1 /*
   2  * Copyright (c) 2003, 2012, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 #include "precompiled.hpp"
  26 #include "asm/macroAssembler.hpp"
  27 #include "interpreter/bytecodeHistogram.hpp"
  28 #include "interpreter/interpreter.hpp"
  29 #include "interpreter/interpreterGenerator.hpp"
  30 #include "interpreter/interpreterRuntime.hpp"
  31 #include "interpreter/templateTable.hpp"
  32 #include "oops/arrayOop.hpp"
  33 #include "oops/methodData.hpp"
  34 #include "oops/method.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 #include "utilities/macros.hpp"
  48 
  49 #define __ _masm->
  50 
  51 #ifndef CC_INTERP
  52 
  53 const int method_offset = frame::interpreter_frame_method_offset * wordSize;
  54 const int bci_offset    = frame::interpreter_frame_bcx_offset    * wordSize;
  55 const int locals_offset = frame::interpreter_frame_locals_offset * wordSize;
  56 
  57 //-----------------------------------------------------------------------------
  58 
  59 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() {
  60   address entry = __ pc();
  61 
  62 #ifdef ASSERT
  63   {
  64     Label L;
  65     __ lea(rax, Address(rbp,
  66                         frame::interpreter_frame_monitor_block_top_offset *
  67                         wordSize));
  68     __ cmpptr(rax, rsp); // rax = maximal rsp for current rbp (stack
  69                          // grows negative)
  70     __ jcc(Assembler::aboveEqual, L); // check if frame is complete
  71     __ stop ("interpreter frame not set up");
  72     __ bind(L);
  73   }
  74 #endif // ASSERT
  75   // Restore bcp under the assumption that the current frame is still
  76   // interpreted
  77   __ restore_bcp();
  78 
  79   // expression stack must be empty before entering the VM if an
  80   // exception happened
  81   __ empty_expression_stack();
  82   // throw exception
  83   __ call_VM(noreg,
  84              CAST_FROM_FN_PTR(address,
  85                               InterpreterRuntime::throw_StackOverflowError));
  86   return entry;
  87 }
  88 
  89 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler(
  90         const char* name) {
  91   address entry = __ pc();
  92   // expression stack must be empty before entering the VM if an
  93   // exception happened
  94   __ empty_expression_stack();
  95   // setup parameters
  96   // ??? convention: expect aberrant index in register ebx
  97   __ lea(c_rarg1, ExternalAddress((address)name));
  98   __ call_VM(noreg,
  99              CAST_FROM_FN_PTR(address,
 100                               InterpreterRuntime::
 101                               throw_ArrayIndexOutOfBoundsException),
 102              c_rarg1, rbx);
 103   return entry;
 104 }
 105 
 106 address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
 107   address entry = __ pc();
 108 
 109   // object is at TOS
 110   __ pop(c_rarg1);
 111 
 112   // expression stack must be empty before entering the VM if an
 113   // exception happened
 114   __ empty_expression_stack();
 115 
 116   __ call_VM(noreg,
 117              CAST_FROM_FN_PTR(address,
 118                               InterpreterRuntime::
 119                               throw_ClassCastException),
 120              c_rarg1);
 121   return entry;
 122 }
 123 
 124 address TemplateInterpreterGenerator::generate_exception_handler_common(
 125         const char* name, const char* message, bool pass_oop) {
 126   assert(!pass_oop || message == NULL, "either oop or message but not both");
 127   address entry = __ pc();
 128   if (pass_oop) {
 129     // object is at TOS
 130     __ pop(c_rarg2);
 131   }
 132   // expression stack must be empty before entering the VM if an
 133   // exception happened
 134   __ empty_expression_stack();
 135   // setup parameters
 136   __ lea(c_rarg1, ExternalAddress((address)name));
 137   if (pass_oop) {
 138     __ call_VM(rax, CAST_FROM_FN_PTR(address,
 139                                      InterpreterRuntime::
 140                                      create_klass_exception),
 141                c_rarg1, c_rarg2);
 142   } else {
 143     // kind of lame ExternalAddress can't take NULL because
 144     // external_word_Relocation will assert.
 145     if (message != NULL) {
 146       __ lea(c_rarg2, ExternalAddress((address)message));
 147     } else {
 148       __ movptr(c_rarg2, NULL_WORD);
 149     }
 150     __ call_VM(rax,
 151                CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception),
 152                c_rarg1, c_rarg2);
 153   }
 154   // throw exception
 155   __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
 156   return entry;
 157 }
 158 
 159 
 160 address TemplateInterpreterGenerator::generate_continuation_for(TosState state) {
 161   address entry = __ pc();
 162   // NULL last_sp until next java call
 163   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
 164   __ dispatch_next(state);
 165   return entry;
 166 }
 167 
 168 
 169 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step) {
 170   address entry = __ pc();
 171 
 172   // Restore stack bottom in case i2c adjusted stack
 173   __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
 174   // and NULL it as marker that esp is now tos until next java call
 175   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
 176 
 177   __ restore_bcp();
 178   __ restore_locals();
 179 
 180   Label L_got_cache, L_giant_index;
 181   if (EnableInvokeDynamic) {
 182     __ cmpb(Address(r13, 0), Bytecodes::_invokedynamic);
 183     __ jcc(Assembler::equal, L_giant_index);
 184   }
 185   __ get_cache_and_index_at_bcp(rbx, rcx, 1, sizeof(u2));
 186   __ bind(L_got_cache);
 187   __ movl(rbx, Address(rbx, rcx,
 188                        Address::times_ptr,
 189                        in_bytes(ConstantPoolCache::base_offset()) +
 190                        3 * wordSize));
 191   __ andl(rbx, 0xFF);
 192   __ lea(rsp, Address(rsp, rbx, Address::times_8));
 193   __ dispatch_next(state, step);
 194 
 195   // out of the main line of code...
 196   if (EnableInvokeDynamic) {
 197     __ bind(L_giant_index);
 198     __ get_cache_and_index_at_bcp(rbx, rcx, 1, sizeof(u4));
 199     __ jmp(L_got_cache);
 200   }
 201 
 202   return entry;
 203 }
 204 
 205 
 206 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state,
 207                                                                int step) {
 208   address entry = __ pc();
 209   // NULL last_sp until next java call
 210   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
 211   __ restore_bcp();
 212   __ restore_locals();
 213   // handle exceptions
 214   {
 215     Label L;
 216     __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t) NULL_WORD);
 217     __ jcc(Assembler::zero, L);
 218     __ call_VM(noreg,
 219                CAST_FROM_FN_PTR(address,
 220                                 InterpreterRuntime::throw_pending_exception));
 221     __ should_not_reach_here();
 222     __ bind(L);
 223   }
 224   __ dispatch_next(state, step);
 225   return entry;
 226 }
 227 
 228 int AbstractInterpreter::BasicType_as_index(BasicType type) {
 229   int i = 0;
 230   switch (type) {
 231     case T_BOOLEAN: i = 0; break;
 232     case T_CHAR   : i = 1; break;
 233     case T_BYTE   : i = 2; break;
 234     case T_SHORT  : i = 3; break;
 235     case T_INT    : i = 4; break;
 236     case T_LONG   : i = 5; break;
 237     case T_VOID   : i = 6; break;
 238     case T_FLOAT  : i = 7; break;
 239     case T_DOUBLE : i = 8; break;
 240     case T_OBJECT : i = 9; break;
 241     case T_ARRAY  : i = 9; break;
 242     default       : ShouldNotReachHere();
 243   }
 244   assert(0 <= i && i < AbstractInterpreter::number_of_result_handlers,
 245          "index out of bounds");
 246   return i;
 247 }
 248 
 249 
 250 address TemplateInterpreterGenerator::generate_result_handler_for(
 251         BasicType type) {
 252   address entry = __ pc();
 253   switch (type) {
 254   case T_BOOLEAN: __ c2bool(rax);            break;
 255   case T_CHAR   : __ movzwl(rax, rax);       break;
 256   case T_BYTE   : __ sign_extend_byte(rax);  break;
 257   case T_SHORT  : __ sign_extend_short(rax); break;
 258   case T_INT    : /* nothing to do */        break;
 259   case T_LONG   : /* nothing to do */        break;
 260   case T_VOID   : /* nothing to do */        break;
 261   case T_FLOAT  : /* nothing to do */        break;
 262   case T_DOUBLE : /* nothing to do */        break;
 263   case T_OBJECT :
 264     // retrieve result from frame
 265     __ movptr(rax, Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize));
 266     // and verify it
 267     __ verify_oop(rax);
 268     break;
 269   default       : ShouldNotReachHere();
 270   }
 271   __ ret(0);                                   // return from result handler
 272   return entry;
 273 }
 274 
 275 address TemplateInterpreterGenerator::generate_safept_entry_for(
 276         TosState state,
 277         address runtime_entry) {
 278   address entry = __ pc();
 279   __ push(state);
 280   __ call_VM(noreg, runtime_entry);
 281   __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
 282   return entry;
 283 }
 284 
 285 
 286 
 287 // Helpers for commoning out cases in the various type of method entries.
 288 //
 289 
 290 
 291 // increment invocation count & check for overflow
 292 //
 293 // Note: checking for negative value instead of overflow
 294 //       so we have a 'sticky' overflow test
 295 //
 296 // rbx: method
 297 // ecx: invocation counter
 298 //
 299 void InterpreterGenerator::generate_counter_incr(
 300         Label* overflow,
 301         Label* profile_method,
 302         Label* profile_method_continue) {
 303   const Address invocation_counter(rbx, in_bytes(Method::invocation_counter_offset()) +
 304                                         in_bytes(InvocationCounter::counter_offset()));
 305   // Note: In tiered we increment either counters in Method* or in MDO depending if we're profiling or not.
 306   if (TieredCompilation) {
 307     int increment = InvocationCounter::count_increment;
 308     int mask = ((1 << Tier0InvokeNotifyFreqLog)  - 1) << InvocationCounter::count_shift;
 309     Label no_mdo, done;
 310     if (ProfileInterpreter) {
 311       // Are we profiling?
 312       __ movptr(rax, Address(rbx, Method::method_data_offset()));
 313       __ testptr(rax, rax);
 314       __ jccb(Assembler::zero, no_mdo);
 315       // Increment counter in the MDO
 316       const Address mdo_invocation_counter(rax, in_bytes(MethodData::invocation_counter_offset()) +
 317                                                 in_bytes(InvocationCounter::counter_offset()));
 318       __ increment_mask_and_jump(mdo_invocation_counter, increment, mask, rcx, false, Assembler::zero, overflow);
 319       __ jmpb(done);
 320     }
 321     __ bind(no_mdo);
 322     // Increment counter in Method* (we don't need to load it, it's in ecx).
 323     __ increment_mask_and_jump(invocation_counter, increment, mask, rcx, true, Assembler::zero, overflow);
 324     __ bind(done);
 325   } else {
 326     const Address backedge_counter(rbx,
 327                                    Method::backedge_counter_offset() +
 328                                    InvocationCounter::counter_offset());
 329 
 330     if (ProfileInterpreter) { // %%% Merge this into MethodData*
 331       __ incrementl(Address(rbx,
 332                             Method::interpreter_invocation_counter_offset()));
 333     }
 334     // Update standard invocation counters
 335     __ movl(rax, backedge_counter);   // load backedge counter
 336 
 337     __ incrementl(rcx, InvocationCounter::count_increment);
 338     __ andl(rax, InvocationCounter::count_mask_value); // mask out the status bits
 339 
 340     __ movl(invocation_counter, rcx); // save invocation count
 341     __ addl(rcx, rax);                // add both counters
 342 
 343     // profile_method is non-null only for interpreted method so
 344     // profile_method != NULL == !native_call
 345 
 346     if (ProfileInterpreter && profile_method != NULL) {
 347       // Test to see if we should create a method data oop
 348       __ cmp32(rcx, ExternalAddress((address)&InvocationCounter::InterpreterProfileLimit));
 349       __ jcc(Assembler::less, *profile_method_continue);
 350 
 351       // if no method data exists, go to profile_method
 352       __ test_method_data_pointer(rax, *profile_method);
 353     }
 354 
 355     __ cmp32(rcx, ExternalAddress((address)&InvocationCounter::InterpreterInvocationLimit));
 356     __ jcc(Assembler::aboveEqual, *overflow);
 357   }
 358 }
 359 
 360 void InterpreterGenerator::generate_counter_overflow(Label* do_continue) {
 361 
 362   // Asm interpreter on entry
 363   // r14 - locals
 364   // r13 - bcp
 365   // rbx - method
 366   // edx - cpool --- DOES NOT APPEAR TO BE TRUE
 367   // rbp - interpreter frame
 368 
 369   // On return (i.e. jump to entry_point) [ back to invocation of interpreter ]
 370   // Everything as it was on entry
 371   // rdx is not restored. Doesn't appear to really be set.
 372 
 373   // InterpreterRuntime::frequency_counter_overflow takes two
 374   // arguments, the first (thread) is passed by call_VM, the second
 375   // indicates if the counter overflow occurs at a backwards branch
 376   // (NULL bcp).  We pass zero for it.  The call returns the address
 377   // of the verified entry point for the method or NULL if the
 378   // compilation did not complete (either went background or bailed
 379   // out).
 380   __ movl(c_rarg1, 0);
 381   __ call_VM(noreg,
 382              CAST_FROM_FN_PTR(address,
 383                               InterpreterRuntime::frequency_counter_overflow),
 384              c_rarg1);
 385 
 386   __ movptr(rbx, Address(rbp, method_offset));   // restore Method*
 387   // Preserve invariant that r13/r14 contain bcp/locals of sender frame
 388   // and jump to the interpreted entry.
 389   __ jmp(*do_continue, relocInfo::none);
 390 }
 391 
 392 // See if we've got enough room on the stack for locals plus overhead.
 393 // The expression stack grows down incrementally, so the normal guard
 394 // page mechanism will work for that.
 395 //
 396 // NOTE: Since the additional locals are also always pushed (wasn't
 397 // obvious in generate_method_entry) so the guard should work for them
 398 // too.
 399 //
 400 // Args:
 401 //      rdx: number of additional locals this frame needs (what we must check)
 402 //      rbx: Method*
 403 //
 404 // Kills:
 405 //      rax
 406 void InterpreterGenerator::generate_stack_overflow_check(void) {
 407 
 408   // monitor entry size: see picture of stack set
 409   // (generate_method_entry) and frame_amd64.hpp
 410   const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
 411 
 412   // total overhead size: entry_size + (saved rbp through expr stack
 413   // bottom).  be sure to change this if you add/subtract anything
 414   // to/from the overhead area
 415   const int overhead_size =
 416     -(frame::interpreter_frame_initial_sp_offset * wordSize) + entry_size;
 417 
 418   const int page_size = os::vm_page_size();
 419 
 420   Label after_frame_check;
 421 
 422   // see if the frame is greater than one page in size. If so,
 423   // then we need to verify there is enough stack space remaining
 424   // for the additional locals.
 425   __ cmpl(rdx, (page_size - overhead_size) / Interpreter::stackElementSize);
 426   __ jcc(Assembler::belowEqual, after_frame_check);
 427 
 428   // compute rsp as if this were going to be the last frame on
 429   // the stack before the red zone
 430 
 431   const Address stack_base(r15_thread, Thread::stack_base_offset());
 432   const Address stack_size(r15_thread, Thread::stack_size_offset());
 433 
 434   // locals + overhead, in bytes
 435   __ mov(rax, rdx);
 436   __ shlptr(rax, Interpreter::logStackElementSize);  // 2 slots per parameter.
 437   __ addptr(rax, overhead_size);
 438 
 439 #ifdef ASSERT
 440   Label stack_base_okay, stack_size_okay;
 441   // verify that thread stack base is non-zero
 442   __ cmpptr(stack_base, (int32_t)NULL_WORD);
 443   __ jcc(Assembler::notEqual, stack_base_okay);
 444   __ stop("stack base is zero");
 445   __ bind(stack_base_okay);
 446   // verify that thread stack size is non-zero
 447   __ cmpptr(stack_size, 0);
 448   __ jcc(Assembler::notEqual, stack_size_okay);
 449   __ stop("stack size is zero");
 450   __ bind(stack_size_okay);
 451 #endif
 452 
 453   // Add stack base to locals and subtract stack size
 454   __ addptr(rax, stack_base);
 455   __ subptr(rax, stack_size);
 456 
 457   // Use the maximum number of pages we might bang.
 458   const int max_pages = StackShadowPages > (StackRedPages+StackYellowPages) ? StackShadowPages :
 459                                                                               (StackRedPages+StackYellowPages);
 460 
 461   // add in the red and yellow zone sizes
 462   __ addptr(rax, max_pages * page_size);
 463 
 464   // check against the current stack bottom
 465   __ cmpptr(rsp, rax);
 466   __ jcc(Assembler::above, after_frame_check);
 467 
 468   // Restore sender's sp as SP. This is necessary if the sender's
 469   // frame is an extended compiled frame (see gen_c2i_adapter())
 470   // and safer anyway in case of JSR292 adaptations.
 471 
 472   __ pop(rax); // return address must be moved if SP is changed
 473   __ mov(rsp, r13);
 474   __ push(rax);
 475 
 476   // Note: the restored frame is not necessarily interpreted.
 477   // Use the shared runtime version of the StackOverflowError.
 478   assert(StubRoutines::throw_StackOverflowError_entry() != NULL, "stub not yet generated");
 479   __ jump(ExternalAddress(StubRoutines::throw_StackOverflowError_entry()));
 480 
 481   // all done with frame size check
 482   __ bind(after_frame_check);
 483 }
 484 
 485 // Allocate monitor and lock method (asm interpreter)
 486 //
 487 // Args:
 488 //      rbx: Method*
 489 //      r14: locals
 490 //
 491 // Kills:
 492 //      rax
 493 //      c_rarg0, c_rarg1, c_rarg2, c_rarg3, ...(param regs)
 494 //      rscratch1, rscratch2 (scratch regs)
 495 void InterpreterGenerator::lock_method(void) {
 496   // synchronize method
 497   const Address access_flags(rbx, Method::access_flags_offset());
 498   const Address monitor_block_top(
 499         rbp,
 500         frame::interpreter_frame_monitor_block_top_offset * wordSize);
 501   const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
 502 
 503 #ifdef ASSERT
 504   {
 505     Label L;
 506     __ movl(rax, access_flags);
 507     __ testl(rax, JVM_ACC_SYNCHRONIZED);
 508     __ jcc(Assembler::notZero, L);
 509     __ stop("method doesn't need synchronization");
 510     __ bind(L);
 511   }
 512 #endif // ASSERT
 513 
 514   // get synchronization object
 515   {
 516     const int mirror_offset = in_bytes(Klass::java_mirror_offset());
 517     Label done;
 518     __ movl(rax, access_flags);
 519     __ testl(rax, JVM_ACC_STATIC);
 520     // get receiver (assume this is frequent case)
 521     __ movptr(rax, Address(r14, Interpreter::local_offset_in_bytes(0)));
 522     __ jcc(Assembler::zero, done);
 523     __ movptr(rax, Address(rbx, Method::const_offset()));
 524     __ movptr(rax, Address(rax, ConstMethod::constants_offset()));
 525     __ movptr(rax, Address(rax,
 526                            ConstantPool::pool_holder_offset_in_bytes()));
 527     __ movptr(rax, Address(rax, mirror_offset));
 528 
 529 #ifdef ASSERT
 530     {
 531       Label L;
 532       __ testptr(rax, rax);
 533       __ jcc(Assembler::notZero, L);
 534       __ stop("synchronization object is NULL");
 535       __ bind(L);
 536     }
 537 #endif // ASSERT
 538 
 539     __ bind(done);
 540   }
 541 
 542   // add space for monitor & lock
 543   __ subptr(rsp, entry_size); // add space for a monitor entry
 544   __ movptr(monitor_block_top, rsp);  // set new monitor block top
 545   // store object
 546   __ movptr(Address(rsp, BasicObjectLock::obj_offset_in_bytes()), rax);
 547   __ movptr(c_rarg1, rsp); // object address
 548   __ lock_object(c_rarg1);
 549 }
 550 
 551 // Generate a fixed interpreter frame. This is identical setup for
 552 // interpreted methods and for native methods hence the shared code.
 553 //
 554 // Args:
 555 //      rax: return address
 556 //      rbx: Method*
 557 //      r14: pointer to locals
 558 //      r13: sender sp
 559 //      rdx: cp cache
 560 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
 561   // initialize fixed part of activation frame
 562   __ push(rax);        // save return address
 563   __ enter();          // save old & set new rbp
 564   __ push(r13);        // set sender sp
 565   __ push((int)NULL_WORD); // leave last_sp as null
 566   __ movptr(r13, Address(rbx, Method::const_offset()));      // get ConstMethod*
 567   __ lea(r13, Address(r13, ConstMethod::codes_offset())); // get codebase
 568   __ push(rbx);        // save Method*
 569   if (ProfileInterpreter) {
 570     Label method_data_continue;
 571     __ movptr(rdx, Address(rbx, in_bytes(Method::method_data_offset())));
 572     __ testptr(rdx, rdx);
 573     __ jcc(Assembler::zero, method_data_continue);
 574     __ addptr(rdx, in_bytes(MethodData::data_offset()));
 575     __ bind(method_data_continue);
 576     __ push(rdx);      // set the mdp (method data pointer)
 577   } else {
 578     __ push(0);
 579   }
 580 
 581   __ movptr(rdx, Address(rbx, Method::const_offset()));
 582   __ movptr(rdx, Address(rdx, ConstMethod::constants_offset()));
 583   __ movptr(rdx, Address(rdx, ConstantPool::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: Method*
 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     // read first instruction word and extract bytecode @ 1 and index @ 2
 633     __ movptr(rdx, Address(rbx, Method::const_offset()));
 634     __ movptr(rdi, Address(rdx, ConstMethod::constants_offset()));
 635     __ movl(rdx, Address(rdx, ConstMethod::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, ConstantPool::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                     ConstantPoolCache::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                       ConstantPoolCache::base_offset() +
 669                       ConstantPoolCacheEntry::f2_offset()));
 670     // edx: flags
 671     __ movl(rdx,
 672             Address(rdi,
 673                     rdx,
 674                     Address::times_8,
 675                     ConstantPoolCache::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::tos_state_shift);
 685     // Make sure we don't need to mask edx after the above shift
 686     ConstantPoolCacheEntry::verify_tos_state_shift();
 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 #if INCLUDE_ALL_GCS
 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: Method*
 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 // INCLUDE_ALL_GCS
 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: Method*
 841   // r13: sender sp
 842 
 843   address entry_point = __ pc();
 844 
 845   const Address constMethod       (rbx, Method::const_offset());
 846   const Address invocation_counter(rbx, Method::
 847                                         invocation_counter_offset() +
 848                                         InvocationCounter::counter_offset());
 849   const Address access_flags      (rbx, Method::access_flags_offset());
 850   const Address size_of_parameters(rcx, ConstMethod::
 851                                         size_of_parameters_offset());
 852 
 853 
 854   // get parameter size (always needed)
 855   __ movptr(rcx, constMethod);
 856   __ load_unsigned_short(rcx, size_of_parameters);
 857 
 858   // native calls don't need the stack size check since they have no
 859   // expression stack and the arguments are already on the stack and
 860   // we only add a handful of words to the stack
 861 
 862   // rbx: Method*
 863   // rcx: size of parameters
 864   // r13: sender sp
 865   __ pop(rax);                                       // get return address
 866 
 867   // for natives the size of locals is zero
 868 
 869   // compute beginning of parameters (r14)
 870   __ lea(r14, Address(rsp, rcx, Address::times_8, -wordSize));
 871 
 872   // add 2 zero-initialized slots for native calls
 873   // initialize result_handler slot
 874   __ push((int) NULL_WORD);
 875   // slot for oop temp
 876   // (static native method holder mirror/jni oop result)
 877   __ push((int) NULL_WORD);
 878 
 879   if (inc_counter) {
 880     __ movl(rcx, invocation_counter);  // (pre-)fetch invocation count
 881   }
 882 
 883   // initialize fixed part of activation frame
 884   generate_fixed_frame(true);
 885 
 886   // make sure method is native & not abstract
 887 #ifdef ASSERT
 888   __ movl(rax, access_flags);
 889   {
 890     Label L;
 891     __ testl(rax, JVM_ACC_NATIVE);
 892     __ jcc(Assembler::notZero, L);
 893     __ stop("tried to execute non-native method as native");
 894     __ bind(L);
 895   }
 896   {
 897     Label L;
 898     __ testl(rax, JVM_ACC_ABSTRACT);
 899     __ jcc(Assembler::zero, L);
 900     __ stop("tried to execute abstract method in interpreter");
 901     __ bind(L);
 902   }
 903 #endif
 904 
 905   // Since at this point in the method invocation the exception handler
 906   // would try to exit the monitor of synchronized methods which hasn't
 907   // been entered yet, we set the thread local variable
 908   // _do_not_unlock_if_synchronized to true. The remove_activation will
 909   // check this flag.
 910 
 911   const Address do_not_unlock_if_synchronized(r15_thread,
 912         in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
 913   __ movbool(do_not_unlock_if_synchronized, true);
 914 
 915   // increment invocation count & check for overflow
 916   Label invocation_counter_overflow;
 917   if (inc_counter) {
 918     generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
 919   }
 920 
 921   Label continue_after_compile;
 922   __ bind(continue_after_compile);
 923 
 924   bang_stack_shadow_pages(true);
 925 
 926   // reset the _do_not_unlock_if_synchronized flag
 927   __ movbool(do_not_unlock_if_synchronized, false);
 928 
 929   // check for synchronized methods
 930   // Must happen AFTER invocation_counter check and stack overflow check,
 931   // so method is not locked if overflows.
 932   if (synchronized) {
 933     lock_method();
 934   } else {
 935     // no synchronization necessary
 936 #ifdef ASSERT
 937     {
 938       Label L;
 939       __ movl(rax, access_flags);
 940       __ testl(rax, JVM_ACC_SYNCHRONIZED);
 941       __ jcc(Assembler::zero, L);
 942       __ stop("method needs synchronization");
 943       __ bind(L);
 944     }
 945 #endif
 946   }
 947 
 948   // start execution
 949 #ifdef ASSERT
 950   {
 951     Label L;
 952     const Address monitor_block_top(rbp,
 953                  frame::interpreter_frame_monitor_block_top_offset * wordSize);
 954     __ movptr(rax, monitor_block_top);
 955     __ cmpptr(rax, rsp);
 956     __ jcc(Assembler::equal, L);
 957     __ stop("broken stack frame setup in interpreter");
 958     __ bind(L);
 959   }
 960 #endif
 961 
 962   // jvmti support
 963   __ notify_method_entry();
 964 
 965   // work registers
 966   const Register method = rbx;
 967   const Register t      = r11;
 968 
 969   // allocate space for parameters
 970   __ get_method(method);
 971   __ movptr(t, Address(method, Method::const_offset()));
 972   __ load_unsigned_short(t, Address(t, ConstMethod::size_of_parameters_offset()));
 973   __ shll(t, Interpreter::logStackElementSize);
 974 
 975   __ subptr(rsp, t);
 976   __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
 977   __ andptr(rsp, -16); // must be 16 byte boundary (see amd64 ABI)
 978 
 979   // get signature handler
 980   {
 981     Label L;
 982     __ movptr(t, Address(method, Method::signature_handler_offset()));
 983     __ testptr(t, t);
 984     __ jcc(Assembler::notZero, L);
 985     __ call_VM(noreg,
 986                CAST_FROM_FN_PTR(address,
 987                                 InterpreterRuntime::prepare_native_call),
 988                method);
 989     __ get_method(method);
 990     __ movptr(t, Address(method, Method::signature_handler_offset()));
 991     __ bind(L);
 992   }
 993 
 994   // call signature handler
 995   assert(InterpreterRuntime::SignatureHandlerGenerator::from() == r14,
 996          "adjust this code");
 997   assert(InterpreterRuntime::SignatureHandlerGenerator::to() == rsp,
 998          "adjust this code");
 999   assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == rscratch1,
1000           "adjust this code");
1001 
1002   // The generated handlers do not touch RBX (the method oop).
1003   // However, large signatures cannot be cached and are generated
1004   // each time here.  The slow-path generator can do a GC on return,
1005   // so we must reload it after the call.
1006   __ call(t);
1007   __ get_method(method);        // slow path can do a GC, reload RBX
1008 
1009 
1010   // result handler is in rax
1011   // set result handler
1012   __ movptr(Address(rbp,
1013                     (frame::interpreter_frame_result_handler_offset) * wordSize),
1014             rax);
1015 
1016   // pass mirror handle if static call
1017   {
1018     Label L;
1019     const int mirror_offset = in_bytes(Klass::java_mirror_offset());
1020     __ movl(t, Address(method, Method::access_flags_offset()));
1021     __ testl(t, JVM_ACC_STATIC);
1022     __ jcc(Assembler::zero, L);
1023     // get mirror
1024     __ movptr(t, Address(method, Method::const_offset()));
1025     __ movptr(t, Address(t, ConstMethod::constants_offset()));
1026     __ movptr(t, Address(t, ConstantPool::pool_holder_offset_in_bytes()));
1027     __ movptr(t, Address(t, mirror_offset));
1028     // copy mirror into activation frame
1029     __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize),
1030             t);
1031     // pass handle to mirror
1032     __ lea(c_rarg1,
1033            Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize));
1034     __ bind(L);
1035   }
1036 
1037   // get native function entry point
1038   {
1039     Label L;
1040     __ movptr(rax, Address(method, Method::native_function_offset()));
1041     ExternalAddress unsatisfied(SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
1042     __ movptr(rscratch2, unsatisfied.addr());
1043     __ cmpptr(rax, rscratch2);
1044     __ jcc(Assembler::notEqual, L);
1045     __ call_VM(noreg,
1046                CAST_FROM_FN_PTR(address,
1047                                 InterpreterRuntime::prepare_native_call),
1048                method);
1049     __ get_method(method);
1050     __ movptr(rax, Address(method, Method::native_function_offset()));
1051     __ bind(L);
1052   }
1053 
1054   // pass JNIEnv
1055   __ lea(c_rarg0, Address(r15_thread, JavaThread::jni_environment_offset()));
1056 
1057   // It is enough that the pc() points into the right code
1058   // segment. It does not have to be the correct return pc.
1059   __ set_last_Java_frame(rsp, rbp, (address) __ pc());
1060 
1061   // change thread state
1062 #ifdef ASSERT
1063   {
1064     Label L;
1065     __ movl(t, Address(r15_thread, JavaThread::thread_state_offset()));
1066     __ cmpl(t, _thread_in_Java);
1067     __ jcc(Assembler::equal, L);
1068     __ stop("Wrong thread state in native stub");
1069     __ bind(L);
1070   }
1071 #endif
1072 
1073   // Change state to native
1074 
1075   __ movl(Address(r15_thread, JavaThread::thread_state_offset()),
1076           _thread_in_native);
1077 
1078   // Call the native method.
1079   __ call(rax);
1080   // result potentially in rax or xmm0
1081 
1082   // Depending on runtime options, either restore the MXCSR
1083   // register after returning from the JNI Call or verify that
1084   // it wasn't changed during -Xcheck:jni.
1085   if (RestoreMXCSROnJNICalls) {
1086     __ ldmxcsr(ExternalAddress(StubRoutines::x86::mxcsr_std()));
1087   }
1088   else if (CheckJNICalls) {
1089     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::x86::verify_mxcsr_entry())));
1090   }
1091 
1092   // NOTE: The order of these pushes is known to frame::interpreter_frame_result
1093   // in order to extract the result of a method call. If the order of these
1094   // pushes change or anything else is added to the stack then the code in
1095   // interpreter_frame_result must also change.
1096 
1097   __ push(dtos);
1098   __ push(ltos);
1099 
1100   // change thread state
1101   __ movl(Address(r15_thread, JavaThread::thread_state_offset()),
1102           _thread_in_native_trans);
1103 
1104   if (os::is_MP()) {
1105     if (UseMembar) {
1106       // Force this write out before the read below
1107       __ membar(Assembler::Membar_mask_bits(
1108            Assembler::LoadLoad | Assembler::LoadStore |
1109            Assembler::StoreLoad | Assembler::StoreStore));
1110     } else {
1111       // Write serialization page so VM thread can do a pseudo remote membar.
1112       // We use the current thread pointer to calculate a thread specific
1113       // offset to write to within the page. This minimizes bus traffic
1114       // due to cache line collision.
1115       __ serialize_memory(r15_thread, rscratch2);
1116     }
1117   }
1118 
1119   // check for safepoint operation in progress and/or pending suspend requests
1120   {
1121     Label Continue;
1122     __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
1123              SafepointSynchronize::_not_synchronized);
1124 
1125     Label L;
1126     __ jcc(Assembler::notEqual, L);
1127     __ cmpl(Address(r15_thread, JavaThread::suspend_flags_offset()), 0);
1128     __ jcc(Assembler::equal, Continue);
1129     __ bind(L);
1130 
1131     // Don't use call_VM as it will see a possible pending exception
1132     // and forward it and never return here preventing us from
1133     // clearing _last_native_pc down below.  Also can't use
1134     // call_VM_leaf either as it will check to see if r13 & r14 are
1135     // preserved and correspond to the bcp/locals pointers. So we do a
1136     // runtime call by hand.
1137     //
1138     __ mov(c_rarg0, r15_thread);
1139     __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1140     __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1141     __ andptr(rsp, -16); // align stack as required by ABI
1142     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans)));
1143     __ mov(rsp, r12); // restore sp
1144     __ reinit_heapbase();
1145     __ bind(Continue);
1146   }
1147 
1148   // change thread state
1149   __ movl(Address(r15_thread, JavaThread::thread_state_offset()), _thread_in_Java);
1150 
1151   // reset_last_Java_frame
1152   __ reset_last_Java_frame(true, true);
1153 
1154   // reset handle block
1155   __ movptr(t, Address(r15_thread, JavaThread::active_handles_offset()));
1156   __ movptr(Address(t, JNIHandleBlock::top_offset_in_bytes()), (int32_t)NULL_WORD);
1157 
1158   // If result is an oop unbox and store it in frame where gc will see it
1159   // and result handler will pick it up
1160 
1161   {
1162     Label no_oop, store_result;
1163     __ lea(t, ExternalAddress(AbstractInterpreter::result_handler(T_OBJECT)));
1164     __ cmpptr(t, Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize));
1165     __ jcc(Assembler::notEqual, no_oop);
1166     // retrieve result
1167     __ pop(ltos);
1168     __ testptr(rax, rax);
1169     __ jcc(Assembler::zero, store_result);
1170     __ movptr(rax, Address(rax, 0));
1171     __ bind(store_result);
1172     __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize), rax);
1173     // keep stack depth as expected by pushing oop which will eventually be discarde
1174     __ push(ltos);
1175     __ bind(no_oop);
1176   }
1177 
1178 
1179   {
1180     Label no_reguard;
1181     __ cmpl(Address(r15_thread, JavaThread::stack_guard_state_offset()),
1182             JavaThread::stack_guard_yellow_disabled);
1183     __ jcc(Assembler::notEqual, no_reguard);
1184 
1185     __ pusha(); // XXX only save smashed registers
1186     __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1187     __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1188     __ andptr(rsp, -16); // align stack as required by ABI
1189     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
1190     __ mov(rsp, r12); // restore sp
1191     __ popa(); // XXX only restore smashed registers
1192     __ reinit_heapbase();
1193 
1194     __ bind(no_reguard);
1195   }
1196 
1197 
1198   // The method register is junk from after the thread_in_native transition
1199   // until here.  Also can't call_VM until the bcp has been
1200   // restored.  Need bcp for throwing exception below so get it now.
1201   __ get_method(method);
1202 
1203   // restore r13 to have legal interpreter frame, i.e., bci == 0 <=>
1204   // r13 == code_base()
1205   __ movptr(r13, Address(method, Method::const_offset()));   // get ConstMethod*
1206   __ lea(r13, Address(r13, ConstMethod::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, Method::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: Method*
1302   // r13: sender sp
1303   address entry_point = __ pc();
1304 
1305   const Address constMethod(rbx, Method::const_offset());
1306   const Address invocation_counter(rbx,
1307                                    Method::invocation_counter_offset() +
1308                                    InvocationCounter::counter_offset());
1309   const Address access_flags(rbx, Method::access_flags_offset());
1310   const Address size_of_parameters(rdx,
1311                                    ConstMethod::size_of_parameters_offset());
1312   const Address size_of_locals(rdx, ConstMethod::size_of_locals_offset());
1313 
1314 
1315   // get parameter size (always needed)
1316   __ movptr(rdx, constMethod);
1317   __ load_unsigned_short(rcx, size_of_parameters);
1318 
1319   // rbx: Method*
1320   // rcx: size of parameters
1321   // r13: sender_sp (could differ from sp+wordSize if we were called via c2i )
1322 
1323   __ load_unsigned_short(rdx, size_of_locals); // get size of locals in words
1324   __ subl(rdx, rcx); // rdx = no. of additional locals
1325 
1326   // YYY
1327 //   __ incrementl(rdx);
1328 //   __ andl(rdx, -2);
1329 
1330   // see if we've got enough room on the stack for locals plus overhead.
1331   generate_stack_overflow_check();
1332 
1333   // get return address
1334   __ pop(rax);
1335 
1336   // compute beginning of parameters (r14)
1337   __ lea(r14, Address(rsp, rcx, Address::times_8, -wordSize));
1338 
1339   // rdx - # of additional locals
1340   // allocate space for locals
1341   // explicitly initialize locals
1342   {
1343     Label exit, loop;
1344     __ testl(rdx, rdx);
1345     __ jcc(Assembler::lessEqual, exit); // do nothing if rdx <= 0
1346     __ bind(loop);
1347     __ push((int) NULL_WORD); // initialize local variables
1348     __ decrementl(rdx); // until everything initialized
1349     __ jcc(Assembler::greater, loop);
1350     __ bind(exit);
1351   }
1352 
1353   // (pre-)fetch invocation count
1354   if (inc_counter) {
1355     __ movl(rcx, invocation_counter);
1356   }
1357   // initialize fixed part of activation frame
1358   generate_fixed_frame(false);
1359 
1360   // make sure method is not native & not abstract
1361 #ifdef ASSERT
1362   __ movl(rax, access_flags);
1363   {
1364     Label L;
1365     __ testl(rax, JVM_ACC_NATIVE);
1366     __ jcc(Assembler::zero, L);
1367     __ stop("tried to execute native method as non-native");
1368     __ bind(L);
1369   }
1370   {
1371     Label L;
1372     __ testl(rax, JVM_ACC_ABSTRACT);
1373     __ jcc(Assembler::zero, L);
1374     __ stop("tried to execute abstract method in interpreter");
1375     __ bind(L);
1376   }
1377 #endif
1378 
1379   // Since at this point in the method invocation the exception
1380   // handler would try to exit the monitor of synchronized methods
1381   // which hasn't been entered yet, we set the thread local variable
1382   // _do_not_unlock_if_synchronized to true. The remove_activation
1383   // will check this flag.
1384 
1385   const Address do_not_unlock_if_synchronized(r15_thread,
1386         in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1387   __ movbool(do_not_unlock_if_synchronized, true);
1388 
1389   // increment invocation count & check for overflow
1390   Label invocation_counter_overflow;
1391   Label profile_method;
1392   Label profile_method_continue;
1393   if (inc_counter) {
1394     generate_counter_incr(&invocation_counter_overflow,
1395                           &profile_method,
1396                           &profile_method_continue);
1397     if (ProfileInterpreter) {
1398       __ bind(profile_method_continue);
1399     }
1400   }
1401 
1402   Label continue_after_compile;
1403   __ bind(continue_after_compile);
1404 
1405   // check for synchronized interpreted methods
1406   bang_stack_shadow_pages(false);
1407 
1408   // reset the _do_not_unlock_if_synchronized flag
1409   __ movbool(do_not_unlock_if_synchronized, false);
1410 
1411   // check for synchronized methods
1412   // Must happen AFTER invocation_counter check and stack overflow check,
1413   // so method is not locked if overflows.
1414   if (synchronized) {
1415     // Allocate monitor and lock method
1416     lock_method();
1417   } else {
1418     // no synchronization necessary
1419 #ifdef ASSERT
1420     {
1421       Label L;
1422       __ movl(rax, access_flags);
1423       __ testl(rax, JVM_ACC_SYNCHRONIZED);
1424       __ jcc(Assembler::zero, L);
1425       __ stop("method needs synchronization");
1426       __ bind(L);
1427     }
1428 #endif
1429   }
1430 
1431   // start execution
1432 #ifdef ASSERT
1433   {
1434     Label L;
1435      const Address monitor_block_top (rbp,
1436                  frame::interpreter_frame_monitor_block_top_offset * wordSize);
1437     __ movptr(rax, monitor_block_top);
1438     __ cmpptr(rax, rsp);
1439     __ jcc(Assembler::equal, L);
1440     __ stop("broken stack frame setup in interpreter");
1441     __ bind(L);
1442   }
1443 #endif
1444 
1445   // jvmti support
1446   __ notify_method_entry();
1447 
1448   __ dispatch_next(vtos);
1449 
1450   // invocation counter overflow
1451   if (inc_counter) {
1452     if (ProfileInterpreter) {
1453       // We have decided to profile this method in the interpreter
1454       __ bind(profile_method);
1455       __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1456       __ set_method_data_pointer_for_bcp();
1457       __ get_method(rbx);
1458       __ jmp(profile_method_continue);
1459     }
1460     // Handle overflow of counter and compile method
1461     __ bind(invocation_counter_overflow);
1462     generate_counter_overflow(&continue_after_compile);
1463   }
1464 
1465   return entry_point;
1466 }
1467 
1468 // Entry points
1469 //
1470 // Here we generate the various kind of entries into the interpreter.
1471 // The two main entry type are generic bytecode methods and native
1472 // call method.  These both come in synchronized and non-synchronized
1473 // versions but the frame layout they create is very similar. The
1474 // other method entry types are really just special purpose entries
1475 // that are really entry and interpretation all in one. These are for
1476 // trivial methods like accessor, empty, or special math methods.
1477 //
1478 // When control flow reaches any of the entry types for the interpreter
1479 // the following holds ->
1480 //
1481 // Arguments:
1482 //
1483 // rbx: Method*
1484 //
1485 // Stack layout immediately at entry
1486 //
1487 // [ return address     ] <--- rsp
1488 // [ parameter n        ]
1489 //   ...
1490 // [ parameter 1        ]
1491 // [ expression stack   ] (caller's java expression stack)
1492 
1493 // Assuming that we don't go to one of the trivial specialized entries
1494 // the stack will look like below when we are ready to execute the
1495 // first bytecode (or call the native routine). The register usage
1496 // will be as the template based interpreter expects (see
1497 // interpreter_amd64.hpp).
1498 //
1499 // local variables follow incoming parameters immediately; i.e.
1500 // the return address is moved to the end of the locals).
1501 //
1502 // [ monitor entry      ] <--- rsp
1503 //   ...
1504 // [ monitor entry      ]
1505 // [ expr. stack bottom ]
1506 // [ saved r13          ]
1507 // [ current r14        ]
1508 // [ Method*            ]
1509 // [ saved ebp          ] <--- rbp
1510 // [ return address     ]
1511 // [ local variable m   ]
1512 //   ...
1513 // [ local variable 1   ]
1514 // [ parameter n        ]
1515 //   ...
1516 // [ parameter 1        ] <--- r14
1517 
1518 address AbstractInterpreterGenerator::generate_method_entry(
1519                                         AbstractInterpreter::MethodKind kind) {
1520   // determine code generation flags
1521   bool synchronized = false;
1522   address entry_point = NULL;
1523 
1524   switch (kind) {
1525   case Interpreter::zerolocals             :                                                                             break;
1526   case Interpreter::zerolocals_synchronized: synchronized = true;                                                        break;
1527   case Interpreter::native                 : entry_point = ((InterpreterGenerator*)this)->generate_native_entry(false); break;
1528   case Interpreter::native_synchronized    : entry_point = ((InterpreterGenerator*)this)->generate_native_entry(true);  break;
1529   case Interpreter::empty                  : entry_point = ((InterpreterGenerator*)this)->generate_empty_entry();       break;
1530   case Interpreter::accessor               : entry_point = ((InterpreterGenerator*)this)->generate_accessor_entry();    break;
1531   case Interpreter::abstract               : entry_point = ((InterpreterGenerator*)this)->generate_abstract_entry();    break;
1532 
1533   case Interpreter::java_lang_math_sin     : // fall thru
1534   case Interpreter::java_lang_math_cos     : // fall thru
1535   case Interpreter::java_lang_math_tan     : // fall thru
1536   case Interpreter::java_lang_math_abs     : // fall thru
1537   case Interpreter::java_lang_math_log     : // fall thru
1538   case Interpreter::java_lang_math_log10   : // fall thru
1539   case Interpreter::java_lang_math_sqrt    : // fall thru
1540   case Interpreter::java_lang_math_pow     : // fall thru
1541   case Interpreter::java_lang_math_exp     : entry_point = ((InterpreterGenerator*)this)->generate_math_entry(kind);    break;
1542   case Interpreter::java_lang_ref_reference_get
1543                                            : entry_point = ((InterpreterGenerator*)this)->generate_Reference_get_entry(); break;
1544   default:
1545     fatal(err_msg("unexpected method kind: %d", kind));
1546     break;
1547   }
1548 
1549   if (entry_point) {
1550     return entry_point;
1551   }
1552 
1553   return ((InterpreterGenerator*) this)->
1554                                 generate_normal_entry(synchronized);
1555 }
1556 
1557 // These should never be compiled since the interpreter will prefer
1558 // the compiled version to the intrinsic version.
1559 bool AbstractInterpreter::can_be_compiled(methodHandle m) {
1560   switch (method_kind(m)) {
1561     case Interpreter::java_lang_math_sin     : // fall thru
1562     case Interpreter::java_lang_math_cos     : // fall thru
1563     case Interpreter::java_lang_math_tan     : // fall thru
1564     case Interpreter::java_lang_math_abs     : // fall thru
1565     case Interpreter::java_lang_math_log     : // fall thru
1566     case Interpreter::java_lang_math_log10   : // fall thru
1567     case Interpreter::java_lang_math_sqrt    : // fall thru
1568     case Interpreter::java_lang_math_pow     : // fall thru
1569     case Interpreter::java_lang_math_exp     :
1570       return false;
1571     default:
1572       return true;
1573   }
1574 }
1575 
1576 // How much stack a method activation needs in words.
1577 int AbstractInterpreter::size_top_interpreter_activation(Method* method) {
1578   const int entry_size = frame::interpreter_frame_monitor_size();
1579 
1580   // total overhead size: entry_size + (saved rbp thru expr stack
1581   // bottom).  be sure to change this if you add/subtract anything
1582   // to/from the overhead area
1583   const int overhead_size =
1584     -(frame::interpreter_frame_initial_sp_offset) + entry_size;
1585 
1586   const int stub_code = frame::entry_frame_after_call_words;
1587   const int extra_stack = Method::extra_stack_entries();
1588   const int method_stack = (method->max_locals() + method->max_stack() + extra_stack) *
1589                            Interpreter::stackElementWords;
1590   return (overhead_size + method_stack + stub_code);
1591 }
1592 
1593 int AbstractInterpreter::layout_activation(Method* method,
1594                                            int tempcount,
1595                                            int popframe_extra_args,
1596                                            int moncount,
1597                                            int caller_actual_parameters,
1598                                            int callee_param_count,
1599                                            int callee_locals,
1600                                            frame* caller,
1601                                            frame* interpreter_frame,
1602                                            bool is_top_frame,
1603                                            bool is_bottom_frame) {
1604   // Note: This calculation must exactly parallel the frame setup
1605   // in AbstractInterpreterGenerator::generate_method_entry.
1606   // If interpreter_frame!=NULL, set up the method, locals, and monitors.
1607   // The frame interpreter_frame, if not NULL, is guaranteed to be the
1608   // right size, as determined by a previous call to this method.
1609   // It is also guaranteed to be walkable even though it is in a skeletal state
1610 
1611   // fixed size of an interpreter frame:
1612   int max_locals = method->max_locals() * Interpreter::stackElementWords;
1613   int extra_locals = (method->max_locals() - method->size_of_parameters()) *
1614                      Interpreter::stackElementWords;
1615 
1616   int overhead = frame::sender_sp_offset -
1617                  frame::interpreter_frame_initial_sp_offset;
1618   // Our locals were accounted for by the caller (or last_frame_adjust
1619   // on the transistion) Since the callee parameters already account
1620   // for the callee's params we only need to account for the extra
1621   // locals.
1622   int size = overhead +
1623          (callee_locals - callee_param_count)*Interpreter::stackElementWords +
1624          moncount * frame::interpreter_frame_monitor_size() +
1625          tempcount* Interpreter::stackElementWords + popframe_extra_args;
1626   if (interpreter_frame != NULL) {
1627 #ifdef ASSERT
1628     if (!EnableInvokeDynamic)
1629       // @@@ FIXME: Should we correct interpreter_frame_sender_sp in the calling sequences?
1630       // Probably, since deoptimization doesn't work yet.
1631       assert(caller->unextended_sp() == interpreter_frame->interpreter_frame_sender_sp(), "Frame not properly walkable");
1632     assert(caller->sp() == interpreter_frame->sender_sp(), "Frame not properly walkable(2)");
1633 #endif
1634 
1635     interpreter_frame->interpreter_frame_set_method(method);
1636     // NOTE the difference in using sender_sp and
1637     // interpreter_frame_sender_sp interpreter_frame_sender_sp is
1638     // the original sp of the caller (the unextended_sp) and
1639     // sender_sp is fp+16 XXX
1640     intptr_t* locals = interpreter_frame->sender_sp() + max_locals - 1;
1641 
1642 #ifdef ASSERT
1643     if (caller->is_interpreted_frame()) {
1644       assert(locals < caller->fp() + frame::interpreter_frame_initial_sp_offset, "bad placement");
1645     }
1646 #endif
1647 
1648     interpreter_frame->interpreter_frame_set_locals(locals);
1649     BasicObjectLock* montop = interpreter_frame->interpreter_frame_monitor_begin();
1650     BasicObjectLock* monbot = montop - moncount;
1651     interpreter_frame->interpreter_frame_set_monitor_end(monbot);
1652 
1653     // Set last_sp
1654     intptr_t*  esp = (intptr_t*) monbot -
1655                      tempcount*Interpreter::stackElementWords -
1656                      popframe_extra_args;
1657     interpreter_frame->interpreter_frame_set_last_sp(esp);
1658 
1659     // All frames but the initial (oldest) interpreter frame we fill in have
1660     // a value for sender_sp that allows walking the stack but isn't
1661     // truly correct. Correct the value here.
1662     if (extra_locals != 0 &&
1663         interpreter_frame->sender_sp() ==
1664         interpreter_frame->interpreter_frame_sender_sp()) {
1665       interpreter_frame->set_interpreter_frame_sender_sp(caller->sp() +
1666                                                          extra_locals);
1667     }
1668     *interpreter_frame->interpreter_frame_cache_addr() =
1669       method->constants()->cache();
1670   }
1671   return size;
1672 }
1673 
1674 //-----------------------------------------------------------------------------
1675 // Exceptions
1676 
1677 void TemplateInterpreterGenerator::generate_throw_exception() {
1678   // Entry point in previous activation (i.e., if the caller was
1679   // interpreted)
1680   Interpreter::_rethrow_exception_entry = __ pc();
1681   // Restore sp to interpreter_frame_last_sp even though we are going
1682   // to empty the expression stack for the exception processing.
1683   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
1684   // rax: exception
1685   // rdx: return address/pc that threw exception
1686   __ restore_bcp();    // r13 points to call/send
1687   __ restore_locals();
1688   __ reinit_heapbase();  // restore r12 as heapbase.
1689   // Entry point for exceptions thrown within interpreter code
1690   Interpreter::_throw_exception_entry = __ pc();
1691   // expression stack is undefined here
1692   // rax: exception
1693   // r13: exception bcp
1694   __ verify_oop(rax);
1695   __ mov(c_rarg1, rax);
1696 
1697   // expression stack must be empty before entering the VM in case of
1698   // an exception
1699   __ empty_expression_stack();
1700   // find exception handler address and preserve exception oop
1701   __ call_VM(rdx,
1702              CAST_FROM_FN_PTR(address,
1703                           InterpreterRuntime::exception_handler_for_exception),
1704              c_rarg1);
1705   // rax: exception handler entry point
1706   // rdx: preserved exception oop
1707   // r13: bcp for exception handler
1708   __ push_ptr(rdx); // push exception which is now the only value on the stack
1709   __ jmp(rax); // jump to exception handler (may be _remove_activation_entry!)
1710 
1711   // If the exception is not handled in the current frame the frame is
1712   // removed and the exception is rethrown (i.e. exception
1713   // continuation is _rethrow_exception).
1714   //
1715   // Note: At this point the bci is still the bxi for the instruction
1716   // which caused the exception and the expression stack is
1717   // empty. Thus, for any VM calls at this point, GC will find a legal
1718   // oop map (with empty expression stack).
1719 
1720   // In current activation
1721   // tos: exception
1722   // esi: exception bcp
1723 
1724   //
1725   // JVMTI PopFrame support
1726   //
1727 
1728   Interpreter::_remove_activation_preserving_args_entry = __ pc();
1729   __ empty_expression_stack();
1730   // Set the popframe_processing bit in pending_popframe_condition
1731   // indicating that we are currently handling popframe, so that
1732   // call_VMs that may happen later do not trigger new popframe
1733   // handling cycles.
1734   __ movl(rdx, Address(r15_thread, JavaThread::popframe_condition_offset()));
1735   __ orl(rdx, JavaThread::popframe_processing_bit);
1736   __ movl(Address(r15_thread, JavaThread::popframe_condition_offset()), rdx);
1737 
1738   {
1739     // Check to see whether we are returning to a deoptimized frame.
1740     // (The PopFrame call ensures that the caller of the popped frame is
1741     // either interpreted or compiled and deoptimizes it if compiled.)
1742     // In this case, we can't call dispatch_next() after the frame is
1743     // popped, but instead must save the incoming arguments and restore
1744     // them after deoptimization has occurred.
1745     //
1746     // Note that we don't compare the return PC against the
1747     // deoptimization blob's unpack entry because of the presence of
1748     // adapter frames in C2.
1749     Label caller_not_deoptimized;
1750     __ movptr(c_rarg1, Address(rbp, frame::return_addr_offset * wordSize));
1751     __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1752                                InterpreterRuntime::interpreter_contains), c_rarg1);
1753     __ testl(rax, rax);
1754     __ jcc(Assembler::notZero, caller_not_deoptimized);
1755 
1756     // Compute size of arguments for saving when returning to
1757     // deoptimized caller
1758     __ get_method(rax);
1759     __ movptr(rax, Address(rax, Method::const_offset()));
1760     __ load_unsigned_short(rax, Address(rax, in_bytes(ConstMethod::
1761                                                 size_of_parameters_offset())));
1762     __ shll(rax, Interpreter::logStackElementSize);
1763     __ restore_locals(); // XXX do we need this?
1764     __ subptr(r14, rax);
1765     __ addptr(r14, wordSize);
1766     // Save these arguments
1767     __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1768                                            Deoptimization::
1769                                            popframe_preserve_args),
1770                           r15_thread, rax, r14);
1771 
1772     __ remove_activation(vtos, rdx,
1773                          /* throw_monitor_exception */ false,
1774                          /* install_monitor_exception */ false,
1775                          /* notify_jvmdi */ false);
1776 
1777     // Inform deoptimization that it is responsible for restoring
1778     // these arguments
1779     __ movl(Address(r15_thread, JavaThread::popframe_condition_offset()),
1780             JavaThread::popframe_force_deopt_reexecution_bit);
1781 
1782     // Continue in deoptimization handler
1783     __ jmp(rdx);
1784 
1785     __ bind(caller_not_deoptimized);
1786   }
1787 
1788   __ remove_activation(vtos, rdx, /* rdx result (retaddr) is not used */
1789                        /* throw_monitor_exception */ false,
1790                        /* install_monitor_exception */ false,
1791                        /* notify_jvmdi */ false);
1792 
1793   // Finish with popframe handling
1794   // A previous I2C followed by a deoptimization might have moved the
1795   // outgoing arguments further up the stack. PopFrame expects the
1796   // mutations to those outgoing arguments to be preserved and other
1797   // constraints basically require this frame to look exactly as
1798   // though it had previously invoked an interpreted activation with
1799   // no space between the top of the expression stack (current
1800   // last_sp) and the top of stack. Rather than force deopt to
1801   // maintain this kind of invariant all the time we call a small
1802   // fixup routine to move the mutated arguments onto the top of our
1803   // expression stack if necessary.
1804   __ mov(c_rarg1, rsp);
1805   __ movptr(c_rarg2, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1806   // PC must point into interpreter here
1807   __ set_last_Java_frame(noreg, rbp, __ pc());
1808   __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), r15_thread, c_rarg1, c_rarg2);
1809   __ reset_last_Java_frame(true, true);
1810   // Restore the last_sp and null it out
1811   __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1812   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
1813 
1814   __ restore_bcp();  // XXX do we need this?
1815   __ restore_locals(); // XXX do we need this?
1816   // The method data pointer was incremented already during
1817   // call profiling. We have to restore the mdp for the current bcp.
1818   if (ProfileInterpreter) {
1819     __ set_method_data_pointer_for_bcp();
1820   }
1821 
1822   // Clear the popframe condition flag
1823   __ movl(Address(r15_thread, JavaThread::popframe_condition_offset()),
1824           JavaThread::popframe_inactive);
1825 
1826   __ dispatch_next(vtos);
1827   // end of PopFrame support
1828 
1829   Interpreter::_remove_activation_entry = __ pc();
1830 
1831   // preserve exception over this code sequence
1832   __ pop_ptr(rax);
1833   __ movptr(Address(r15_thread, JavaThread::vm_result_offset()), rax);
1834   // remove the activation (without doing throws on illegalMonitorExceptions)
1835   __ remove_activation(vtos, rdx, false, true, false);
1836   // restore exception
1837   __ get_vm_result(rax, r15_thread);
1838 
1839   // In between activations - previous activation type unknown yet
1840   // compute continuation point - the continuation point expects the
1841   // following registers set up:
1842   //
1843   // rax: exception
1844   // rdx: return address/pc that threw exception
1845   // rsp: expression stack of caller
1846   // rbp: ebp of caller
1847   __ push(rax);                                  // save exception
1848   __ push(rdx);                                  // save return address
1849   __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1850                           SharedRuntime::exception_handler_for_return_address),
1851                         r15_thread, rdx);
1852   __ mov(rbx, rax);                              // save exception handler
1853   __ pop(rdx);                                   // restore return address
1854   __ pop(rax);                                   // restore exception
1855   // Note that an "issuing PC" is actually the next PC after the call
1856   __ jmp(rbx);                                   // jump to exception
1857                                                  // handler of caller
1858 }
1859 
1860 
1861 //
1862 // JVMTI ForceEarlyReturn support
1863 //
1864 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
1865   address entry = __ pc();
1866 
1867   __ restore_bcp();
1868   __ restore_locals();
1869   __ empty_expression_stack();
1870   __ load_earlyret_value(state);
1871 
1872   __ movptr(rdx, Address(r15_thread, JavaThread::jvmti_thread_state_offset()));
1873   Address cond_addr(rdx, JvmtiThreadState::earlyret_state_offset());
1874 
1875   // Clear the earlyret state
1876   __ movl(cond_addr, JvmtiThreadState::earlyret_inactive);
1877 
1878   __ remove_activation(state, rsi,
1879                        false, /* throw_monitor_exception */
1880                        false, /* install_monitor_exception */
1881                        true); /* notify_jvmdi */
1882   __ jmp(rsi);
1883 
1884   return entry;
1885 } // end of ForceEarlyReturn support
1886 
1887 
1888 //-----------------------------------------------------------------------------
1889 // Helper for vtos entry point generation
1890 
1891 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
1892                                                          address& bep,
1893                                                          address& cep,
1894                                                          address& sep,
1895                                                          address& aep,
1896                                                          address& iep,
1897                                                          address& lep,
1898                                                          address& fep,
1899                                                          address& dep,
1900                                                          address& vep) {
1901   assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
1902   Label L;
1903   aep = __ pc();  __ push_ptr();  __ jmp(L);
1904   fep = __ pc();  __ push_f();    __ jmp(L);
1905   dep = __ pc();  __ push_d();    __ jmp(L);
1906   lep = __ pc();  __ push_l();    __ jmp(L);
1907   bep = cep = sep =
1908   iep = __ pc();  __ push_i();
1909   vep = __ pc();
1910   __ bind(L);
1911   generate_and_dispatch(t);
1912 }
1913 
1914 
1915 //-----------------------------------------------------------------------------
1916 // Generation of individual instructions
1917 
1918 // helpers for generate_and_dispatch
1919 
1920 
1921 InterpreterGenerator::InterpreterGenerator(StubQueue* code)
1922   : TemplateInterpreterGenerator(code) {
1923    generate_all(); // down here so it can be "virtual"
1924 }
1925 
1926 //-----------------------------------------------------------------------------
1927 
1928 // Non-product code
1929 #ifndef PRODUCT
1930 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
1931   address entry = __ pc();
1932 
1933   __ push(state);
1934   __ push(c_rarg0);
1935   __ push(c_rarg1);
1936   __ push(c_rarg2);
1937   __ push(c_rarg3);
1938   __ mov(c_rarg2, rax);  // Pass itos
1939 #ifdef _WIN64
1940   __ movflt(xmm3, xmm0); // Pass ftos
1941 #endif
1942   __ call_VM(noreg,
1943              CAST_FROM_FN_PTR(address, SharedRuntime::trace_bytecode),
1944              c_rarg1, c_rarg2, c_rarg3);
1945   __ pop(c_rarg3);
1946   __ pop(c_rarg2);
1947   __ pop(c_rarg1);
1948   __ pop(c_rarg0);
1949   __ pop(state);
1950   __ ret(0);                                   // return from result handler
1951 
1952   return entry;
1953 }
1954 
1955 void TemplateInterpreterGenerator::count_bytecode() {
1956   __ incrementl(ExternalAddress((address) &BytecodeCounter::_counter_value));
1957 }
1958 
1959 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
1960   __ incrementl(ExternalAddress((address) &BytecodeHistogram::_counters[t->bytecode()]));
1961 }
1962 
1963 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
1964   __ mov32(rbx, ExternalAddress((address) &BytecodePairHistogram::_index));
1965   __ shrl(rbx, BytecodePairHistogram::log2_number_of_codes);
1966   __ orl(rbx,
1967          ((int) t->bytecode()) <<
1968          BytecodePairHistogram::log2_number_of_codes);
1969   __ mov32(ExternalAddress((address) &BytecodePairHistogram::_index), rbx);
1970   __ lea(rscratch1, ExternalAddress((address) BytecodePairHistogram::_counters));
1971   __ incrementl(Address(rscratch1, rbx, Address::times_4));
1972 }
1973 
1974 
1975 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
1976   // Call a little run-time stub to avoid blow-up for each bytecode.
1977   // The run-time runtime saves the right registers, depending on
1978   // the tosca in-state for the given template.
1979 
1980   assert(Interpreter::trace_code(t->tos_in()) != NULL,
1981          "entry must have been generated");
1982   __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1983   __ andptr(rsp, -16); // align stack as required by ABI
1984   __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in())));
1985   __ mov(rsp, r12); // restore sp
1986   __ reinit_heapbase();
1987 }
1988 
1989 
1990 void TemplateInterpreterGenerator::stop_interpreter_at() {
1991   Label L;
1992   __ cmp32(ExternalAddress((address) &BytecodeCounter::_counter_value),
1993            StopInterpreterAt);
1994   __ jcc(Assembler::notEqual, L);
1995   __ int3();
1996   __ bind(L);
1997 }
1998 #endif // !PRODUCT
1999 #endif // ! CC_INTERP