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/interp_masm.hpp"
  29 #include "interpreter/interpreter.hpp"
  30 #include "interpreter/interpreterRuntime.hpp"
  31 #include "interpreter/templateInterpreterGenerator.hpp"
  32 #include "interpreter/templateTable.hpp"
  33 #include "oops/arrayOop.hpp"
  34 #include "oops/methodData.hpp"
  35 #include "oops/method.hpp"
  36 #include "oops/oop.inline.hpp"
  37 #include "prims/jvmtiExport.hpp"
  38 #include "prims/jvmtiThreadState.hpp"
  39 #include "runtime/arguments.hpp"
  40 #include "runtime/deoptimization.hpp"
  41 #include "runtime/frame.inline.hpp"
  42 #include "runtime/sharedRuntime.hpp"
  43 #include "runtime/stubRoutines.hpp"
  44 #include "runtime/synchronizer.hpp"
  45 #include "runtime/timer.hpp"
  46 #include "runtime/vframeArray.hpp"
  47 #include "utilities/debug.hpp"
  48 #include "utilities/macros.hpp"
  49 
  50 #define __ _masm->
  51 
  52 // Size of interpreter code.  Increase if too small.  Interpreter will
  53 // fail with a guarantee ("not enough space for interpreter generation");
  54 // if too small.
  55 // Run with +PrintInterpreter to get the VM to print out the size.
  56 // Max size with JVMTI
  57 #ifdef AMD64
  58 int TemplateInterpreter::InterpreterCodeSize = JVMCI_ONLY(268) NOT_JVMCI(256) * 1024;
  59 #else
  60 int TemplateInterpreter::InterpreterCodeSize = 224 * 1024;
  61 #endif // AMD64
  62 
  63 // Global Register Names
  64 static const Register rbcp     = LP64_ONLY(r13) NOT_LP64(rsi);
  65 static const Register rlocals  = LP64_ONLY(r14) NOT_LP64(rdi);
  66 
  67 const int method_offset = frame::interpreter_frame_method_offset * wordSize;
  68 const int bcp_offset    = frame::interpreter_frame_bcp_offset    * wordSize;
  69 const int locals_offset = frame::interpreter_frame_locals_offset * wordSize;
  70 
  71 
  72 //-----------------------------------------------------------------------------
  73 
  74 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() {
  75   address entry = __ pc();
  76 
  77 #ifdef ASSERT
  78   {
  79     Label L;
  80     __ lea(rax, Address(rbp,
  81                         frame::interpreter_frame_monitor_block_top_offset *
  82                         wordSize));
  83     __ cmpptr(rax, rsp); // rax = maximal rsp for current rbp (stack
  84                          // grows negative)
  85     __ jcc(Assembler::aboveEqual, L); // check if frame is complete
  86     __ stop ("interpreter frame not set up");
  87     __ bind(L);
  88   }
  89 #endif // ASSERT
  90   // Restore bcp under the assumption that the current frame is still
  91   // interpreted
  92   __ restore_bcp();
  93 
  94   // expression stack must be empty before entering the VM if an
  95   // exception happened
  96   __ empty_expression_stack();
  97   // throw exception
  98   __ call_VM(noreg,
  99              CAST_FROM_FN_PTR(address,
 100                               InterpreterRuntime::throw_StackOverflowError));
 101   return entry;
 102 }
 103 
 104 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler(
 105         const char* name) {
 106   address entry = __ pc();
 107   // expression stack must be empty before entering the VM if an
 108   // exception happened
 109   __ empty_expression_stack();
 110   // setup parameters
 111   // ??? convention: expect aberrant index in register ebx
 112   Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
 113   __ lea(rarg, ExternalAddress((address)name));
 114   __ call_VM(noreg,
 115              CAST_FROM_FN_PTR(address,
 116                               InterpreterRuntime::
 117                               throw_ArrayIndexOutOfBoundsException),
 118              rarg, rbx);
 119   return entry;
 120 }
 121 
 122 address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
 123   address entry = __ pc();
 124 
 125   // object is at TOS
 126   Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
 127   __ pop(rarg);
 128 
 129   // expression stack must be empty before entering the VM if an
 130   // exception happened
 131   __ empty_expression_stack();
 132 
 133   __ call_VM(noreg,
 134              CAST_FROM_FN_PTR(address,
 135                               InterpreterRuntime::
 136                               throw_ClassCastException),
 137              rarg);
 138   return entry;
 139 }
 140 
 141 address TemplateInterpreterGenerator::generate_exception_handler_common(
 142         const char* name, const char* message, bool pass_oop) {
 143   assert(!pass_oop || message == NULL, "either oop or message but not both");
 144   address entry = __ pc();
 145 
 146   Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
 147   Register rarg2 = NOT_LP64(rbx) LP64_ONLY(c_rarg2);
 148 
 149   if (pass_oop) {
 150     // object is at TOS
 151     __ pop(rarg2);
 152   }
 153   // expression stack must be empty before entering the VM if an
 154   // exception happened
 155   __ empty_expression_stack();
 156   // setup parameters
 157   __ lea(rarg, ExternalAddress((address)name));
 158   if (pass_oop) {
 159     __ call_VM(rax, CAST_FROM_FN_PTR(address,
 160                                      InterpreterRuntime::
 161                                      create_klass_exception),
 162                rarg, rarg2);
 163   } else {
 164     __ lea(rarg2, ExternalAddress((address)message));
 165     __ call_VM(rax,
 166                CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception),
 167                rarg, rarg2);
 168   }
 169   // throw exception
 170   __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
 171   return entry;
 172 }
 173 
 174 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step, size_t index_size) {
 175   address entry = __ pc();
 176 
 177 #ifndef _LP64
 178 #ifdef COMPILER2
 179   // The FPU stack is clean if UseSSE >= 2 but must be cleaned in other cases
 180   if ((state == ftos && UseSSE < 1) || (state == dtos && UseSSE < 2)) {
 181     for (int i = 1; i < 8; i++) {
 182         __ ffree(i);
 183     }
 184   } else if (UseSSE < 2) {
 185     __ empty_FPU_stack();
 186   }
 187 #endif // COMPILER2
 188   if ((state == ftos && UseSSE < 1) || (state == dtos && UseSSE < 2)) {
 189     __ MacroAssembler::verify_FPU(1, "generate_return_entry_for compiled");
 190   } else {
 191     __ MacroAssembler::verify_FPU(0, "generate_return_entry_for compiled");
 192   }
 193 
 194   if (state == ftos) {
 195     __ MacroAssembler::verify_FPU(UseSSE >= 1 ? 0 : 1, "generate_return_entry_for in interpreter");
 196   } else if (state == dtos) {
 197     __ MacroAssembler::verify_FPU(UseSSE >= 2 ? 0 : 1, "generate_return_entry_for in interpreter");
 198   }
 199 #endif // _LP64
 200 
 201   // Restore stack bottom in case i2c adjusted stack
 202   __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
 203   // and NULL it as marker that esp is now tos until next java call
 204   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
 205 
 206   __ restore_bcp();
 207   __ restore_locals();
 208 
 209   if (state == atos) {
 210     Register mdp = rbx;
 211     Register tmp = rcx;
 212     __ profile_return_type(mdp, rax, tmp);
 213   }
 214 
 215   const Register cache = rbx;
 216   const Register index = rcx;
 217   __ get_cache_and_index_at_bcp(cache, index, 1, index_size);
 218 
 219   const Register flags = cache;
 220   __ movl(flags, Address(cache, index, Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()));
 221   __ andl(flags, ConstantPoolCacheEntry::parameter_size_mask);
 222   __ lea(rsp, Address(rsp, flags, Interpreter::stackElementScale()));
 223 
 224    const Register java_thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
 225    if (JvmtiExport::can_pop_frame()) {
 226      NOT_LP64(__ get_thread(java_thread));
 227      __ check_and_handle_popframe(java_thread);
 228    }
 229    if (JvmtiExport::can_force_early_return()) {
 230      NOT_LP64(__ get_thread(java_thread));
 231      __ check_and_handle_earlyret(java_thread);
 232    }
 233 
 234   __ dispatch_next(state, step);
 235 
 236   return entry;
 237 }
 238 
 239 
 240 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state, int step) {
 241   address entry = __ pc();
 242 
 243 #ifndef _LP64
 244   if (state == ftos) {
 245     __ MacroAssembler::verify_FPU(UseSSE >= 1 ? 0 : 1, "generate_deopt_entry_for in interpreter");
 246   } else if (state == dtos) {
 247     __ MacroAssembler::verify_FPU(UseSSE >= 2 ? 0 : 1, "generate_deopt_entry_for in interpreter");
 248   }
 249 #endif // _LP64
 250 
 251   // NULL last_sp until next java call
 252   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
 253   __ restore_bcp();
 254   __ restore_locals();
 255   const Register thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
 256   NOT_LP64(__ get_thread(thread));
 257 #if INCLUDE_JVMCI
 258   // Check if we need to take lock at entry of synchronized method.  This can
 259   // only occur on method entry so emit it only for vtos with step 0.
 260   if ((UseJVMCICompiler || UseAOT) && state == vtos && step == 0) {
 261     Label L;
 262     __ cmpb(Address(thread, JavaThread::pending_monitorenter_offset()), 0);
 263     __ jcc(Assembler::zero, L);
 264     // Clear flag.
 265     __ movb(Address(thread, JavaThread::pending_monitorenter_offset()), 0);
 266     // Satisfy calling convention for lock_method().
 267     __ get_method(rbx);
 268     // Take lock.
 269     lock_method();
 270     __ bind(L);
 271   } else {
 272 #ifdef ASSERT
 273     if (UseJVMCICompiler) {
 274       Label L;
 275       __ cmpb(Address(r15_thread, JavaThread::pending_monitorenter_offset()), 0);
 276       __ jccb(Assembler::zero, L);
 277       __ stop("unexpected pending monitor in deopt entry");
 278       __ bind(L);
 279     }
 280 #endif
 281   }
 282 #endif
 283   // handle exceptions
 284   {
 285     Label L;
 286     __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t) NULL_WORD);
 287     __ jcc(Assembler::zero, L);
 288     __ call_VM(noreg,
 289                CAST_FROM_FN_PTR(address,
 290                                 InterpreterRuntime::throw_pending_exception));
 291     __ should_not_reach_here();
 292     __ bind(L);
 293   }
 294   __ dispatch_next(state, step);
 295   return entry;
 296 }
 297 
 298 address TemplateInterpreterGenerator::generate_result_handler_for(
 299         BasicType type) {
 300   address entry = __ pc();
 301   switch (type) {
 302   case T_BOOLEAN: __ c2bool(rax);            break;
 303 #ifndef _LP64
 304   case T_CHAR   : __ andptr(rax, 0xFFFF);    break;
 305 #else
 306   case T_CHAR   : __ movzwl(rax, rax);       break;
 307 #endif // _LP64
 308   case T_BYTE   : __ sign_extend_byte(rax);  break;
 309   case T_SHORT  : __ sign_extend_short(rax); break;
 310   case T_INT    : /* nothing to do */        break;
 311   case T_LONG   : /* nothing to do */        break;
 312   case T_VOID   : /* nothing to do */        break;
 313 #ifndef _LP64
 314   case T_DOUBLE :
 315   case T_FLOAT  :
 316     { const Register t = InterpreterRuntime::SignatureHandlerGenerator::temp();
 317       __ pop(t);                            // remove return address first
 318       // Must return a result for interpreter or compiler. In SSE
 319       // mode, results are returned in xmm0 and the FPU stack must
 320       // be empty.
 321       if (type == T_FLOAT && UseSSE >= 1) {
 322         // Load ST0
 323         __ fld_d(Address(rsp, 0));
 324         // Store as float and empty fpu stack
 325         __ fstp_s(Address(rsp, 0));
 326         // and reload
 327         __ movflt(xmm0, Address(rsp, 0));
 328       } else if (type == T_DOUBLE && UseSSE >= 2 ) {
 329         __ movdbl(xmm0, Address(rsp, 0));
 330       } else {
 331         // restore ST0
 332         __ fld_d(Address(rsp, 0));
 333       }
 334       // and pop the temp
 335       __ addptr(rsp, 2 * wordSize);
 336       __ push(t);                           // restore return address
 337     }
 338     break;
 339 #else
 340   case T_FLOAT  : /* nothing to do */        break;
 341   case T_DOUBLE : /* nothing to do */        break;
 342 #endif // _LP64
 343 
 344   case T_OBJECT :
 345     // retrieve result from frame
 346     __ movptr(rax, Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize));
 347     // and verify it
 348     __ verify_oop(rax);
 349     break;
 350   default       : ShouldNotReachHere();
 351   }
 352   __ ret(0);                                   // return from result handler
 353   return entry;
 354 }
 355 
 356 address TemplateInterpreterGenerator::generate_safept_entry_for(
 357         TosState state,
 358         address runtime_entry) {
 359   address entry = __ pc();
 360   __ push(state);
 361   __ call_VM(noreg, runtime_entry);
 362   __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
 363   return entry;
 364 }
 365 
 366 
 367 
 368 // Helpers for commoning out cases in the various type of method entries.
 369 //
 370 
 371 
 372 // increment invocation count & check for overflow
 373 //
 374 // Note: checking for negative value instead of overflow
 375 //       so we have a 'sticky' overflow test
 376 //
 377 // rbx: method
 378 // rcx: invocation counter
 379 //
 380 void TemplateInterpreterGenerator::generate_counter_incr(
 381         Label* overflow,
 382         Label* profile_method,
 383         Label* profile_method_continue) {
 384   Label done;
 385   // Note: In tiered we increment either counters in Method* or in MDO depending if we're profiling or not.
 386   if (TieredCompilation) {
 387     int increment = InvocationCounter::count_increment;
 388     Label no_mdo;
 389     if (ProfileInterpreter) {
 390       // Are we profiling?
 391       __ movptr(rax, Address(rbx, Method::method_data_offset()));
 392       __ testptr(rax, rax);
 393       __ jccb(Assembler::zero, no_mdo);
 394       // Increment counter in the MDO
 395       const Address mdo_invocation_counter(rax, in_bytes(MethodData::invocation_counter_offset()) +
 396                                                 in_bytes(InvocationCounter::counter_offset()));
 397       const Address mask(rax, in_bytes(MethodData::invoke_mask_offset()));
 398       __ increment_mask_and_jump(mdo_invocation_counter, increment, mask, rcx, false, Assembler::zero, overflow);
 399       __ jmp(done);
 400     }
 401     __ bind(no_mdo);
 402     // Increment counter in MethodCounters
 403     const Address invocation_counter(rax,
 404                   MethodCounters::invocation_counter_offset() +
 405                   InvocationCounter::counter_offset());
 406     __ get_method_counters(rbx, rax, done);
 407     const Address mask(rax, in_bytes(MethodCounters::invoke_mask_offset()));
 408     __ increment_mask_and_jump(invocation_counter, increment, mask, rcx,
 409                                false, Assembler::zero, overflow);
 410     __ bind(done);
 411   } else { // not TieredCompilation
 412     const Address backedge_counter(rax,
 413                   MethodCounters::backedge_counter_offset() +
 414                   InvocationCounter::counter_offset());
 415     const Address invocation_counter(rax,
 416                   MethodCounters::invocation_counter_offset() +
 417                   InvocationCounter::counter_offset());
 418 
 419     __ get_method_counters(rbx, rax, done);
 420 
 421     if (ProfileInterpreter) {
 422       __ incrementl(Address(rax,
 423               MethodCounters::interpreter_invocation_counter_offset()));
 424     }
 425     // Update standard invocation counters
 426     __ movl(rcx, invocation_counter);
 427     __ incrementl(rcx, InvocationCounter::count_increment);
 428     __ movl(invocation_counter, rcx); // save invocation count
 429 
 430     __ movl(rax, backedge_counter);   // load backedge counter
 431     __ andl(rax, InvocationCounter::count_mask_value); // mask out the status bits
 432 
 433     __ addl(rcx, rax);                // add both counters
 434 
 435     // profile_method is non-null only for interpreted method so
 436     // profile_method != NULL == !native_call
 437 
 438     if (ProfileInterpreter && profile_method != NULL) {
 439       // Test to see if we should create a method data oop
 440       __ movptr(rax, Address(rbx, Method::method_counters_offset()));
 441       __ cmp32(rcx, Address(rax, in_bytes(MethodCounters::interpreter_profile_limit_offset())));
 442       __ jcc(Assembler::less, *profile_method_continue);
 443 
 444       // if no method data exists, go to profile_method
 445       __ test_method_data_pointer(rax, *profile_method);
 446     }
 447 
 448     __ movptr(rax, Address(rbx, Method::method_counters_offset()));
 449     __ cmp32(rcx, Address(rax, in_bytes(MethodCounters::interpreter_invocation_limit_offset())));
 450     __ jcc(Assembler::aboveEqual, *overflow);
 451     __ bind(done);
 452   }
 453 }
 454 
 455 void TemplateInterpreterGenerator::generate_counter_overflow(Label& do_continue) {
 456 
 457   // Asm interpreter on entry
 458   // r14/rdi - locals
 459   // r13/rsi - bcp
 460   // rbx - method
 461   // rdx - cpool --- DOES NOT APPEAR TO BE TRUE
 462   // rbp - interpreter frame
 463 
 464   // On return (i.e. jump to entry_point) [ back to invocation of interpreter ]
 465   // Everything as it was on entry
 466   // rdx is not restored. Doesn't appear to really be set.
 467 
 468   // InterpreterRuntime::frequency_counter_overflow takes two
 469   // arguments, the first (thread) is passed by call_VM, the second
 470   // indicates if the counter overflow occurs at a backwards branch
 471   // (NULL bcp).  We pass zero for it.  The call returns the address
 472   // of the verified entry point for the method or NULL if the
 473   // compilation did not complete (either went background or bailed
 474   // out).
 475   Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
 476   __ movl(rarg, 0);
 477   __ call_VM(noreg,
 478              CAST_FROM_FN_PTR(address,
 479                               InterpreterRuntime::frequency_counter_overflow),
 480              rarg);
 481 
 482   __ movptr(rbx, Address(rbp, method_offset));   // restore Method*
 483   // Preserve invariant that r13/r14 contain bcp/locals of sender frame
 484   // and jump to the interpreted entry.
 485   __ jmp(do_continue, relocInfo::none);
 486 }
 487 
 488 // See if we've got enough room on the stack for locals plus overhead below
 489 // JavaThread::stack_overflow_limit(). If not, throw a StackOverflowError
 490 // without going through the signal handler, i.e., reserved and yellow zones
 491 // will not be made usable. The shadow zone must suffice to handle the
 492 // overflow.
 493 // The expression stack grows down incrementally, so the normal guard
 494 // page mechanism will work for that.
 495 //
 496 // NOTE: Since the additional locals are also always pushed (wasn't
 497 // obvious in generate_fixed_frame) so the guard should work for them
 498 // too.
 499 //
 500 // Args:
 501 //      rdx: number of additional locals this frame needs (what we must check)
 502 //      rbx: Method*
 503 //
 504 // Kills:
 505 //      rax
 506 void TemplateInterpreterGenerator::generate_stack_overflow_check(void) {
 507 
 508   // monitor entry size: see picture of stack in frame_x86.hpp
 509   const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
 510 
 511   // total overhead size: entry_size + (saved rbp through expr stack
 512   // bottom).  be sure to change this if you add/subtract anything
 513   // to/from the overhead area
 514   const int overhead_size =
 515     -(frame::interpreter_frame_initial_sp_offset * wordSize) + entry_size;
 516 
 517   const int page_size = os::vm_page_size();
 518 
 519   Label after_frame_check;
 520 
 521   // see if the frame is greater than one page in size. If so,
 522   // then we need to verify there is enough stack space remaining
 523   // for the additional locals.
 524   __ cmpl(rdx, (page_size - overhead_size) / Interpreter::stackElementSize);
 525   __ jcc(Assembler::belowEqual, after_frame_check);
 526 
 527   // compute rsp as if this were going to be the last frame on
 528   // the stack before the red zone
 529 
 530   Label after_frame_check_pop;
 531   const Register thread = NOT_LP64(rsi) LP64_ONLY(r15_thread);
 532 #ifndef _LP64
 533   __ push(thread);
 534   __ get_thread(thread);
 535 #endif
 536 
 537   const Address stack_limit(thread, JavaThread::stack_overflow_limit_offset());
 538 
 539   // locals + overhead, in bytes
 540   __ mov(rax, rdx);
 541   __ shlptr(rax, Interpreter::logStackElementSize); // Convert parameter count to bytes.
 542   __ addptr(rax, overhead_size);
 543 
 544 #ifdef ASSERT
 545   Label limit_okay;
 546   // Verify that thread stack overflow limit is non-zero.
 547   __ cmpptr(stack_limit, (int32_t)NULL_WORD);
 548   __ jcc(Assembler::notEqual, limit_okay);
 549   __ stop("stack overflow limit is zero");
 550   __ bind(limit_okay);
 551 #endif
 552 
 553   // Add locals/frame size to stack limit.
 554   __ addptr(rax, stack_limit);
 555 
 556   // Check against the current stack bottom.
 557   __ cmpptr(rsp, rax);
 558 
 559   __ jcc(Assembler::above, after_frame_check_pop);
 560   NOT_LP64(__ pop(rsi));  // get saved bcp
 561 
 562   // Restore sender's sp as SP. This is necessary if the sender's
 563   // frame is an extended compiled frame (see gen_c2i_adapter())
 564   // and safer anyway in case of JSR292 adaptations.
 565 
 566   __ pop(rax); // return address must be moved if SP is changed
 567   __ mov(rsp, rbcp);
 568   __ push(rax);
 569 
 570   // Note: the restored frame is not necessarily interpreted.
 571   // Use the shared runtime version of the StackOverflowError.
 572   assert(StubRoutines::throw_StackOverflowError_entry() != NULL, "stub not yet generated");
 573   __ jump(ExternalAddress(StubRoutines::throw_StackOverflowError_entry()));
 574   // all done with frame size check
 575   __ bind(after_frame_check_pop);
 576   NOT_LP64(__ pop(rsi));
 577 
 578   // all done with frame size check
 579   __ bind(after_frame_check);
 580 }
 581 
 582 // Allocate monitor and lock method (asm interpreter)
 583 //
 584 // Args:
 585 //      rbx: Method*
 586 //      r14/rdi: locals
 587 //
 588 // Kills:
 589 //      rax
 590 //      c_rarg0, c_rarg1, c_rarg2, c_rarg3, ...(param regs)
 591 //      rscratch1, rscratch2 (scratch regs)
 592 void TemplateInterpreterGenerator::lock_method() {
 593   // synchronize method
 594   const Address access_flags(rbx, Method::access_flags_offset());
 595   const Address monitor_block_top(
 596         rbp,
 597         frame::interpreter_frame_monitor_block_top_offset * wordSize);
 598   const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
 599 
 600 #ifdef ASSERT
 601   {
 602     Label L;
 603     __ movl(rax, access_flags);
 604     __ testl(rax, JVM_ACC_SYNCHRONIZED);
 605     __ jcc(Assembler::notZero, L);
 606     __ stop("method doesn't need synchronization");
 607     __ bind(L);
 608   }
 609 #endif // ASSERT
 610 
 611   // get synchronization object
 612   {
 613     Label done;
 614     __ movl(rax, access_flags);
 615     __ testl(rax, JVM_ACC_STATIC);
 616     // get receiver (assume this is frequent case)
 617     __ movptr(rax, Address(rlocals, Interpreter::local_offset_in_bytes(0)));
 618     __ jcc(Assembler::zero, done);
 619     __ load_mirror(rax, rbx);
 620 
 621 #ifdef ASSERT
 622     {
 623       Label L;
 624       __ testptr(rax, rax);
 625       __ jcc(Assembler::notZero, L);
 626       __ stop("synchronization object is NULL");
 627       __ bind(L);
 628     }
 629 #endif // ASSERT
 630 
 631     __ bind(done);
 632     oopDesc::bs()->interpreter_write_barrier(_masm, rax);
 633   }
 634 
 635   // add space for monitor & lock
 636   __ subptr(rsp, entry_size); // add space for a monitor entry
 637   __ movptr(monitor_block_top, rsp);  // set new monitor block top
 638   // store object
 639   __ shenandoah_store_addr_check(rax);
 640   __ movptr(Address(rsp, BasicObjectLock::obj_offset_in_bytes()), rax);
 641   const Register lockreg = NOT_LP64(rdx) LP64_ONLY(c_rarg1);
 642   __ movptr(lockreg, rsp); // object address
 643   __ lock_object(lockreg);
 644 }
 645 
 646 // Generate a fixed interpreter frame. This is identical setup for
 647 // interpreted methods and for native methods hence the shared code.
 648 //
 649 // Args:
 650 //      rax: return address
 651 //      rbx: Method*
 652 //      r14/rdi: pointer to locals
 653 //      r13/rsi: sender sp
 654 //      rdx: cp cache
 655 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
 656   // initialize fixed part of activation frame
 657   __ push(rax);        // save return address
 658   __ enter();          // save old & set new rbp
 659   __ push(rbcp);        // set sender sp
 660   __ push((int)NULL_WORD); // leave last_sp as null
 661   __ movptr(rbcp, Address(rbx, Method::const_offset()));      // get ConstMethod*
 662   __ lea(rbcp, Address(rbcp, ConstMethod::codes_offset())); // get codebase
 663   __ push(rbx);        // save Method*
 664   // Get mirror and store it in the frame as GC root for this Method*
 665   __ load_mirror(rdx, rbx);
 666   __ push(rdx);
 667   if (ProfileInterpreter) {
 668     Label method_data_continue;
 669     __ movptr(rdx, Address(rbx, in_bytes(Method::method_data_offset())));
 670     __ testptr(rdx, rdx);
 671     __ jcc(Assembler::zero, method_data_continue);
 672     __ addptr(rdx, in_bytes(MethodData::data_offset()));
 673     __ bind(method_data_continue);
 674     __ push(rdx);      // set the mdp (method data pointer)
 675   } else {
 676     __ push(0);
 677   }
 678 
 679   __ movptr(rdx, Address(rbx, Method::const_offset()));
 680   __ movptr(rdx, Address(rdx, ConstMethod::constants_offset()));
 681   __ movptr(rdx, Address(rdx, ConstantPool::cache_offset_in_bytes()));
 682   __ push(rdx); // set constant pool cache
 683   __ push(rlocals); // set locals pointer
 684   if (native_call) {
 685     __ push(0); // no bcp
 686   } else {
 687     __ push(rbcp); // set bcp
 688   }
 689   __ push(0); // reserve word for pointer to expression stack bottom
 690   __ movptr(Address(rsp, 0), rsp); // set expression stack bottom
 691 }
 692 
 693 // End of helpers
 694 
 695 // Method entry for java.lang.ref.Reference.get.
 696 address TemplateInterpreterGenerator::generate_Reference_get_entry(void) {
 697 #if INCLUDE_ALL_GCS
 698   // Code: _aload_0, _getfield, _areturn
 699   // parameter size = 1
 700   //
 701   // The code that gets generated by this routine is split into 2 parts:
 702   //    1. The "intrinsified" code for G1 (or any SATB based GC),
 703   //    2. The slow path - which is an expansion of the regular method entry.
 704   //
 705   // Notes:-
 706   // * In the G1 code we do not check whether we need to block for
 707   //   a safepoint. If G1 is enabled then we must execute the specialized
 708   //   code for Reference.get (except when the Reference object is null)
 709   //   so that we can log the value in the referent field with an SATB
 710   //   update buffer.
 711   //   If the code for the getfield template is modified so that the
 712   //   G1 pre-barrier code is executed when the current method is
 713   //   Reference.get() then going through the normal method entry
 714   //   will be fine.
 715   // * The G1 code can, however, check the receiver object (the instance
 716   //   of java.lang.Reference) and jump to the slow path if null. If the
 717   //   Reference object is null then we obviously cannot fetch the referent
 718   //   and so we don't need to call the G1 pre-barrier. Thus we can use the
 719   //   regular method entry code to generate the NPE.
 720   //
 721   // rbx: Method*
 722 
 723   // r13: senderSP must preserve for slow path, set SP to it on fast path
 724 
 725   address entry = __ pc();
 726 
 727   const int referent_offset = java_lang_ref_Reference::referent_offset;
 728   guarantee(referent_offset > 0, "referent offset not initialized");
 729 
 730   if (UseG1GC || UseShenandoahGC) {
 731     Label slow_path;
 732     // rbx: method
 733 
 734     // Check if local 0 != NULL
 735     // If the receiver is null then it is OK to jump to the slow path.
 736     __ movptr(rax, Address(rsp, wordSize));
 737 
 738     __ testptr(rax, rax);
 739     __ jcc(Assembler::zero, slow_path);
 740 
 741     oopDesc::bs()->interpreter_read_barrier_not_null(_masm, rax);
 742 
 743     // rax: local 0
 744     // rbx: method (but can be used as scratch now)
 745     // rdx: scratch
 746     // rdi: scratch
 747 
 748     // Preserve the sender sp in case the pre-barrier
 749     // calls the runtime
 750     NOT_LP64(__ push(rsi));
 751 
 752     // Generate the G1 pre-barrier code to log the value of
 753     // the referent field in an SATB buffer.
 754 
 755     // Load the value of the referent field.
 756     const Address field_address(rax, referent_offset);
 757     __ load_heap_oop(rax, field_address);
 758 
 759     const Register sender_sp = NOT_LP64(rsi) LP64_ONLY(r13);
 760     const Register thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
 761     NOT_LP64(__ get_thread(thread));
 762 
 763     // Generate the G1 pre-barrier code to log the value of
 764     // the referent field in an SATB buffer.
 765     __ keep_alive_barrier(rax /* pre_val */,
 766                           thread /* thread */,
 767                           rbx /* tmp */);
 768 
 769     // _areturn
 770     NOT_LP64(__ pop(rsi));      // get sender sp
 771     __ pop(rdi);                // get return address
 772     __ mov(rsp, sender_sp);     // set sp to sender sp
 773     __ jmp(rdi);
 774     __ ret(0);
 775 
 776     // generate a vanilla interpreter entry as the slow path
 777     __ bind(slow_path);
 778     __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::zerolocals));
 779     return entry;
 780   }
 781 #endif // INCLUDE_ALL_GCS
 782 
 783   // If G1 is not enabled then attempt to go through the accessor entry point
 784   // Reference.get is an accessor
 785   return NULL;
 786 }
 787 
 788 void TemplateInterpreterGenerator::bang_stack_shadow_pages(bool native_call) {
 789   // Quick & dirty stack overflow checking: bang the stack & handle trap.
 790   // Note that we do the banging after the frame is setup, since the exception
 791   // handling code expects to find a valid interpreter frame on the stack.
 792   // Doing the banging earlier fails if the caller frame is not an interpreter
 793   // frame.
 794   // (Also, the exception throwing code expects to unlock any synchronized
 795   // method receiever, so do the banging after locking the receiver.)
 796 
 797   // Bang each page in the shadow zone. We can't assume it's been done for
 798   // an interpreter frame with greater than a page of locals, so each page
 799   // needs to be checked.  Only true for non-native.
 800   if (UseStackBanging) {
 801     const int page_size = os::vm_page_size();
 802     const int n_shadow_pages = ((int)JavaThread::stack_shadow_zone_size()) / page_size;
 803     const int start_page = native_call ? n_shadow_pages : 1;
 804     for (int pages = start_page; pages <= n_shadow_pages; pages++) {
 805       __ bang_stack_with_offset(pages*page_size);
 806     }
 807   }
 808 }
 809 
 810 // Interpreter stub for calling a native method. (asm interpreter)
 811 // This sets up a somewhat different looking stack for calling the
 812 // native method than the typical interpreter frame setup.
 813 address TemplateInterpreterGenerator::generate_native_entry(bool synchronized) {
 814   // determine code generation flags
 815   bool inc_counter  = UseCompiler || CountCompiledCalls || LogTouchedMethods;
 816 
 817   // rbx: Method*
 818   // rbcp: sender sp
 819 
 820   address entry_point = __ pc();
 821 
 822   const Address constMethod       (rbx, Method::const_offset());
 823   const Address access_flags      (rbx, Method::access_flags_offset());
 824   const Address size_of_parameters(rcx, ConstMethod::
 825                                         size_of_parameters_offset());
 826 
 827 
 828   // get parameter size (always needed)
 829   __ movptr(rcx, constMethod);
 830   __ load_unsigned_short(rcx, size_of_parameters);
 831 
 832   // native calls don't need the stack size check since they have no
 833   // expression stack and the arguments are already on the stack and
 834   // we only add a handful of words to the stack
 835 
 836   // rbx: Method*
 837   // rcx: size of parameters
 838   // rbcp: sender sp
 839   __ pop(rax);                                       // get return address
 840 
 841   // for natives the size of locals is zero
 842 
 843   // compute beginning of parameters
 844   __ lea(rlocals, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize));
 845 
 846   // add 2 zero-initialized slots for native calls
 847   // initialize result_handler slot
 848   __ push((int) NULL_WORD);
 849   // slot for oop temp
 850   // (static native method holder mirror/jni oop result)
 851   __ push((int) NULL_WORD);
 852 
 853   // initialize fixed part of activation frame
 854   generate_fixed_frame(true);
 855 
 856   // make sure method is native & not abstract
 857 #ifdef ASSERT
 858   __ movl(rax, access_flags);
 859   {
 860     Label L;
 861     __ testl(rax, JVM_ACC_NATIVE);
 862     __ jcc(Assembler::notZero, L);
 863     __ stop("tried to execute non-native method as native");
 864     __ bind(L);
 865   }
 866   {
 867     Label L;
 868     __ testl(rax, JVM_ACC_ABSTRACT);
 869     __ jcc(Assembler::zero, L);
 870     __ stop("tried to execute abstract method in interpreter");
 871     __ bind(L);
 872   }
 873 #endif
 874 
 875   // Since at this point in the method invocation the exception handler
 876   // would try to exit the monitor of synchronized methods which hasn't
 877   // been entered yet, we set the thread local variable
 878   // _do_not_unlock_if_synchronized to true. The remove_activation will
 879   // check this flag.
 880 
 881   const Register thread1 = NOT_LP64(rax) LP64_ONLY(r15_thread);
 882   NOT_LP64(__ get_thread(thread1));
 883   const Address do_not_unlock_if_synchronized(thread1,
 884         in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
 885   __ movbool(do_not_unlock_if_synchronized, true);
 886 
 887   // increment invocation count & check for overflow
 888   Label invocation_counter_overflow;
 889   if (inc_counter) {
 890     generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
 891   }
 892 
 893   Label continue_after_compile;
 894   __ bind(continue_after_compile);
 895 
 896   bang_stack_shadow_pages(true);
 897 
 898   // reset the _do_not_unlock_if_synchronized flag
 899   NOT_LP64(__ get_thread(thread1));
 900   __ movbool(do_not_unlock_if_synchronized, false);
 901 
 902   // check for synchronized methods
 903   // Must happen AFTER invocation_counter check and stack overflow check,
 904   // so method is not locked if overflows.
 905   if (synchronized) {
 906     lock_method();
 907   } else {
 908     // no synchronization necessary
 909 #ifdef ASSERT
 910     {
 911       Label L;
 912       __ movl(rax, access_flags);
 913       __ testl(rax, JVM_ACC_SYNCHRONIZED);
 914       __ jcc(Assembler::zero, L);
 915       __ stop("method needs synchronization");
 916       __ bind(L);
 917     }
 918 #endif
 919   }
 920 
 921   // start execution
 922 #ifdef ASSERT
 923   {
 924     Label L;
 925     const Address monitor_block_top(rbp,
 926                  frame::interpreter_frame_monitor_block_top_offset * wordSize);
 927     __ movptr(rax, monitor_block_top);
 928     __ cmpptr(rax, rsp);
 929     __ jcc(Assembler::equal, L);
 930     __ stop("broken stack frame setup in interpreter");
 931     __ bind(L);
 932   }
 933 #endif
 934 
 935   // jvmti support
 936   __ notify_method_entry();
 937 
 938   // work registers
 939   const Register method = rbx;
 940   const Register thread = NOT_LP64(rdi) LP64_ONLY(r15_thread);
 941   const Register t      = NOT_LP64(rcx) LP64_ONLY(r11);
 942 
 943   // allocate space for parameters
 944   __ get_method(method);
 945   __ movptr(t, Address(method, Method::const_offset()));
 946   __ load_unsigned_short(t, Address(t, ConstMethod::size_of_parameters_offset()));
 947 
 948 #ifndef _LP64
 949   __ shlptr(t, Interpreter::logStackElementSize); // Convert parameter count to bytes.
 950   __ addptr(t, 2*wordSize);     // allocate two more slots for JNIEnv and possible mirror
 951   __ subptr(rsp, t);
 952   __ andptr(rsp, -(StackAlignmentInBytes)); // gcc needs 16 byte aligned stacks to do XMM intrinsics
 953 #else
 954   __ shll(t, Interpreter::logStackElementSize);
 955 
 956   __ subptr(rsp, t);
 957   __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
 958   __ andptr(rsp, -16); // must be 16 byte boundary (see amd64 ABI)
 959 #endif // _LP64
 960 
 961   // get signature handler
 962   {
 963     Label L;
 964     __ movptr(t, Address(method, Method::signature_handler_offset()));
 965     __ testptr(t, t);
 966     __ jcc(Assembler::notZero, L);
 967     __ call_VM(noreg,
 968                CAST_FROM_FN_PTR(address,
 969                                 InterpreterRuntime::prepare_native_call),
 970                method);
 971     __ get_method(method);
 972     __ movptr(t, Address(method, Method::signature_handler_offset()));
 973     __ bind(L);
 974   }
 975 
 976   // call signature handler
 977   assert(InterpreterRuntime::SignatureHandlerGenerator::from() == rlocals,
 978          "adjust this code");
 979   assert(InterpreterRuntime::SignatureHandlerGenerator::to() == rsp,
 980          "adjust this code");
 981   assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == NOT_LP64(t) LP64_ONLY(rscratch1),
 982          "adjust this code");
 983 
 984   // The generated handlers do not touch RBX (the method oop).
 985   // However, large signatures cannot be cached and are generated
 986   // each time here.  The slow-path generator can do a GC on return,
 987   // so we must reload it after the call.
 988   __ call(t);
 989   __ get_method(method);        // slow path can do a GC, reload RBX
 990 
 991 
 992   // result handler is in rax
 993   // set result handler
 994   __ movptr(Address(rbp,
 995                     (frame::interpreter_frame_result_handler_offset) * wordSize),
 996             rax);
 997 
 998   // pass mirror handle if static call
 999   {
1000     Label L;
1001     __ movl(t, Address(method, Method::access_flags_offset()));
1002     __ testl(t, JVM_ACC_STATIC);
1003     __ jcc(Assembler::zero, L);
1004     // get mirror
1005     __ load_mirror(t, method);
1006     // copy mirror into activation frame
1007     __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize),
1008             t);
1009     // pass handle to mirror
1010 #ifndef _LP64
1011     __ lea(t, Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize));
1012     __ movptr(Address(rsp, wordSize), t);
1013 #else
1014     __ lea(c_rarg1,
1015            Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize));
1016 #endif // _LP64
1017     __ bind(L);
1018   }
1019 
1020   // get native function entry point
1021   {
1022     Label L;
1023     __ movptr(rax, Address(method, Method::native_function_offset()));
1024     ExternalAddress unsatisfied(SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
1025     __ cmpptr(rax, unsatisfied.addr());
1026     __ jcc(Assembler::notEqual, L);
1027     __ call_VM(noreg,
1028                CAST_FROM_FN_PTR(address,
1029                                 InterpreterRuntime::prepare_native_call),
1030                method);
1031     __ get_method(method);
1032     __ movptr(rax, Address(method, Method::native_function_offset()));
1033     __ bind(L);
1034   }
1035 
1036   // pass JNIEnv
1037 #ifndef _LP64
1038    __ get_thread(thread);
1039    __ lea(t, Address(thread, JavaThread::jni_environment_offset()));
1040    __ movptr(Address(rsp, 0), t);
1041 
1042    // set_last_Java_frame_before_call
1043    // It is enough that the pc()
1044    // points into the right code segment. It does not have to be the correct return pc.
1045    __ set_last_Java_frame(thread, noreg, rbp, __ pc());
1046 #else
1047    __ lea(c_rarg0, Address(r15_thread, JavaThread::jni_environment_offset()));
1048 
1049    // It is enough that the pc() points into the right code
1050    // segment. It does not have to be the correct return pc.
1051    __ set_last_Java_frame(rsp, rbp, (address) __ pc());
1052 #endif // _LP64
1053 
1054   // change thread state
1055 #ifdef ASSERT
1056   {
1057     Label L;
1058     __ movl(t, Address(thread, JavaThread::thread_state_offset()));
1059     __ cmpl(t, _thread_in_Java);
1060     __ jcc(Assembler::equal, L);
1061     __ stop("Wrong thread state in native stub");
1062     __ bind(L);
1063   }
1064 #endif
1065 
1066   // Change state to native
1067 
1068   __ movl(Address(thread, JavaThread::thread_state_offset()),
1069           _thread_in_native);
1070 
1071   // Call the native method.
1072   __ call(rax);
1073   // 32: result potentially in rdx:rax or ST0
1074   // 64: result potentially in rax or xmm0
1075 
1076   // Verify or restore cpu control state after JNI call
1077   __ restore_cpu_control_state_after_jni();
1078 
1079   // NOTE: The order of these pushes is known to frame::interpreter_frame_result
1080   // in order to extract the result of a method call. If the order of these
1081   // pushes change or anything else is added to the stack then the code in
1082   // interpreter_frame_result must also change.
1083 
1084 #ifndef _LP64
1085   // save potential result in ST(0) & rdx:rax
1086   // (if result handler is the T_FLOAT or T_DOUBLE handler, result must be in ST0 -
1087   // the check is necessary to avoid potential Intel FPU overflow problems by saving/restoring 'empty' FPU registers)
1088   // It is safe to do this push because state is _thread_in_native and return address will be found
1089   // via _last_native_pc and not via _last_jave_sp
1090 
1091   // NOTE: the order of theses push(es) is known to frame::interpreter_frame_result.
1092   // If the order changes or anything else is added to the stack the code in
1093   // interpreter_frame_result will have to be changed.
1094 
1095   { Label L;
1096     Label push_double;
1097     ExternalAddress float_handler(AbstractInterpreter::result_handler(T_FLOAT));
1098     ExternalAddress double_handler(AbstractInterpreter::result_handler(T_DOUBLE));
1099     __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize),
1100               float_handler.addr());
1101     __ jcc(Assembler::equal, push_double);
1102     __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize),
1103               double_handler.addr());
1104     __ jcc(Assembler::notEqual, L);
1105     __ bind(push_double);
1106     __ push_d(); // FP values are returned using the FPU, so push FPU contents (even if UseSSE > 0).
1107     __ bind(L);
1108   }
1109 #else
1110   __ push(dtos);
1111 #endif // _LP64
1112 
1113   __ push(ltos);
1114 
1115   // change thread state
1116   NOT_LP64(__ get_thread(thread));
1117   __ movl(Address(thread, JavaThread::thread_state_offset()),
1118           _thread_in_native_trans);
1119 
1120   if (os::is_MP()) {
1121     if (UseMembar) {
1122       // Force this write out before the read below
1123       __ membar(Assembler::Membar_mask_bits(
1124            Assembler::LoadLoad | Assembler::LoadStore |
1125            Assembler::StoreLoad | Assembler::StoreStore));
1126     } else {
1127       // Write serialization page so VM thread can do a pseudo remote membar.
1128       // We use the current thread pointer to calculate a thread specific
1129       // offset to write to within the page. This minimizes bus traffic
1130       // due to cache line collision.
1131       __ serialize_memory(thread, rcx);
1132     }
1133   }
1134 
1135 #ifndef _LP64
1136   if (AlwaysRestoreFPU) {
1137     //  Make sure the control word is correct.
1138     __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
1139   }
1140 #endif // _LP64
1141 
1142   // check for safepoint operation in progress and/or pending suspend requests
1143   {
1144     Label Continue;
1145     __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
1146              SafepointSynchronize::_not_synchronized);
1147 
1148     Label L;
1149     __ jcc(Assembler::notEqual, L);
1150     __ cmpl(Address(thread, JavaThread::suspend_flags_offset()), 0);
1151     __ jcc(Assembler::equal, Continue);
1152     __ bind(L);
1153 
1154     // Don't use call_VM as it will see a possible pending exception
1155     // and forward it and never return here preventing us from
1156     // clearing _last_native_pc down below.  Also can't use
1157     // call_VM_leaf either as it will check to see if r13 & r14 are
1158     // preserved and correspond to the bcp/locals pointers. So we do a
1159     // runtime call by hand.
1160     //
1161 #ifndef _LP64
1162     __ push(thread);
1163     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address,
1164                                             JavaThread::check_special_condition_for_native_trans)));
1165     __ increment(rsp, wordSize);
1166     __ get_thread(thread);
1167 #else
1168     __ mov(c_rarg0, r15_thread);
1169     __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1170     __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1171     __ andptr(rsp, -16); // align stack as required by ABI
1172     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans)));
1173     __ mov(rsp, r12); // restore sp
1174     __ reinit_heapbase();
1175 #endif // _LP64
1176     __ bind(Continue);
1177   }
1178 
1179   // change thread state
1180   __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_Java);
1181 
1182   // reset_last_Java_frame
1183   __ reset_last_Java_frame(thread, true);
1184 
1185   if (CheckJNICalls) {
1186     // clear_pending_jni_exception_check
1187     __ movptr(Address(thread, JavaThread::pending_jni_exception_check_fn_offset()), NULL_WORD);
1188   }
1189 
1190   // reset handle block
1191   __ movptr(t, Address(thread, JavaThread::active_handles_offset()));
1192   __ movl(Address(t, JNIHandleBlock::top_offset_in_bytes()), (int32_t)NULL_WORD);
1193 
1194   // If result is an oop unbox and store it in frame where gc will see it
1195   // and result handler will pick it up
1196 
1197   {
1198     Label no_oop, not_weak, store_result;
1199     __ lea(t, ExternalAddress(AbstractInterpreter::result_handler(T_OBJECT)));
1200     __ cmpptr(t, Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize));
1201     __ jcc(Assembler::notEqual, no_oop);
1202     // retrieve result
1203     __ pop(ltos);
1204     // Unbox oop result, e.g. JNIHandles::resolve value.
1205     __ resolve_jobject(rax /* value */,
1206                        thread /* thread */,
1207                        t /* tmp */);
1208     __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize), rax);
1209     // keep stack depth as expected by pushing oop which will eventually be discarded
1210     __ push(ltos);
1211     __ bind(no_oop);
1212   }
1213 
1214 
1215   {
1216     Label no_reguard;
1217     __ cmpl(Address(thread, JavaThread::stack_guard_state_offset()),
1218             JavaThread::stack_guard_yellow_reserved_disabled);
1219     __ jcc(Assembler::notEqual, no_reguard);
1220 
1221     __ pusha(); // XXX only save smashed registers
1222 #ifndef _LP64
1223     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
1224     __ popa();
1225 #else
1226     __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1227     __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1228     __ andptr(rsp, -16); // align stack as required by ABI
1229     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
1230     __ mov(rsp, r12); // restore sp
1231     __ popa(); // XXX only restore smashed registers
1232     __ reinit_heapbase();
1233 #endif // _LP64
1234 
1235     __ bind(no_reguard);
1236   }
1237 
1238 
1239   // The method register is junk from after the thread_in_native transition
1240   // until here.  Also can't call_VM until the bcp has been
1241   // restored.  Need bcp for throwing exception below so get it now.
1242   __ get_method(method);
1243 
1244   // restore to have legal interpreter frame, i.e., bci == 0 <=> code_base()
1245   __ movptr(rbcp, Address(method, Method::const_offset()));   // get ConstMethod*
1246   __ lea(rbcp, Address(rbcp, ConstMethod::codes_offset()));    // get codebase
1247 
1248   // handle exceptions (exception handling will handle unlocking!)
1249   {
1250     Label L;
1251     __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t) NULL_WORD);
1252     __ jcc(Assembler::zero, L);
1253     // Note: At some point we may want to unify this with the code
1254     // used in call_VM_base(); i.e., we should use the
1255     // StubRoutines::forward_exception code. For now this doesn't work
1256     // here because the rsp is not correctly set at this point.
1257     __ MacroAssembler::call_VM(noreg,
1258                                CAST_FROM_FN_PTR(address,
1259                                InterpreterRuntime::throw_pending_exception));
1260     __ should_not_reach_here();
1261     __ bind(L);
1262   }
1263 
1264   // do unlocking if necessary
1265   {
1266     Label L;
1267     __ movl(t, Address(method, Method::access_flags_offset()));
1268     __ testl(t, JVM_ACC_SYNCHRONIZED);
1269     __ jcc(Assembler::zero, L);
1270     // the code below should be shared with interpreter macro
1271     // assembler implementation
1272     {
1273       Label unlock;
1274       // BasicObjectLock will be first in list, since this is a
1275       // synchronized method. However, need to check that the object
1276       // has not been unlocked by an explicit monitorexit bytecode.
1277       const Address monitor(rbp,
1278                             (intptr_t)(frame::interpreter_frame_initial_sp_offset *
1279                                        wordSize - (int)sizeof(BasicObjectLock)));
1280 
1281       const Register regmon = NOT_LP64(rdx) LP64_ONLY(c_rarg1);
1282 
1283       // monitor expect in c_rarg1 for slow unlock path
1284       __ lea(regmon, monitor); // address of first monitor
1285 
1286       __ movptr(t, Address(regmon, BasicObjectLock::obj_offset_in_bytes()));
1287       __ shenandoah_store_addr_check(t); // Invariant
1288       __ testptr(t, t);
1289       __ jcc(Assembler::notZero, unlock);
1290 
1291       // Entry already unlocked, need to throw exception
1292       __ MacroAssembler::call_VM(noreg,
1293                                  CAST_FROM_FN_PTR(address,
1294                    InterpreterRuntime::throw_illegal_monitor_state_exception));
1295       __ should_not_reach_here();
1296 
1297       __ bind(unlock);
1298       __ unlock_object(regmon);
1299     }
1300     __ bind(L);
1301   }
1302 
1303   // jvmti support
1304   // Note: This must happen _after_ handling/throwing any exceptions since
1305   //       the exception handler code notifies the runtime of method exits
1306   //       too. If this happens before, method entry/exit notifications are
1307   //       not properly paired (was bug - gri 11/22/99).
1308   __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI);
1309 
1310   // restore potential result in edx:eax, call result handler to
1311   // restore potential result in ST0 & handle result
1312 
1313   __ pop(ltos);
1314   LP64_ONLY( __ pop(dtos));
1315 
1316   __ movptr(t, Address(rbp,
1317                        (frame::interpreter_frame_result_handler_offset) * wordSize));
1318   __ call(t);
1319 
1320   // remove activation
1321   __ movptr(t, Address(rbp,
1322                        frame::interpreter_frame_sender_sp_offset *
1323                        wordSize)); // get sender sp
1324   __ leave();                                // remove frame anchor
1325   __ pop(rdi);                               // get return address
1326   __ mov(rsp, t);                            // set sp to sender sp
1327   __ jmp(rdi);
1328 
1329   if (inc_counter) {
1330     // Handle overflow of counter and compile method
1331     __ bind(invocation_counter_overflow);
1332     generate_counter_overflow(continue_after_compile);
1333   }
1334 
1335   return entry_point;
1336 }
1337 
1338 // Abstract method entry
1339 // Attempt to execute abstract method. Throw exception
1340 address TemplateInterpreterGenerator::generate_abstract_entry(void) {
1341 
1342   address entry_point = __ pc();
1343 
1344   // abstract method entry
1345 
1346   //  pop return address, reset last_sp to NULL
1347   __ empty_expression_stack();
1348   __ restore_bcp();      // rsi must be correct for exception handler   (was destroyed)
1349   __ restore_locals();   // make sure locals pointer is correct as well (was destroyed)
1350 
1351   // throw exception
1352   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError));
1353   // the call_VM checks for exception, so we should never return here.
1354   __ should_not_reach_here();
1355 
1356   return entry_point;
1357 }
1358 
1359 //
1360 // Generic interpreted method entry to (asm) interpreter
1361 //
1362 address TemplateInterpreterGenerator::generate_normal_entry(bool synchronized) {
1363   // determine code generation flags
1364   bool inc_counter  = UseCompiler || CountCompiledCalls || LogTouchedMethods;
1365 
1366   // ebx: Method*
1367   // rbcp: sender sp
1368   address entry_point = __ pc();
1369 
1370   const Address constMethod(rbx, Method::const_offset());
1371   const Address access_flags(rbx, Method::access_flags_offset());
1372   const Address size_of_parameters(rdx,
1373                                    ConstMethod::size_of_parameters_offset());
1374   const Address size_of_locals(rdx, ConstMethod::size_of_locals_offset());
1375 
1376 
1377   // get parameter size (always needed)
1378   __ movptr(rdx, constMethod);
1379   __ load_unsigned_short(rcx, size_of_parameters);
1380 
1381   // rbx: Method*
1382   // rcx: size of parameters
1383   // rbcp: sender_sp (could differ from sp+wordSize if we were called via c2i )
1384 
1385   __ load_unsigned_short(rdx, size_of_locals); // get size of locals in words
1386   __ subl(rdx, rcx); // rdx = no. of additional locals
1387 
1388   // YYY
1389 //   __ incrementl(rdx);
1390 //   __ andl(rdx, -2);
1391 
1392   // see if we've got enough room on the stack for locals plus overhead.
1393   generate_stack_overflow_check();
1394 
1395   // get return address
1396   __ pop(rax);
1397 
1398   // compute beginning of parameters
1399   __ lea(rlocals, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize));
1400 
1401   // rdx - # of additional locals
1402   // allocate space for locals
1403   // explicitly initialize locals
1404   {
1405     Label exit, loop;
1406     __ testl(rdx, rdx);
1407     __ jcc(Assembler::lessEqual, exit); // do nothing if rdx <= 0
1408     __ bind(loop);
1409     __ push((int) NULL_WORD); // initialize local variables
1410     __ decrementl(rdx); // until everything initialized
1411     __ jcc(Assembler::greater, loop);
1412     __ bind(exit);
1413   }
1414 
1415   // initialize fixed part of activation frame
1416   generate_fixed_frame(false);
1417 
1418   // make sure method is not native & not abstract
1419 #ifdef ASSERT
1420   __ movl(rax, access_flags);
1421   {
1422     Label L;
1423     __ testl(rax, JVM_ACC_NATIVE);
1424     __ jcc(Assembler::zero, L);
1425     __ stop("tried to execute native method as non-native");
1426     __ bind(L);
1427   }
1428   {
1429     Label L;
1430     __ testl(rax, JVM_ACC_ABSTRACT);
1431     __ jcc(Assembler::zero, L);
1432     __ stop("tried to execute abstract method in interpreter");
1433     __ bind(L);
1434   }
1435 #endif
1436 
1437   // Since at this point in the method invocation the exception
1438   // handler would try to exit the monitor of synchronized methods
1439   // which hasn't been entered yet, we set the thread local variable
1440   // _do_not_unlock_if_synchronized to true. The remove_activation
1441   // will check this flag.
1442 
1443   const Register thread = NOT_LP64(rax) LP64_ONLY(r15_thread);
1444   NOT_LP64(__ get_thread(thread));
1445   const Address do_not_unlock_if_synchronized(thread,
1446         in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1447   __ movbool(do_not_unlock_if_synchronized, true);
1448 
1449   __ profile_parameters_type(rax, rcx, rdx);
1450   // increment invocation count & check for overflow
1451   Label invocation_counter_overflow;
1452   Label profile_method;
1453   Label profile_method_continue;
1454   if (inc_counter) {
1455     generate_counter_incr(&invocation_counter_overflow,
1456                           &profile_method,
1457                           &profile_method_continue);
1458     if (ProfileInterpreter) {
1459       __ bind(profile_method_continue);
1460     }
1461   }
1462 
1463   Label continue_after_compile;
1464   __ bind(continue_after_compile);
1465 
1466   // check for synchronized interpreted methods
1467   bang_stack_shadow_pages(false);
1468 
1469   // reset the _do_not_unlock_if_synchronized flag
1470   NOT_LP64(__ get_thread(thread));
1471   __ movbool(do_not_unlock_if_synchronized, false);
1472 
1473   // check for synchronized methods
1474   // Must happen AFTER invocation_counter check and stack overflow check,
1475   // so method is not locked if overflows.
1476   if (synchronized) {
1477     // Allocate monitor and lock method
1478     lock_method();
1479   } else {
1480     // no synchronization necessary
1481 #ifdef ASSERT
1482     {
1483       Label L;
1484       __ movl(rax, access_flags);
1485       __ testl(rax, JVM_ACC_SYNCHRONIZED);
1486       __ jcc(Assembler::zero, L);
1487       __ stop("method needs synchronization");
1488       __ bind(L);
1489     }
1490 #endif
1491   }
1492 
1493   // start execution
1494 #ifdef ASSERT
1495   {
1496     Label L;
1497      const Address monitor_block_top (rbp,
1498                  frame::interpreter_frame_monitor_block_top_offset * wordSize);
1499     __ movptr(rax, monitor_block_top);
1500     __ cmpptr(rax, rsp);
1501     __ jcc(Assembler::equal, L);
1502     __ stop("broken stack frame setup in interpreter");
1503     __ bind(L);
1504   }
1505 #endif
1506 
1507   // jvmti support
1508   __ notify_method_entry();
1509 
1510   __ dispatch_next(vtos);
1511 
1512   // invocation counter overflow
1513   if (inc_counter) {
1514     if (ProfileInterpreter) {
1515       // We have decided to profile this method in the interpreter
1516       __ bind(profile_method);
1517       __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1518       __ set_method_data_pointer_for_bcp();
1519       __ get_method(rbx);
1520       __ jmp(profile_method_continue);
1521     }
1522     // Handle overflow of counter and compile method
1523     __ bind(invocation_counter_overflow);
1524     generate_counter_overflow(continue_after_compile);
1525   }
1526 
1527   return entry_point;
1528 }
1529 
1530 //-----------------------------------------------------------------------------
1531 // Exceptions
1532 
1533 void TemplateInterpreterGenerator::generate_throw_exception() {
1534   // Entry point in previous activation (i.e., if the caller was
1535   // interpreted)
1536   Interpreter::_rethrow_exception_entry = __ pc();
1537   // Restore sp to interpreter_frame_last_sp even though we are going
1538   // to empty the expression stack for the exception processing.
1539   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
1540   // rax: exception
1541   // rdx: return address/pc that threw exception
1542   __ restore_bcp();    // r13/rsi points to call/send
1543   __ restore_locals();
1544   LP64_ONLY(__ reinit_heapbase());  // restore r12 as heapbase.
1545   // Entry point for exceptions thrown within interpreter code
1546   Interpreter::_throw_exception_entry = __ pc();
1547   // expression stack is undefined here
1548   // rax: exception
1549   // r13/rsi: exception bcp
1550   __ verify_oop(rax);
1551   Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
1552   LP64_ONLY(__ mov(c_rarg1, rax));
1553 
1554   // expression stack must be empty before entering the VM in case of
1555   // an exception
1556   __ empty_expression_stack();
1557   // find exception handler address and preserve exception oop
1558   __ call_VM(rdx,
1559              CAST_FROM_FN_PTR(address,
1560                           InterpreterRuntime::exception_handler_for_exception),
1561              rarg);
1562   // rax: exception handler entry point
1563   // rdx: preserved exception oop
1564   // r13/rsi: bcp for exception handler
1565   __ push_ptr(rdx); // push exception which is now the only value on the stack
1566   __ jmp(rax); // jump to exception handler (may be _remove_activation_entry!)
1567 
1568   // If the exception is not handled in the current frame the frame is
1569   // removed and the exception is rethrown (i.e. exception
1570   // continuation is _rethrow_exception).
1571   //
1572   // Note: At this point the bci is still the bxi for the instruction
1573   // which caused the exception and the expression stack is
1574   // empty. Thus, for any VM calls at this point, GC will find a legal
1575   // oop map (with empty expression stack).
1576 
1577   // In current activation
1578   // tos: exception
1579   // esi: exception bcp
1580 
1581   //
1582   // JVMTI PopFrame support
1583   //
1584 
1585   Interpreter::_remove_activation_preserving_args_entry = __ pc();
1586   __ empty_expression_stack();
1587   // Set the popframe_processing bit in pending_popframe_condition
1588   // indicating that we are currently handling popframe, so that
1589   // call_VMs that may happen later do not trigger new popframe
1590   // handling cycles.
1591   const Register thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
1592   NOT_LP64(__ get_thread(thread));
1593   __ movl(rdx, Address(thread, JavaThread::popframe_condition_offset()));
1594   __ orl(rdx, JavaThread::popframe_processing_bit);
1595   __ movl(Address(thread, JavaThread::popframe_condition_offset()), rdx);
1596 
1597   {
1598     // Check to see whether we are returning to a deoptimized frame.
1599     // (The PopFrame call ensures that the caller of the popped frame is
1600     // either interpreted or compiled and deoptimizes it if compiled.)
1601     // In this case, we can't call dispatch_next() after the frame is
1602     // popped, but instead must save the incoming arguments and restore
1603     // them after deoptimization has occurred.
1604     //
1605     // Note that we don't compare the return PC against the
1606     // deoptimization blob's unpack entry because of the presence of
1607     // adapter frames in C2.
1608     Label caller_not_deoptimized;
1609     Register rarg = NOT_LP64(rdx) LP64_ONLY(c_rarg1);
1610     __ movptr(rarg, Address(rbp, frame::return_addr_offset * wordSize));
1611     __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1612                                InterpreterRuntime::interpreter_contains), rarg);
1613     __ testl(rax, rax);
1614     __ jcc(Assembler::notZero, caller_not_deoptimized);
1615 
1616     // Compute size of arguments for saving when returning to
1617     // deoptimized caller
1618     __ get_method(rax);
1619     __ movptr(rax, Address(rax, Method::const_offset()));
1620     __ load_unsigned_short(rax, Address(rax, in_bytes(ConstMethod::
1621                                                 size_of_parameters_offset())));
1622     __ shll(rax, Interpreter::logStackElementSize);
1623     __ restore_locals();
1624     __ subptr(rlocals, rax);
1625     __ addptr(rlocals, wordSize);
1626     // Save these arguments
1627     NOT_LP64(__ get_thread(thread));
1628     __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1629                                            Deoptimization::
1630                                            popframe_preserve_args),
1631                           thread, rax, rlocals);
1632 
1633     __ remove_activation(vtos, rdx,
1634                          /* throw_monitor_exception */ false,
1635                          /* install_monitor_exception */ false,
1636                          /* notify_jvmdi */ false);
1637 
1638     // Inform deoptimization that it is responsible for restoring
1639     // these arguments
1640     NOT_LP64(__ get_thread(thread));
1641     __ movl(Address(thread, JavaThread::popframe_condition_offset()),
1642             JavaThread::popframe_force_deopt_reexecution_bit);
1643 
1644     // Continue in deoptimization handler
1645     __ jmp(rdx);
1646 
1647     __ bind(caller_not_deoptimized);
1648   }
1649 
1650   __ remove_activation(vtos, rdx, /* rdx result (retaddr) is not used */
1651                        /* throw_monitor_exception */ false,
1652                        /* install_monitor_exception */ false,
1653                        /* notify_jvmdi */ false);
1654 
1655   // Finish with popframe handling
1656   // A previous I2C followed by a deoptimization might have moved the
1657   // outgoing arguments further up the stack. PopFrame expects the
1658   // mutations to those outgoing arguments to be preserved and other
1659   // constraints basically require this frame to look exactly as
1660   // though it had previously invoked an interpreted activation with
1661   // no space between the top of the expression stack (current
1662   // last_sp) and the top of stack. Rather than force deopt to
1663   // maintain this kind of invariant all the time we call a small
1664   // fixup routine to move the mutated arguments onto the top of our
1665   // expression stack if necessary.
1666 #ifndef _LP64
1667   __ mov(rax, rsp);
1668   __ movptr(rbx, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1669   __ get_thread(thread);
1670   // PC must point into interpreter here
1671   __ set_last_Java_frame(thread, noreg, rbp, __ pc());
1672   __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), thread, rax, rbx);
1673   __ get_thread(thread);
1674 #else
1675   __ mov(c_rarg1, rsp);
1676   __ movptr(c_rarg2, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1677   // PC must point into interpreter here
1678   __ set_last_Java_frame(noreg, rbp, __ pc());
1679   __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), r15_thread, c_rarg1, c_rarg2);
1680 #endif
1681   __ reset_last_Java_frame(thread, true);
1682 
1683   // Restore the last_sp and null it out
1684   __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1685   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
1686 
1687   __ restore_bcp();
1688   __ restore_locals();
1689   // The method data pointer was incremented already during
1690   // call profiling. We have to restore the mdp for the current bcp.
1691   if (ProfileInterpreter) {
1692     __ set_method_data_pointer_for_bcp();
1693   }
1694 
1695   // Clear the popframe condition flag
1696   NOT_LP64(__ get_thread(thread));
1697   __ movl(Address(thread, JavaThread::popframe_condition_offset()),
1698           JavaThread::popframe_inactive);
1699 
1700 #if INCLUDE_JVMTI
1701   {
1702     Label L_done;
1703     const Register local0 = rlocals;
1704 
1705     __ cmpb(Address(rbcp, 0), Bytecodes::_invokestatic);
1706     __ jcc(Assembler::notEqual, L_done);
1707 
1708     // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call.
1709     // Detect such a case in the InterpreterRuntime function and return the member name argument, or NULL.
1710 
1711     __ get_method(rdx);
1712     __ movptr(rax, Address(local0, 0));
1713     __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), rax, rdx, rbcp);
1714 
1715     __ testptr(rax, rax);
1716     __ jcc(Assembler::zero, L_done);
1717 
1718     __ movptr(Address(rbx, 0), rax);
1719     __ bind(L_done);
1720   }
1721 #endif // INCLUDE_JVMTI
1722 
1723   __ dispatch_next(vtos);
1724   // end of PopFrame support
1725 
1726   Interpreter::_remove_activation_entry = __ pc();
1727 
1728   // preserve exception over this code sequence
1729   __ pop_ptr(rax);
1730   NOT_LP64(__ get_thread(thread));
1731   __ movptr(Address(thread, JavaThread::vm_result_offset()), rax);
1732   // remove the activation (without doing throws on illegalMonitorExceptions)
1733   __ remove_activation(vtos, rdx, false, true, false);
1734   // restore exception
1735   NOT_LP64(__ get_thread(thread));
1736   __ get_vm_result(rax, thread);
1737 
1738   // In between activations - previous activation type unknown yet
1739   // compute continuation point - the continuation point expects the
1740   // following registers set up:
1741   //
1742   // rax: exception
1743   // rdx: return address/pc that threw exception
1744   // rsp: expression stack of caller
1745   // rbp: ebp of caller
1746   __ push(rax);                                  // save exception
1747   __ push(rdx);                                  // save return address
1748   __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1749                           SharedRuntime::exception_handler_for_return_address),
1750                         thread, rdx);
1751   __ mov(rbx, rax);                              // save exception handler
1752   __ pop(rdx);                                   // restore return address
1753   __ pop(rax);                                   // restore exception
1754   // Note that an "issuing PC" is actually the next PC after the call
1755   __ jmp(rbx);                                   // jump to exception
1756                                                  // handler of caller
1757 }
1758 
1759 
1760 //
1761 // JVMTI ForceEarlyReturn support
1762 //
1763 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
1764   address entry = __ pc();
1765 
1766   __ restore_bcp();
1767   __ restore_locals();
1768   __ empty_expression_stack();
1769   __ load_earlyret_value(state);  // 32 bits returns value in rdx, so don't reuse
1770 
1771   const Register thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
1772   NOT_LP64(__ get_thread(thread));
1773   __ movptr(rcx, Address(thread, JavaThread::jvmti_thread_state_offset()));
1774   Address cond_addr(rcx, JvmtiThreadState::earlyret_state_offset());
1775 
1776   // Clear the earlyret state
1777   __ movl(cond_addr, JvmtiThreadState::earlyret_inactive);
1778 
1779   __ remove_activation(state, rsi,
1780                        false, /* throw_monitor_exception */
1781                        false, /* install_monitor_exception */
1782                        true); /* notify_jvmdi */
1783   __ jmp(rsi);
1784 
1785   return entry;
1786 } // end of ForceEarlyReturn support
1787 
1788 
1789 //-----------------------------------------------------------------------------
1790 // Helper for vtos entry point generation
1791 
1792 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
1793                                                          address& bep,
1794                                                          address& cep,
1795                                                          address& sep,
1796                                                          address& aep,
1797                                                          address& iep,
1798                                                          address& lep,
1799                                                          address& fep,
1800                                                          address& dep,
1801                                                          address& vep) {
1802   assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
1803   Label L;
1804   aep = __ pc();  __ push_ptr();   __ jmp(L);
1805 #ifndef _LP64
1806   fep = __ pc(); __ push(ftos); __ jmp(L);
1807   dep = __ pc(); __ push(dtos); __ jmp(L);
1808 #else
1809   fep = __ pc();  __ push_f(xmm0); __ jmp(L);
1810   dep = __ pc();  __ push_d(xmm0); __ jmp(L);
1811 #endif // _LP64
1812   lep = __ pc();  __ push_l();     __ jmp(L);
1813   bep = cep = sep =
1814   iep = __ pc();  __ push_i();
1815   vep = __ pc();
1816   __ bind(L);
1817   generate_and_dispatch(t);
1818 }
1819 
1820 //-----------------------------------------------------------------------------
1821 
1822 // Non-product code
1823 #ifndef PRODUCT
1824 
1825 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
1826   address entry = __ pc();
1827 
1828 #ifndef _LP64
1829   // prepare expression stack
1830   __ pop(rcx);          // pop return address so expression stack is 'pure'
1831   __ push(state);       // save tosca
1832 
1833   // pass tosca registers as arguments & call tracer
1834   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::trace_bytecode), rcx, rax, rdx);
1835   __ mov(rcx, rax);     // make sure return address is not destroyed by pop(state)
1836   __ pop(state);        // restore tosca
1837 
1838   // return
1839   __ jmp(rcx);
1840 #else
1841   __ push(state);
1842   __ push(c_rarg0);
1843   __ push(c_rarg1);
1844   __ push(c_rarg2);
1845   __ push(c_rarg3);
1846   __ mov(c_rarg2, rax);  // Pass itos
1847 #ifdef _WIN64
1848   __ movflt(xmm3, xmm0); // Pass ftos
1849 #endif
1850   __ call_VM(noreg,
1851              CAST_FROM_FN_PTR(address, InterpreterRuntime::trace_bytecode),
1852              c_rarg1, c_rarg2, c_rarg3);
1853   __ pop(c_rarg3);
1854   __ pop(c_rarg2);
1855   __ pop(c_rarg1);
1856   __ pop(c_rarg0);
1857   __ pop(state);
1858   __ ret(0);                                   // return from result handler
1859 #endif // _LP64
1860 
1861   return entry;
1862 }
1863 
1864 void TemplateInterpreterGenerator::count_bytecode() {
1865   __ incrementl(ExternalAddress((address) &BytecodeCounter::_counter_value));
1866 }
1867 
1868 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
1869   __ incrementl(ExternalAddress((address) &BytecodeHistogram::_counters[t->bytecode()]));
1870 }
1871 
1872 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
1873   __ mov32(rbx, ExternalAddress((address) &BytecodePairHistogram::_index));
1874   __ shrl(rbx, BytecodePairHistogram::log2_number_of_codes);
1875   __ orl(rbx,
1876          ((int) t->bytecode()) <<
1877          BytecodePairHistogram::log2_number_of_codes);
1878   __ mov32(ExternalAddress((address) &BytecodePairHistogram::_index), rbx);
1879   __ lea(rscratch1, ExternalAddress((address) BytecodePairHistogram::_counters));
1880   __ incrementl(Address(rscratch1, rbx, Address::times_4));
1881 }
1882 
1883 
1884 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
1885   // Call a little run-time stub to avoid blow-up for each bytecode.
1886   // The run-time runtime saves the right registers, depending on
1887   // the tosca in-state for the given template.
1888 
1889   assert(Interpreter::trace_code(t->tos_in()) != NULL,
1890          "entry must have been generated");
1891 #ifndef _LP64
1892   __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in())));
1893 #else
1894   __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1895   __ andptr(rsp, -16); // align stack as required by ABI
1896   __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in())));
1897   __ mov(rsp, r12); // restore sp
1898   __ reinit_heapbase();
1899 #endif // _LP64
1900 }
1901 
1902 
1903 void TemplateInterpreterGenerator::stop_interpreter_at() {
1904   Label L;
1905   __ cmp32(ExternalAddress((address) &BytecodeCounter::_counter_value),
1906            StopInterpreterAt);
1907   __ jcc(Assembler::notEqual, L);
1908   __ int3();
1909   __ bind(L);
1910 }
1911 #endif // !PRODUCT