1 /*
   2  * Copyright (c) 2003, 2017, 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, size_t index_size) {
 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   if (state == atos) {
 181     Register mdp = rbx;
 182     Register tmp = rcx;
 183     __ profile_return_type(mdp, rax, tmp);
 184   }
 185 
 186   const Register cache = rbx;
 187   const Register index = rcx;
 188   __ get_cache_and_index_at_bcp(cache, index, 1, index_size);
 189 
 190   const Register flags = cache;
 191   __ movl(flags, Address(cache, index, Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()));
 192   __ andl(flags, ConstantPoolCacheEntry::parameter_size_mask);
 193   __ lea(rsp, Address(rsp, flags, Interpreter::stackElementScale()));
 194   __ dispatch_next(state, step);
 195 
 196   return entry;
 197 }
 198 
 199 
 200 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state,
 201                                                                int step) {
 202   address entry = __ pc();
 203   // NULL last_sp until next java call
 204   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
 205   __ restore_bcp();
 206   __ restore_locals();
 207   // handle exceptions
 208   {
 209     Label L;
 210     __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t) NULL_WORD);
 211     __ jcc(Assembler::zero, L);
 212     __ call_VM(noreg,
 213                CAST_FROM_FN_PTR(address,
 214                                 InterpreterRuntime::throw_pending_exception));
 215     __ should_not_reach_here();
 216     __ bind(L);
 217   }
 218   __ dispatch_next(state, step);
 219   return entry;
 220 }
 221 
 222 int AbstractInterpreter::BasicType_as_index(BasicType type) {
 223   int i = 0;
 224   switch (type) {
 225     case T_BOOLEAN: i = 0; break;
 226     case T_CHAR   : i = 1; break;
 227     case T_BYTE   : i = 2; break;
 228     case T_SHORT  : i = 3; break;
 229     case T_INT    : i = 4; break;
 230     case T_LONG   : i = 5; break;
 231     case T_VOID   : i = 6; break;
 232     case T_FLOAT  : i = 7; break;
 233     case T_DOUBLE : i = 8; break;
 234     case T_OBJECT : i = 9; break;
 235     case T_ARRAY  : i = 9; break;
 236     default       : ShouldNotReachHere();
 237   }
 238   assert(0 <= i && i < AbstractInterpreter::number_of_result_handlers,
 239          "index out of bounds");
 240   return i;
 241 }
 242 
 243 
 244 address TemplateInterpreterGenerator::generate_result_handler_for(
 245         BasicType type) {
 246   address entry = __ pc();
 247   switch (type) {
 248   case T_BOOLEAN: __ c2bool(rax);            break;
 249   case T_CHAR   : __ movzwl(rax, rax);       break;
 250   case T_BYTE   : __ sign_extend_byte(rax);  break;
 251   case T_SHORT  : __ sign_extend_short(rax); break;
 252   case T_INT    : /* nothing to do */        break;
 253   case T_LONG   : /* nothing to do */        break;
 254   case T_VOID   : /* nothing to do */        break;
 255   case T_FLOAT  : /* nothing to do */        break;
 256   case T_DOUBLE : /* nothing to do */        break;
 257   case T_OBJECT :
 258     // retrieve result from frame
 259     __ movptr(rax, Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize));
 260     // and verify it
 261     __ verify_oop(rax);
 262     break;
 263   default       : ShouldNotReachHere();
 264   }
 265   __ ret(0);                                   // return from result handler
 266   return entry;
 267 }
 268 
 269 address TemplateInterpreterGenerator::generate_safept_entry_for(
 270         TosState state,
 271         address runtime_entry) {
 272   address entry = __ pc();
 273   __ push(state);
 274   __ call_VM(noreg, runtime_entry);
 275   __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
 276   return entry;
 277 }
 278 
 279 
 280 
 281 // Helpers for commoning out cases in the various type of method entries.
 282 //
 283 
 284 
 285 // increment invocation count & check for overflow
 286 //
 287 // Note: checking for negative value instead of overflow
 288 //       so we have a 'sticky' overflow test
 289 //
 290 // rbx: method
 291 // ecx: invocation counter
 292 //
 293 void InterpreterGenerator::generate_counter_incr(
 294         Label* overflow,
 295         Label* profile_method,
 296         Label* profile_method_continue) {
 297   Label done;
 298   // Note: In tiered we increment either counters in Method* or in MDO depending if we're profiling or not.
 299   if (TieredCompilation) {
 300     int increment = InvocationCounter::count_increment;
 301     int mask = ((1 << Tier0InvokeNotifyFreqLog)  - 1) << InvocationCounter::count_shift;
 302     Label no_mdo;
 303     if (ProfileInterpreter) {
 304       // Are we profiling?
 305       __ movptr(rax, Address(rbx, Method::method_data_offset()));
 306       __ testptr(rax, rax);
 307       __ jccb(Assembler::zero, no_mdo);
 308       // Increment counter in the MDO
 309       const Address mdo_invocation_counter(rax, in_bytes(MethodData::invocation_counter_offset()) +
 310                                                 in_bytes(InvocationCounter::counter_offset()));
 311       __ increment_mask_and_jump(mdo_invocation_counter, increment, mask, rcx, false, Assembler::zero, overflow);
 312       __ jmp(done);
 313     }
 314     __ bind(no_mdo);
 315     // Increment counter in MethodCounters
 316     const Address invocation_counter(rax,
 317                   MethodCounters::invocation_counter_offset() +
 318                   InvocationCounter::counter_offset());
 319     __ get_method_counters(rbx, rax, done);
 320     __ increment_mask_and_jump(invocation_counter, increment, mask, rcx,
 321                                false, Assembler::zero, overflow);
 322     __ bind(done);
 323   } else {
 324     const Address backedge_counter(rax,
 325                   MethodCounters::backedge_counter_offset() +
 326                   InvocationCounter::counter_offset());
 327     const Address invocation_counter(rax,
 328                   MethodCounters::invocation_counter_offset() +
 329                   InvocationCounter::counter_offset());
 330 
 331     __ get_method_counters(rbx, rax, done);
 332 
 333     if (ProfileInterpreter) {
 334       __ incrementl(Address(rax,
 335               MethodCounters::interpreter_invocation_counter_offset()));
 336     }
 337     // Update standard invocation counters
 338     __ movl(rcx, invocation_counter);
 339     __ incrementl(rcx, InvocationCounter::count_increment);
 340     __ movl(invocation_counter, rcx); // save invocation count
 341 
 342     __ movl(rax, backedge_counter);   // load backedge counter
 343     __ andl(rax, InvocationCounter::count_mask_value); // mask out the status bits
 344 
 345     __ addl(rcx, rax);                // add both counters
 346 
 347     // profile_method is non-null only for interpreted method so
 348     // profile_method != NULL == !native_call
 349 
 350     if (ProfileInterpreter && profile_method != NULL) {
 351       // Test to see if we should create a method data oop
 352       __ cmp32(rcx, ExternalAddress((address)&InvocationCounter::InterpreterProfileLimit));
 353       __ jcc(Assembler::less, *profile_method_continue);
 354 
 355       // if no method data exists, go to profile_method
 356       __ test_method_data_pointer(rax, *profile_method);
 357     }
 358 
 359     __ cmp32(rcx, ExternalAddress((address)&InvocationCounter::InterpreterInvocationLimit));
 360     __ jcc(Assembler::aboveEqual, *overflow);
 361     __ bind(done);
 362   }
 363 }
 364 
 365 void InterpreterGenerator::generate_counter_overflow(Label* do_continue) {
 366 
 367   // Asm interpreter on entry
 368   // r14 - locals
 369   // r13 - bcp
 370   // rbx - method
 371   // edx - cpool --- DOES NOT APPEAR TO BE TRUE
 372   // rbp - interpreter frame
 373 
 374   // On return (i.e. jump to entry_point) [ back to invocation of interpreter ]
 375   // Everything as it was on entry
 376   // rdx is not restored. Doesn't appear to really be set.
 377 
 378   // InterpreterRuntime::frequency_counter_overflow takes two
 379   // arguments, the first (thread) is passed by call_VM, the second
 380   // indicates if the counter overflow occurs at a backwards branch
 381   // (NULL bcp).  We pass zero for it.  The call returns the address
 382   // of the verified entry point for the method or NULL if the
 383   // compilation did not complete (either went background or bailed
 384   // out).
 385   __ movl(c_rarg1, 0);
 386   __ call_VM(noreg,
 387              CAST_FROM_FN_PTR(address,
 388                               InterpreterRuntime::frequency_counter_overflow),
 389              c_rarg1);
 390 
 391   __ movptr(rbx, Address(rbp, method_offset));   // restore Method*
 392   // Preserve invariant that r13/r14 contain bcp/locals of sender frame
 393   // and jump to the interpreted entry.
 394   __ jmp(*do_continue, relocInfo::none);
 395 }
 396 
 397 // See if we've got enough room on the stack for locals plus overhead.
 398 // The expression stack grows down incrementally, so the normal guard
 399 // page mechanism will work for that.
 400 //
 401 // NOTE: Since the additional locals are also always pushed (wasn't
 402 // obvious in generate_method_entry) so the guard should work for them
 403 // too.
 404 //
 405 // Args:
 406 //      rdx: number of additional locals this frame needs (what we must check)
 407 //      rbx: Method*
 408 //
 409 // Kills:
 410 //      rax
 411 void InterpreterGenerator::generate_stack_overflow_check(void) {
 412 
 413   // monitor entry size: see picture of stack set
 414   // (generate_method_entry) and frame_amd64.hpp
 415   const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
 416 
 417   // total overhead size: entry_size + (saved rbp through expr stack
 418   // bottom).  be sure to change this if you add/subtract anything
 419   // to/from the overhead area
 420   const int overhead_size =
 421     -(frame::interpreter_frame_initial_sp_offset * wordSize) + entry_size;
 422 
 423   const int page_size = os::vm_page_size();
 424 
 425   Label after_frame_check;
 426 
 427   // see if the frame is greater than one page in size. If so,
 428   // then we need to verify there is enough stack space remaining
 429   // for the additional locals.
 430   __ cmpl(rdx, (page_size - overhead_size) / Interpreter::stackElementSize);
 431   __ jcc(Assembler::belowEqual, after_frame_check);
 432 
 433   // compute rsp as if this were going to be the last frame on
 434   // the stack before the red zone
 435 
 436   const Address stack_base(r15_thread, Thread::stack_base_offset());
 437   const Address stack_size(r15_thread, Thread::stack_size_offset());
 438 
 439   // locals + overhead, in bytes
 440   __ mov(rax, rdx);
 441   __ shlptr(rax, Interpreter::logStackElementSize);  // 2 slots per parameter.
 442   __ addptr(rax, overhead_size);
 443 
 444 #ifdef ASSERT
 445   Label stack_base_okay, stack_size_okay;
 446   // verify that thread stack base is non-zero
 447   __ cmpptr(stack_base, (int32_t)NULL_WORD);
 448   __ jcc(Assembler::notEqual, stack_base_okay);
 449   __ stop("stack base is zero");
 450   __ bind(stack_base_okay);
 451   // verify that thread stack size is non-zero
 452   __ cmpptr(stack_size, 0);
 453   __ jcc(Assembler::notEqual, stack_size_okay);
 454   __ stop("stack size is zero");
 455   __ bind(stack_size_okay);
 456 #endif
 457 
 458   // Add stack base to locals and subtract stack size
 459   __ addptr(rax, stack_base);
 460   __ subptr(rax, stack_size);
 461 
 462   // Use the maximum number of pages we might bang.
 463   const int max_pages = StackShadowPages > (StackRedPages+StackYellowPages) ? StackShadowPages :
 464                                                                               (StackRedPages+StackYellowPages);
 465 
 466   // add in the red and yellow zone sizes
 467   __ addptr(rax, max_pages * page_size);
 468 
 469   // check against the current stack bottom
 470   __ cmpptr(rsp, rax);
 471   __ jcc(Assembler::above, after_frame_check);
 472 
 473   // Restore sender's sp as SP. This is necessary if the sender's
 474   // frame is an extended compiled frame (see gen_c2i_adapter())
 475   // and safer anyway in case of JSR292 adaptations.
 476 
 477   __ pop(rax); // return address must be moved if SP is changed
 478   __ mov(rsp, r13);
 479   __ push(rax);
 480 
 481   // Note: the restored frame is not necessarily interpreted.
 482   // Use the shared runtime version of the StackOverflowError.
 483   assert(StubRoutines::throw_StackOverflowError_entry() != NULL, "stub not yet generated");
 484   __ jump(ExternalAddress(StubRoutines::throw_StackOverflowError_entry()));
 485 
 486   // all done with frame size check
 487   __ bind(after_frame_check);
 488 }
 489 
 490 // Allocate monitor and lock method (asm interpreter)
 491 //
 492 // Args:
 493 //      rbx: Method*
 494 //      r14: locals
 495 //
 496 // Kills:
 497 //      rax
 498 //      c_rarg0, c_rarg1, c_rarg2, c_rarg3, ...(param regs)
 499 //      rscratch1, rscratch2 (scratch regs)
 500 void InterpreterGenerator::lock_method(void) {
 501   // synchronize method
 502   const Address access_flags(rbx, Method::access_flags_offset());
 503   const Address monitor_block_top(
 504         rbp,
 505         frame::interpreter_frame_monitor_block_top_offset * wordSize);
 506   const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
 507 
 508 #ifdef ASSERT
 509   {
 510     Label L;
 511     __ movl(rax, access_flags);
 512     __ testl(rax, JVM_ACC_SYNCHRONIZED);
 513     __ jcc(Assembler::notZero, L);
 514     __ stop("method doesn't need synchronization");
 515     __ bind(L);
 516   }
 517 #endif // ASSERT
 518 
 519   // get synchronization object
 520   {
 521     const int mirror_offset = in_bytes(Klass::java_mirror_offset());
 522     Label done;
 523     __ movl(rax, access_flags);
 524     __ testl(rax, JVM_ACC_STATIC);
 525     // get receiver (assume this is frequent case)
 526     __ movptr(rax, Address(r14, Interpreter::local_offset_in_bytes(0)));
 527     __ jcc(Assembler::zero, done);
 528     __ movptr(rax, Address(rbx, Method::const_offset()));
 529     __ movptr(rax, Address(rax, ConstMethod::constants_offset()));
 530     __ movptr(rax, Address(rax,
 531                            ConstantPool::pool_holder_offset_in_bytes()));
 532     __ movptr(rax, Address(rax, mirror_offset));
 533 
 534 #ifdef ASSERT
 535     {
 536       Label L;
 537       __ testptr(rax, rax);
 538       __ jcc(Assembler::notZero, L);
 539       __ stop("synchronization object is NULL");
 540       __ bind(L);
 541     }
 542 #endif // ASSERT
 543 
 544     __ bind(done);
 545   }
 546 
 547   // add space for monitor & lock
 548   __ subptr(rsp, entry_size); // add space for a monitor entry
 549   __ movptr(monitor_block_top, rsp);  // set new monitor block top
 550   // store object
 551   __ movptr(Address(rsp, BasicObjectLock::obj_offset_in_bytes()), rax);
 552   __ movptr(c_rarg1, rsp); // object address
 553   __ lock_object(c_rarg1);
 554 }
 555 
 556 // Generate a fixed interpreter frame. This is identical setup for
 557 // interpreted methods and for native methods hence the shared code.
 558 //
 559 // Args:
 560 //      rax: return address
 561 //      rbx: Method*
 562 //      r14: pointer to locals
 563 //      r13: sender sp
 564 //      rdx: cp cache
 565 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
 566   // initialize fixed part of activation frame
 567   __ push(rax);        // save return address
 568   __ enter();          // save old & set new rbp
 569   __ push(r13);        // set sender sp
 570   __ push((int)NULL_WORD); // leave last_sp as null
 571   __ movptr(r13, Address(rbx, Method::const_offset()));      // get ConstMethod*
 572   __ lea(r13, Address(r13, ConstMethod::codes_offset())); // get codebase
 573   __ push(rbx);        // save Method*
 574   if (ProfileInterpreter) {
 575     Label method_data_continue;
 576     __ movptr(rdx, Address(rbx, in_bytes(Method::method_data_offset())));
 577     __ testptr(rdx, rdx);
 578     __ jcc(Assembler::zero, method_data_continue);
 579     __ addptr(rdx, in_bytes(MethodData::data_offset()));
 580     __ bind(method_data_continue);
 581     __ push(rdx);      // set the mdp (method data pointer)
 582   } else {
 583     __ push(0);
 584   }
 585 
 586   __ movptr(rdx, Address(rbx, Method::const_offset()));
 587   __ movptr(rdx, Address(rdx, ConstMethod::constants_offset()));
 588   __ movptr(rdx, Address(rdx, ConstantPool::cache_offset_in_bytes()));
 589   __ push(rdx); // set constant pool cache
 590   __ push(r14); // set locals pointer
 591   if (native_call) {
 592     __ push(0); // no bcp
 593   } else {
 594     __ push(r13); // set bcp
 595   }
 596   __ push(0); // reserve word for pointer to expression stack bottom
 597   __ movptr(Address(rsp, 0), rsp); // set expression stack bottom
 598 }
 599 
 600 // End of helpers
 601 
 602 // Various method entries
 603 //------------------------------------------------------------------------------------------------------------------------
 604 //
 605 //
 606 
 607 // Call an accessor method (assuming it is resolved, otherwise drop
 608 // into vanilla (slow path) entry
 609 address InterpreterGenerator::generate_accessor_entry(void) {
 610   // rbx: Method*
 611 
 612   // r13: senderSP must preserver for slow path, set SP to it on fast path
 613 
 614   address entry_point = __ pc();
 615   Label xreturn_path;
 616 
 617   // do fastpath for resolved accessor methods
 618   if (UseFastAccessorMethods) {
 619     // Code: _aload_0, _(i|a)getfield, _(i|a)return or any rewrites
 620     //       thereof; parameter size = 1
 621     // Note: We can only use this code if the getfield has been resolved
 622     //       and if we don't have a null-pointer exception => check for
 623     //       these conditions first and use slow path if necessary.
 624     Label slow_path;
 625     // If we need a safepoint check, generate full interpreter entry.
 626     __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
 627              SafepointSynchronize::_not_synchronized);
 628 
 629     __ jcc(Assembler::notEqual, slow_path);
 630     // rbx: method
 631     __ movptr(rax, Address(rsp, wordSize));
 632 
 633     // check if local 0 != NULL and read field
 634     __ testptr(rax, rax);
 635     __ jcc(Assembler::zero, slow_path);
 636 
 637     // read first instruction word and extract bytecode @ 1 and index @ 2
 638     __ movptr(rdx, Address(rbx, Method::const_offset()));
 639     __ movptr(rdi, Address(rdx, ConstMethod::constants_offset()));
 640     __ movl(rdx, Address(rdx, ConstMethod::codes_offset()));
 641     // Shift codes right to get the index on the right.
 642     // The bytecode fetched looks like <index><0xb4><0x2a>
 643     __ shrl(rdx, 2 * BitsPerByte);
 644     __ shll(rdx, exact_log2(in_words(ConstantPoolCacheEntry::size())));
 645     __ movptr(rdi, Address(rdi, ConstantPool::cache_offset_in_bytes()));
 646 
 647     // rax: local 0
 648     // rbx: method
 649     // rdx: constant pool cache index
 650     // rdi: constant pool cache
 651 
 652     // check if getfield has been resolved and read constant pool cache entry
 653     // check the validity of the cache entry by testing whether _indices field
 654     // contains Bytecode::_getfield in b1 byte.
 655     assert(in_words(ConstantPoolCacheEntry::size()) == 4,
 656            "adjust shift below");
 657     __ movl(rcx,
 658             Address(rdi,
 659                     rdx,
 660                     Address::times_8,
 661                     ConstantPoolCache::base_offset() +
 662                     ConstantPoolCacheEntry::indices_offset()));
 663     __ shrl(rcx, 2 * BitsPerByte);
 664     __ andl(rcx, 0xFF);
 665     __ cmpl(rcx, Bytecodes::_getfield);
 666     __ jcc(Assembler::notEqual, slow_path);
 667 
 668     // Note: constant pool entry is not valid before bytecode is resolved
 669     __ movptr(rcx,
 670               Address(rdi,
 671                       rdx,
 672                       Address::times_8,
 673                       ConstantPoolCache::base_offset() +
 674                       ConstantPoolCacheEntry::f2_offset()));
 675     // edx: flags
 676     __ movl(rdx,
 677             Address(rdi,
 678                     rdx,
 679                     Address::times_8,
 680                     ConstantPoolCache::base_offset() +
 681                     ConstantPoolCacheEntry::flags_offset()));
 682 
 683     Label notObj, notInt, notByte, notBool, notShort;
 684     const Address field_address(rax, rcx, Address::times_1);
 685 
 686     // Need to differentiate between igetfield, agetfield, bgetfield etc.
 687     // because they are different sizes.
 688     // Use the type from the constant pool cache
 689     __ shrl(rdx, ConstantPoolCacheEntry::tos_state_shift);
 690     // Make sure we don't need to mask edx after the above shift
 691     ConstantPoolCacheEntry::verify_tos_state_shift();
 692 
 693     __ cmpl(rdx, atos);
 694     __ jcc(Assembler::notEqual, notObj);
 695     // atos
 696     __ load_heap_oop(rax, field_address);
 697     __ jmp(xreturn_path);
 698 
 699     __ bind(notObj);
 700     __ cmpl(rdx, itos);
 701     __ jcc(Assembler::notEqual, notInt);
 702     // itos
 703     __ movl(rax, field_address);
 704     __ jmp(xreturn_path);
 705 
 706     __ bind(notInt);
 707     __ cmpl(rdx, btos);
 708     __ jcc(Assembler::notEqual, notByte);
 709     // btos
 710     __ load_signed_byte(rax, field_address);
 711     __ jmp(xreturn_path);
 712 
 713     __ bind(notByte);
 714     __ cmpl(rdx, ztos);
 715     __ jcc(Assembler::notEqual, notBool);
 716     // ztos
 717     __ load_signed_byte(rax, field_address);
 718     __ jmp(xreturn_path);
 719 
 720     __ bind(notBool);
 721     __ cmpl(rdx, stos);
 722     __ jcc(Assembler::notEqual, notShort);
 723     // stos
 724     __ load_signed_short(rax, field_address);
 725     __ jmp(xreturn_path);
 726 
 727     __ bind(notShort);
 728 #ifdef ASSERT
 729     Label okay;
 730     __ cmpl(rdx, ctos);
 731     __ jcc(Assembler::equal, okay);
 732     __ stop("what type is this?");
 733     __ bind(okay);
 734 #endif
 735     // ctos
 736     __ load_unsigned_short(rax, field_address);
 737 
 738     __ bind(xreturn_path);
 739 
 740     // _ireturn/_areturn
 741     __ pop(rdi);
 742     __ mov(rsp, r13);
 743     __ jmp(rdi);
 744     __ ret(0);
 745 
 746     // generate a vanilla interpreter entry as the slow path
 747     __ bind(slow_path);
 748     (void) generate_normal_entry(false);
 749   } else {
 750     (void) generate_normal_entry(false);
 751   }
 752 
 753   return entry_point;
 754 }
 755 
 756 // Method entry for java.lang.ref.Reference.get.
 757 address InterpreterGenerator::generate_Reference_get_entry(void) {
 758 #if INCLUDE_ALL_GCS
 759   // Code: _aload_0, _getfield, _areturn
 760   // parameter size = 1
 761   //
 762   // The code that gets generated by this routine is split into 2 parts:
 763   //    1. The "intrinsified" code for G1 (or any SATB based GC),
 764   //    2. The slow path - which is an expansion of the regular method entry.
 765   //
 766   // Notes:-
 767   // * In the G1 code we do not check whether we need to block for
 768   //   a safepoint. If G1 is enabled then we must execute the specialized
 769   //   code for Reference.get (except when the Reference object is null)
 770   //   so that we can log the value in the referent field with an SATB
 771   //   update buffer.
 772   //   If the code for the getfield template is modified so that the
 773   //   G1 pre-barrier code is executed when the current method is
 774   //   Reference.get() then going through the normal method entry
 775   //   will be fine.
 776   // * The G1 code can, however, check the receiver object (the instance
 777   //   of java.lang.Reference) and jump to the slow path if null. If the
 778   //   Reference object is null then we obviously cannot fetch the referent
 779   //   and so we don't need to call the G1 pre-barrier. Thus we can use the
 780   //   regular method entry code to generate the NPE.
 781   //
 782   // This code is based on generate_accessor_enty.
 783   //
 784   // rbx: Method*
 785 
 786   // r13: senderSP must preserve for slow path, set SP to it on fast path
 787 
 788   address entry = __ pc();
 789 
 790   const int referent_offset = java_lang_ref_Reference::referent_offset;
 791   guarantee(referent_offset > 0, "referent offset not initialized");
 792 
 793   if (UseG1GC || UseShenandoahGC) {
 794     Label slow_path;
 795     // rbx: method
 796 
 797     // Check if local 0 != NULL
 798     // If the receiver is null then it is OK to jump to the slow path.
 799     __ movptr(rax, Address(rsp, wordSize));
 800 
 801     __ testptr(rax, rax);
 802     __ jcc(Assembler::zero, slow_path);
 803 
 804     // rax: local 0
 805     // rbx: method (but can be used as scratch now)
 806     // rdx: scratch
 807     // rdi: scratch
 808 
 809     // Generate the G1 pre-barrier code to log the value of
 810     // the referent field in an SATB buffer.
 811 
 812     // Load the value of the referent field.
 813     const Address field_address(rax, referent_offset);
 814     __ load_heap_oop(rax, field_address);
 815 
 816     // Generate the G1 pre-barrier code to log the value of
 817     // the referent field in an SATB buffer.
 818     if (!UseShenandoahGC || ShenandoahSATBBarrier) {
 819       if (UseShenandoahGC) __ push_IU_state();
 820     __ g1_write_barrier_pre(noreg /* obj */,
 821                             rax /* pre_val */,
 822                             r15_thread /* thread */,
 823                             rbx /* tmp */,
 824                             true /* tosca_live */,
 825                             true /* expand_call */);
 826       if (UseShenandoahGC) __ pop_IU_state();
 827     }
 828 
 829     // _areturn
 830     __ pop(rdi);                // get return address
 831     __ mov(rsp, r13);           // set sp to sender sp
 832     __ jmp(rdi);
 833     __ ret(0);
 834 
 835     // generate a vanilla interpreter entry as the slow path
 836     __ bind(slow_path);
 837     (void) generate_normal_entry(false);
 838 
 839     return entry;
 840   }
 841 #endif // INCLUDE_ALL_GCS
 842 
 843   // If G1 is not enabled then attempt to go through the accessor entry point
 844   // Reference.get is an accessor
 845   return generate_accessor_entry();
 846 }
 847 
 848 /**
 849  * Method entry for static native methods:
 850  *   int java.util.zip.CRC32.update(int crc, int b)
 851  */
 852 address InterpreterGenerator::generate_CRC32_update_entry() {
 853   if (UseCRC32Intrinsics) {
 854     address entry = __ pc();
 855 
 856     // rbx,: Method*
 857     // r13: senderSP must preserved for slow path, set SP to it on fast path
 858     // c_rarg0: scratch (rdi on non-Win64, rcx on Win64)
 859     // c_rarg1: scratch (rsi on non-Win64, rdx on Win64)
 860 
 861     Label slow_path;
 862     // If we need a safepoint check, generate full interpreter entry.
 863     ExternalAddress state(SafepointSynchronize::address_of_state());
 864     __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
 865              SafepointSynchronize::_not_synchronized);
 866     __ jcc(Assembler::notEqual, slow_path);
 867 
 868     // We don't generate local frame and don't align stack because
 869     // we call stub code and there is no safepoint on this path.
 870 
 871     // Load parameters
 872     const Register crc = rax;  // crc
 873     const Register val = c_rarg0;  // source java byte value
 874     const Register tbl = c_rarg1;  // scratch
 875 
 876     // Arguments are reversed on java expression stack
 877     __ movl(val, Address(rsp,   wordSize)); // byte value
 878     __ movl(crc, Address(rsp, 2*wordSize)); // Initial CRC
 879 
 880     __ lea(tbl, ExternalAddress(StubRoutines::crc_table_addr()));
 881     __ notl(crc); // ~crc
 882     __ update_byte_crc32(crc, val, tbl);
 883     __ notl(crc); // ~crc
 884     // result in rax
 885 
 886     // _areturn
 887     __ pop(rdi);                // get return address
 888     __ mov(rsp, r13);           // set sp to sender sp
 889     __ jmp(rdi);
 890 
 891     // generate a vanilla native entry as the slow path
 892     __ bind(slow_path);
 893 
 894     (void) generate_native_entry(false);
 895 
 896     return entry;
 897   }
 898   return generate_native_entry(false);
 899 }
 900 
 901 /**
 902  * Method entry for static native methods:
 903  *   int java.util.zip.CRC32.updateBytes(int crc, byte[] b, int off, int len)
 904  *   int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
 905  */
 906 address InterpreterGenerator::generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
 907   if (UseCRC32Intrinsics) {
 908     address entry = __ pc();
 909 
 910     // rbx,: Method*
 911     // r13: senderSP must preserved for slow path, set SP to it on fast path
 912 
 913     Label slow_path;
 914     // If we need a safepoint check, generate full interpreter entry.
 915     ExternalAddress state(SafepointSynchronize::address_of_state());
 916     __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
 917              SafepointSynchronize::_not_synchronized);
 918     __ jcc(Assembler::notEqual, slow_path);
 919 
 920     // We don't generate local frame and don't align stack because
 921     // we call stub code and there is no safepoint on this path.
 922 
 923     // Load parameters
 924     const Register crc = c_rarg0;  // crc
 925     const Register buf = c_rarg1;  // source java byte array address
 926     const Register len = c_rarg2;  // length
 927     const Register off = len;      // offset (never overlaps with 'len')
 928 
 929     // Arguments are reversed on java expression stack
 930     // Calculate address of start element
 931     if (kind == Interpreter::java_util_zip_CRC32_updateByteBuffer) {
 932       __ movptr(buf, Address(rsp, 3*wordSize)); // long buf
 933       __ movl2ptr(off, Address(rsp, 2*wordSize)); // offset
 934       __ addq(buf, off); // + offset
 935       __ movl(crc,   Address(rsp, 5*wordSize)); // Initial CRC
 936     } else {
 937       __ movptr(buf, Address(rsp, 3*wordSize)); // byte[] array
 938       __ addptr(buf, arrayOopDesc::base_offset_in_bytes(T_BYTE)); // + header size
 939       __ movl2ptr(off, Address(rsp, 2*wordSize)); // offset
 940       __ addq(buf, off); // + offset
 941       __ movl(crc,   Address(rsp, 4*wordSize)); // Initial CRC
 942     }
 943     // Can now load 'len' since we're finished with 'off'
 944     __ movl(len, Address(rsp, wordSize)); // Length
 945 
 946     __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, StubRoutines::updateBytesCRC32()), crc, buf, len);
 947     // result in rax
 948 
 949     // _areturn
 950     __ pop(rdi);                // get return address
 951     __ mov(rsp, r13);           // set sp to sender sp
 952     __ jmp(rdi);
 953 
 954     // generate a vanilla native entry as the slow path
 955     __ bind(slow_path);
 956 
 957     (void) generate_native_entry(false);
 958 
 959     return entry;
 960   }
 961   return generate_native_entry(false);
 962 }
 963 
 964 // Interpreter stub for calling a native method. (asm interpreter)
 965 // This sets up a somewhat different looking stack for calling the
 966 // native method than the typical interpreter frame setup.
 967 address InterpreterGenerator::generate_native_entry(bool synchronized) {
 968   // determine code generation flags
 969   bool inc_counter  = UseCompiler || CountCompiledCalls;
 970 
 971   // rbx: Method*
 972   // r13: sender sp
 973 
 974   address entry_point = __ pc();
 975 
 976   const Address constMethod       (rbx, Method::const_offset());
 977   const Address access_flags      (rbx, Method::access_flags_offset());
 978   const Address size_of_parameters(rcx, ConstMethod::
 979                                         size_of_parameters_offset());
 980 
 981 
 982   // get parameter size (always needed)
 983   __ movptr(rcx, constMethod);
 984   __ load_unsigned_short(rcx, size_of_parameters);
 985 
 986   // native calls don't need the stack size check since they have no
 987   // expression stack and the arguments are already on the stack and
 988   // we only add a handful of words to the stack
 989 
 990   // rbx: Method*
 991   // rcx: size of parameters
 992   // r13: sender sp
 993   __ pop(rax);                                       // get return address
 994 
 995   // for natives the size of locals is zero
 996 
 997   // compute beginning of parameters (r14)
 998   __ lea(r14, Address(rsp, rcx, Address::times_8, -wordSize));
 999 
1000   // add 2 zero-initialized slots for native calls
1001   // initialize result_handler slot
1002   __ push((int) NULL_WORD);
1003   // slot for oop temp
1004   // (static native method holder mirror/jni oop result)
1005   __ push((int) NULL_WORD);
1006 
1007   // initialize fixed part of activation frame
1008   generate_fixed_frame(true);
1009 
1010   // make sure method is native & not abstract
1011 #ifdef ASSERT
1012   __ movl(rax, access_flags);
1013   {
1014     Label L;
1015     __ testl(rax, JVM_ACC_NATIVE);
1016     __ jcc(Assembler::notZero, L);
1017     __ stop("tried to execute non-native method as native");
1018     __ bind(L);
1019   }
1020   {
1021     Label L;
1022     __ testl(rax, JVM_ACC_ABSTRACT);
1023     __ jcc(Assembler::zero, L);
1024     __ stop("tried to execute abstract method in interpreter");
1025     __ bind(L);
1026   }
1027 #endif
1028 
1029   // Since at this point in the method invocation the exception handler
1030   // would try to exit the monitor of synchronized methods which hasn't
1031   // been entered yet, we set the thread local variable
1032   // _do_not_unlock_if_synchronized to true. The remove_activation will
1033   // check this flag.
1034 
1035   const Address do_not_unlock_if_synchronized(r15_thread,
1036         in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1037   __ movbool(do_not_unlock_if_synchronized, true);
1038 
1039   // increment invocation count & check for overflow
1040   Label invocation_counter_overflow;
1041   if (inc_counter) {
1042     generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
1043   }
1044 
1045   Label continue_after_compile;
1046   __ bind(continue_after_compile);
1047 
1048   bang_stack_shadow_pages(true);
1049 
1050   // reset the _do_not_unlock_if_synchronized flag
1051   __ movbool(do_not_unlock_if_synchronized, false);
1052 
1053   // check for synchronized methods
1054   // Must happen AFTER invocation_counter check and stack overflow check,
1055   // so method is not locked if overflows.
1056   if (synchronized) {
1057     lock_method();
1058   } else {
1059     // no synchronization necessary
1060 #ifdef ASSERT
1061     {
1062       Label L;
1063       __ movl(rax, access_flags);
1064       __ testl(rax, JVM_ACC_SYNCHRONIZED);
1065       __ jcc(Assembler::zero, L);
1066       __ stop("method needs synchronization");
1067       __ bind(L);
1068     }
1069 #endif
1070   }
1071 
1072   // start execution
1073 #ifdef ASSERT
1074   {
1075     Label L;
1076     const Address monitor_block_top(rbp,
1077                  frame::interpreter_frame_monitor_block_top_offset * wordSize);
1078     __ movptr(rax, monitor_block_top);
1079     __ cmpptr(rax, rsp);
1080     __ jcc(Assembler::equal, L);
1081     __ stop("broken stack frame setup in interpreter");
1082     __ bind(L);
1083   }
1084 #endif
1085 
1086   // jvmti support
1087   __ notify_method_entry();
1088 
1089   // work registers
1090   const Register method = rbx;
1091   const Register t      = r11;
1092 
1093   // allocate space for parameters
1094   __ get_method(method);
1095   __ movptr(t, Address(method, Method::const_offset()));
1096   __ load_unsigned_short(t, Address(t, ConstMethod::size_of_parameters_offset()));
1097   __ shll(t, Interpreter::logStackElementSize);
1098 
1099   __ subptr(rsp, t);
1100   __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1101   __ andptr(rsp, -16); // must be 16 byte boundary (see amd64 ABI)
1102 
1103   // get signature handler
1104   {
1105     Label L;
1106     __ movptr(t, Address(method, Method::signature_handler_offset()));
1107     __ testptr(t, t);
1108     __ jcc(Assembler::notZero, L);
1109     __ call_VM(noreg,
1110                CAST_FROM_FN_PTR(address,
1111                                 InterpreterRuntime::prepare_native_call),
1112                method);
1113     __ get_method(method);
1114     __ movptr(t, Address(method, Method::signature_handler_offset()));
1115     __ bind(L);
1116   }
1117 
1118   // call signature handler
1119   assert(InterpreterRuntime::SignatureHandlerGenerator::from() == r14,
1120          "adjust this code");
1121   assert(InterpreterRuntime::SignatureHandlerGenerator::to() == rsp,
1122          "adjust this code");
1123   assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == rscratch1,
1124           "adjust this code");
1125 
1126   // The generated handlers do not touch RBX (the method oop).
1127   // However, large signatures cannot be cached and are generated
1128   // each time here.  The slow-path generator can do a GC on return,
1129   // so we must reload it after the call.
1130   __ call(t);
1131   __ get_method(method);        // slow path can do a GC, reload RBX
1132 
1133 
1134   // result handler is in rax
1135   // set result handler
1136   __ movptr(Address(rbp,
1137                     (frame::interpreter_frame_result_handler_offset) * wordSize),
1138             rax);
1139 
1140   // pass mirror handle if static call
1141   {
1142     Label L;
1143     const int mirror_offset = in_bytes(Klass::java_mirror_offset());
1144     __ movl(t, Address(method, Method::access_flags_offset()));
1145     __ testl(t, JVM_ACC_STATIC);
1146     __ jcc(Assembler::zero, L);
1147     // get mirror
1148     __ movptr(t, Address(method, Method::const_offset()));
1149     __ movptr(t, Address(t, ConstMethod::constants_offset()));
1150     __ movptr(t, Address(t, ConstantPool::pool_holder_offset_in_bytes()));
1151     __ movptr(t, Address(t, mirror_offset));
1152     // copy mirror into activation frame
1153     __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize),
1154             t);
1155     // pass handle to mirror
1156     __ lea(c_rarg1,
1157            Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize));
1158     __ bind(L);
1159   }
1160 
1161   // get native function entry point
1162   {
1163     Label L;
1164     __ movptr(rax, Address(method, Method::native_function_offset()));
1165     ExternalAddress unsatisfied(SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
1166     __ movptr(rscratch2, unsatisfied.addr());
1167     __ cmpptr(rax, rscratch2);
1168     __ jcc(Assembler::notEqual, L);
1169     __ call_VM(noreg,
1170                CAST_FROM_FN_PTR(address,
1171                                 InterpreterRuntime::prepare_native_call),
1172                method);
1173     __ get_method(method);
1174     __ movptr(rax, Address(method, Method::native_function_offset()));
1175     __ bind(L);
1176   }
1177 
1178   // pass JNIEnv
1179   __ lea(c_rarg0, Address(r15_thread, JavaThread::jni_environment_offset()));
1180 
1181   // It is enough that the pc() points into the right code
1182   // segment. It does not have to be the correct return pc.
1183   __ set_last_Java_frame(rsp, rbp, (address) __ pc());
1184 
1185   // change thread state
1186 #ifdef ASSERT
1187   {
1188     Label L;
1189     __ movl(t, Address(r15_thread, JavaThread::thread_state_offset()));
1190     __ cmpl(t, _thread_in_Java);
1191     __ jcc(Assembler::equal, L);
1192     __ stop("Wrong thread state in native stub");
1193     __ bind(L);
1194   }
1195 #endif
1196 
1197   // Change state to native
1198 
1199   __ movl(Address(r15_thread, JavaThread::thread_state_offset()),
1200           _thread_in_native);
1201 
1202   // Call the native method.
1203   __ call(rax);
1204   // result potentially in rax or xmm0
1205 
1206   // Verify or restore cpu control state after JNI call
1207   __ restore_cpu_control_state_after_jni();
1208 
1209   // NOTE: The order of these pushes is known to frame::interpreter_frame_result
1210   // in order to extract the result of a method call. If the order of these
1211   // pushes change or anything else is added to the stack then the code in
1212   // interpreter_frame_result must also change.
1213 
1214   __ push(dtos);
1215   __ push(ltos);
1216 
1217   // change thread state
1218   __ movl(Address(r15_thread, JavaThread::thread_state_offset()),
1219           _thread_in_native_trans);
1220 
1221   if (os::is_MP()) {
1222     if (UseMembar) {
1223       // Force this write out before the read below
1224       __ membar(Assembler::Membar_mask_bits(
1225            Assembler::LoadLoad | Assembler::LoadStore |
1226            Assembler::StoreLoad | Assembler::StoreStore));
1227     } else {
1228       // Write serialization page so VM thread can do a pseudo remote membar.
1229       // We use the current thread pointer to calculate a thread specific
1230       // offset to write to within the page. This minimizes bus traffic
1231       // due to cache line collision.
1232       __ serialize_memory(r15_thread, rscratch2);
1233     }
1234   }
1235 
1236   // check for safepoint operation in progress and/or pending suspend requests
1237   {
1238     Label Continue;
1239     __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
1240              SafepointSynchronize::_not_synchronized);
1241 
1242     Label L;
1243     __ jcc(Assembler::notEqual, L);
1244     __ cmpl(Address(r15_thread, JavaThread::suspend_flags_offset()), 0);
1245     __ jcc(Assembler::equal, Continue);
1246     __ bind(L);
1247 
1248     // Don't use call_VM as it will see a possible pending exception
1249     // and forward it and never return here preventing us from
1250     // clearing _last_native_pc down below.  Also can't use
1251     // call_VM_leaf either as it will check to see if r13 & r14 are
1252     // preserved and correspond to the bcp/locals pointers. So we do a
1253     // runtime call by hand.
1254     //
1255     __ mov(c_rarg0, r15_thread);
1256     __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1257     __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1258     __ andptr(rsp, -16); // align stack as required by ABI
1259     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans)));
1260     __ mov(rsp, r12); // restore sp
1261     __ reinit_heapbase();
1262     __ bind(Continue);
1263   }
1264 
1265   // change thread state
1266   __ movl(Address(r15_thread, JavaThread::thread_state_offset()), _thread_in_Java);
1267 
1268   // reset_last_Java_frame
1269   __ reset_last_Java_frame(r15_thread, true);
1270 
1271   // reset handle block
1272   __ movptr(t, Address(r15_thread, JavaThread::active_handles_offset()));
1273   __ movl(Address(t, JNIHandleBlock::top_offset_in_bytes()), (int32_t)NULL_WORD);
1274 
1275   // If result is an oop unbox and store it in frame where gc will see it
1276   // and result handler will pick it up
1277 
1278   {
1279     Label no_oop;
1280     __ lea(t, ExternalAddress(AbstractInterpreter::result_handler(T_OBJECT)));
1281     __ cmpptr(t, Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize));
1282     __ jcc(Assembler::notEqual, no_oop);
1283     // retrieve result
1284     __ pop(ltos);
1285     // Unbox oop result, e.g. JNIHandles::resolve value.
1286     __ resolve_jobject(rax /* value */,
1287                        r15_thread /* thread */,
1288                        t /* tmp */);
1289     __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize), rax);
1290     // keep stack depth as expected by pushing oop which will eventually be discarde
1291     __ push(ltos);
1292     __ bind(no_oop);
1293   }
1294 
1295 
1296   {
1297     Label no_reguard;
1298     __ cmpl(Address(r15_thread, JavaThread::stack_guard_state_offset()),
1299             JavaThread::stack_guard_yellow_disabled);
1300     __ jcc(Assembler::notEqual, no_reguard);
1301 
1302     __ pusha(); // XXX only save smashed registers
1303     __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1304     __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1305     __ andptr(rsp, -16); // align stack as required by ABI
1306     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
1307     __ mov(rsp, r12); // restore sp
1308     __ popa(); // XXX only restore smashed registers
1309     __ reinit_heapbase();
1310 
1311     __ bind(no_reguard);
1312   }
1313 
1314 
1315   // The method register is junk from after the thread_in_native transition
1316   // until here.  Also can't call_VM until the bcp has been
1317   // restored.  Need bcp for throwing exception below so get it now.
1318   __ get_method(method);
1319 
1320   // restore r13 to have legal interpreter frame, i.e., bci == 0 <=>
1321   // r13 == code_base()
1322   __ movptr(r13, Address(method, Method::const_offset()));   // get ConstMethod*
1323   __ lea(r13, Address(r13, ConstMethod::codes_offset()));    // get codebase
1324   // handle exceptions (exception handling will handle unlocking!)
1325   {
1326     Label L;
1327     __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t) NULL_WORD);
1328     __ jcc(Assembler::zero, L);
1329     // Note: At some point we may want to unify this with the code
1330     // used in call_VM_base(); i.e., we should use the
1331     // StubRoutines::forward_exception code. For now this doesn't work
1332     // here because the rsp is not correctly set at this point.
1333     __ MacroAssembler::call_VM(noreg,
1334                                CAST_FROM_FN_PTR(address,
1335                                InterpreterRuntime::throw_pending_exception));
1336     __ should_not_reach_here();
1337     __ bind(L);
1338   }
1339 
1340   // do unlocking if necessary
1341   {
1342     Label L;
1343     __ movl(t, Address(method, Method::access_flags_offset()));
1344     __ testl(t, JVM_ACC_SYNCHRONIZED);
1345     __ jcc(Assembler::zero, L);
1346     // the code below should be shared with interpreter macro
1347     // assembler implementation
1348     {
1349       Label unlock;
1350       // BasicObjectLock will be first in list, since this is a
1351       // synchronized method. However, need to check that the object
1352       // has not been unlocked by an explicit monitorexit bytecode.
1353       const Address monitor(rbp,
1354                             (intptr_t)(frame::interpreter_frame_initial_sp_offset *
1355                                        wordSize - sizeof(BasicObjectLock)));
1356 
1357       // monitor expect in c_rarg1 for slow unlock path
1358       __ lea(c_rarg1, monitor); // address of first monitor
1359 
1360       __ movptr(t, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
1361       __ testptr(t, t);
1362       __ jcc(Assembler::notZero, unlock);
1363 
1364       // Entry already unlocked, need to throw exception
1365       __ MacroAssembler::call_VM(noreg,
1366                                  CAST_FROM_FN_PTR(address,
1367                    InterpreterRuntime::throw_illegal_monitor_state_exception));
1368       __ should_not_reach_here();
1369 
1370       __ bind(unlock);
1371       __ unlock_object(c_rarg1);
1372     }
1373     __ bind(L);
1374   }
1375 
1376   // jvmti support
1377   // Note: This must happen _after_ handling/throwing any exceptions since
1378   //       the exception handler code notifies the runtime of method exits
1379   //       too. If this happens before, method entry/exit notifications are
1380   //       not properly paired (was bug - gri 11/22/99).
1381   __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI);
1382 
1383   // restore potential result in edx:eax, call result handler to
1384   // restore potential result in ST0 & handle result
1385 
1386   __ pop(ltos);
1387   __ pop(dtos);
1388 
1389   __ movptr(t, Address(rbp,
1390                        (frame::interpreter_frame_result_handler_offset) * wordSize));
1391   __ call(t);
1392 
1393   // remove activation
1394   __ movptr(t, Address(rbp,
1395                        frame::interpreter_frame_sender_sp_offset *
1396                        wordSize)); // get sender sp
1397   __ leave();                                // remove frame anchor
1398   __ pop(rdi);                               // get return address
1399   __ mov(rsp, t);                            // set sp to sender sp
1400   __ jmp(rdi);
1401 
1402   if (inc_counter) {
1403     // Handle overflow of counter and compile method
1404     __ bind(invocation_counter_overflow);
1405     generate_counter_overflow(&continue_after_compile);
1406   }
1407 
1408   return entry_point;
1409 }
1410 
1411 //
1412 // Generic interpreted method entry to (asm) interpreter
1413 //
1414 address InterpreterGenerator::generate_normal_entry(bool synchronized) {
1415   // determine code generation flags
1416   bool inc_counter  = UseCompiler || CountCompiledCalls;
1417 
1418   // ebx: Method*
1419   // r13: sender sp
1420   address entry_point = __ pc();
1421 
1422   const Address constMethod(rbx, Method::const_offset());
1423   const Address access_flags(rbx, Method::access_flags_offset());
1424   const Address size_of_parameters(rdx,
1425                                    ConstMethod::size_of_parameters_offset());
1426   const Address size_of_locals(rdx, ConstMethod::size_of_locals_offset());
1427 
1428 
1429   // get parameter size (always needed)
1430   __ movptr(rdx, constMethod);
1431   __ load_unsigned_short(rcx, size_of_parameters);
1432 
1433   // rbx: Method*
1434   // rcx: size of parameters
1435   // r13: sender_sp (could differ from sp+wordSize if we were called via c2i )
1436 
1437   __ load_unsigned_short(rdx, size_of_locals); // get size of locals in words
1438   __ subl(rdx, rcx); // rdx = no. of additional locals
1439 
1440   // YYY
1441 //   __ incrementl(rdx);
1442 //   __ andl(rdx, -2);
1443 
1444   // see if we've got enough room on the stack for locals plus overhead.
1445   generate_stack_overflow_check();
1446 
1447   // get return address
1448   __ pop(rax);
1449 
1450   // compute beginning of parameters (r14)
1451   __ lea(r14, Address(rsp, rcx, Address::times_8, -wordSize));
1452 
1453   // rdx - # of additional locals
1454   // allocate space for locals
1455   // explicitly initialize locals
1456   {
1457     Label exit, loop;
1458     __ testl(rdx, rdx);
1459     __ jcc(Assembler::lessEqual, exit); // do nothing if rdx <= 0
1460     __ bind(loop);
1461     __ push((int) NULL_WORD); // initialize local variables
1462     __ decrementl(rdx); // until everything initialized
1463     __ jcc(Assembler::greater, loop);
1464     __ bind(exit);
1465   }
1466 
1467   // initialize fixed part of activation frame
1468   generate_fixed_frame(false);
1469 
1470   // make sure method is not native & not abstract
1471 #ifdef ASSERT
1472   __ movl(rax, access_flags);
1473   {
1474     Label L;
1475     __ testl(rax, JVM_ACC_NATIVE);
1476     __ jcc(Assembler::zero, L);
1477     __ stop("tried to execute native method as non-native");
1478     __ bind(L);
1479   }
1480   {
1481     Label L;
1482     __ testl(rax, JVM_ACC_ABSTRACT);
1483     __ jcc(Assembler::zero, L);
1484     __ stop("tried to execute abstract method in interpreter");
1485     __ bind(L);
1486   }
1487 #endif
1488 
1489   // Since at this point in the method invocation the exception
1490   // handler would try to exit the monitor of synchronized methods
1491   // which hasn't been entered yet, we set the thread local variable
1492   // _do_not_unlock_if_synchronized to true. The remove_activation
1493   // will check this flag.
1494 
1495   const Address do_not_unlock_if_synchronized(r15_thread,
1496         in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1497   __ movbool(do_not_unlock_if_synchronized, true);
1498 
1499   __ profile_parameters_type(rax, rcx, rdx);
1500   // increment invocation count & check for overflow
1501   Label invocation_counter_overflow;
1502   Label profile_method;
1503   Label profile_method_continue;
1504   if (inc_counter) {
1505     generate_counter_incr(&invocation_counter_overflow,
1506                           &profile_method,
1507                           &profile_method_continue);
1508     if (ProfileInterpreter) {
1509       __ bind(profile_method_continue);
1510     }
1511   }
1512 
1513   Label continue_after_compile;
1514   __ bind(continue_after_compile);
1515 
1516   // check for synchronized interpreted methods
1517   bang_stack_shadow_pages(false);
1518 
1519   // reset the _do_not_unlock_if_synchronized flag
1520   __ movbool(do_not_unlock_if_synchronized, false);
1521 
1522   // check for synchronized methods
1523   // Must happen AFTER invocation_counter check and stack overflow check,
1524   // so method is not locked if overflows.
1525   if (synchronized) {
1526     // Allocate monitor and lock method
1527     lock_method();
1528   } else {
1529     // no synchronization necessary
1530 #ifdef ASSERT
1531     {
1532       Label L;
1533       __ movl(rax, access_flags);
1534       __ testl(rax, JVM_ACC_SYNCHRONIZED);
1535       __ jcc(Assembler::zero, L);
1536       __ stop("method needs synchronization");
1537       __ bind(L);
1538     }
1539 #endif
1540   }
1541 
1542   // start execution
1543 #ifdef ASSERT
1544   {
1545     Label L;
1546      const Address monitor_block_top (rbp,
1547                  frame::interpreter_frame_monitor_block_top_offset * wordSize);
1548     __ movptr(rax, monitor_block_top);
1549     __ cmpptr(rax, rsp);
1550     __ jcc(Assembler::equal, L);
1551     __ stop("broken stack frame setup in interpreter");
1552     __ bind(L);
1553   }
1554 #endif
1555 
1556   // jvmti support
1557   __ notify_method_entry();
1558 
1559   __ dispatch_next(vtos);
1560 
1561   // invocation counter overflow
1562   if (inc_counter) {
1563     if (ProfileInterpreter) {
1564       // We have decided to profile this method in the interpreter
1565       __ bind(profile_method);
1566       __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1567       __ set_method_data_pointer_for_bcp();
1568       __ get_method(rbx);
1569       __ jmp(profile_method_continue);
1570     }
1571     // Handle overflow of counter and compile method
1572     __ bind(invocation_counter_overflow);
1573     generate_counter_overflow(&continue_after_compile);
1574   }
1575 
1576   return entry_point;
1577 }
1578 
1579 // Entry points
1580 //
1581 // Here we generate the various kind of entries into the interpreter.
1582 // The two main entry type are generic bytecode methods and native
1583 // call method.  These both come in synchronized and non-synchronized
1584 // versions but the frame layout they create is very similar. The
1585 // other method entry types are really just special purpose entries
1586 // that are really entry and interpretation all in one. These are for
1587 // trivial methods like accessor, empty, or special math methods.
1588 //
1589 // When control flow reaches any of the entry types for the interpreter
1590 // the following holds ->
1591 //
1592 // Arguments:
1593 //
1594 // rbx: Method*
1595 //
1596 // Stack layout immediately at entry
1597 //
1598 // [ return address     ] <--- rsp
1599 // [ parameter n        ]
1600 //   ...
1601 // [ parameter 1        ]
1602 // [ expression stack   ] (caller's java expression stack)
1603 
1604 // Assuming that we don't go to one of the trivial specialized entries
1605 // the stack will look like below when we are ready to execute the
1606 // first bytecode (or call the native routine). The register usage
1607 // will be as the template based interpreter expects (see
1608 // interpreter_amd64.hpp).
1609 //
1610 // local variables follow incoming parameters immediately; i.e.
1611 // the return address is moved to the end of the locals).
1612 //
1613 // [ monitor entry      ] <--- rsp
1614 //   ...
1615 // [ monitor entry      ]
1616 // [ expr. stack bottom ]
1617 // [ saved r13          ]
1618 // [ current r14        ]
1619 // [ Method*            ]
1620 // [ saved ebp          ] <--- rbp
1621 // [ return address     ]
1622 // [ local variable m   ]
1623 //   ...
1624 // [ local variable 1   ]
1625 // [ parameter n        ]
1626 //   ...
1627 // [ parameter 1        ] <--- r14
1628 
1629 address AbstractInterpreterGenerator::generate_method_entry(
1630                                         AbstractInterpreter::MethodKind kind) {
1631   // determine code generation flags
1632   bool synchronized = false;
1633   address entry_point = NULL;
1634   InterpreterGenerator* ig_this = (InterpreterGenerator*)this;
1635 
1636   switch (kind) {
1637   case Interpreter::zerolocals             :                                                      break;
1638   case Interpreter::zerolocals_synchronized: synchronized = true;                                 break;
1639   case Interpreter::native                 : entry_point = ig_this->generate_native_entry(false); break;
1640   case Interpreter::native_synchronized    : entry_point = ig_this->generate_native_entry(true);  break;
1641   case Interpreter::empty                  : entry_point = ig_this->generate_empty_entry();       break;
1642   case Interpreter::accessor               : entry_point = ig_this->generate_accessor_entry();    break;
1643   case Interpreter::abstract               : entry_point = ig_this->generate_abstract_entry();    break;
1644 
1645   case Interpreter::java_lang_math_sin     : // fall thru
1646   case Interpreter::java_lang_math_cos     : // fall thru
1647   case Interpreter::java_lang_math_tan     : // fall thru
1648   case Interpreter::java_lang_math_abs     : // fall thru
1649   case Interpreter::java_lang_math_log     : // fall thru
1650   case Interpreter::java_lang_math_log10   : // fall thru
1651   case Interpreter::java_lang_math_sqrt    : // fall thru
1652   case Interpreter::java_lang_math_pow     : // fall thru
1653   case Interpreter::java_lang_math_exp     : entry_point = ig_this->generate_math_entry(kind);      break;
1654   case Interpreter::java_lang_ref_reference_get
1655                                            : entry_point = ig_this->generate_Reference_get_entry(); break;
1656   case Interpreter::java_util_zip_CRC32_update
1657                                            : entry_point = ig_this->generate_CRC32_update_entry();  break;
1658   case Interpreter::java_util_zip_CRC32_updateBytes
1659                                            : // fall thru
1660   case Interpreter::java_util_zip_CRC32_updateByteBuffer
1661                                            : entry_point = ig_this->generate_CRC32_updateBytes_entry(kind); break;
1662   default:
1663     fatal(err_msg("unexpected method kind: %d", kind));
1664     break;
1665   }
1666 
1667   if (entry_point) {
1668     return entry_point;
1669   }
1670 
1671   return ig_this->generate_normal_entry(synchronized);
1672 }
1673 
1674 // These should never be compiled since the interpreter will prefer
1675 // the compiled version to the intrinsic version.
1676 bool AbstractInterpreter::can_be_compiled(methodHandle m) {
1677   switch (method_kind(m)) {
1678     case Interpreter::java_lang_math_sin     : // fall thru
1679     case Interpreter::java_lang_math_cos     : // fall thru
1680     case Interpreter::java_lang_math_tan     : // fall thru
1681     case Interpreter::java_lang_math_abs     : // fall thru
1682     case Interpreter::java_lang_math_log     : // fall thru
1683     case Interpreter::java_lang_math_log10   : // fall thru
1684     case Interpreter::java_lang_math_sqrt    : // fall thru
1685     case Interpreter::java_lang_math_pow     : // fall thru
1686     case Interpreter::java_lang_math_exp     :
1687       return false;
1688     default:
1689       return true;
1690   }
1691 }
1692 
1693 // How much stack a method activation needs in words.
1694 int AbstractInterpreter::size_top_interpreter_activation(Method* method) {
1695   const int entry_size = frame::interpreter_frame_monitor_size();
1696 
1697   // total overhead size: entry_size + (saved rbp thru expr stack
1698   // bottom).  be sure to change this if you add/subtract anything
1699   // to/from the overhead area
1700   const int overhead_size =
1701     -(frame::interpreter_frame_initial_sp_offset) + entry_size;
1702 
1703   const int stub_code = frame::entry_frame_after_call_words;
1704   const int method_stack = (method->max_locals() + method->max_stack()) *
1705                            Interpreter::stackElementWords;
1706   return (overhead_size + method_stack + stub_code);
1707 }
1708 
1709 //-----------------------------------------------------------------------------
1710 // Exceptions
1711 
1712 void TemplateInterpreterGenerator::generate_throw_exception() {
1713   // Entry point in previous activation (i.e., if the caller was
1714   // interpreted)
1715   Interpreter::_rethrow_exception_entry = __ pc();
1716   // Restore sp to interpreter_frame_last_sp even though we are going
1717   // to empty the expression stack for the exception processing.
1718   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
1719   // rax: exception
1720   // rdx: return address/pc that threw exception
1721   __ restore_bcp();    // r13 points to call/send
1722   __ restore_locals();
1723   __ reinit_heapbase();  // restore r12 as heapbase.
1724   // Entry point for exceptions thrown within interpreter code
1725   Interpreter::_throw_exception_entry = __ pc();
1726   // expression stack is undefined here
1727   // rax: exception
1728   // r13: exception bcp
1729   __ verify_oop(rax);
1730   __ mov(c_rarg1, rax);
1731 
1732   // expression stack must be empty before entering the VM in case of
1733   // an exception
1734   __ empty_expression_stack();
1735   // find exception handler address and preserve exception oop
1736   __ call_VM(rdx,
1737              CAST_FROM_FN_PTR(address,
1738                           InterpreterRuntime::exception_handler_for_exception),
1739              c_rarg1);
1740   // rax: exception handler entry point
1741   // rdx: preserved exception oop
1742   // r13: bcp for exception handler
1743   __ push_ptr(rdx); // push exception which is now the only value on the stack
1744   __ jmp(rax); // jump to exception handler (may be _remove_activation_entry!)
1745 
1746   // If the exception is not handled in the current frame the frame is
1747   // removed and the exception is rethrown (i.e. exception
1748   // continuation is _rethrow_exception).
1749   //
1750   // Note: At this point the bci is still the bxi for the instruction
1751   // which caused the exception and the expression stack is
1752   // empty. Thus, for any VM calls at this point, GC will find a legal
1753   // oop map (with empty expression stack).
1754 
1755   // In current activation
1756   // tos: exception
1757   // esi: exception bcp
1758 
1759   //
1760   // JVMTI PopFrame support
1761   //
1762 
1763   Interpreter::_remove_activation_preserving_args_entry = __ pc();
1764   __ empty_expression_stack();
1765   // Set the popframe_processing bit in pending_popframe_condition
1766   // indicating that we are currently handling popframe, so that
1767   // call_VMs that may happen later do not trigger new popframe
1768   // handling cycles.
1769   __ movl(rdx, Address(r15_thread, JavaThread::popframe_condition_offset()));
1770   __ orl(rdx, JavaThread::popframe_processing_bit);
1771   __ movl(Address(r15_thread, JavaThread::popframe_condition_offset()), rdx);
1772 
1773   {
1774     // Check to see whether we are returning to a deoptimized frame.
1775     // (The PopFrame call ensures that the caller of the popped frame is
1776     // either interpreted or compiled and deoptimizes it if compiled.)
1777     // In this case, we can't call dispatch_next() after the frame is
1778     // popped, but instead must save the incoming arguments and restore
1779     // them after deoptimization has occurred.
1780     //
1781     // Note that we don't compare the return PC against the
1782     // deoptimization blob's unpack entry because of the presence of
1783     // adapter frames in C2.
1784     Label caller_not_deoptimized;
1785     __ movptr(c_rarg1, Address(rbp, frame::return_addr_offset * wordSize));
1786     __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1787                                InterpreterRuntime::interpreter_contains), c_rarg1);
1788     __ testl(rax, rax);
1789     __ jcc(Assembler::notZero, caller_not_deoptimized);
1790 
1791     // Compute size of arguments for saving when returning to
1792     // deoptimized caller
1793     __ get_method(rax);
1794     __ movptr(rax, Address(rax, Method::const_offset()));
1795     __ load_unsigned_short(rax, Address(rax, in_bytes(ConstMethod::
1796                                                 size_of_parameters_offset())));
1797     __ shll(rax, Interpreter::logStackElementSize);
1798     __ restore_locals(); // XXX do we need this?
1799     __ subptr(r14, rax);
1800     __ addptr(r14, wordSize);
1801     // Save these arguments
1802     __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1803                                            Deoptimization::
1804                                            popframe_preserve_args),
1805                           r15_thread, rax, r14);
1806 
1807     __ remove_activation(vtos, rdx,
1808                          /* throw_monitor_exception */ false,
1809                          /* install_monitor_exception */ false,
1810                          /* notify_jvmdi */ false);
1811 
1812     // Inform deoptimization that it is responsible for restoring
1813     // these arguments
1814     __ movl(Address(r15_thread, JavaThread::popframe_condition_offset()),
1815             JavaThread::popframe_force_deopt_reexecution_bit);
1816 
1817     // Continue in deoptimization handler
1818     __ jmp(rdx);
1819 
1820     __ bind(caller_not_deoptimized);
1821   }
1822 
1823   __ remove_activation(vtos, rdx, /* rdx result (retaddr) is not used */
1824                        /* throw_monitor_exception */ false,
1825                        /* install_monitor_exception */ false,
1826                        /* notify_jvmdi */ false);
1827 
1828   // Finish with popframe handling
1829   // A previous I2C followed by a deoptimization might have moved the
1830   // outgoing arguments further up the stack. PopFrame expects the
1831   // mutations to those outgoing arguments to be preserved and other
1832   // constraints basically require this frame to look exactly as
1833   // though it had previously invoked an interpreted activation with
1834   // no space between the top of the expression stack (current
1835   // last_sp) and the top of stack. Rather than force deopt to
1836   // maintain this kind of invariant all the time we call a small
1837   // fixup routine to move the mutated arguments onto the top of our
1838   // expression stack if necessary.
1839   __ mov(c_rarg1, rsp);
1840   __ movptr(c_rarg2, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1841   // PC must point into interpreter here
1842   __ set_last_Java_frame(noreg, rbp, __ pc());
1843   __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), r15_thread, c_rarg1, c_rarg2);
1844   __ reset_last_Java_frame(r15_thread, true);
1845   // Restore the last_sp and null it out
1846   __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1847   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
1848 
1849   __ restore_bcp();  // XXX do we need this?
1850   __ restore_locals(); // XXX do we need this?
1851   // The method data pointer was incremented already during
1852   // call profiling. We have to restore the mdp for the current bcp.
1853   if (ProfileInterpreter) {
1854     __ set_method_data_pointer_for_bcp();
1855   }
1856 
1857   // Clear the popframe condition flag
1858   __ movl(Address(r15_thread, JavaThread::popframe_condition_offset()),
1859           JavaThread::popframe_inactive);
1860 
1861 #if INCLUDE_JVMTI
1862   if (EnableInvokeDynamic) {
1863     Label L_done;
1864     const Register local0 = r14;
1865 
1866     __ cmpb(Address(r13, 0), Bytecodes::_invokestatic);
1867     __ jcc(Assembler::notEqual, L_done);
1868 
1869     // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call.
1870     // Detect such a case in the InterpreterRuntime function and return the member name argument, or NULL.
1871 
1872     __ get_method(rdx);
1873     __ movptr(rax, Address(local0, 0));
1874     __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), rax, rdx, r13);
1875 
1876     __ testptr(rax, rax);
1877     __ jcc(Assembler::zero, L_done);
1878 
1879     __ movptr(Address(rbx, 0), rax);
1880     __ bind(L_done);
1881   }
1882 #endif // INCLUDE_JVMTI
1883 
1884   __ dispatch_next(vtos);
1885   // end of PopFrame support
1886 
1887   Interpreter::_remove_activation_entry = __ pc();
1888 
1889   // preserve exception over this code sequence
1890   __ pop_ptr(rax);
1891   __ movptr(Address(r15_thread, JavaThread::vm_result_offset()), rax);
1892   // remove the activation (without doing throws on illegalMonitorExceptions)
1893   __ remove_activation(vtos, rdx, false, true, false);
1894   // restore exception
1895   __ get_vm_result(rax, r15_thread);
1896 
1897   // In between activations - previous activation type unknown yet
1898   // compute continuation point - the continuation point expects the
1899   // following registers set up:
1900   //
1901   // rax: exception
1902   // rdx: return address/pc that threw exception
1903   // rsp: expression stack of caller
1904   // rbp: ebp of caller
1905   __ push(rax);                                  // save exception
1906   __ push(rdx);                                  // save return address
1907   __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1908                           SharedRuntime::exception_handler_for_return_address),
1909                         r15_thread, rdx);
1910   __ mov(rbx, rax);                              // save exception handler
1911   __ pop(rdx);                                   // restore return address
1912   __ pop(rax);                                   // restore exception
1913   // Note that an "issuing PC" is actually the next PC after the call
1914   __ jmp(rbx);                                   // jump to exception
1915                                                  // handler of caller
1916 }
1917 
1918 
1919 //
1920 // JVMTI ForceEarlyReturn support
1921 //
1922 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
1923   address entry = __ pc();
1924 
1925   __ restore_bcp();
1926   __ restore_locals();
1927   __ empty_expression_stack();
1928   __ load_earlyret_value(state);
1929 
1930   __ movptr(rdx, Address(r15_thread, JavaThread::jvmti_thread_state_offset()));
1931   Address cond_addr(rdx, JvmtiThreadState::earlyret_state_offset());
1932 
1933   // Clear the earlyret state
1934   __ movl(cond_addr, JvmtiThreadState::earlyret_inactive);
1935 
1936   __ remove_activation(state, rsi,
1937                        false, /* throw_monitor_exception */
1938                        false, /* install_monitor_exception */
1939                        true); /* notify_jvmdi */
1940   __ jmp(rsi);
1941 
1942   return entry;
1943 } // end of ForceEarlyReturn support
1944 
1945 
1946 //-----------------------------------------------------------------------------
1947 // Helper for vtos entry point generation
1948 
1949 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
1950                                                          address& bep,
1951                                                          address& cep,
1952                                                          address& sep,
1953                                                          address& aep,
1954                                                          address& iep,
1955                                                          address& lep,
1956                                                          address& fep,
1957                                                          address& dep,
1958                                                          address& vep) {
1959   assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
1960   Label L;
1961   aep = __ pc();  __ push_ptr();  __ jmp(L);
1962   fep = __ pc();  __ push_f();    __ jmp(L);
1963   dep = __ pc();  __ push_d();    __ jmp(L);
1964   lep = __ pc();  __ push_l();    __ jmp(L);
1965   bep = cep = sep =
1966   iep = __ pc();  __ push_i();
1967   vep = __ pc();
1968   __ bind(L);
1969   generate_and_dispatch(t);
1970 }
1971 
1972 
1973 //-----------------------------------------------------------------------------
1974 // Generation of individual instructions
1975 
1976 // helpers for generate_and_dispatch
1977 
1978 
1979 InterpreterGenerator::InterpreterGenerator(StubQueue* code)
1980   : TemplateInterpreterGenerator(code) {
1981    generate_all(); // down here so it can be "virtual"
1982 }
1983 
1984 //-----------------------------------------------------------------------------
1985 
1986 // Non-product code
1987 #ifndef PRODUCT
1988 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
1989   address entry = __ pc();
1990 
1991   __ push(state);
1992   __ push(c_rarg0);
1993   __ push(c_rarg1);
1994   __ push(c_rarg2);
1995   __ push(c_rarg3);
1996   __ mov(c_rarg2, rax);  // Pass itos
1997 #ifdef _WIN64
1998   __ movflt(xmm3, xmm0); // Pass ftos
1999 #endif
2000   __ call_VM(noreg,
2001              CAST_FROM_FN_PTR(address, SharedRuntime::trace_bytecode),
2002              c_rarg1, c_rarg2, c_rarg3);
2003   __ pop(c_rarg3);
2004   __ pop(c_rarg2);
2005   __ pop(c_rarg1);
2006   __ pop(c_rarg0);
2007   __ pop(state);
2008   __ ret(0);                                   // return from result handler
2009 
2010   return entry;
2011 }
2012 
2013 void TemplateInterpreterGenerator::count_bytecode() {
2014   __ incrementl(ExternalAddress((address) &BytecodeCounter::_counter_value));
2015 }
2016 
2017 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
2018   __ incrementl(ExternalAddress((address) &BytecodeHistogram::_counters[t->bytecode()]));
2019 }
2020 
2021 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
2022   __ mov32(rbx, ExternalAddress((address) &BytecodePairHistogram::_index));
2023   __ shrl(rbx, BytecodePairHistogram::log2_number_of_codes);
2024   __ orl(rbx,
2025          ((int) t->bytecode()) <<
2026          BytecodePairHistogram::log2_number_of_codes);
2027   __ mov32(ExternalAddress((address) &BytecodePairHistogram::_index), rbx);
2028   __ lea(rscratch1, ExternalAddress((address) BytecodePairHistogram::_counters));
2029   __ incrementl(Address(rscratch1, rbx, Address::times_4));
2030 }
2031 
2032 
2033 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
2034   // Call a little run-time stub to avoid blow-up for each bytecode.
2035   // The run-time runtime saves the right registers, depending on
2036   // the tosca in-state for the given template.
2037 
2038   assert(Interpreter::trace_code(t->tos_in()) != NULL,
2039          "entry must have been generated");
2040   __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
2041   __ andptr(rsp, -16); // align stack as required by ABI
2042   __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in())));
2043   __ mov(rsp, r12); // restore sp
2044   __ reinit_heapbase();
2045 }
2046 
2047 
2048 void TemplateInterpreterGenerator::stop_interpreter_at() {
2049   Label L;
2050   __ cmp32(ExternalAddress((address) &BytecodeCounter::_counter_value),
2051            StopInterpreterAt);
2052   __ jcc(Assembler::notEqual, L);
2053   __ int3();
2054   __ bind(L);
2055 }
2056 #endif // !PRODUCT
2057 #endif // ! CC_INTERP