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   }
 633 
 634   // add space for monitor & lock
 635   __ subptr(rsp, entry_size); // add space for a monitor entry
 636   __ movptr(monitor_block_top, rsp);  // set new monitor block top
 637   // store object
 638   __ movptr(Address(rsp, BasicObjectLock::obj_offset_in_bytes()), rax);
 639   const Register lockreg = NOT_LP64(rdx) LP64_ONLY(c_rarg1);
 640   __ movptr(lockreg, rsp); // object address
 641   __ lock_object(lockreg);
 642 }
 643 
 644 // Generate a fixed interpreter frame. This is identical setup for
 645 // interpreted methods and for native methods hence the shared code.
 646 //
 647 // Args:
 648 //      rax: return address
 649 //      rbx: Method*
 650 //      r14/rdi: pointer to locals
 651 //      r13/rsi: sender sp
 652 //      rdx: cp cache
 653 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
 654   // initialize fixed part of activation frame
 655   __ push(rax);        // save return address
 656   __ enter();          // save old & set new rbp
 657   __ push(rbcp);        // set sender sp
 658   __ push((int)NULL_WORD); // leave last_sp as null
 659   __ movptr(rbcp, Address(rbx, Method::const_offset()));      // get ConstMethod*
 660   __ lea(rbcp, Address(rbcp, ConstMethod::codes_offset())); // get codebase
 661   __ push(rbx);        // save Method*
 662   // Get mirror and store it in the frame as GC root for this Method*
 663   __ load_mirror(rdx, rbx);
 664   __ push(rdx);
 665   if (ProfileInterpreter) {
 666     Label method_data_continue;
 667     __ movptr(rdx, Address(rbx, in_bytes(Method::method_data_offset())));
 668     __ testptr(rdx, rdx);
 669     __ jcc(Assembler::zero, method_data_continue);
 670     __ addptr(rdx, in_bytes(MethodData::data_offset()));
 671     __ bind(method_data_continue);
 672     __ push(rdx);      // set the mdp (method data pointer)
 673   } else {
 674     __ push(0);
 675   }
 676 
 677   __ movptr(rdx, Address(rbx, Method::const_offset()));
 678   __ movptr(rdx, Address(rdx, ConstMethod::constants_offset()));
 679   __ movptr(rdx, Address(rdx, ConstantPool::cache_offset_in_bytes()));
 680   __ push(rdx); // set constant pool cache
 681   __ push(rlocals); // set locals pointer
 682   if (native_call) {
 683     __ push(0); // no bcp
 684   } else {
 685     __ push(rbcp); // set bcp
 686   }
 687   __ push(0); // reserve word for pointer to expression stack bottom
 688   __ movptr(Address(rsp, 0), rsp); // set expression stack bottom
 689 }
 690 
 691 // End of helpers
 692 
















































































 693 // Method entry for java.lang.ref.Reference.get.
 694 address TemplateInterpreterGenerator::generate_Reference_get_entry(void) {
 695 #if INCLUDE_ALL_GCS
 696   // Code: _aload_0, _getfield, _areturn
 697   // parameter size = 1
 698   //
 699   // The code that gets generated by this routine is split into 2 parts:
 700   //    1. The "intrinsified" code for G1 (or any SATB based GC),
 701   //    2. The slow path - which is an expansion of the regular method entry.
 702   //
 703   // Notes:-
 704   // * In the G1 code we do not check whether we need to block for
 705   //   a safepoint. If G1 is enabled then we must execute the specialized
 706   //   code for Reference.get (except when the Reference object is null)
 707   //   so that we can log the value in the referent field with an SATB
 708   //   update buffer.
 709   //   If the code for the getfield template is modified so that the
 710   //   G1 pre-barrier code is executed when the current method is
 711   //   Reference.get() then going through the normal method entry
 712   //   will be fine.
 713   // * The G1 code can, however, check the receiver object (the instance
 714   //   of java.lang.Reference) and jump to the slow path if null. If the
 715   //   Reference object is null then we obviously cannot fetch the referent
 716   //   and so we don't need to call the G1 pre-barrier. Thus we can use the
 717   //   regular method entry code to generate the NPE.
 718   //
 719   // rbx: Method*
 720 
 721   // r13: senderSP must preserve for slow path, set SP to it on fast path
 722 
 723   address entry = __ pc();
 724 
 725   const int referent_offset = java_lang_ref_Reference::referent_offset;
 726   guarantee(referent_offset > 0, "referent offset not initialized");
 727 
 728   if (UseG1GC) {
 729     Label slow_path;
 730     // rbx: method
 731 
 732     // Check if local 0 != NULL
 733     // If the receiver is null then it is OK to jump to the slow path.
 734     __ movptr(rax, Address(rsp, wordSize));
 735 
 736     __ testptr(rax, rax);
 737     __ jcc(Assembler::zero, slow_path);
 738 
 739     // rax: local 0
 740     // rbx: method (but can be used as scratch now)
 741     // rdx: scratch
 742     // rdi: scratch
 743 
 744     // Preserve the sender sp in case the pre-barrier
 745     // calls the runtime
 746     NOT_LP64(__ push(rsi));
 747 
 748     // Generate the G1 pre-barrier code to log the value of
 749     // the referent field in an SATB buffer.
 750 
 751     // Load the value of the referent field.
 752     const Address field_address(rax, referent_offset);
 753     __ load_heap_oop(rax, field_address);
 754 
 755     const Register sender_sp = NOT_LP64(rsi) LP64_ONLY(r13);
 756     const Register thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
 757     NOT_LP64(__ get_thread(thread));
 758 
 759     // Generate the G1 pre-barrier code to log the value of
 760     // the referent field in an SATB buffer.
 761     __ g1_write_barrier_pre(noreg /* obj */,
 762                             rax /* pre_val */,
 763                             thread /* thread */,
 764                             rbx /* tmp */,
 765                             true /* tosca_live */,
 766                             true /* expand_call */);
 767 
 768     // _areturn
 769     NOT_LP64(__ pop(rsi));      // get sender sp
 770     __ pop(rdi);                // get return address
 771     __ mov(rsp, sender_sp);     // set sp to sender sp
 772     __ jmp(rdi);
 773     __ ret(0);
 774 
 775     // generate a vanilla interpreter entry as the slow path
 776     __ bind(slow_path);
 777     __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::zerolocals));
 778     return entry;
 779   }
 780 #endif // INCLUDE_ALL_GCS
 781 
 782   // If G1 is not enabled then attempt to go through the accessor entry point
 783   // Reference.get is an accessor
 784   return NULL;
 785 }
 786 
 787 void TemplateInterpreterGenerator::bang_stack_shadow_pages(bool native_call) {
 788   // Quick & dirty stack overflow checking: bang the stack & handle trap.
 789   // Note that we do the banging after the frame is setup, since the exception
 790   // handling code expects to find a valid interpreter frame on the stack.
 791   // Doing the banging earlier fails if the caller frame is not an interpreter
 792   // frame.
 793   // (Also, the exception throwing code expects to unlock any synchronized
 794   // method receiever, so do the banging after locking the receiver.)
 795 
 796   // Bang each page in the shadow zone. We can't assume it's been done for
 797   // an interpreter frame with greater than a page of locals, so each page
 798   // needs to be checked.  Only true for non-native.
 799   if (UseStackBanging) {
 800     const int page_size = os::vm_page_size();
 801     const int n_shadow_pages = ((int)JavaThread::stack_shadow_zone_size()) / page_size;
 802     const int start_page = native_call ? n_shadow_pages : 1;
 803     for (int pages = start_page; pages <= n_shadow_pages; pages++) {
 804       __ bang_stack_with_offset(pages*page_size);
 805     }
 806   }
 807 }
 808 
 809 // Interpreter stub for calling a native method. (asm interpreter)
 810 // This sets up a somewhat different looking stack for calling the
 811 // native method than the typical interpreter frame setup.
 812 address TemplateInterpreterGenerator::generate_native_entry(bool synchronized) {
 813   // determine code generation flags
 814   bool inc_counter  = UseCompiler || CountCompiledCalls || LogTouchedMethods;
 815 
 816   // rbx: Method*
 817   // rbcp: sender sp
 818 
 819   address entry_point = __ pc();
 820 
 821   const Address constMethod       (rbx, Method::const_offset());
 822   const Address access_flags      (rbx, Method::access_flags_offset());
 823   const Address size_of_parameters(rcx, ConstMethod::
 824                                         size_of_parameters_offset());
 825 
 826 
 827   // get parameter size (always needed)
 828   __ movptr(rcx, constMethod);
 829   __ load_unsigned_short(rcx, size_of_parameters);
 830 
 831   // native calls don't need the stack size check since they have no
 832   // expression stack and the arguments are already on the stack and
 833   // we only add a handful of words to the stack
 834 
 835   // rbx: Method*
 836   // rcx: size of parameters
 837   // rbcp: sender sp
 838   __ pop(rax);                                       // get return address
 839 
 840   // for natives the size of locals is zero
 841 
 842   // compute beginning of parameters
 843   __ lea(rlocals, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize));
 844 
 845   // add 2 zero-initialized slots for native calls
 846   // initialize result_handler slot
 847   __ push((int) NULL_WORD);
 848   // slot for oop temp
 849   // (static native method holder mirror/jni oop result)
 850   __ push((int) NULL_WORD);
 851 
 852   // initialize fixed part of activation frame
 853   generate_fixed_frame(true);
 854 
 855   // make sure method is native & not abstract
 856 #ifdef ASSERT
 857   __ movl(rax, access_flags);
 858   {
 859     Label L;
 860     __ testl(rax, JVM_ACC_NATIVE);
 861     __ jcc(Assembler::notZero, L);
 862     __ stop("tried to execute non-native method as native");
 863     __ bind(L);
 864   }
 865   {
 866     Label L;
 867     __ testl(rax, JVM_ACC_ABSTRACT);
 868     __ jcc(Assembler::zero, L);
 869     __ stop("tried to execute abstract method in interpreter");
 870     __ bind(L);
 871   }
 872 #endif
 873 
 874   // Since at this point in the method invocation the exception handler
 875   // would try to exit the monitor of synchronized methods which hasn't
 876   // been entered yet, we set the thread local variable
 877   // _do_not_unlock_if_synchronized to true. The remove_activation will
 878   // check this flag.
 879 
 880   const Register thread1 = NOT_LP64(rax) LP64_ONLY(r15_thread);
 881   NOT_LP64(__ get_thread(thread1));
 882   const Address do_not_unlock_if_synchronized(thread1,
 883         in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
 884   __ movbool(do_not_unlock_if_synchronized, true);
 885 
 886   // increment invocation count & check for overflow
 887   Label invocation_counter_overflow;
 888   if (inc_counter) {
 889     generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
 890   }
 891 
 892   Label continue_after_compile;
 893   __ bind(continue_after_compile);
 894 
 895   bang_stack_shadow_pages(true);
 896 
 897   // reset the _do_not_unlock_if_synchronized flag
 898   NOT_LP64(__ get_thread(thread1));
 899   __ movbool(do_not_unlock_if_synchronized, false);
 900 
 901   // check for synchronized methods
 902   // Must happen AFTER invocation_counter check and stack overflow check,
 903   // so method is not locked if overflows.
 904   if (synchronized) {
 905     lock_method();
 906   } else {
 907     // no synchronization necessary
 908 #ifdef ASSERT
 909     {
 910       Label L;
 911       __ movl(rax, access_flags);
 912       __ testl(rax, JVM_ACC_SYNCHRONIZED);
 913       __ jcc(Assembler::zero, L);
 914       __ stop("method needs synchronization");
 915       __ bind(L);
 916     }
 917 #endif
 918   }
 919 
 920   // start execution
 921 #ifdef ASSERT
 922   {
 923     Label L;
 924     const Address monitor_block_top(rbp,
 925                  frame::interpreter_frame_monitor_block_top_offset * wordSize);
 926     __ movptr(rax, monitor_block_top);
 927     __ cmpptr(rax, rsp);
 928     __ jcc(Assembler::equal, L);
 929     __ stop("broken stack frame setup in interpreter");
 930     __ bind(L);
 931   }
 932 #endif
 933 
 934   // jvmti support
 935   __ notify_method_entry();
 936 
 937   // work registers
 938   const Register method = rbx;
 939   const Register thread = NOT_LP64(rdi) LP64_ONLY(r15_thread);
 940   const Register t      = NOT_LP64(rcx) LP64_ONLY(r11);
 941 
 942   // allocate space for parameters
 943   __ get_method(method);
 944   __ movptr(t, Address(method, Method::const_offset()));
 945   __ load_unsigned_short(t, Address(t, ConstMethod::size_of_parameters_offset()));
 946 
 947 #ifndef _LP64
 948   __ shlptr(t, Interpreter::logStackElementSize); // Convert parameter count to bytes.
 949   __ addptr(t, 2*wordSize);     // allocate two more slots for JNIEnv and possible mirror
 950   __ subptr(rsp, t);
 951   __ andptr(rsp, -(StackAlignmentInBytes)); // gcc needs 16 byte aligned stacks to do XMM intrinsics
 952 #else
 953   __ shll(t, Interpreter::logStackElementSize);
 954 
 955   __ subptr(rsp, t);
 956   __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
 957   __ andptr(rsp, -16); // must be 16 byte boundary (see amd64 ABI)
 958 #endif // _LP64
 959 
 960   // get signature handler
 961   {
 962     Label L;
 963     __ movptr(t, Address(method, Method::signature_handler_offset()));
 964     __ testptr(t, t);
 965     __ jcc(Assembler::notZero, L);
 966     __ call_VM(noreg,
 967                CAST_FROM_FN_PTR(address,
 968                                 InterpreterRuntime::prepare_native_call),
 969                method);
 970     __ get_method(method);
 971     __ movptr(t, Address(method, Method::signature_handler_offset()));
 972     __ bind(L);
 973   }
 974 
 975   // call signature handler
 976   assert(InterpreterRuntime::SignatureHandlerGenerator::from() == rlocals,
 977          "adjust this code");
 978   assert(InterpreterRuntime::SignatureHandlerGenerator::to() == rsp,
 979          "adjust this code");
 980   assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == NOT_LP64(t) LP64_ONLY(rscratch1),
 981          "adjust this code");
 982 
 983   // The generated handlers do not touch RBX (the method oop).
 984   // However, large signatures cannot be cached and are generated
 985   // each time here.  The slow-path generator can do a GC on return,
 986   // so we must reload it after the call.
 987   __ call(t);
 988   __ get_method(method);        // slow path can do a GC, reload RBX
 989 
 990 
 991   // result handler is in rax
 992   // set result handler
 993   __ movptr(Address(rbp,
 994                     (frame::interpreter_frame_result_handler_offset) * wordSize),
 995             rax);
 996 
 997   // pass mirror handle if static call
 998   {
 999     Label L;
1000     __ movl(t, Address(method, Method::access_flags_offset()));
1001     __ testl(t, JVM_ACC_STATIC);
1002     __ jcc(Assembler::zero, L);
1003     // get mirror
1004     __ load_mirror(t, method);
1005     // copy mirror into activation frame
1006     __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize),
1007             t);
1008     // pass handle to mirror
1009 #ifndef _LP64
1010     __ lea(t, Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize));
1011     __ movptr(Address(rsp, wordSize), t);
1012 #else
1013     __ lea(c_rarg1,
1014            Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize));
1015 #endif // _LP64
1016     __ bind(L);
1017   }
1018 
1019   // get native function entry point
1020   {
1021     Label L;
1022     __ movptr(rax, Address(method, Method::native_function_offset()));
1023     ExternalAddress unsatisfied(SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
1024     __ cmpptr(rax, unsatisfied.addr());
1025     __ jcc(Assembler::notEqual, L);
1026     __ call_VM(noreg,
1027                CAST_FROM_FN_PTR(address,
1028                                 InterpreterRuntime::prepare_native_call),
1029                method);
1030     __ get_method(method);
1031     __ movptr(rax, Address(method, Method::native_function_offset()));
1032     __ bind(L);
1033   }
1034 
1035   // pass JNIEnv
1036 #ifndef _LP64
1037    __ get_thread(thread);
1038    __ lea(t, Address(thread, JavaThread::jni_environment_offset()));
1039    __ movptr(Address(rsp, 0), t);
1040 
1041    // set_last_Java_frame_before_call
1042    // It is enough that the pc()
1043    // points into the right code segment. It does not have to be the correct return pc.
1044    __ set_last_Java_frame(thread, noreg, rbp, __ pc());
1045 #else
1046    __ lea(c_rarg0, Address(r15_thread, JavaThread::jni_environment_offset()));
1047 
1048    // It is enough that the pc() points into the right code
1049    // segment. It does not have to be the correct return pc.
1050    __ set_last_Java_frame(rsp, rbp, (address) __ pc());
1051 #endif // _LP64
1052 
1053   // change thread state
1054 #ifdef ASSERT
1055   {
1056     Label L;
1057     __ movl(t, Address(thread, JavaThread::thread_state_offset()));
1058     __ cmpl(t, _thread_in_Java);
1059     __ jcc(Assembler::equal, L);
1060     __ stop("Wrong thread state in native stub");
1061     __ bind(L);
1062   }
1063 #endif
1064 
1065   // Change state to native
1066 
1067   __ movl(Address(thread, JavaThread::thread_state_offset()),
1068           _thread_in_native);
1069 
1070   // Call the native method.
1071   __ call(rax);
1072   // 32: result potentially in rdx:rax or ST0
1073   // 64: result potentially in rax or xmm0
1074 
1075   // Verify or restore cpu control state after JNI call
1076   __ restore_cpu_control_state_after_jni();
1077 
1078   // NOTE: The order of these pushes is known to frame::interpreter_frame_result
1079   // in order to extract the result of a method call. If the order of these
1080   // pushes change or anything else is added to the stack then the code in
1081   // interpreter_frame_result must also change.
1082 
1083 #ifndef _LP64
1084   // save potential result in ST(0) & rdx:rax
1085   // (if result handler is the T_FLOAT or T_DOUBLE handler, result must be in ST0 -
1086   // the check is necessary to avoid potential Intel FPU overflow problems by saving/restoring 'empty' FPU registers)
1087   // It is safe to do this push because state is _thread_in_native and return address will be found
1088   // via _last_native_pc and not via _last_jave_sp
1089 
1090   // NOTE: the order of theses push(es) is known to frame::interpreter_frame_result.
1091   // If the order changes or anything else is added to the stack the code in
1092   // interpreter_frame_result will have to be changed.
1093 
1094   { Label L;
1095     Label push_double;
1096     ExternalAddress float_handler(AbstractInterpreter::result_handler(T_FLOAT));
1097     ExternalAddress double_handler(AbstractInterpreter::result_handler(T_DOUBLE));
1098     __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize),
1099               float_handler.addr());
1100     __ jcc(Assembler::equal, push_double);
1101     __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize),
1102               double_handler.addr());
1103     __ jcc(Assembler::notEqual, L);
1104     __ bind(push_double);
1105     __ push_d(); // FP values are returned using the FPU, so push FPU contents (even if UseSSE > 0).
1106     __ bind(L);
1107   }
1108 #else
1109   __ push(dtos);
1110 #endif // _LP64
1111 
1112   __ push(ltos);
1113 
1114   // change thread state
1115   NOT_LP64(__ get_thread(thread));
1116   __ movl(Address(thread, JavaThread::thread_state_offset()),
1117           _thread_in_native_trans);
1118 
1119   if (os::is_MP()) {
1120     if (UseMembar) {
1121       // Force this write out before the read below
1122       __ membar(Assembler::Membar_mask_bits(
1123            Assembler::LoadLoad | Assembler::LoadStore |
1124            Assembler::StoreLoad | Assembler::StoreStore));
1125     } else {
1126       // Write serialization page so VM thread can do a pseudo remote membar.
1127       // We use the current thread pointer to calculate a thread specific
1128       // offset to write to within the page. This minimizes bus traffic
1129       // due to cache line collision.
1130       __ serialize_memory(thread, rcx);
1131     }
1132   }
1133 
1134 #ifndef _LP64
1135   if (AlwaysRestoreFPU) {
1136     //  Make sure the control word is correct.
1137     __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
1138   }
1139 #endif // _LP64
1140 
1141   // check for safepoint operation in progress and/or pending suspend requests
1142   {
1143     Label Continue;
1144     __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
1145              SafepointSynchronize::_not_synchronized);
1146 
1147     Label L;
1148     __ jcc(Assembler::notEqual, L);
1149     __ cmpl(Address(thread, JavaThread::suspend_flags_offset()), 0);
1150     __ jcc(Assembler::equal, Continue);
1151     __ bind(L);
1152 
1153     // Don't use call_VM as it will see a possible pending exception
1154     // and forward it and never return here preventing us from
1155     // clearing _last_native_pc down below.  Also can't use
1156     // call_VM_leaf either as it will check to see if r13 & r14 are
1157     // preserved and correspond to the bcp/locals pointers. So we do a
1158     // runtime call by hand.
1159     //
1160 #ifndef _LP64
1161     __ push(thread);
1162     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address,
1163                                             JavaThread::check_special_condition_for_native_trans)));
1164     __ increment(rsp, wordSize);
1165     __ get_thread(thread);
1166 #else
1167     __ mov(c_rarg0, r15_thread);
1168     __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1169     __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1170     __ andptr(rsp, -16); // align stack as required by ABI
1171     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans)));
1172     __ mov(rsp, r12); // restore sp
1173     __ reinit_heapbase();
1174 #endif // _LP64
1175     __ bind(Continue);
1176   }
1177 
1178   // change thread state
1179   __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_Java);
1180 
1181   // reset_last_Java_frame
1182   __ reset_last_Java_frame(thread, true);
1183 
1184   if (CheckJNICalls) {
1185     // clear_pending_jni_exception_check
1186     __ movptr(Address(thread, JavaThread::pending_jni_exception_check_fn_offset()), NULL_WORD);
1187   }
1188 
1189   // reset handle block
1190   __ movptr(t, Address(thread, JavaThread::active_handles_offset()));
1191   __ movl(Address(t, JNIHandleBlock::top_offset_in_bytes()), (int32_t)NULL_WORD);
1192 
1193   // If result is an oop unbox and store it in frame where gc will see it
1194   // and result handler will pick it up
1195 
1196   {
1197     Label no_oop, not_weak, store_result;
1198     __ lea(t, ExternalAddress(AbstractInterpreter::result_handler(T_OBJECT)));
1199     __ cmpptr(t, Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize));
1200     __ jcc(Assembler::notEqual, no_oop);
1201     // retrieve result
1202     __ pop(ltos);
1203     // Unbox oop result, e.g. JNIHandles::resolve value.
1204     __ resolve_jobject(rax /* value */,
1205                        thread /* thread */,
1206                        t /* tmp */);
1207     __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize), rax);
1208     // keep stack depth as expected by pushing oop which will eventually be discarded
1209     __ push(ltos);
1210     __ bind(no_oop);
1211   }
1212 
1213 
1214   {
1215     Label no_reguard;
1216     __ cmpl(Address(thread, JavaThread::stack_guard_state_offset()),
1217             JavaThread::stack_guard_yellow_reserved_disabled);
1218     __ jcc(Assembler::notEqual, no_reguard);
1219 
1220     __ pusha(); // XXX only save smashed registers
1221 #ifndef _LP64
1222     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
1223     __ popa();
1224 #else
1225     __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1226     __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1227     __ andptr(rsp, -16); // align stack as required by ABI
1228     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
1229     __ mov(rsp, r12); // restore sp
1230     __ popa(); // XXX only restore smashed registers
1231     __ reinit_heapbase();
1232 #endif // _LP64
1233 
1234     __ bind(no_reguard);
1235   }
1236 
1237 
1238   // The method register is junk from after the thread_in_native transition
1239   // until here.  Also can't call_VM until the bcp has been
1240   // restored.  Need bcp for throwing exception below so get it now.
1241   __ get_method(method);
1242 
1243   // restore to have legal interpreter frame, i.e., bci == 0 <=> code_base()
1244   __ movptr(rbcp, Address(method, Method::const_offset()));   // get ConstMethod*
1245   __ lea(rbcp, Address(rbcp, ConstMethod::codes_offset()));    // get codebase
1246 
1247   // handle exceptions (exception handling will handle unlocking!)
1248   {
1249     Label L;
1250     __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t) NULL_WORD);
1251     __ jcc(Assembler::zero, L);
1252     // Note: At some point we may want to unify this with the code
1253     // used in call_VM_base(); i.e., we should use the
1254     // StubRoutines::forward_exception code. For now this doesn't work
1255     // here because the rsp is not correctly set at this point.
1256     __ MacroAssembler::call_VM(noreg,
1257                                CAST_FROM_FN_PTR(address,
1258                                InterpreterRuntime::throw_pending_exception));
1259     __ should_not_reach_here();
1260     __ bind(L);
1261   }
1262 
1263   // do unlocking if necessary
1264   {
1265     Label L;
1266     __ movl(t, Address(method, Method::access_flags_offset()));
1267     __ testl(t, JVM_ACC_SYNCHRONIZED);
1268     __ jcc(Assembler::zero, L);
1269     // the code below should be shared with interpreter macro
1270     // assembler implementation
1271     {
1272       Label unlock;
1273       // BasicObjectLock will be first in list, since this is a
1274       // synchronized method. However, need to check that the object
1275       // has not been unlocked by an explicit monitorexit bytecode.
1276       const Address monitor(rbp,
1277                             (intptr_t)(frame::interpreter_frame_initial_sp_offset *
1278                                        wordSize - (int)sizeof(BasicObjectLock)));
1279 
1280       const Register regmon = NOT_LP64(rdx) LP64_ONLY(c_rarg1);
1281 
1282       // monitor expect in c_rarg1 for slow unlock path
1283       __ lea(regmon, monitor); // address of first monitor
1284 
1285       __ movptr(t, Address(regmon, BasicObjectLock::obj_offset_in_bytes()));
1286       __ testptr(t, t);
1287       __ jcc(Assembler::notZero, unlock);
1288 
1289       // Entry already unlocked, need to throw exception
1290       __ MacroAssembler::call_VM(noreg,
1291                                  CAST_FROM_FN_PTR(address,
1292                    InterpreterRuntime::throw_illegal_monitor_state_exception));
1293       __ should_not_reach_here();
1294 
1295       __ bind(unlock);
1296       __ unlock_object(regmon);
1297     }
1298     __ bind(L);
1299   }
1300 
1301   // jvmti support
1302   // Note: This must happen _after_ handling/throwing any exceptions since
1303   //       the exception handler code notifies the runtime of method exits
1304   //       too. If this happens before, method entry/exit notifications are
1305   //       not properly paired (was bug - gri 11/22/99).
1306   __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI);
1307 
1308   // restore potential result in edx:eax, call result handler to
1309   // restore potential result in ST0 & handle result
1310 
1311   __ pop(ltos);
1312   LP64_ONLY( __ pop(dtos));
1313 
1314   __ movptr(t, Address(rbp,
1315                        (frame::interpreter_frame_result_handler_offset) * wordSize));
1316   __ call(t);
1317 
1318   // remove activation
1319   __ movptr(t, Address(rbp,
1320                        frame::interpreter_frame_sender_sp_offset *
1321                        wordSize)); // get sender sp
1322   __ leave();                                // remove frame anchor
1323   __ pop(rdi);                               // get return address
1324   __ mov(rsp, t);                            // set sp to sender sp
1325   __ jmp(rdi);
1326 
1327   if (inc_counter) {
1328     // Handle overflow of counter and compile method
1329     __ bind(invocation_counter_overflow);
1330     generate_counter_overflow(continue_after_compile);
1331   }
1332 
1333   return entry_point;
1334 }
1335 
1336 // Abstract method entry
1337 // Attempt to execute abstract method. Throw exception
1338 address TemplateInterpreterGenerator::generate_abstract_entry(void) {
1339 
1340   address entry_point = __ pc();
1341 
1342   // abstract method entry
1343 
1344   //  pop return address, reset last_sp to NULL
1345   __ empty_expression_stack();
1346   __ restore_bcp();      // rsi must be correct for exception handler   (was destroyed)
1347   __ restore_locals();   // make sure locals pointer is correct as well (was destroyed)
1348 
1349   // throw exception
1350   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError));
1351   // the call_VM checks for exception, so we should never return here.
1352   __ should_not_reach_here();
1353 
1354   return entry_point;
1355 }
1356 
1357 //
1358 // Generic interpreted method entry to (asm) interpreter
1359 //
1360 address TemplateInterpreterGenerator::generate_normal_entry(bool synchronized) {
1361   // determine code generation flags
1362   bool inc_counter  = UseCompiler || CountCompiledCalls || LogTouchedMethods;
1363 
1364   // ebx: Method*
1365   // rbcp: sender sp
1366   address entry_point = __ pc();
1367 
1368   const Address constMethod(rbx, Method::const_offset());
1369   const Address access_flags(rbx, Method::access_flags_offset());
1370   const Address size_of_parameters(rdx,
1371                                    ConstMethod::size_of_parameters_offset());
1372   const Address size_of_locals(rdx, ConstMethod::size_of_locals_offset());
1373 
1374 
1375   // get parameter size (always needed)
1376   __ movptr(rdx, constMethod);
1377   __ load_unsigned_short(rcx, size_of_parameters);
1378 
1379   // rbx: Method*
1380   // rcx: size of parameters
1381   // rbcp: sender_sp (could differ from sp+wordSize if we were called via c2i )
1382 
1383   __ load_unsigned_short(rdx, size_of_locals); // get size of locals in words
1384   __ subl(rdx, rcx); // rdx = no. of additional locals
1385 
1386   // YYY
1387 //   __ incrementl(rdx);
1388 //   __ andl(rdx, -2);
1389 
1390   // see if we've got enough room on the stack for locals plus overhead.
1391   generate_stack_overflow_check();
1392 
1393   // get return address
1394   __ pop(rax);
1395 
1396   // compute beginning of parameters
1397   __ lea(rlocals, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize));
1398 
1399   // rdx - # of additional locals
1400   // allocate space for locals
1401   // explicitly initialize locals
1402   {
1403     Label exit, loop;
1404     __ testl(rdx, rdx);
1405     __ jcc(Assembler::lessEqual, exit); // do nothing if rdx <= 0
1406     __ bind(loop);
1407     __ push((int) NULL_WORD); // initialize local variables
1408     __ decrementl(rdx); // until everything initialized
1409     __ jcc(Assembler::greater, loop);
1410     __ bind(exit);
1411   }
1412 
1413   // initialize fixed part of activation frame
1414   generate_fixed_frame(false);
1415 
1416   // make sure method is not native & not abstract
1417 #ifdef ASSERT
1418   __ movl(rax, access_flags);
1419   {
1420     Label L;
1421     __ testl(rax, JVM_ACC_NATIVE);
1422     __ jcc(Assembler::zero, L);
1423     __ stop("tried to execute native method as non-native");
1424     __ bind(L);
1425   }
1426   {
1427     Label L;
1428     __ testl(rax, JVM_ACC_ABSTRACT);
1429     __ jcc(Assembler::zero, L);
1430     __ stop("tried to execute abstract method in interpreter");
1431     __ bind(L);
1432   }
1433 #endif
1434 
1435   // Since at this point in the method invocation the exception
1436   // handler would try to exit the monitor of synchronized methods
1437   // which hasn't been entered yet, we set the thread local variable
1438   // _do_not_unlock_if_synchronized to true. The remove_activation
1439   // will check this flag.
1440 
1441   const Register thread = NOT_LP64(rax) LP64_ONLY(r15_thread);
1442   NOT_LP64(__ get_thread(thread));
1443   const Address do_not_unlock_if_synchronized(thread,
1444         in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1445   __ movbool(do_not_unlock_if_synchronized, true);
1446 
1447   __ profile_parameters_type(rax, rcx, rdx);
1448   // increment invocation count & check for overflow
1449   Label invocation_counter_overflow;
1450   Label profile_method;
1451   Label profile_method_continue;
1452   if (inc_counter) {
1453     generate_counter_incr(&invocation_counter_overflow,
1454                           &profile_method,
1455                           &profile_method_continue);
1456     if (ProfileInterpreter) {
1457       __ bind(profile_method_continue);
1458     }
1459   }
1460 
1461   Label continue_after_compile;
1462   __ bind(continue_after_compile);
1463 
1464   // check for synchronized interpreted methods
1465   bang_stack_shadow_pages(false);
1466 
1467   // reset the _do_not_unlock_if_synchronized flag
1468   NOT_LP64(__ get_thread(thread));
1469   __ movbool(do_not_unlock_if_synchronized, false);
1470 
1471   // check for synchronized methods
1472   // Must happen AFTER invocation_counter check and stack overflow check,
1473   // so method is not locked if overflows.
1474   if (synchronized) {
1475     // Allocate monitor and lock method
1476     lock_method();
1477   } else {
1478     // no synchronization necessary
1479 #ifdef ASSERT
1480     {
1481       Label L;
1482       __ movl(rax, access_flags);
1483       __ testl(rax, JVM_ACC_SYNCHRONIZED);
1484       __ jcc(Assembler::zero, L);
1485       __ stop("method needs synchronization");
1486       __ bind(L);
1487     }
1488 #endif
1489   }
1490 
1491   // start execution
1492 #ifdef ASSERT
1493   {
1494     Label L;
1495      const Address monitor_block_top (rbp,
1496                  frame::interpreter_frame_monitor_block_top_offset * wordSize);
1497     __ movptr(rax, monitor_block_top);
1498     __ cmpptr(rax, rsp);
1499     __ jcc(Assembler::equal, L);
1500     __ stop("broken stack frame setup in interpreter");
1501     __ bind(L);
1502   }
1503 #endif
1504 
1505   // jvmti support
1506   __ notify_method_entry();
1507 
1508   __ dispatch_next(vtos);
1509 
1510   // invocation counter overflow
1511   if (inc_counter) {
1512     if (ProfileInterpreter) {
1513       // We have decided to profile this method in the interpreter
1514       __ bind(profile_method);
1515       __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1516       __ set_method_data_pointer_for_bcp();
1517       __ get_method(rbx);
1518       __ jmp(profile_method_continue);
1519     }
1520     // Handle overflow of counter and compile method
1521     __ bind(invocation_counter_overflow);
1522     generate_counter_overflow(continue_after_compile);
1523   }
1524 
1525   return entry_point;
1526 }
1527 
1528 //-----------------------------------------------------------------------------
1529 // Exceptions
1530 
1531 void TemplateInterpreterGenerator::generate_throw_exception() {
1532   // Entry point in previous activation (i.e., if the caller was
1533   // interpreted)
1534   Interpreter::_rethrow_exception_entry = __ pc();
1535   // Restore sp to interpreter_frame_last_sp even though we are going
1536   // to empty the expression stack for the exception processing.
1537   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
1538   // rax: exception
1539   // rdx: return address/pc that threw exception
1540   __ restore_bcp();    // r13/rsi points to call/send
1541   __ restore_locals();
1542   LP64_ONLY(__ reinit_heapbase());  // restore r12 as heapbase.
1543   // Entry point for exceptions thrown within interpreter code
1544   Interpreter::_throw_exception_entry = __ pc();
1545   // expression stack is undefined here
1546   // rax: exception
1547   // r13/rsi: exception bcp
1548   __ verify_oop(rax);
1549   Register rarg = NOT_LP64(rax) LP64_ONLY(c_rarg1);
1550   LP64_ONLY(__ mov(c_rarg1, rax));
1551 
1552   // expression stack must be empty before entering the VM in case of
1553   // an exception
1554   __ empty_expression_stack();
1555   // find exception handler address and preserve exception oop
1556   __ call_VM(rdx,
1557              CAST_FROM_FN_PTR(address,
1558                           InterpreterRuntime::exception_handler_for_exception),
1559              rarg);
1560   // rax: exception handler entry point
1561   // rdx: preserved exception oop
1562   // r13/rsi: bcp for exception handler
1563   __ push_ptr(rdx); // push exception which is now the only value on the stack
1564   __ jmp(rax); // jump to exception handler (may be _remove_activation_entry!)
1565 
1566   // If the exception is not handled in the current frame the frame is
1567   // removed and the exception is rethrown (i.e. exception
1568   // continuation is _rethrow_exception).
1569   //
1570   // Note: At this point the bci is still the bxi for the instruction
1571   // which caused the exception and the expression stack is
1572   // empty. Thus, for any VM calls at this point, GC will find a legal
1573   // oop map (with empty expression stack).
1574 
1575   // In current activation
1576   // tos: exception
1577   // esi: exception bcp
1578 
1579   //
1580   // JVMTI PopFrame support
1581   //
1582 
1583   Interpreter::_remove_activation_preserving_args_entry = __ pc();
1584   __ empty_expression_stack();
1585   // Set the popframe_processing bit in pending_popframe_condition
1586   // indicating that we are currently handling popframe, so that
1587   // call_VMs that may happen later do not trigger new popframe
1588   // handling cycles.
1589   const Register thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
1590   NOT_LP64(__ get_thread(thread));
1591   __ movl(rdx, Address(thread, JavaThread::popframe_condition_offset()));
1592   __ orl(rdx, JavaThread::popframe_processing_bit);
1593   __ movl(Address(thread, JavaThread::popframe_condition_offset()), rdx);
1594 
1595   {
1596     // Check to see whether we are returning to a deoptimized frame.
1597     // (The PopFrame call ensures that the caller of the popped frame is
1598     // either interpreted or compiled and deoptimizes it if compiled.)
1599     // In this case, we can't call dispatch_next() after the frame is
1600     // popped, but instead must save the incoming arguments and restore
1601     // them after deoptimization has occurred.
1602     //
1603     // Note that we don't compare the return PC against the
1604     // deoptimization blob's unpack entry because of the presence of
1605     // adapter frames in C2.
1606     Label caller_not_deoptimized;
1607     Register rarg = NOT_LP64(rdx) LP64_ONLY(c_rarg1);
1608     __ movptr(rarg, Address(rbp, frame::return_addr_offset * wordSize));
1609     __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1610                                InterpreterRuntime::interpreter_contains), rarg);
1611     __ testl(rax, rax);
1612     __ jcc(Assembler::notZero, caller_not_deoptimized);
1613 
1614     // Compute size of arguments for saving when returning to
1615     // deoptimized caller
1616     __ get_method(rax);
1617     __ movptr(rax, Address(rax, Method::const_offset()));
1618     __ load_unsigned_short(rax, Address(rax, in_bytes(ConstMethod::
1619                                                 size_of_parameters_offset())));
1620     __ shll(rax, Interpreter::logStackElementSize);
1621     __ restore_locals();
1622     __ subptr(rlocals, rax);
1623     __ addptr(rlocals, wordSize);
1624     // Save these arguments
1625     NOT_LP64(__ get_thread(thread));
1626     __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1627                                            Deoptimization::
1628                                            popframe_preserve_args),
1629                           thread, rax, rlocals);
1630 
1631     __ remove_activation(vtos, rdx,
1632                          /* throw_monitor_exception */ false,
1633                          /* install_monitor_exception */ false,
1634                          /* notify_jvmdi */ false);
1635 
1636     // Inform deoptimization that it is responsible for restoring
1637     // these arguments
1638     NOT_LP64(__ get_thread(thread));
1639     __ movl(Address(thread, JavaThread::popframe_condition_offset()),
1640             JavaThread::popframe_force_deopt_reexecution_bit);
1641 
1642     // Continue in deoptimization handler
1643     __ jmp(rdx);
1644 
1645     __ bind(caller_not_deoptimized);
1646   }
1647 
1648   __ remove_activation(vtos, rdx, /* rdx result (retaddr) is not used */
1649                        /* throw_monitor_exception */ false,
1650                        /* install_monitor_exception */ false,
1651                        /* notify_jvmdi */ false);
1652 
1653   // Finish with popframe handling
1654   // A previous I2C followed by a deoptimization might have moved the
1655   // outgoing arguments further up the stack. PopFrame expects the
1656   // mutations to those outgoing arguments to be preserved and other
1657   // constraints basically require this frame to look exactly as
1658   // though it had previously invoked an interpreted activation with
1659   // no space between the top of the expression stack (current
1660   // last_sp) and the top of stack. Rather than force deopt to
1661   // maintain this kind of invariant all the time we call a small
1662   // fixup routine to move the mutated arguments onto the top of our
1663   // expression stack if necessary.
1664 #ifndef _LP64
1665   __ mov(rax, rsp);
1666   __ movptr(rbx, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1667   __ get_thread(thread);
1668   // PC must point into interpreter here
1669   __ set_last_Java_frame(thread, noreg, rbp, __ pc());
1670   __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), thread, rax, rbx);
1671   __ get_thread(thread);
1672 #else
1673   __ mov(c_rarg1, rsp);
1674   __ movptr(c_rarg2, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1675   // PC must point into interpreter here
1676   __ set_last_Java_frame(noreg, rbp, __ pc());
1677   __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), r15_thread, c_rarg1, c_rarg2);
1678 #endif
1679   __ reset_last_Java_frame(thread, true);
1680 
1681   // Restore the last_sp and null it out
1682   __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1683   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
1684 
1685   __ restore_bcp();
1686   __ restore_locals();
1687   // The method data pointer was incremented already during
1688   // call profiling. We have to restore the mdp for the current bcp.
1689   if (ProfileInterpreter) {
1690     __ set_method_data_pointer_for_bcp();
1691   }
1692 
1693   // Clear the popframe condition flag
1694   NOT_LP64(__ get_thread(thread));
1695   __ movl(Address(thread, JavaThread::popframe_condition_offset()),
1696           JavaThread::popframe_inactive);
1697 
1698 #if INCLUDE_JVMTI
1699   {
1700     Label L_done;
1701     const Register local0 = rlocals;
1702 
1703     __ cmpb(Address(rbcp, 0), Bytecodes::_invokestatic);
1704     __ jcc(Assembler::notEqual, L_done);
1705 
1706     // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call.
1707     // Detect such a case in the InterpreterRuntime function and return the member name argument, or NULL.
1708 
1709     __ get_method(rdx);
1710     __ movptr(rax, Address(local0, 0));
1711     __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), rax, rdx, rbcp);
1712 
1713     __ testptr(rax, rax);
1714     __ jcc(Assembler::zero, L_done);
1715 
1716     __ movptr(Address(rbx, 0), rax);
1717     __ bind(L_done);
1718   }
1719 #endif // INCLUDE_JVMTI
1720 
1721   __ dispatch_next(vtos);
1722   // end of PopFrame support
1723 
1724   Interpreter::_remove_activation_entry = __ pc();
1725 
1726   // preserve exception over this code sequence
1727   __ pop_ptr(rax);
1728   NOT_LP64(__ get_thread(thread));
1729   __ movptr(Address(thread, JavaThread::vm_result_offset()), rax);
1730   // remove the activation (without doing throws on illegalMonitorExceptions)
1731   __ remove_activation(vtos, rdx, false, true, false);
1732   // restore exception
1733   NOT_LP64(__ get_thread(thread));
1734   __ get_vm_result(rax, thread);
1735 
1736   // In between activations - previous activation type unknown yet
1737   // compute continuation point - the continuation point expects the
1738   // following registers set up:
1739   //
1740   // rax: exception
1741   // rdx: return address/pc that threw exception
1742   // rsp: expression stack of caller
1743   // rbp: ebp of caller
1744   __ push(rax);                                  // save exception
1745   __ push(rdx);                                  // save return address
1746   __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1747                           SharedRuntime::exception_handler_for_return_address),
1748                         thread, rdx);
1749   __ mov(rbx, rax);                              // save exception handler
1750   __ pop(rdx);                                   // restore return address
1751   __ pop(rax);                                   // restore exception
1752   // Note that an "issuing PC" is actually the next PC after the call
1753   __ jmp(rbx);                                   // jump to exception
1754                                                  // handler of caller
1755 }
1756 
1757 
1758 //
1759 // JVMTI ForceEarlyReturn support
1760 //
1761 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
1762   address entry = __ pc();
1763 
1764   __ restore_bcp();
1765   __ restore_locals();
1766   __ empty_expression_stack();
1767   __ load_earlyret_value(state);  // 32 bits returns value in rdx, so don't reuse
1768 
1769   const Register thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
1770   NOT_LP64(__ get_thread(thread));
1771   __ movptr(rcx, Address(thread, JavaThread::jvmti_thread_state_offset()));
1772   Address cond_addr(rcx, JvmtiThreadState::earlyret_state_offset());
1773 
1774   // Clear the earlyret state
1775   __ movl(cond_addr, JvmtiThreadState::earlyret_inactive);
1776 
1777   __ remove_activation(state, rsi,
1778                        false, /* throw_monitor_exception */
1779                        false, /* install_monitor_exception */
1780                        true); /* notify_jvmdi */
1781   __ jmp(rsi);
1782 
1783   return entry;
1784 } // end of ForceEarlyReturn support
1785 
1786 
1787 //-----------------------------------------------------------------------------
1788 // Helper for vtos entry point generation
1789 
1790 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
1791                                                          address& bep,
1792                                                          address& cep,
1793                                                          address& sep,
1794                                                          address& aep,
1795                                                          address& iep,
1796                                                          address& lep,
1797                                                          address& fep,
1798                                                          address& dep,
1799                                                          address& vep) {
1800   assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
1801   Label L;
1802   aep = __ pc();  __ push_ptr();   __ jmp(L);
1803 #ifndef _LP64
1804   fep = __ pc(); __ push(ftos); __ jmp(L);
1805   dep = __ pc(); __ push(dtos); __ jmp(L);
1806 #else
1807   fep = __ pc();  __ push_f(xmm0); __ jmp(L);
1808   dep = __ pc();  __ push_d(xmm0); __ jmp(L);
1809 #endif // _LP64
1810   lep = __ pc();  __ push_l();     __ jmp(L);
1811   bep = cep = sep =
1812   iep = __ pc();  __ push_i();
1813   vep = __ pc();
1814   __ bind(L);
1815   generate_and_dispatch(t);
1816 }
1817 
1818 //-----------------------------------------------------------------------------
1819 
1820 // Non-product code
1821 #ifndef PRODUCT
1822 
1823 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
1824   address entry = __ pc();
1825 
1826 #ifndef _LP64
1827   // prepare expression stack
1828   __ pop(rcx);          // pop return address so expression stack is 'pure'
1829   __ push(state);       // save tosca
1830 
1831   // pass tosca registers as arguments & call tracer
1832   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::trace_bytecode), rcx, rax, rdx);
1833   __ mov(rcx, rax);     // make sure return address is not destroyed by pop(state)
1834   __ pop(state);        // restore tosca
1835 
1836   // return
1837   __ jmp(rcx);
1838 #else
1839   __ push(state);
1840   __ push(c_rarg0);
1841   __ push(c_rarg1);
1842   __ push(c_rarg2);
1843   __ push(c_rarg3);
1844   __ mov(c_rarg2, rax);  // Pass itos
1845 #ifdef _WIN64
1846   __ movflt(xmm3, xmm0); // Pass ftos
1847 #endif
1848   __ call_VM(noreg,
1849              CAST_FROM_FN_PTR(address, InterpreterRuntime::trace_bytecode),
1850              c_rarg1, c_rarg2, c_rarg3);
1851   __ pop(c_rarg3);
1852   __ pop(c_rarg2);
1853   __ pop(c_rarg1);
1854   __ pop(c_rarg0);
1855   __ pop(state);
1856   __ ret(0);                                   // return from result handler
1857 #endif // _LP64
1858 
1859   return entry;
1860 }
1861 
1862 void TemplateInterpreterGenerator::count_bytecode() {
1863   __ incrementl(ExternalAddress((address) &BytecodeCounter::_counter_value));
1864 }
1865 
1866 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
1867   __ incrementl(ExternalAddress((address) &BytecodeHistogram::_counters[t->bytecode()]));
1868 }
1869 
1870 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
1871   __ mov32(rbx, ExternalAddress((address) &BytecodePairHistogram::_index));
1872   __ shrl(rbx, BytecodePairHistogram::log2_number_of_codes);
1873   __ orl(rbx,
1874          ((int) t->bytecode()) <<
1875          BytecodePairHistogram::log2_number_of_codes);
1876   __ mov32(ExternalAddress((address) &BytecodePairHistogram::_index), rbx);
1877   __ lea(rscratch1, ExternalAddress((address) BytecodePairHistogram::_counters));
1878   __ incrementl(Address(rscratch1, rbx, Address::times_4));
1879 }
1880 
1881 
1882 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
1883   // Call a little run-time stub to avoid blow-up for each bytecode.
1884   // The run-time runtime saves the right registers, depending on
1885   // the tosca in-state for the given template.
1886 
1887   assert(Interpreter::trace_code(t->tos_in()) != NULL,
1888          "entry must have been generated");
1889 #ifndef _LP64
1890   __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in())));
1891 #else
1892   __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1893   __ andptr(rsp, -16); // align stack as required by ABI
1894   __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in())));
1895   __ mov(rsp, r12); // restore sp
1896   __ reinit_heapbase();
1897 #endif // _LP64
1898 }
1899 
1900 
1901 void TemplateInterpreterGenerator::stop_interpreter_at() {
1902   Label L;
1903   __ cmp32(ExternalAddress((address) &BytecodeCounter::_counter_value),
1904            StopInterpreterAt);
1905   __ jcc(Assembler::notEqual, L);
1906   __ int3();
1907   __ bind(L);
1908 }
1909 #endif // !PRODUCT
--- EOF ---