitor & lock
 544   __ subptr(rsp, entry_size); // add space for a monitor entry
 545   __ movptr(monitor_block_top, rsp);  // set new monitor block top
 546   // store object
 547   __ movptr(Address(rsp, BasicObjectLock::obj_offset_in_bytes()), rax);
 548   __ movptr(c_rarg1, rsp); // object address
 549   __ lock_object(c_rarg1);
 550 }
 551 
 552 // Generate a fixed interpreter frame. This is identical setup for
 553 // interpreted methods and for native methods hence the shared code.
 554 //
 555 // Args:
 556 //      rax: return address
 557 //      rbx: methodOop
 558 //      r14: pointer to locals
 559 //      r13: sender sp
 560 //      rdx: cp cache
 561 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
 562   // initialize fixed part of activation frame
 563   __ push(rax);        // save return address
 564   __ enter();          // save old & set new rbp
 565   __ push(r13);        // set sender sp
 566   __ push((int)NULL_WORD); // leave last_sp as null
 567   __ movptr(r13, Address(rbx, methodOopDesc::const_offset()));      // get constMethodOop
 568   __ lea(r13, Address(r13, constMethodOopDesc::codes_offset())); // get codebase
 569   __ push(rbx);        // save methodOop
 570   if (ProfileInterpreter) {
 571     Label method_data_continue;
 572     __ movptr(rdx, Address(rbx, in_bytes(methodOopDesc::method_data_offset())));
 573     __ testptr(rdx, rdx);
 574     __ jcc(Assembler::zero, method_data_continue);
 575     __ addptr(rdx, in_bytes(methodDataOopDesc::data_offset()));
 576     __ bind(method_data_continue);
 577     __ push(rdx);      // set the mdp (method data pointer)
 578   } else {
 579     __ push(0);
 580   }
 581 
 582   __ movptr(rdx, Address(rbx, methodOopDesc::constants_offset()));
 583   __ movptr(rdx, Address(rdx, constantPoolOopDesc::cache_offset_in_bytes()));
 584   __ push(rdx); // set constant pool cache
 585   __ push(r14); // set locals pointer
 586   if (native_call) {
 587     __ push(0); // no bcp
 588   } else {
 589     __ push(r13); // set bcp
 590   }
 591   __ push(0); // reserve word for pointer to expression stack bottom
 592   __ movptr(Address(rsp, 0), rsp); // set expression stack bottom
 593 }
 594 
 595 // End of helpers
 596 
 597 // Various method entries
 598 //------------------------------------------------------------------------------------------------------------------------
 599 //
 600 //
 601 
 602 // Call an accessor method (assuming it is resolved, otherwise drop
 603 // into vanilla (slow path) entry
 604 address InterpreterGenerator::generate_accessor_entry(void) {
 605   // rbx: methodOop
 606 
 607   // r13: senderSP must preserver for slow path, set SP to it on fast path
 608 
 609   address entry_point = __ pc();
 610   Label xreturn_path;
 611 
 612   // do fastpath for resolved accessor methods
 613   if (UseFastAccessorMethods) {
 614     // Code: _aload_0, _(i|a)getfield, _(i|a)return or any rewrites
 615     //       thereof; parameter size = 1
 616     // Note: We can only use this code if the getfield has been resolved
 617     //       and if we don't have a null-pointer exception => check for
 618     //       these conditions first and use slow path if necessary.
 619     Label slow_path;
 620     // If we need a safepoint check, generate full interpreter entry.
 621     __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
 622              SafepointSynchronize::_not_synchronized);
 623 
 624     __ jcc(Assembler::notEqual, slow_path);
 625     // rbx: method
 626     __ movptr(rax, Address(rsp, wordSize));
 627 
 628     // check if local 0 != NULL and read field
 629     __ testptr(rax, rax);
 630     __ jcc(Assembler::zero, slow_path);
 631 
 632     __ movptr(rdi, Address(rbx, methodOopDesc::constants_offset()));
 633     // read first instruction word and extract bytecode @ 1 and index @ 2
 634     __ movptr(rdx, Address(rbx, methodOopDesc::const_offset()));
 635     __ movl(rdx, Address(rdx, constMethodOopDesc::codes_offset()));
 636     // Shift codes right to get the index on the right.
 637     // The bytecode fetched looks like <index><0xb4><0x2a>
 638     __ shrl(rdx, 2 * BitsPerByte);
 639     __ shll(rdx, exact_log2(in_words(ConstantPoolCacheEntry::size())));
 640     __ movptr(rdi, Address(rdi, constantPoolOopDesc::cache_offset_in_bytes()));
 641 
 642     // rax: local 0
 643     // rbx: method
 644     // rdx: constant pool cache index
 645     // rdi: constant pool cache
 646 
 647     // check if getfield has been resolved and read constant pool cache entry
 648     // check the validity of the cache entry by testing whether _indices field
 649     // contains Bytecode::_getfield in b1 byte.
 650     assert(in_words(ConstantPoolCacheEntry::size()) == 4,
 651            "adjust shift below");
 652     __ movl(rcx,
 653             Address(rdi,
 654                     rdx,
 655                     Address::times_8,
 656                     constantPoolCacheOopDesc::base_offset() +
 657                     ConstantPoolCacheEntry::indices_offset()));
 658     __ shrl(rcx, 2 * BitsPerByte);
 659     __ andl(rcx, 0xFF);
 660     __ cmpl(rcx, Bytecodes::_getfield);
 661     __ jcc(Assembler::notEqual, slow_path);
 662 
 663     // Note: constant pool entry is not valid before bytecode is resolved
 664     __ movptr(rcx,
 665               Address(rdi,
 666                       rdx,
 667                       Address::times_8,
 668                       constantPoolCacheOopDesc::base_offset() +
 669                       ConstantPoolCacheEntry::f2_offset()));
 670     // edx: flags
 671     __ movl(rdx,
 672             Address(rdi,
 673                     rdx,
 674                     Address::times_8,
 675                     constantPoolCacheOopDesc::base_offset() +
 676                     ConstantPoolCacheEntry::flags_offset()));
 677 
 678     Label notObj, notInt, notByte, notShort;
 679     const Address field_address(rax, rcx, Address::times_1);
 680 
 681     // Need to differentiate between igetfield, agetfield, bgetfield etc.
 682     // because they are different sizes.
 683     // Use the type from the constant pool cache
 684     __ shrl(rdx, ConstantPoolCacheEntry::tosBits);
 685     // Make sure we don't need to mask edx for tosBits after the above shift
 686     ConstantPoolCacheEntry::verify_tosBits();
 687 
 688     __ cmpl(rdx, atos);
 689     __ jcc(Assembler::notEqual, notObj);
 690     // atos
 691     __ load_heap_oop(rax, field_address);
 692     __ jmp(xreturn_path);
 693 
 694     __ bind(notObj);
 695     __ cmpl(rdx, itos);
 696     __ jcc(Assembler::notEqual, notInt);
 697     // itos
 698     __ movl(rax, field_address);
 699     __ jmp(xreturn_path);
 700 
 701     __ bind(notInt);
 702     __ cmpl(rdx, btos);
 703     __ jcc(Assembler::notEqual, notByte);
 704     // btos
 705     __ load_signed_byte(rax, field_address);
 706     __ jmp(xreturn_path);
 707 
 708     __ bind(notByte);
 709     __ cmpl(rdx, stos);
 710     __ jcc(Assembler::notEqual, notShort);
 711     // stos
 712     __ load_signed_short(rax, field_address);
 713     __ jmp(xreturn_path);
 714 
 715     __ bind(notShort);
 716 #ifdef ASSERT
 717     Label okay;
 718     __ cmpl(rdx, ctos);
 719     __ jcc(Assembler::equal, okay);
 720     __ stop("what type is this?");
 721     __ bind(okay);
 722 #endif
 723     // ctos
 724     __ load_unsigned_short(rax, field_address);
 725 
 726     __ bind(xreturn_path);
 727 
 728     // _ireturn/_areturn
 729     __ pop(rdi);
 730     __ mov(rsp, r13);
 731     __ jmp(rdi);
 732     __ ret(0);
 733 
 734     // generate a vanilla interpreter entry as the slow path
 735     __ bind(slow_path);
 736     (void) generate_normal_entry(false);
 737   } else {
 738     (void) generate_normal_entry(false);
 739   }
 740 
 741   return entry_point;
 742 }
 743 
 744 // Method entry for java.lang.ref.Reference.get.
 745 address InterpreterGenerator::generate_Reference_get_entry(void) {
 746 #ifndef SERIALGC
 747   // Code: _aload_0, _getfield, _areturn
 748   // parameter size = 1
 749   //
 750   // The code that gets generated by this routine is split into 2 parts:
 751   //    1. The "intrinsified" code for G1 (or any SATB based GC),
 752   //    2. The slow path - which is an expansion of the regular method entry.
 753   //
 754   // Notes:-
 755   // * In the G1 code we do not check whether we need to block for
 756   //   a safepoint. If G1 is enabled then we must execute the specialized
 757   //   code for Reference.get (except when the Reference object is null)
 758   //   so that we can log the value in the referent field with an SATB
 759   //   update buffer.
 760   //   If the code for the getfield template is modified so that the
 761   //   G1 pre-barrier code is executed when the current method is
 762   //   Reference.get() then going through the normal method entry
 763   //   will be fine.
 764   // * The G1 code can, however, check the receiver object (the instance
 765   //   of java.lang.Reference) and jump to the slow path if null. If the
 766   //   Reference object is null then we obviously cannot fetch the referent
 767   //   and so we don't need to call the G1 pre-barrier. Thus we can use the
 768   //   regular method entry code to generate the NPE.
 769   //
 770   // This code is based on generate_accessor_enty.
 771   //
 772   // rbx: methodOop
 773 
 774   // r13: senderSP must preserve for slow path, set SP to it on fast path
 775 
 776   address entry = __ pc();
 777 
 778   const int referent_offset = java_lang_ref_Reference::referent_offset;
 779   guarantee(referent_offset > 0, "referent offset not initialized");
 780 
 781   if (UseG1GC) {
 782     Label slow_path;
 783     // rbx: method
 784 
 785     // Check if local 0 != NULL
 786     // If the receiver is null then it is OK to jump to the slow path.
 787     __ movptr(rax, Address(rsp, wordSize));
 788 
 789     __ testptr(rax, rax);
 790     __ jcc(Assembler::zero, slow_path);
 791 
 792     // rax: local 0
 793     // rbx: method (but can be used as scratch now)
 794     // rdx: scratch
 795     // rdi: scratch
 796 
 797     // Generate the G1 pre-barrier code to log the value of
 798     // the referent field in an SATB buffer.
 799 
 800     // Load the value of the referent field.
 801     const Address field_address(rax, referent_offset);
 802     __ load_heap_oop(rax, field_address);
 803 
 804     // Generate the G1 pre-barrier code to log the value of
 805     // the referent field in an SATB buffer.
 806     __ g1_write_barrier_pre(noreg /* obj */,
 807                             rax /* pre_val */,
 808                             r15_thread /* thread */,
 809                             rbx /* tmp */,
 810                             true /* tosca_live */,
 811                             true /* expand_call */);
 812 
 813     // _areturn
 814     __ pop(rdi);                // get return address
 815     __ mov(rsp, r13);           // set sp to sender sp
 816     __ jmp(rdi);
 817     __ ret(0);
 818 
 819     // generate a vanilla interpreter entry as the slow path
 820     __ bind(slow_path);
 821     (void) generate_normal_entry(false);
 822 
 823     return entry;
 824   }
 825 #endif // SERIALGC
 826 
 827   // If G1 is not enabled then attempt to go through the accessor entry point
 828   // Reference.get is an accessor
 829   return generate_accessor_entry();
 830 }
 831 
 832 
 833 // Interpreter stub for calling a native method. (asm interpreter)
 834 // This sets up a somewhat different looking stack for calling the
 835 // native method than the typical interpreter frame setup.
 836 address InterpreterGenerator::generate_native_entry(bool synchronized) {
 837   // determine code generation flags
 838   bool inc_counter  = UseCompiler || CountCompiledCalls;
 839 
 840   // rbx: methodOop
 841   // r13: sender sp
 842 
 843   address entry_point = __ pc();
 844 
 845   const Address size_of_parameters(rbx, methodOopDesc::
 846                                         size_of_parameters_offset());
 847   const Address invocation_counter(rbx, methodOopDesc::
 848                                         invocation_counter_offset() +
 849                                         InvocationCounter::counter_offset());
 850   const Address access_flags      (rbx, methodOopDesc::access_flags_offset());
 851 
 852   // get parameter size (always needed)
 853   __ load_unsigned_short(rcx, size_of_parameters);
 854 
 855   // native calls don't need the stack size check since they have no
 856   // expression stack and the arguments are already on the stack and
 857   // we only add a handful of words to the stack
 858 
 859   // rbx: methodOop
 860   // rcx: size of parameters
 861   // r13: sender sp
 862   __ pop(rax);                                       // get return address
 863 
 864   // for natives the size of locals is zero
 865 
 866   // compute beginning of parameters (r14)
 867   __ lea(r14, Address(rsp, rcx, Address::times_8, -wordSize));
 868 
 869   // add 2 zero-initialized slots for native calls
 870   // initialize result_handler slot
 871   __ push((int) NULL_WORD);
 872   // slot for oop temp
 873   // (static native method holder mirror/jni oop result)
 874   __ push((int) NULL_WORD);
 875 
 876   if (inc_counter) {
 877     __ movl(rcx, invocation_counter);  // (pre-)fetch invocation count
 878   }
 879 
 880   // initialize fixed part of activation frame
 881   generate_fixed_frame(true);
 882 
 883   // make sure method is native & not abstract
 884 #ifdef ASSERT
 885   __ movl(rax, access_flags);
 886   {
 887     Label L;
 888     __ testl(rax, JVM_ACC_NATIVE);
 889     __ jcc(Assembler::notZero, L);
 890     __ stop("tried to execute non-native method as native");
 891     __ bind(L);
 892   }
 893   {
 894     Label L;
 895     __ testl(rax, JVM_ACC_ABSTRACT);
 896     __ jcc(Assembler::zero, L);
 897     __ stop("tried to execute abstract method in interpreter");
 898     __ bind(L);
 899   }
 900 #endif
 901 
 902   // Since at this point in the method invocation the exception handler
 903   // would try to exit the monitor of synchronized methods which hasn't
 904   // been entered yet, we set the thread local variable
 905   // _do_not_unlock_if_synchronized to true. The remove_activation will
 906   // check this flag.
 907 
 908   const Address do_not_unlock_if_synchronized(r15_thread,
 909         in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
 910   __ movbool(do_not_unlock_if_synchronized, true);
 911 
 912   // increment invocation count & check for overflow
 913   Label invocation_counter_overflow;
 914   if (inc_counter) {
 915     generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
 916   }
 917 
 918   Label continue_after_compile;
 919   __ bind(continue_after_compile);
 920 
 921   bang_stack_shadow_pages(true);
 922 
 923   // reset the _do_not_unlock_if_synchronized flag
 924   __ movbool(do_not_unlock_if_synchronized, false);
 925 
 926   // check for synchronized methods
 927   // Must happen AFTER invocation_counter check and stack overflow check,
 928   // so method is not locked if overflows.
 929   if (synchronized) {
 930     lock_method();
 931   } else {
 932     // no synchronization necessary
 933 #ifdef ASSERT
 934     {
 935       Label L;
 936       __ movl(rax, access_flags);
 937       __ testl(rax, JVM_ACC_SYNCHRONIZED);
 938       __ jcc(Assembler::zero, L);
 939       __ stop("method needs synchronization");
 940       __ bind(L);
 941     }
 942 #endif
 943   }
 944 
 945   // start execution
 946 #ifdef ASSERT
 947   {
 948     Label L;
 949     const Address monitor_block_top(rbp,
 950                  frame::interpreter_frame_monitor_block_top_offset * wordSize);
 951     __ movptr(rax, monitor_block_top);
 952     __ cmpptr(rax, rsp);
 953     __ jcc(Assembler::equal, L);
 954     __ stop("broken stack frame setup in interpreter");
 955     __ bind(L);
 956   }
 957 #endif
 958 
 959   // jvmti support
 960   __ notify_method_entry();
 961 
 962   // work registers
 963   const Register method = rbx;
 964   const Register t      = r11;
 965 
 966   // allocate space for parameters
 967   __ get_method(method);
 968   __ verify_oop(method);
 969   __ load_unsigned_short(t,
 970                          Address(method,
 971                                  methodOopDesc::size_of_parameters_offset()));
 972   __ shll(t, Interpreter::logStackElementSize);
 973 
 974   __ subptr(rsp, t);
 975   __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
 976   __ andptr(rsp, -16); // must be 16 byte boundary (see amd64 ABI)
 977 
 978   // get signature handler
 979   {
 980     Label L;
 981     __ movptr(t, Address(method, methodOopDesc::signature_handler_offset()));
 982     __ testptr(t, t);
 983     __ jcc(Assembler::notZero, L);
 984     __ call_VM(noreg,
 985                CAST_FROM_FN_PTR(address,
 986                                 InterpreterRuntime::prepare_native_call),
 987                method);
 988     __ get_method(method);
 989     __ movptr(t, Address(method, methodOopDesc::signature_handler_offset()));
 990     __ bind(L);
 991   }
 992 
 993   // call signature handler
 994   assert(InterpreterRuntime::SignatureHandlerGenerator::from() == r14,
 995          "adjust this code");
 996   assert(InterpreterRuntime::SignatureHandlerGenerator::to() == rsp,
 997          "adjust this code");
 998   assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == rscratch1,
 999           "adjust this code");
1000 
1001   // The generated handlers do not touch RBX (the method oop).
1002   // However, large signatures cannot be cached and are generated
1003   // each time here.  The slow-path generator can do a GC on return,
1004   // so we must reload it after the call.
1005   __ call(t);
1006   __ get_method(method);        // slow path can do a GC, reload RBX
1007 
1008 
1009   // result handler is in rax
1010   // set result handler
1011   __ movptr(Address(rbp,
1012                     (frame::interpreter_frame_result_handler_offset) * wordSize),
1013             rax);
1014 
1015   // pass mirror handle if static call
1016   {
1017     Label L;
1018     const int mirror_offset = in_bytes(Klass::java_mirror_offset());
1019     __ movl(t, Address(method, methodOopDesc::access_flags_offset()));
1020     __ testl(t, JVM_ACC_STATIC);
1021     __ jcc(Assembler::zero, L);
1022     // get mirror
1023     __ movptr(t, Address(method, methodOopDesc::constants_offset()));
1024     __ movptr(t, Address(t, constantPoolOopDesc::pool_holder_offset_in_bytes()));
1025     __ movptr(t, Address(t, mirror_offset));
1026     // copy mirror into activation frame
1027     __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize),
1028             t);
1029     // pass handle to mirror
1030     __ lea(c_rarg1,
1031            Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize));
1032     __ bind(L);
1033   }
1034 
1035   // get native function entry point
1036   {
1037     Label L;
1038     __ movptr(rax, Address(method, methodOopDesc::native_function_offset()));
1039     ExternalAddress unsatisfied(SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
1040     __ movptr(rscratch2, unsatisfied.addr());
1041     __ cmpptr(rax, rscratch2);
1042     __ jcc(Assembler::notEqual, L);
1043     __ call_VM(noreg,
1044                CAST_FROM_FN_PTR(address,
1045                                 InterpreterRuntime::prepare_native_call),
1046                method);
1047     __ get_method(method);
1048     __ verify_oop(method);
1049     __ movptr(rax, Address(method, methodOopDesc::native_function_offset()));
1050     __ bind(L);
1051   }
1052 
1053   // pass JNIEnv
1054   __ lea(c_rarg0, Address(r15_thread, JavaThread::jni_environment_offset()));
1055 
1056   // It is enough that the pc() points into the right code
1057   // segment. It does not have to be the correct return pc.
1058   __ set_last_Java_frame(rsp, rbp, (address) __ pc());
1059 
1060   // change thread state
1061 #ifdef ASSERT
1062   {
1063     Label L;
1064     __ movl(t, Address(r15_thread, JavaThread::thread_state_offset()));
1065     __ cmpl(t, _thread_in_Java);
1066     __ jcc(Assembler::equal, L);
1067     __ stop("Wrong thread state in native stub");
1068     __ bind(L);
1069   }
1070 #endif
1071 
1072   // Change state to native
1073 
1074   __ movl(Address(r15_thread, JavaThread::thread_state_offset()),
1075           _thread_in_native);
1076 
1077   // Call the native method.
1078   __ call(rax);
1079   // result potentially in rax or xmm0
1080 
1081   // Depending on runtime options, either restore the MXCSR
1082   // register after returning from the JNI Call or verify that
1083   // it wasn't changed during -Xcheck:jni.
1084   if (RestoreMXCSROnJNICalls) {
1085     __ ldmxcsr(ExternalAddress(StubRoutines::x86::mxcsr_std()));
1086   }
1087   else if (CheckJNICalls) {
1088     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::x86::verify_mxcsr_entry())));
1089   }
1090 
1091   // NOTE: The order of these pushes is known to frame::interpreter_frame_result
1092   // in order to extract the result of a method call. If the order of these
1093   // pushes change or anything else is added to the stack then the code in
1094   // interpreter_frame_result must also change.
1095 
1096   __ push(dtos);
1097   __ push(ltos);
1098 
1099   // change thread state
1100   __ movl(Address(r15_thread, JavaThread::thread_state_offset()),
1101           _thread_in_native_trans);
1102 
1103   if (os::is_MP()) {
1104     if (UseMembar) {
1105       // Force this write out before the read below
1106       __ membar(Assembler::Membar_mask_bits(
1107            Assembler::LoadLoad | Assembler::LoadStore |
1108            Assembler::StoreLoad | Assembler::StoreStore));
1109     } else {
1110       // Write serialization page so VM thread can do a pseudo remote membar.
1111       // We use the current thread pointer to calculate a thread specific
1112       // offset to write to within the page. This minimizes bus traffic
1113       // due to cache line collision.
1114       __ serialize_memory(r15_thread, rscratch2);
1115     }
1116   }
1117 
1118   // check for safepoint operation in progress and/or pending suspend requests
1119   {
1120     Label Continue;
1121     __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
1122              SafepointSynchronize::_not_synchronized);
1123 
1124     Label L;
1125     __ jcc(Assembler::notEqual, L);
1126     __ cmpl(Address(r15_thread, JavaThread::suspend_flags_offset()), 0);
1127     __ jcc(Assembler::equal, Continue);
1128     __ bind(L);
1129 
1130     // Don't use call_VM as it will see a possible pending exception
1131     // and forward it and never return here preventing us from
1132     // clearing _last_native_pc down below.  Also can't use
1133     // call_VM_leaf either as it will check to see if r13 & r14 are
1134     // preserved and correspond to the bcp/locals pointers. So we do a
1135     // runtime call by hand.
1136     //
1137     __ mov(c_rarg0, r15_thread);
1138     __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1139     __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1140     __ andptr(rsp, -16); // align stack as required by ABI
1141     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans)));
1142     __ mov(rsp, r12); // restore sp
1143     __ reinit_heapbase();
1144     __ bind(Continue);
1145   }
1146 
1147   // change thread state
1148   __ movl(Address(r15_thread, JavaThread::thread_state_offset()), _thread_in_Java);
1149 
1150   // reset_last_Java_frame
1151   __ reset_last_Java_frame(true, true);
1152 
1153   // reset handle block
1154   __ movptr(t, Address(r15_thread, JavaThread::active_handles_offset()));
1155   __ movptr(Address(t, JNIHandleBlock::top_offset_in_bytes()), (int32_t)NULL_WORD);
1156 
1157   // If result is an oop unbox and store it in frame where gc will see it
1158   // and result handler will pick it up
1159 
1160   {
1161     Label no_oop, store_result;
1162     __ lea(t, ExternalAddress(AbstractInterpreter::result_handler(T_OBJECT)));
1163     __ cmpptr(t, Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize));
1164     __ jcc(Assembler::notEqual, no_oop);
1165     // retrieve result
1166     __ pop(ltos);
1167     __ testptr(rax, rax);
1168     __ jcc(Assembler::zero, store_result);
1169     __ movptr(rax, Address(rax, 0));
1170     __ bind(store_result);
1171     __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize), rax);
1172     // keep stack depth as expected by pushing oop which will eventually be discarde
1173     __ push(ltos);
1174     __ bind(no_oop);
1175   }
1176 
1177 
1178   {
1179     Label no_reguard;
1180     __ cmpl(Address(r15_thread, JavaThread::stack_guard_state_offset()),
1181             JavaThread::stack_guard_yellow_disabled);
1182     __ jcc(Assembler::notEqual, no_reguard);
1183 
1184     __ pusha(); // XXX only save smashed registers
1185     __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1186     __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1187     __ andptr(rsp, -16); // align stack as required by ABI
1188     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
1189     __ mov(rsp, r12); // restore sp
1190     __ popa(); // XXX only restore smashed registers
1191     __ reinit_heapbase();
1192 
1193     __ bind(no_reguard);
1194   }
1195 
1196 
1197   // The method register is junk from after the thread_in_native transition
1198   // until here.  Also can't call_VM until the bcp has been
1199   // restored.  Need bcp for throwing exception below so get it now.
1200   __ get_method(method);
1201   __ verify_oop(method);
1202 
1203   // restore r13 to have legal interpreter frame, i.e., bci == 0 <=>
1204   // r13 == code_base()
1205   __ movptr(r13, Address(method, methodOopDesc::const_offset()));   // get constMethodOop
1206   __ lea(r13, Address(r13, constMethodOopDesc::codes_offset()));    // get codebase
1207   // handle exceptions (exception handling will handle unlocking!)
1208   {
1209     Label L;
1210     __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t) NULL_WORD);
1211     __ jcc(Assembler::zero, L);
1212     // Note: At some point we may want to unify this with the code
1213     // used in call_VM_base(); i.e., we should use the
1214     // StubRoutines::forward_exception code. For now this doesn't work
1215     // here because the rsp is not correctly set at this point.
1216     __ MacroAssembler::call_VM(noreg,
1217                                CAST_FROM_FN_PTR(address,
1218                                InterpreterRuntime::throw_pending_exception));
1219     __ should_not_reach_here();
1220     __ bind(L);
1221   }
1222 
1223   // do unlocking if necessary
1224   {
1225     Label L;
1226     __ movl(t, Address(method, methodOopDesc::access_flags_offset()));
1227     __ testl(t, JVM_ACC_SYNCHRONIZED);
1228     __ jcc(Assembler::zero, L);
1229     // the code below should be shared with interpreter macro
1230     // assembler implementation
1231     {
1232       Label unlock;
1233       // BasicObjectLock will be first in list, since this is a
1234       // synchronized method. However, need to check that the object
1235       // has not been unlocked by an explicit monitorexit bytecode.
1236       const Address monitor(rbp,
1237                             (intptr_t)(frame::interpreter_frame_initial_sp_offset *
1238                                        wordSize - sizeof(BasicObjectLock)));
1239 
1240       // monitor expect in c_rarg1 for slow unlock path
1241       __ lea(c_rarg1, monitor); // address of first monitor
1242 
1243       __ movptr(t, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
1244       __ testptr(t, t);
1245       __ jcc(Assembler::notZero, unlock);
1246 
1247       // Entry already unlocked, need to throw exception
1248       __ MacroAssembler::call_VM(noreg,
1249                                  CAST_FROM_FN_PTR(address,
1250                    InterpreterRuntime::throw_illegal_monitor_state_exception));
1251       __ should_not_reach_here();
1252 
1253       __ bind(unlock);
1254       __ unlock_object(c_rarg1);
1255     }
1256     __ bind(L);
1257   }
1258 
1259   // jvmti support
1260   // Note: This must happen _after_ handling/throwing any exceptions since
1261   //       the exception handler code notifies the runtime of method exits
1262   //       too. If this happens before, method entry/exit notifications are
1263   //       not properly paired (was bug - gri 11/22/99).
1264   __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI);
1265 
1266   // restore potential result in edx:eax, call result handler to
1267   // restore potential result in ST0 & handle result
1268 
1269   __ pop(ltos);
1270   __ pop(dtos);
1271 
1272   __ movptr(t, Address(rbp,
1273                        (frame::interpreter_frame_result_handler_offset) * wordSize));
1274   __ call(t);
1275 
1276   // remove activation
1277   __ movptr(t, Address(rbp,
1278                        frame::interpreter_frame_sender_sp_offset *
1279                        wordSize)); // get sender sp
1280   __ leave();                                // remove frame anchor
1281   __ pop(rdi);                               // get return address
1282   __ mov(rsp, t);                            // set sp to sender sp
1283   __ jmp(rdi);
1284 
1285   if (inc_counter) {
1286     // Handle overflow of counter and compile method
1287     __ bind(invocation_counter_overflow);
1288     generate_counter_overflow(&continue_after_compile);
1289   }
1290 
1291   return entry_point;
1292 }
1293 
1294 //
1295 // Generic interpreted method entry to (asm) interpreter
1296 //
1297 address InterpreterGenerator::generate_normal_entry(bool synchronized) {
1298   // determine code generation flags
1299   bool inc_counter  = UseCompiler || CountCompiledCalls;
1300 
1301   // ebx: methodOop
1302   // r13: sender sp
1303   address entry_point = __ pc();
1304 
1305   const Address size_of_parameters(rbx,
1306                                    methodOopDesc::size_of_parameters_offset());
1307   const Address size_of_locals(rbx, methodOopDesc::size_of_locals_offset());
1308   const Address invocation_counter(rbx,
1309                                    methodOopDesc::invocation_counter_offset() +
1310                                    InvocationCounter::counter_offset());
1311   const Address access_flags(rbx, methodOopDesc::access_flags_offset());
1312 
1313   // get parameter size (always needed)
1314   __ load_unsigned_short(rcx, size_of_parameters);
1315 
1316   // rbx: methodOop
1317   // rcx: size of parameters
1318   // r13: sender_sp (could differ from sp+wordSize if we were called via c2i )
1319 
1320   __ load_unsigned_short(rdx, size_of_locals); // get size of locals in words
1321   __ subl(rdx, rcx); // rdx = no. of additional locals
1322 
1323   // YYY
1324 //   __ incrementl(rdx);
1325 //   __ andl(rdx, -2);
1326 
1327   // see if we've got enough room on the stack for locals plus overhead.
1328   generate_stack_overflow_check();
1329 
1330   // get return address
1331   __ pop(rax);
1332 
1333   // compute beginning of parameters (r14)
1334   __ lea(r14, Address(rsp, rcx, Address::times_8, -wordSize));
1335 
1336   // rdx - # of additional locals
1337   // allocate space for locals
1338   // explicitly initialize locals
1339   {
1340     Label exit, loop;
1341     __ testl(rdx, rdx);
1342     __ jcc(Assembler::lessEqual, exit); // do nothing if rdx <= 0
1343     __ bind(loop);
1344     __ push((int) NULL_WORD); // initialize local variables
1345     __ decrementl(rdx); // until everything initialized
1346     __ jcc(Assembler::greater, loop);
1347     __ bind(exit);
1348   }
1349 
1350   // (pre-)fetch invocation count
1351   if (inc_counter) {
1352     __ movl(rcx, invocation_counter);
1353   }
1354   // initialize fixed part of activation frame
1355   generate_fixed_frame(false);
1356 
1357   // make sure method is not native & not abstract
1358 #ifdef ASSERT
1359   __ movl(rax, access_flags);
1360   {
1361     Label L;
1362     __ testl(rax, JVM_ACC_NATIVE);
1363     __ jcc(Assembler::zero, L);
1364     __ stop("tried to execute native method as non-native");
1365     __ bind(L);
1366   }
1367   {
1368     Label L;
1369     __ testl(rax, JVM_ACC_ABSTRACT);
1370     __ jcc(Assembler::zero, L);
1371     __ stop("tried to execute abstract method in interpreter");
1372     __ bind(L);
1373   }
1374 #endif
1375 
1376   // Since at this point in the method invocation the exception
1377   // handler would try to exit the monitor of synchronized methods
1378   // which hasn't been entered yet, we set the thread local variable
1379   // _do_not_unlock_if_synchronized to true. The remove_activation
1380   // will check this flag.
1381 
1382   const Address do_not_unlock_if_synchronized(r15_thread,
1383         in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1384   __ movbool(do_not_unlock_if_synchronized, true);
1385 
1386   // increment invocation count & check for overflow
1387   Label invocation_counter_overflow;
1388   Label profile_method;
1389   Label profile_method_continue;
1390   if (inc_counter) {
1391     generate_counter_incr(&invocation_counter_overflow,
1392                           &profile_method,
1393                           &profile_method_continue);
1394     if (ProfileInterpreter) {
1395       __ bind(profile_method_continue);
1396     }
1397   }
1398 
1399   Label continue_after_compile;
1400   __ bind(continue_after_compile);
1401 
1402   // check for synchronized interpreted methods
1403   bang_stack_shadow_pages(false);
1404 
1405   // reset the _do_not_unlock_if_synchronized flag
1406   __ movbool(do_not_unlock_if_synchronized, false);
1407 
1408   // check for synchronized methods
1409   // Must happen AFTER invocation_counter check and stack overflow check,
1410   // so method is not locked if overflows.
1411   if (synchronized) {
1412     // Allocate monitor and lock method
1413     lock_method();
1414   } else {
1415     // no synchronization necessary
1416 #ifdef ASSERT
1417     {
1418       Label L;
1419       __ movl(rax, access_flags);
1420       __ testl(rax, JVM_ACC_SYNCHRONIZED);
1421       __ jcc(Assembler::zero, L);
1422       __ stop("method needs synchronization");
1423       __ bind(L);
1424     }
1425 #endif
1426   }
1427 
1428   // start execution
1429 #ifdef ASSERT
1430   {
1431     Label L;
1432      const Address monitor_block_top (rbp,
1433                  frame::interpreter_frame_monitor_block_top_offset * wordSize);
1434     __ movptr(rax, monitor_block_top);
1435     __ cmpptr(rax, rsp);
1436     __ jcc(Assembler::equal, L);
1437     __ stop("broken stack frame setup in interpreter");
1438     __ bind(L);
1439   }
1440 #endif
1441 
1442   // jvmti support
1443   __ notify_method_entry();
1444 
1445   __ dispatch_next(vtos);
1446 
1447   // invocation counter overflow
1448   if (inc_counter) {
1449     if (ProfileInterpreter) {
1450       // We have decided to profile this method in the interpreter
1451       __ bind(profile_method);
1452       __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1453       __ set_method_data_pointer_for_bcp();
1454       __ get_method(rbx);
1455       __ jmp(profile_method_continue);
1456     }
1457     // Handle overflow of counter and compile method
1458     __ bind(invocation_counter_overflow);
1459     generate_counter_overflow(&continue_after_compile);
1460   }
1461 
1462   return entry_point;
1463 }
1464 
1465 // Entry points
1466 //
1467 // Here we generate the various kind of entries into the interpreter.
1468 // The two main entry type are generic bytecode methods and native
1469 // call method.  These both come in synchronized and non-synchronized
1470 // versions but the frame layout they create is very similar. The
1471 // other method entry types are really just special purpose entries
1472 // that are really entry and interpretation all in one. These are for
1473 // trivial methods like accessor, empty, or special math methods.
1474 //
1475 // When control flow reaches any of the entry types for the interpreter
1476 // the following holds ->
1477 //
1478 // Arguments:
1479 //
1480 // rbx: methodOop
1481 //
1482 // Stack layout immediately at entry
1483 //
1484 // [ return address     ] <--- rsp
1485 // [ parameter n        ]
1486 //   ...
1487 // [ parameter 1        ]
1488 // [ expression stack   ] (caller's java expression stack)
1489 
1490 // Assuming that we don't go to one of the trivial specialized entries
1491 // the stack will look like below when we are ready to execute the
1492 // first bytecode (or call the native routine). The register usage
1493 // will be as the template based interpreter expects (see
1494 // interpreter_amd64.hpp).
1495 //
1496 // local variables follow incoming parameters immediately; i.e.
1497 // the return address is moved to the end of the locals).
1498 //
1499 // [ monitor entry      ] <--- rsp
1500 //   ...
1501 // [ monitor entry      ]
1502 // [ expr. stack bottom ]
1503 // [ saved r13          ]
1504 // [ current r14        ]
1505 // [ methodOop          ]
1506 // [ saved ebp          ] <--- rbp
1507 // [ return address     ]
1508 // [ local variable m   ]
1509 //   ...
1510 // [ local variable 1   ]
1511 // [ parameter n        ]
1512 //   ...
1513 // [ parameter 1        ] <--- r14
1514 
1515 address AbstractInterpreterGenerator::generate_method_entry(
1516                                         AbstractInterpreter::MethodKind kind) {
1517   // determine code generation flags
1518   bool synchronized = false;
1519   address entry_point = NULL;
1520 
1521   switch (kind) {
1522   case Interpreter::zerolocals             :                                                                             break;
1523   case Interpreter::zerolocals_synchronized: synchronized = true;                                                        break;
1524   case Interpreter::native                 : entry_point = ((InterpreterGenerator*) this)->generate_native_entry(false); break;
1525   case Interpreter::native_synchronized    : entry_point = ((InterpreterGenerator*) this)->generate_native_entry(true);  break;
1526   case Interpreter::empty                  : entry_point = ((InterpreterGenerator*) this)->generate_empty_entry();       break;
1527   case Interpreter::accessor               : entry_point = ((InterpreterGenerator*) this)->generate_accessor_entry();    break;
1528   case Interpreter::abstract               : entry_point = ((InterpreterGenerator*) this)->generate_abstract_entry();    break;
1529   case Interpreter::method_handle          : entry_point = ((InterpreterGenerator*) this)->generate_method_handle_entry();break;
1530 
1531   case Interpreter::java_lang_math_sin     : // fall thru
1532   case Interpreter::java_lang_math_cos     : // fall thru
1533   case Interpreter::java_lang_math_tan     : // fall thru
1534   case Interpreter::java_lang_math_abs     : // fall thru
1535   case Interpreter::java_lang_math_log     : // fall thru
1536   case Interpreter::java_lang_math_log10   : // fall thru
1537   case Interpreter::java_lang_math_sqrt    : entry_point = ((InterpreterGenerator*) this)->generate_math_entry(kind);    break;
1538   case Interpreter::java_lang_ref_reference_get
1539                                            : entry_point = ((InterpreterGenerator*)this)->generate_Reference_get_entry(); break;
1540   default                                  : ShouldNotReachHere();                                                       break;
1541   }
1542 
1543   if (entry_point) {
1544     return entry_point;
1545   }
1546 
1547   return ((InterpreterGenerator*) this)->
1548                                 generate_normal_entry(synchronized);
1549 }
1550 
1551 // These should never be compiled since the interpreter will prefer
1552 // the compiled version to the intrinsic version.
1553 bool AbstractInterpreter::can_be_compiled(methodHandle m) {
1554   switch (method_kind(m)) {
1555     case Interpreter::java_lang_math_sin     : // fall thru
1556     case Interpreter::java_lang_math_cos     : // fall thru
1557     case Interpreter::java_lang_math_tan     : // fall thru
1558     case Interpreter::java_lang_math_abs     : // fall thru
1559     case Interpreter::java_lang_math_log     : // fall thru
1560     case Interpreter::java_lang_math_log10   : // fall thru
1561     case Interpreter::java_lang_math_sqrt    :
1562       return false;
1563     default:
1564       return true;
1565   }
1566 }
1567 
1568 // How much stack a method activation needs in words.
1569 int AbstractInterpreter::size_top_interpreter_activation(methodOop method) {
1570   const int entry_size = frame::interpreter_frame_monitor_size();
1571 
1572   // total overhead size: entry_size + (saved rbp thru expr stack
1573   // bottom).  be sure to change this if you add/subtract anything
1574   // to/from the overhead area
1575   const int overhead_size =
1576     -(frame::interpreter_frame_initial_sp_offset) + entry_size;
1577 
1578   const int stub_code = frame::entry_frame_after_call_words;
1579   const int extra_stack = methodOopDesc::extra_stack_entries();
1580   const int method_stack = (method->max_locals() + method->max_stack() + extra_stack) *
1581                            Interpreter::stackElementWords;
1582   return (overhead_size + method_stack + stub_code);
1583 }
1584 
1585 int AbstractInterpreter::layout_activation(methodOop method,
1586                                            int tempcount,
1587                                            int popframe_extra_args,
1588                                            int moncount,
1589                                            int caller_actual_parameters,
1590                                            int callee_param_count,
1591                                            int callee_locals,
1592                                            frame* caller,
1593                                            frame* interpreter_frame,
1594                                            bool is_top_frame) {
1595   // Note: This calculation must exactly parallel the frame setup
1596   // in AbstractInterpreterGenerator::generate_method_entry.
1597   // If interpreter_frame!=NULL, set up the method, locals, and monitors.
1598   // The frame interpreter_frame, if not NULL, is guaranteed to be the
1599   // right size, as determined by a previous call to this method.
1600   // It is also guaranteed to be walkable even though it is in a skeletal state
1601 
1602   // fixed size of an interpreter frame:
1603   int max_locals = method->max_locals() * Interpreter::stackElementWords;
1604   int extra_locals = (method->max_locals() - method->size_of_parameters()) *
1605                      Interpreter::stackElementWords;
1606 
1607   int overhead = frame::sender_sp_offset -
1608                  frame::interpreter_frame_initial_sp_offset;
1609   // Our locals were accounted for by the caller (or last_frame_adjust
1610   // on the transistion) Since the callee parameters already account
1611   // for the callee's params we only need to account for the extra
1612   // locals.
1613   int size = overhead +
1614          (callee_locals - callee_param_count)*Interpreter::stackElementWords +
1615          moncount * frame::interpreter_frame_monitor_size() +
1616          tempcount* Interpreter::stackElementWords + popframe_extra_args;
1617   if (interpreter_frame != NULL) {
1618 #ifdef ASSERT
1619     if (!EnableInvokeDynamic)
1620       // @@@ FIXME: Should we correct interpreter_frame_sender_sp in the calling sequences?
1621       // Probably, since deoptimization doesn't work yet.
1622       assert(caller->unextended_sp() == interpreter_frame->interpreter_frame_sender_sp(), "Frame not properly walkable");
1623     assert(caller->sp() == interpreter_frame->sender_sp(), "Frame not properly walkable(2)");
1624 #endif
1625 
1626     interpreter_frame->interpreter_frame_set_method(method);
1627     // NOTE the difference in using sender_sp and
1628     // interpreter_frame_sender_sp interpreter_frame_sender_sp is
1629     // the original sp of the caller (the unextended_sp) and
1630     // sender_sp is fp+16 XXX
1631     intptr_t* locals = interpreter_frame->sender_sp() + max_locals - 1;
1632 
1633 #ifdef ASSERT
1634     if (caller->is_interpreted_frame()) {
1635       assert(locals < caller->fp() + frame::interpreter_frame_initial_sp_offset, "bad placement");
1636     }
1637 #endif
1638 
1639     interpreter_frame->interpreter_frame_set_locals(locals);
1640     BasicObjectLock* montop = interpreter_frame->interpreter_frame_monitor_begin();
1641     BasicObjectLock* monbot = montop - moncount;
1642     interpreter_frame->interpreter_frame_set_monitor_end(monbot);
1643 
1644     // Set last_sp
1645     intptr_t*  esp = (intptr_t*) monbot -
1646                      tempcount*Interpreter::stackElementWords -
1647                      popframe_extra_args;
1648     interpreter_frame->interpreter_frame_set_last_sp(esp);
1649 
1650     // All frames but the initial (oldest) interpreter frame we fill in have
1651     // a value for sender_sp that allows walking the stack but isn't
1652     // truly correct. Correct the value here.
1653     if (extra_locals != 0 &&
1654         interpreter_frame->sender_sp() ==
1655         interpreter_frame->interpreter_frame_sender_sp()) {
1656       interpreter_frame->set_interpreter_frame_sender_sp(caller->sp() +
1657                                                          extra_locals);
1658     }
1659     *interpreter_frame->interpreter_frame_cache_addr() =
1660       method->constants()->cache();
1661   }
1662   return size;
1663 }
1664 
1665 //-----------------------------------------------------------------------------
1666 // Exceptions
1667 
1668 void TemplateInterpreterGenerator::generate_throw_exception() {
1669   // Entry point in previous activation (i.e., if the caller was
1670   // interpreted)
1671   Interpreter::_rethrow_exception_entry = __ pc();
1672   // Restore sp to interpreter_frame_last_sp even though we are going
1673   // to empty the expression stack for the exception processing.
1674   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
1675   // rax: exception
1676   // rdx: return address/pc that threw exception
1677   __ restore_bcp();    // r13 points to call/send
1678   __ restore_locals();
1679   __ reinit_heapbase();  // restore r12 as heapbase.
1680   // Entry point for exceptions thrown within interpreter code
1681   Interpreter::_throw_exception_entry = __ pc();
1682   // expression stack is undefined here
1683   // rax: exception
1684   // r13: exception bcp
1685   __ verify_oop(rax);
1686   __ mov(c_rarg1, rax);
1687 
1688   // expression stack must be empty before entering the VM in case of
1689   // an exception
1690   __ empty_expression_stack();
1691   // find exception handler address and preserve exception oop
1692   __ call_VM(rdx,
1693              CAST_FROM_FN_PTR(address,
1694                           InterpreterRuntime::exception_handler_for_exception),
1695              c_rarg1);
1696   // rax: exception handler entry point
1697   // rdx: preserved exception oop
1698   // r13: bcp for exception handler
1699   __ push_ptr(rdx); // push exception which is now the only value on the stack
1700   __ jmp(rax); // jump to exception handler (may be _remove_activation_entry!)
1701 
1702   // If the exception is not handled in the current frame the frame is
1703   // removed and the exception is rethrown (i.e. exception
1704   // continuation is _rethrow_exception).
1705   //
1706   // Note: At this point the bci is still the bxi for the instruction
1707   // which caused the exception and the expression stack is
1708   // empty. Thus, for any VM calls at this point, GC will find a legal
1709   // oop map (with empty expression stack).
1710 
1711   // In current activation
1712   // tos: exception
1713   // esi: exception bcp
1714 
1715   //
1716   // JVMTI PopFrame support
1717   //
1718 
1719   Interpreter::_remove_activation_preserving_args_entry = __ pc();
1720   __ empty_expression_stack();
1721   // Set the popframe_processing bit in pending_popframe_condition
1722   // indicating that we are currently handling popframe, so that
1723   // call_VMs that may happen later do not trigger new popframe
1724   // handling cycles.
1725   __ movl(rdx, Address(r15_thread, JavaThread::popframe_condition_offset()));
1726   __ orl(rdx, JavaThread::popframe_processing_bit);
1727   __ movl(Address(r15_thread, JavaThread::popframe_condition_offset()), rdx);
1728 
1729   {
1730     // Check to see whether we are returning to a deoptimized frame.
1731     // (The PopFrame call ensures that the caller of the popped frame is
1732     // either interpreted or compiled and deoptimizes it if compiled.)
1733     // In this case, we can't call dispatch_next() after the frame is
1734     // popped, but instead must save the incoming arguments and restore
1735     // them after deoptimization has occurred.
1736     //
1737     // Note that we don't compare the return PC against the
1738     // deoptimization blob's unpack entry because of the presence of
1739     // adapter frames in C2.
1740     Label caller_not_deoptimized;
1741     __ movptr(c_rarg1, Address(rbp, frame::return_addr_offset * wordSize));
1742     __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1743                                InterpreterRuntime::interpreter_contains), c_rarg1);
1744     __ testl(rax, rax);
1745     __ jcc(Assembler::notZero, caller_not_deoptimized);
1746 
1747     // Compute size of arguments for saving when returning to
1748     // deoptimized caller
1749     __ get_method(rax);
1750     __ load_unsigned_short(rax, Address(rax, in_bytes(methodOopDesc::
1751                                                 size_of_parameters_offset())));
1752     __ shll(rax, Interpreter::logStackElementSize);
1753     __ restore_locals(); // XXX do we need this?
1754     __ subptr(r14, rax);
1755     __ addptr(r14, wordSize);
1756     // Save these arguments
1757     __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1758                                            Deoptimization::
1759                                            popframe_preserve_args),
1760                           r15_thread, rax, r14);
1761 
1762     __ remove_activation(vtos, rdx,
1763                          /* throw_monitor_exception */ false,
1764                          /* install_monitor_exception */ false,
1765                          /* notify_jvmdi */ false);
1766 
1767     // Inform deoptimization that it is responsible for restoring
1768     // these arguments
1769     __ movl(Address(r15_thread, JavaThread::popframe_condition_offset()),
1770             JavaThread::popframe_force_deopt_reexecution_bit);
1771 
1772     // Continue in deoptimization handler
1773     __ jmp(rdx);
1774 
1775     __ bind(caller_not_deoptimized);
1776   }
1777 
1778   __ remove_activation(vtos, rdx, /* rdx result (retaddr) is not used */
1779                        /* throw_monitor_exception */ false,
1780                        /* install_monitor_exception */ false,
1781                        /* notify_jvmdi */ false);
1782 
1783   // Finish with popframe handling
1784   // A previous I2C followed by a deoptimization might have moved the
1785   // outgoing arguments further up the stack. PopFrame expects the
1786   // mutations to those outgoing arguments to be preserved and other
1787   // constraints basically require this frame to look exactly as
1788   // though it had previously invoked an interpreted activation with
1789   // no space between the top of the expression stack (current
1790   // last_sp) and the top of stack. Rather than force deopt to
1791   // maintain this kind of invariant all the time we call a small
1792   // fixup routine to move the mutated arguments onto the top of our
1793   // expression stack if necessary.
1794   __ mov(c_rarg1, rsp);
1795   __ movptr(c_rarg2, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1796   // PC must point into interpreter here
1797   __ set_last_Java_frame(noreg, rbp, __ pc());
1798   __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), r15_thread, c_rarg1, c_rarg2);
1799   __ reset_last_Java_frame(true, true);
1800   // Restore the last_sp and null it out
1801   __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1802   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
1803 
1804   __ restore_bcp();  // XXX do we need this?
1805   __ restore_locals(); // XXX do we need this?
1806   // The method data pointer was incremented already during
1807   // call profiling. We have to restore the mdp for the current bcp.
1808   if (ProfileInterpreter) {
1809     __ set_method_data_pointer_for_bcp();
1810   }
1811 
1812   // Clear the popframe condition flag
1813   __ movl(Address(r15_thread, JavaThread::popframe_condition_offset()),
1814           JavaThread::popframe_inactive);
1815 
1816   __ dispatch_next(vtos);
1817   // end of PopFrame support
1818 
1819   Interpreter::_remove_activation_entry = __ pc();
1820 
1821   // preserve exception over this code sequence
1822   __ pop_ptr(rax);
1823   __ movptr(Address(r15_thread, JavaThread::vm_result_offset()), rax);
1824   // remove the activation (without doing throws on illegalMonitorExceptions)
1825   __ remove_activation(vtos, rdx, false, true, false);
1826   // restore exception
1827   __ movptr(rax, Address(r15_thread, JavaThread::vm_result_offset()));
1828   __ movptr(Address(r15_thread, JavaThread::vm_result_offset()), (int32_t)NULL_WORD);
1829   __ verify_oop(rax);
1830 
1831   // In between activations - previous activation type unknown yet
1832   // compute continuation point - the continuation point expects the
1833   // following registers set up:
1834   //
1835   // rax: exception
1836   // rdx: return address/pc that threw exception
1837   // rsp: expression stack of caller
1838   // rbp: ebp of caller
1839   __ push(rax);                                  // save exception
1840   __ push(rdx);                                  // save return address
1841   __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1842                           SharedRuntime::exception_handler_for_return_address),
1843                         r15_thread, rdx);
1844   __ mov(rbx, rax);                              // save exception handler
1845   __ pop(rdx);                                   // restore return address
1846   __ pop(rax);                                   // restore exception
1847   // Note that an "issuing PC" is actually the next PC after the call
1848   __ jmp(rbx);                                   // jump to exception
1849                                                  // handler of caller
1850 }
1851 
1852 
1853 //
1854 // JVMTI ForceEarlyReturn support
1855 //
1856 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
1857   address entry = __ pc();
1858 
1859   __ restore_bcp();
1860   __ restore_locals();
1861   __ empty_expression_stack();
1862   __ load_earlyret_value(state);
1863 
1864   __ movptr(rdx, Address(r15_thread, JavaThread::jvmti_thread_state_offset()));
1865   Address cond_addr(rdx, JvmtiThreadState::earlyret_state_offset());
1866 
1867   // Clear the earlyret state
1868   __ movl(cond_addr, JvmtiThreadState::earlyret_inactive);
1869 
1870   __ remove_activation(state, rsi,
1871                        false, /* throw_monitor_exception */
1872                        false, /* install_monitor_exception */
1873                        true); /* notify_jvmdi */
1874   __ jmp(rsi);
1875 
1876   return entry;
1877 } // end of ForceEarlyReturn support
1878 
1879 
1880 //-----------------------------------------------------------------------------
1881 // Helper for vtos entry point generation
1882 
1883 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
1884                                                          address& bep,
1885                                                          address& cep,
1886                                                          address& sep,
1887                                                          address& aep,
1888                                                          address& iep,
1889                                                          address& lep,
1890                                                          address& fep,
1891                                                          address& dep,
1892                                                          address& vep) {
1893   assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
1894   Label L;
1895   aep = __ pc();  __ push_ptr();  __ jmp(L);
1896   fep = __ pc();  __ push_f();    __ jmp(L);
1897   dep = __ pc();  __ push_d();    __ jmp(L);
1898   lep = __ pc();  __ push_l();    __ jmp(L);
1899   bep = cep = sep =
1900   iep = __ pc();  __ push_i();
1901   vep = __ pc();
1902   __ bind(L);
1903   generate_and_dispatch(t);
1904 }
1905 
1906 
1907 //-----------------------------------------------------------------------------
1908 // Generation of individual instructions
1909 
1910 // helpers for generate_and_dispatch
1911 
1912 
1913 InterpreterGenerator::InterpreterGenerator(StubQueue* code)
1914   : TemplateInterpreterGenerator(code) {
1915    generate_all(); // down here so it can be "virtual"
1916 }
1917 
1918 //-----------------------------------------------------------------------------
1919 
1920 // Non-product code
1921 #ifndef PRODUCT
1922 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
1923   address entry = __ pc();
1924 
1925   __ push(state);
1926   __ push(c_rarg0);
1927   __ push(c_rarg1);
1928   __ push(c_rarg2);
1929   __ push(c_rarg3);
1930   __ mov(c_rarg2, rax);  // Pass itos
1931 #ifdef _WIN64
1932   __ movflt(xmm3, xmm0); // Pass ftos
1933 #endif
1934   __ call_VM(noreg,
1935              CAST_FROM_FN_PTR(address, SharedRuntime::trace_bytecode),
1936              c_rarg1, c_rarg2, c_rarg3);
1937   __ pop(c_rarg3);
1938   __ pop(c_rarg2);
1939   __ pop(c_rarg1);
1940   __ pop(c_rarg0);
1941   __ pop(state);
1942   __ ret(0);                                   // return from result handler
1943 
1944   return entry;
1945 }
1946 
1947 void TemplateInterpreterGenerator::count_bytecode() {
1948   __ incrementl(ExternalAddress((address) &BytecodeCounter::_counter_value));
1949 }
1950 
1951 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
1952   __ incrementl(ExternalAddress((address) &BytecodeHistogram::_counters[t->bytecode()]));
1953 }
1954 
1955 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
1956   __ mov32(rbx, ExternalAddress((address) &BytecodePairHistogram::_index));
1957   __ shrl(rbx, BytecodePairHistogram::log2_number_of_codes);
1958   __ orl(rbx,
1959          ((int) t->bytecode()) <<
1960          BytecodePairHistogram::log2_number_of_codes);
1961   __ mov32(ExternalAddress((address) &BytecodePairHistogram::_index), rbx);
1962   __ lea(rscratch1, ExternalAddress((address) BytecodePairHistogram::_counters));
1963   __ incrementl(Address(rscratch1, rbx, Address::times_4));
1964 }
1965 
1966 
1967 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
1968   // Call a little run-time stub to avoid blow-up for each bytecode.
1969   // The run-time runtime saves the right registers, depending on
1970   // the tosca in-state for the given template.
1971 
1972   assert(Interpreter::trace_code(t->tos_in()) != NULL,
1973          "entry must have been generated");
1974   __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1975   __ andptr(rsp, -16); // align stack as required by ABI
1976   __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in())));
1977   __ mov(rsp, r12); // restore sp
1978   __ reinit_heapbase();
1979 }
1980 
1981 
1982 void TemplateInterpreterGenerator::stop_interpreter_at() {
1983   Label L;
1984   __ cmp32(ExternalAddress((address) &BytecodeCounter::_counter_value),
1985            StopInterpreterAt);
1986   __ jcc(Assembler::notEqual, L);
1987   __ int3();
1988   __ bind(L);
1989 }
1990 #endif // !PRODUCT
1991 #endif // ! CC_INTERP