1 /*
   2  * Copyright (c) 1997, 2017, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 #include "precompiled.hpp"
  26 #include "asm/macroAssembler.hpp"
  27 #include "interpreter/bytecodeHistogram.hpp"
  28 #include "interpreter/interpreter.hpp"
  29 #include "interpreter/interpreterGenerator.hpp"
  30 #include "interpreter/interpreterRuntime.hpp"
  31 #include "interpreter/templateTable.hpp"
  32 #include "oops/arrayOop.hpp"
  33 #include "oops/methodData.hpp"
  34 #include "oops/method.hpp"
  35 #include "oops/oop.inline.hpp"
  36 #include "prims/jvmtiExport.hpp"
  37 #include "prims/jvmtiThreadState.hpp"
  38 #include "runtime/arguments.hpp"
  39 #include "runtime/deoptimization.hpp"
  40 #include "runtime/frame.inline.hpp"
  41 #include "runtime/sharedRuntime.hpp"
  42 #include "runtime/stubRoutines.hpp"
  43 #include "runtime/synchronizer.hpp"
  44 #include "runtime/timer.hpp"
  45 #include "runtime/vframeArray.hpp"
  46 #include "utilities/debug.hpp"
  47 #include "utilities/macros.hpp"
  48 
  49 #define __ _masm->
  50 
  51 
  52 #ifndef CC_INTERP
  53 const int method_offset = frame::interpreter_frame_method_offset * wordSize;
  54 const int bci_offset    = frame::interpreter_frame_bcx_offset    * wordSize;
  55 const int locals_offset = frame::interpreter_frame_locals_offset * wordSize;
  56 
  57 //------------------------------------------------------------------------------------------------------------------------
  58 
  59 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() {
  60   address entry = __ pc();
  61 
  62   // Note: There should be a minimal interpreter frame set up when stack
  63   // overflow occurs since we check explicitly for it now.
  64   //
  65 #ifdef ASSERT
  66   { Label L;
  67     __ lea(rax, Address(rbp,
  68                 frame::interpreter_frame_monitor_block_top_offset * wordSize));
  69     __ cmpptr(rax, rsp);  // rax, = maximal rsp for current rbp,
  70                         //  (stack grows negative)
  71     __ jcc(Assembler::aboveEqual, L); // check if frame is complete
  72     __ stop ("interpreter frame not set up");
  73     __ bind(L);
  74   }
  75 #endif // ASSERT
  76   // Restore bcp under the assumption that the current frame is still
  77   // interpreted
  78   __ restore_bcp();
  79 
  80   // expression stack must be empty before entering the VM if an exception
  81   // happened
  82   __ empty_expression_stack();
  83   __ empty_FPU_stack();
  84   // throw exception
  85   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_StackOverflowError));
  86   return entry;
  87 }
  88 
  89 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler(const char* name) {
  90   address entry = __ pc();
  91   // expression stack must be empty before entering the VM if an exception happened
  92   __ empty_expression_stack();
  93   __ empty_FPU_stack();
  94   // setup parameters
  95   // ??? convention: expect aberrant index in register rbx,
  96   __ lea(rax, ExternalAddress((address)name));
  97   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ArrayIndexOutOfBoundsException), rax, rbx);
  98   return entry;
  99 }
 100 
 101 address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
 102   address entry = __ pc();
 103   // object is at TOS
 104   __ pop(rax);
 105   // expression stack must be empty before entering the VM if an exception
 106   // happened
 107   __ empty_expression_stack();
 108   __ empty_FPU_stack();
 109   __ call_VM(noreg,
 110              CAST_FROM_FN_PTR(address,
 111                               InterpreterRuntime::throw_ClassCastException),
 112              rax);
 113   return entry;
 114 }
 115 
 116 address TemplateInterpreterGenerator::generate_exception_handler_common(const char* name, const char* message, bool pass_oop) {
 117   assert(!pass_oop || message == NULL, "either oop or message but not both");
 118   address entry = __ pc();
 119   if (pass_oop) {
 120     // object is at TOS
 121     __ pop(rbx);
 122   }
 123   // expression stack must be empty before entering the VM if an exception happened
 124   __ empty_expression_stack();
 125   __ empty_FPU_stack();
 126   // setup parameters
 127   __ lea(rax, ExternalAddress((address)name));
 128   if (pass_oop) {
 129     __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_klass_exception), rax, rbx);
 130   } else {
 131     if (message != NULL) {
 132       __ lea(rbx, ExternalAddress((address)message));
 133     } else {
 134       __ movptr(rbx, NULL_WORD);
 135     }
 136     __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception), rax, rbx);
 137   }
 138   // throw exception
 139   __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
 140   return entry;
 141 }
 142 
 143 
 144 address TemplateInterpreterGenerator::generate_continuation_for(TosState state) {
 145   address entry = __ pc();
 146   // NULL last_sp until next java call
 147   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
 148   __ dispatch_next(state);
 149   return entry;
 150 }
 151 
 152 
 153 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step, size_t index_size) {
 154   address entry = __ pc();
 155 
 156 #ifdef COMPILER2
 157   // The FPU stack is clean if UseSSE >= 2 but must be cleaned in other cases
 158   if ((state == ftos && UseSSE < 1) || (state == dtos && UseSSE < 2)) {
 159     for (int i = 1; i < 8; i++) {
 160         __ ffree(i);
 161     }
 162   } else if (UseSSE < 2) {
 163     __ empty_FPU_stack();
 164   }
 165 #endif
 166   if ((state == ftos && UseSSE < 1) || (state == dtos && UseSSE < 2)) {
 167     __ MacroAssembler::verify_FPU(1, "generate_return_entry_for compiled");
 168   } else {
 169     __ MacroAssembler::verify_FPU(0, "generate_return_entry_for compiled");
 170   }
 171 
 172   // In SSE mode, interpreter returns FP results in xmm0 but they need
 173   // to end up back on the FPU so it can operate on them.
 174   if (state == ftos && UseSSE >= 1) {
 175     __ subptr(rsp, wordSize);
 176     __ movflt(Address(rsp, 0), xmm0);
 177     __ fld_s(Address(rsp, 0));
 178     __ addptr(rsp, wordSize);
 179   } else if (state == dtos && UseSSE >= 2) {
 180     __ subptr(rsp, 2*wordSize);
 181     __ movdbl(Address(rsp, 0), xmm0);
 182     __ fld_d(Address(rsp, 0));
 183     __ addptr(rsp, 2*wordSize);
 184   }
 185 
 186   __ MacroAssembler::verify_FPU(state == ftos || state == dtos ? 1 : 0, "generate_return_entry_for in interpreter");
 187 
 188   // Restore stack bottom in case i2c adjusted stack
 189   __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
 190   // and NULL it as marker that rsp is now tos until next java call
 191   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
 192 
 193   __ restore_bcp();
 194   __ restore_locals();
 195 
 196   if (state == atos) {
 197     Register mdp = rbx;
 198     Register tmp = rcx;
 199     __ profile_return_type(mdp, rax, tmp);
 200   }
 201 
 202   const Register cache = rbx;
 203   const Register index = rcx;
 204   __ get_cache_and_index_at_bcp(cache, index, 1, index_size);
 205 
 206   const Register flags = cache;
 207   __ movl(flags, Address(cache, index, Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()));
 208   __ andl(flags, ConstantPoolCacheEntry::parameter_size_mask);
 209   __ lea(rsp, Address(rsp, flags, Interpreter::stackElementScale()));
 210   __ dispatch_next(state, step);
 211 
 212   return entry;
 213 }
 214 
 215 
 216 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state, int step) {
 217   address entry = __ pc();
 218 
 219   // In SSE mode, FP results are in xmm0
 220   if (state == ftos && UseSSE > 0) {
 221     __ subptr(rsp, wordSize);
 222     __ movflt(Address(rsp, 0), xmm0);
 223     __ fld_s(Address(rsp, 0));
 224     __ addptr(rsp, wordSize);
 225   } else if (state == dtos && UseSSE >= 2) {
 226     __ subptr(rsp, 2*wordSize);
 227     __ movdbl(Address(rsp, 0), xmm0);
 228     __ fld_d(Address(rsp, 0));
 229     __ addptr(rsp, 2*wordSize);
 230   }
 231 
 232   __ MacroAssembler::verify_FPU(state == ftos || state == dtos ? 1 : 0, "generate_deopt_entry_for in interpreter");
 233 
 234   // The stack is not extended by deopt but we must NULL last_sp as this
 235   // entry is like a "return".
 236   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
 237   __ restore_bcp();
 238   __ restore_locals();
 239   // handle exceptions
 240   { Label L;
 241     const Register thread = rcx;
 242     __ get_thread(thread);
 243     __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
 244     __ jcc(Assembler::zero, L);
 245     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_pending_exception));
 246     __ should_not_reach_here();
 247     __ bind(L);
 248   }
 249   __ dispatch_next(state, step);
 250   return entry;
 251 }
 252 
 253 
 254 int AbstractInterpreter::BasicType_as_index(BasicType type) {
 255   int i = 0;
 256   switch (type) {
 257     case T_BOOLEAN: i = 0; break;
 258     case T_CHAR   : i = 1; break;
 259     case T_BYTE   : i = 2; break;
 260     case T_SHORT  : i = 3; break;
 261     case T_INT    : // fall through
 262     case T_LONG   : // fall through
 263     case T_VOID   : i = 4; break;
 264     case T_FLOAT  : i = 5; break;  // have to treat float and double separately for SSE
 265     case T_DOUBLE : i = 6; break;
 266     case T_OBJECT : // fall through
 267     case T_ARRAY  : i = 7; break;
 268     default       : ShouldNotReachHere();
 269   }
 270   assert(0 <= i && i < AbstractInterpreter::number_of_result_handlers, "index out of bounds");
 271   return i;
 272 }
 273 
 274 
 275 address TemplateInterpreterGenerator::generate_result_handler_for(BasicType type) {
 276   address entry = __ pc();
 277   switch (type) {
 278     case T_BOOLEAN: __ c2bool(rax);            break;
 279     case T_CHAR   : __ andptr(rax, 0xFFFF);    break;
 280     case T_BYTE   : __ sign_extend_byte (rax); break;
 281     case T_SHORT  : __ sign_extend_short(rax); break;
 282     case T_INT    : /* nothing to do */        break;
 283     case T_DOUBLE :
 284     case T_FLOAT  :
 285       { const Register t = InterpreterRuntime::SignatureHandlerGenerator::temp();
 286         __ pop(t);                            // remove return address first
 287         // Must return a result for interpreter or compiler. In SSE
 288         // mode, results are returned in xmm0 and the FPU stack must
 289         // be empty.
 290         if (type == T_FLOAT && UseSSE >= 1) {
 291           // Load ST0
 292           __ fld_d(Address(rsp, 0));
 293           // Store as float and empty fpu stack
 294           __ fstp_s(Address(rsp, 0));
 295           // and reload
 296           __ movflt(xmm0, Address(rsp, 0));
 297         } else if (type == T_DOUBLE && UseSSE >= 2 ) {
 298           __ movdbl(xmm0, Address(rsp, 0));
 299         } else {
 300           // restore ST0
 301           __ fld_d(Address(rsp, 0));
 302         }
 303         // and pop the temp
 304         __ addptr(rsp, 2 * wordSize);
 305         __ push(t);                           // restore return address
 306       }
 307       break;
 308     case T_OBJECT :
 309       // retrieve result from frame
 310       __ movptr(rax, Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize));
 311       // and verify it
 312       __ verify_oop(rax);
 313       break;
 314     default       : ShouldNotReachHere();
 315   }
 316   __ ret(0);                                   // return from result handler
 317   return entry;
 318 }
 319 
 320 address TemplateInterpreterGenerator::generate_safept_entry_for(TosState state, address runtime_entry) {
 321   address entry = __ pc();
 322   __ push(state);
 323   __ call_VM(noreg, runtime_entry);
 324   __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
 325   return entry;
 326 }
 327 
 328 
 329 // Helpers for commoning out cases in the various type of method entries.
 330 //
 331 
 332 // increment invocation count & check for overflow
 333 //
 334 // Note: checking for negative value instead of overflow
 335 //       so we have a 'sticky' overflow test
 336 //
 337 // rbx,: method
 338 // rcx: invocation counter
 339 //
 340 void InterpreterGenerator::generate_counter_incr(Label* overflow, Label* profile_method, Label* profile_method_continue) {
 341   Label done;
 342   // Note: In tiered we increment either counters in MethodCounters* or in MDO
 343   // depending if we're profiling or not.
 344   if (TieredCompilation) {
 345     int increment = InvocationCounter::count_increment;
 346     int mask = ((1 << Tier0InvokeNotifyFreqLog)  - 1) << InvocationCounter::count_shift;
 347     Label no_mdo;
 348     if (ProfileInterpreter) {
 349       // Are we profiling?
 350       __ movptr(rax, Address(rbx, Method::method_data_offset()));
 351       __ testptr(rax, rax);
 352       __ jccb(Assembler::zero, no_mdo);
 353       // Increment counter in the MDO
 354       const Address mdo_invocation_counter(rax, in_bytes(MethodData::invocation_counter_offset()) +
 355                                                 in_bytes(InvocationCounter::counter_offset()));
 356       __ increment_mask_and_jump(mdo_invocation_counter, increment, mask, rcx, false, Assembler::zero, overflow);
 357       __ jmp(done);
 358     }
 359     __ bind(no_mdo);
 360     // Increment counter in MethodCounters
 361     const Address invocation_counter(rax,
 362                   MethodCounters::invocation_counter_offset() +
 363                   InvocationCounter::counter_offset());
 364 
 365     __ get_method_counters(rbx, rax, done);
 366     __ increment_mask_and_jump(invocation_counter, increment, mask,
 367                                rcx, false, Assembler::zero, overflow);
 368     __ bind(done);
 369   } else {
 370     const Address backedge_counter  (rax,
 371                   MethodCounters::backedge_counter_offset() +
 372                   InvocationCounter::counter_offset());
 373     const Address invocation_counter(rax,
 374                   MethodCounters::invocation_counter_offset() +
 375                   InvocationCounter::counter_offset());
 376 
 377     __ get_method_counters(rbx, rax, done);
 378 
 379     if (ProfileInterpreter) {
 380       __ incrementl(Address(rax,
 381               MethodCounters::interpreter_invocation_counter_offset()));
 382     }
 383 
 384     // Update standard invocation counters
 385     __ movl(rcx, invocation_counter);
 386     __ incrementl(rcx, InvocationCounter::count_increment);
 387     __ movl(invocation_counter, rcx);             // save invocation count
 388 
 389     __ movl(rax, backedge_counter);               // load backedge counter
 390     __ andl(rax, InvocationCounter::count_mask_value);  // mask out the status bits
 391 
 392     __ addl(rcx, rax);                            // add both counters
 393 
 394     // profile_method is non-null only for interpreted method so
 395     // profile_method != NULL == !native_call
 396     // BytecodeInterpreter only calls for native so code is elided.
 397 
 398     if (ProfileInterpreter && profile_method != NULL) {
 399       // Test to see if we should create a method data oop
 400       __ cmp32(rcx,
 401                ExternalAddress((address)&InvocationCounter::InterpreterProfileLimit));
 402       __ jcc(Assembler::less, *profile_method_continue);
 403 
 404       // if no method data exists, go to profile_method
 405       __ test_method_data_pointer(rax, *profile_method);
 406     }
 407 
 408     __ cmp32(rcx,
 409              ExternalAddress((address)&InvocationCounter::InterpreterInvocationLimit));
 410     __ jcc(Assembler::aboveEqual, *overflow);
 411     __ bind(done);
 412   }
 413 }
 414 
 415 void InterpreterGenerator::generate_counter_overflow(Label* do_continue) {
 416 
 417   // Asm interpreter on entry
 418   // rdi - locals
 419   // rsi - bcp
 420   // rbx, - method
 421   // rdx - cpool
 422   // rbp, - interpreter frame
 423 
 424   // C++ interpreter on entry
 425   // rsi - new interpreter state pointer
 426   // rbp - interpreter frame pointer
 427   // rbx - method
 428 
 429   // On return (i.e. jump to entry_point) [ back to invocation of interpreter ]
 430   // rbx, - method
 431   // rcx - rcvr (assuming there is one)
 432   // top of stack return address of interpreter caller
 433   // rsp - sender_sp
 434 
 435   // C++ interpreter only
 436   // rsi - previous interpreter state pointer
 437 
 438   // InterpreterRuntime::frequency_counter_overflow takes one argument
 439   // indicating if the counter overflow occurs at a backwards branch (non-NULL bcp).
 440   // The call returns the address of the verified entry point for the method or NULL
 441   // if the compilation did not complete (either went background or bailed out).
 442   __ movptr(rax, (intptr_t)false);
 443   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), rax);
 444 
 445   __ movptr(rbx, Address(rbp, method_offset));   // restore Method*
 446 
 447   // Preserve invariant that rsi/rdi contain bcp/locals of sender frame
 448   // and jump to the interpreted entry.
 449   __ jmp(*do_continue, relocInfo::none);
 450 
 451 }
 452 
 453 void InterpreterGenerator::generate_stack_overflow_check(void) {
 454   // see if we've got enough room on the stack for locals plus overhead.
 455   // the expression stack grows down incrementally, so the normal guard
 456   // page mechanism will work for that.
 457   //
 458   // Registers live on entry:
 459   //
 460   // Asm interpreter
 461   // rdx: number of additional locals this frame needs (what we must check)
 462   // rbx,: Method*
 463 
 464   // destroyed on exit
 465   // rax,
 466 
 467   // NOTE:  since the additional locals are also always pushed (wasn't obvious in
 468   // generate_method_entry) so the guard should work for them too.
 469   //
 470 
 471   // monitor entry size: see picture of stack set (generate_method_entry) and frame_x86.hpp
 472   const int entry_size    = frame::interpreter_frame_monitor_size() * wordSize;
 473 
 474   // total overhead size: entry_size + (saved rbp, thru expr stack bottom).
 475   // be sure to change this if you add/subtract anything to/from the overhead area
 476   const int overhead_size = -(frame::interpreter_frame_initial_sp_offset*wordSize) + entry_size;
 477 
 478   const int page_size = os::vm_page_size();
 479 
 480   Label after_frame_check;
 481 
 482   // see if the frame is greater than one page in size. If so,
 483   // then we need to verify there is enough stack space remaining
 484   // for the additional locals.
 485   __ cmpl(rdx, (page_size - overhead_size)/Interpreter::stackElementSize);
 486   __ jcc(Assembler::belowEqual, after_frame_check);
 487 
 488   // compute rsp as if this were going to be the last frame on
 489   // the stack before the red zone
 490 
 491   Label after_frame_check_pop;
 492 
 493   __ push(rsi);
 494 
 495   const Register thread = rsi;
 496 
 497   __ get_thread(thread);
 498 
 499   const Address stack_base(thread, Thread::stack_base_offset());
 500   const Address stack_size(thread, Thread::stack_size_offset());
 501 
 502   // locals + overhead, in bytes
 503   __ lea(rax, Address(noreg, rdx, Interpreter::stackElementScale(), overhead_size));
 504 
 505 #ifdef ASSERT
 506   Label stack_base_okay, stack_size_okay;
 507   // verify that thread stack base is non-zero
 508   __ cmpptr(stack_base, (int32_t)NULL_WORD);
 509   __ jcc(Assembler::notEqual, stack_base_okay);
 510   __ stop("stack base is zero");
 511   __ bind(stack_base_okay);
 512   // verify that thread stack size is non-zero
 513   __ cmpptr(stack_size, 0);
 514   __ jcc(Assembler::notEqual, stack_size_okay);
 515   __ stop("stack size is zero");
 516   __ bind(stack_size_okay);
 517 #endif
 518 
 519   // Add stack base to locals and subtract stack size
 520   __ addptr(rax, stack_base);
 521   __ subptr(rax, stack_size);
 522 
 523   // Use the maximum number of pages we might bang.
 524   const int max_pages = StackShadowPages > (StackRedPages+StackYellowPages) ? StackShadowPages :
 525                                                                               (StackRedPages+StackYellowPages);
 526   __ addptr(rax, max_pages * page_size);
 527 
 528   // check against the current stack bottom
 529   __ cmpptr(rsp, rax);
 530   __ jcc(Assembler::above, after_frame_check_pop);
 531 
 532   __ pop(rsi);  // get saved bcp / (c++ prev state ).
 533 
 534   // Restore sender's sp as SP. This is necessary if the sender's
 535   // frame is an extended compiled frame (see gen_c2i_adapter())
 536   // and safer anyway in case of JSR292 adaptations.
 537 
 538   __ pop(rax); // return address must be moved if SP is changed
 539   __ mov(rsp, rsi);
 540   __ push(rax);
 541 
 542   // Note: the restored frame is not necessarily interpreted.
 543   // Use the shared runtime version of the StackOverflowError.
 544   assert(StubRoutines::throw_StackOverflowError_entry() != NULL, "stub not yet generated");
 545   __ jump(ExternalAddress(StubRoutines::throw_StackOverflowError_entry()));
 546   // all done with frame size check
 547   __ bind(after_frame_check_pop);
 548   __ pop(rsi);
 549 
 550   __ bind(after_frame_check);
 551 }
 552 
 553 // Allocate monitor and lock method (asm interpreter)
 554 // rbx, - Method*
 555 //
 556 void InterpreterGenerator::lock_method(void) {
 557   // synchronize method
 558   const Address access_flags      (rbx, Method::access_flags_offset());
 559   const Address monitor_block_top (rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
 560   const int entry_size            = frame::interpreter_frame_monitor_size() * wordSize;
 561 
 562   #ifdef ASSERT
 563     { Label L;
 564       __ movl(rax, access_flags);
 565       __ testl(rax, JVM_ACC_SYNCHRONIZED);
 566       __ jcc(Assembler::notZero, L);
 567       __ stop("method doesn't need synchronization");
 568       __ bind(L);
 569     }
 570   #endif // ASSERT
 571   // get synchronization object
 572   { Label done;
 573     const int mirror_offset = in_bytes(Klass::java_mirror_offset());
 574     __ movl(rax, access_flags);
 575     __ testl(rax, JVM_ACC_STATIC);
 576     __ movptr(rax, Address(rdi, Interpreter::local_offset_in_bytes(0)));  // get receiver (assume this is frequent case)
 577     __ jcc(Assembler::zero, done);
 578     __ movptr(rax, Address(rbx, Method::const_offset()));
 579     __ movptr(rax, Address(rax, ConstMethod::constants_offset()));
 580     __ movptr(rax, Address(rax, ConstantPool::pool_holder_offset_in_bytes()));
 581     __ movptr(rax, Address(rax, mirror_offset));
 582     __ bind(done);
 583   }
 584   // add space for monitor & lock
 585   __ subptr(rsp, entry_size);                                           // add space for a monitor entry
 586   __ movptr(monitor_block_top, rsp);                                    // set new monitor block top
 587   __ movptr(Address(rsp, BasicObjectLock::obj_offset_in_bytes()), rax); // store object
 588   __ mov(rdx, rsp);                                                    // object address
 589   __ lock_object(rdx);
 590 }
 591 
 592 //
 593 // Generate a fixed interpreter frame. This is identical setup for interpreted methods
 594 // and for native methods hence the shared code.
 595 
 596 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
 597   // initialize fixed part of activation frame
 598   __ push(rax);                                       // save return address
 599   __ enter();                                         // save old & set new rbp,
 600 
 601 
 602   __ push(rsi);                                       // set sender sp
 603   __ push((int32_t)NULL_WORD);                        // leave last_sp as null
 604   __ movptr(rsi, Address(rbx,Method::const_offset())); // get ConstMethod*
 605   __ lea(rsi, Address(rsi,ConstMethod::codes_offset())); // get codebase
 606   __ push(rbx);                                      // save Method*
 607   if (ProfileInterpreter) {
 608     Label method_data_continue;
 609     __ movptr(rdx, Address(rbx, in_bytes(Method::method_data_offset())));
 610     __ testptr(rdx, rdx);
 611     __ jcc(Assembler::zero, method_data_continue);
 612     __ addptr(rdx, in_bytes(MethodData::data_offset()));
 613     __ bind(method_data_continue);
 614     __ push(rdx);                                       // set the mdp (method data pointer)
 615   } else {
 616     __ push(0);
 617   }
 618 
 619   __ movptr(rdx, Address(rbx, Method::const_offset()));
 620   __ movptr(rdx, Address(rdx, ConstMethod::constants_offset()));
 621   __ movptr(rdx, Address(rdx, ConstantPool::cache_offset_in_bytes()));
 622   __ push(rdx);                                       // set constant pool cache
 623   __ push(rdi);                                       // set locals pointer
 624   if (native_call) {
 625     __ push(0);                                       // no bcp
 626   } else {
 627     __ push(rsi);                                     // set bcp
 628     }
 629   __ push(0);                                         // reserve word for pointer to expression stack bottom
 630   __ movptr(Address(rsp, 0), rsp);                    // set expression stack bottom
 631 }
 632 
 633 // End of helpers
 634 
 635 //
 636 // Various method entries
 637 //------------------------------------------------------------------------------------------------------------------------
 638 //
 639 //
 640 
 641 // Call an accessor method (assuming it is resolved, otherwise drop into vanilla (slow path) entry
 642 
 643 address InterpreterGenerator::generate_accessor_entry(void) {
 644 
 645   // rbx,: Method*
 646   // rcx: receiver (preserve for slow entry into asm interpreter)
 647 
 648   // rsi: senderSP must preserved for slow path, set SP to it on fast path
 649 
 650   address entry_point = __ pc();
 651   Label xreturn_path;
 652 
 653   // do fastpath for resolved accessor methods
 654   if (UseFastAccessorMethods) {
 655     Label slow_path;
 656     // If we need a safepoint check, generate full interpreter entry.
 657     ExternalAddress state(SafepointSynchronize::address_of_state());
 658     __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
 659              SafepointSynchronize::_not_synchronized);
 660 
 661     __ jcc(Assembler::notEqual, slow_path);
 662     // ASM/C++ Interpreter
 663     // Code: _aload_0, _(i|a)getfield, _(i|a)return or any rewrites thereof; parameter size = 1
 664     // Note: We can only use this code if the getfield has been resolved
 665     //       and if we don't have a null-pointer exception => check for
 666     //       these conditions first and use slow path if necessary.
 667     // rbx,: method
 668     // rcx: receiver
 669     __ movptr(rax, Address(rsp, wordSize));
 670 
 671     // check if local 0 != NULL and read field
 672     __ testptr(rax, rax);
 673     __ jcc(Assembler::zero, slow_path);
 674 
 675     // read first instruction word and extract bytecode @ 1 and index @ 2
 676     __ movptr(rdx, Address(rbx, Method::const_offset()));
 677     __ movptr(rdi, Address(rdx, ConstMethod::constants_offset()));
 678     __ movl(rdx, Address(rdx, ConstMethod::codes_offset()));
 679     // Shift codes right to get the index on the right.
 680     // The bytecode fetched looks like <index><0xb4><0x2a>
 681     __ shrl(rdx, 2*BitsPerByte);
 682     __ shll(rdx, exact_log2(in_words(ConstantPoolCacheEntry::size())));
 683     __ movptr(rdi, Address(rdi, ConstantPool::cache_offset_in_bytes()));
 684 
 685     // rax,: local 0
 686     // rbx,: method
 687     // rcx: receiver - do not destroy since it is needed for slow path!
 688     // rcx: scratch
 689     // rdx: constant pool cache index
 690     // rdi: constant pool cache
 691     // rsi: sender sp
 692 
 693     // check if getfield has been resolved and read constant pool cache entry
 694     // check the validity of the cache entry by testing whether _indices field
 695     // contains Bytecode::_getfield in b1 byte.
 696     assert(in_words(ConstantPoolCacheEntry::size()) == 4, "adjust shift below");
 697     __ movl(rcx,
 698             Address(rdi,
 699                     rdx,
 700                     Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset()));
 701     __ shrl(rcx, 2*BitsPerByte);
 702     __ andl(rcx, 0xFF);
 703     __ cmpl(rcx, Bytecodes::_getfield);
 704     __ jcc(Assembler::notEqual, slow_path);
 705 
 706     // Note: constant pool entry is not valid before bytecode is resolved
 707     __ movptr(rcx,
 708               Address(rdi,
 709                       rdx,
 710                       Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::f2_offset()));
 711     __ movl(rdx,
 712             Address(rdi,
 713                     rdx,
 714                     Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()));
 715 
 716     Label notByte, notBool, notShort, notChar;
 717     const Address field_address (rax, rcx, Address::times_1);
 718 
 719     // Need to differentiate between igetfield, agetfield, bgetfield etc.
 720     // because they are different sizes.
 721     // Use the type from the constant pool cache
 722     __ shrl(rdx, ConstantPoolCacheEntry::tos_state_shift);
 723     // Make sure we don't need to mask rdx after the above shift
 724     ConstantPoolCacheEntry::verify_tos_state_shift();
 725     __ cmpl(rdx, btos);
 726     __ jcc(Assembler::notEqual, notByte);
 727     __ load_signed_byte(rax, field_address);
 728     __ jmp(xreturn_path);
 729 
 730     __ bind(notByte);
 731     __ cmpl(rdx, ztos);
 732     __ jcc(Assembler::notEqual, notBool);
 733     __ load_signed_byte(rax, field_address);
 734     __ jmp(xreturn_path);
 735 
 736     __ bind(notBool);
 737     __ cmpl(rdx, stos);
 738     __ jcc(Assembler::notEqual, notShort);
 739     __ load_signed_short(rax, field_address);
 740     __ jmp(xreturn_path);
 741 
 742     __ bind(notShort);
 743     __ cmpl(rdx, ctos);
 744     __ jcc(Assembler::notEqual, notChar);
 745     __ load_unsigned_short(rax, field_address);
 746     __ jmp(xreturn_path);
 747 
 748     __ bind(notChar);
 749 #ifdef ASSERT
 750     Label okay;
 751     __ cmpl(rdx, atos);
 752     __ jcc(Assembler::equal, okay);
 753     __ cmpl(rdx, itos);
 754     __ jcc(Assembler::equal, okay);
 755     __ stop("what type is this?");
 756     __ bind(okay);
 757 #endif // ASSERT
 758     // All the rest are a 32 bit wordsize
 759     // This is ok for now. Since fast accessors should be going away
 760     __ movptr(rax, field_address);
 761 
 762     __ bind(xreturn_path);
 763 
 764     // _ireturn/_areturn
 765     __ pop(rdi);                               // get return address
 766     __ mov(rsp, rsi);                          // set sp to sender sp
 767     __ jmp(rdi);
 768 
 769     // generate a vanilla interpreter entry as the slow path
 770     __ bind(slow_path);
 771 
 772     (void) generate_normal_entry(false);
 773     return entry_point;
 774   }
 775   return NULL;
 776 
 777 }
 778 
 779 // Method entry for java.lang.ref.Reference.get.
 780 address InterpreterGenerator::generate_Reference_get_entry(void) {
 781 #if INCLUDE_ALL_GCS
 782   // Code: _aload_0, _getfield, _areturn
 783   // parameter size = 1
 784   //
 785   // The code that gets generated by this routine is split into 2 parts:
 786   //    1. The "intrinsified" code for G1 (or any SATB based GC),
 787   //    2. The slow path - which is an expansion of the regular method entry.
 788   //
 789   // Notes:-
 790   // * In the G1 code we do not check whether we need to block for
 791   //   a safepoint. If G1 is enabled then we must execute the specialized
 792   //   code for Reference.get (except when the Reference object is null)
 793   //   so that we can log the value in the referent field with an SATB
 794   //   update buffer.
 795   //   If the code for the getfield template is modified so that the
 796   //   G1 pre-barrier code is executed when the current method is
 797   //   Reference.get() then going through the normal method entry
 798   //   will be fine.
 799   // * The G1 code below can, however, check the receiver object (the instance
 800   //   of java.lang.Reference) and jump to the slow path if null. If the
 801   //   Reference object is null then we obviously cannot fetch the referent
 802   //   and so we don't need to call the G1 pre-barrier. Thus we can use the
 803   //   regular method entry code to generate the NPE.
 804   //
 805   // This code is based on generate_accessor_enty.
 806 
 807   // rbx,: Method*
 808   // rcx: receiver (preserve for slow entry into asm interpreter)
 809 
 810   // rsi: senderSP must preserved for slow path, set SP to it on fast path
 811 
 812   address entry = __ pc();
 813 
 814   const int referent_offset = java_lang_ref_Reference::referent_offset;
 815   guarantee(referent_offset > 0, "referent offset not initialized");
 816 
 817   if (UseG1GC) {
 818     Label slow_path;
 819 
 820     // Check if local 0 != NULL
 821     // If the receiver is null then it is OK to jump to the slow path.
 822     __ movptr(rax, Address(rsp, wordSize));
 823     __ testptr(rax, rax);
 824     __ jcc(Assembler::zero, slow_path);
 825 
 826     // rax: local 0 (must be preserved across the G1 barrier call)
 827     //
 828     // rbx: method (at this point it's scratch)
 829     // rcx: receiver (at this point it's scratch)
 830     // rdx: scratch
 831     // rdi: scratch
 832     //
 833     // rsi: sender sp
 834 
 835     // Preserve the sender sp in case the pre-barrier
 836     // calls the runtime
 837     __ push(rsi);
 838 
 839     // Load the value of the referent field.
 840     const Address field_address(rax, referent_offset);
 841     __ movptr(rax, field_address);
 842 
 843     // Generate the G1 pre-barrier code to log the value of
 844     // the referent field in an SATB buffer.
 845     __ get_thread(rcx);
 846     __ g1_write_barrier_pre(noreg /* obj */,
 847                             rax /* pre_val */,
 848                             rcx /* thread */,
 849                             rbx /* tmp */,
 850                             true /* tosca_save */,
 851                             true /* expand_call */);
 852 
 853     // _areturn
 854     __ pop(rsi);                // get sender sp
 855     __ pop(rdi);                // get return address
 856     __ mov(rsp, rsi);           // set sp to sender sp
 857     __ jmp(rdi);
 858 
 859     __ bind(slow_path);
 860     (void) generate_normal_entry(false);
 861 
 862     return entry;
 863   }
 864 #endif // INCLUDE_ALL_GCS
 865 
 866   // If G1 is not enabled then attempt to go through the accessor entry point
 867   // Reference.get is an accessor
 868   return generate_accessor_entry();
 869 }
 870 
 871 /**
 872  * Method entry for static native methods:
 873  *   int java.util.zip.CRC32.update(int crc, int b)
 874  */
 875 address InterpreterGenerator::generate_CRC32_update_entry() {
 876   if (UseCRC32Intrinsics) {
 877     address entry = __ pc();
 878 
 879     // rbx,: Method*
 880     // rsi: senderSP must preserved for slow path, set SP to it on fast path
 881     // rdx: scratch
 882     // rdi: scratch
 883 
 884     Label slow_path;
 885     // If we need a safepoint check, generate full interpreter entry.
 886     ExternalAddress state(SafepointSynchronize::address_of_state());
 887     __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
 888              SafepointSynchronize::_not_synchronized);
 889     __ jcc(Assembler::notEqual, slow_path);
 890 
 891     // We don't generate local frame and don't align stack because
 892     // we call stub code and there is no safepoint on this path.
 893 
 894     // Load parameters
 895     const Register crc = rax;  // crc
 896     const Register val = rdx;  // source java byte value
 897     const Register tbl = rdi;  // scratch
 898 
 899     // Arguments are reversed on java expression stack
 900     __ movl(val, Address(rsp,   wordSize)); // byte value
 901     __ movl(crc, Address(rsp, 2*wordSize)); // Initial CRC
 902 
 903     __ lea(tbl, ExternalAddress(StubRoutines::crc_table_addr()));
 904     __ notl(crc); // ~crc
 905     __ update_byte_crc32(crc, val, tbl);
 906     __ notl(crc); // ~crc
 907     // result in rax
 908 
 909     // _areturn
 910     __ pop(rdi);                // get return address
 911     __ mov(rsp, rsi);           // set sp to sender sp
 912     __ jmp(rdi);
 913 
 914     // generate a vanilla native entry as the slow path
 915     __ bind(slow_path);
 916 
 917     (void) generate_native_entry(false);
 918 
 919     return entry;
 920   }
 921   return generate_native_entry(false);
 922 }
 923 
 924 /**
 925  * Method entry for static native methods:
 926  *   int java.util.zip.CRC32.updateBytes(int crc, byte[] b, int off, int len)
 927  *   int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
 928  */
 929 address InterpreterGenerator::generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
 930   if (UseCRC32Intrinsics) {
 931     address entry = __ pc();
 932 
 933     // rbx,: Method*
 934     // rsi: senderSP must preserved for slow path, set SP to it on fast path
 935     // rdx: scratch
 936     // rdi: scratch
 937 
 938     Label slow_path;
 939     // If we need a safepoint check, generate full interpreter entry.
 940     ExternalAddress state(SafepointSynchronize::address_of_state());
 941     __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
 942              SafepointSynchronize::_not_synchronized);
 943     __ jcc(Assembler::notEqual, slow_path);
 944 
 945     // We don't generate local frame and don't align stack because
 946     // we call stub code and there is no safepoint on this path.
 947 
 948     // Load parameters
 949     const Register crc = rax;  // crc
 950     const Register buf = rdx;  // source java byte array address
 951     const Register len = rdi;  // length
 952 
 953     // Arguments are reversed on java expression stack
 954     __ movl(len,   Address(rsp,   wordSize)); // Length
 955     // Calculate address of start element
 956     if (kind == Interpreter::java_util_zip_CRC32_updateByteBuffer) {
 957       __ movptr(buf, Address(rsp, 3*wordSize)); // long buf
 958       __ addptr(buf, Address(rsp, 2*wordSize)); // + offset
 959       __ movl(crc,   Address(rsp, 5*wordSize)); // Initial CRC
 960     } else {
 961       __ movptr(buf, Address(rsp, 3*wordSize)); // byte[] array
 962       __ addptr(buf, arrayOopDesc::base_offset_in_bytes(T_BYTE)); // + header size
 963       __ addptr(buf, Address(rsp, 2*wordSize)); // + offset
 964       __ movl(crc,   Address(rsp, 4*wordSize)); // Initial CRC
 965     }
 966 
 967     __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, StubRoutines::updateBytesCRC32()), crc, buf, len);
 968     // result in rax
 969 
 970     // _areturn
 971     __ pop(rdi);                // get return address
 972     __ mov(rsp, rsi);           // set sp to sender sp
 973     __ jmp(rdi);
 974 
 975     // generate a vanilla native entry as the slow path
 976     __ bind(slow_path);
 977 
 978     (void) generate_native_entry(false);
 979 
 980     return entry;
 981   }
 982   return generate_native_entry(false);
 983 }
 984 
 985 //
 986 // Interpreter stub for calling a native method. (asm interpreter)
 987 // This sets up a somewhat different looking stack for calling the native method
 988 // than the typical interpreter frame setup.
 989 //
 990 
 991 address InterpreterGenerator::generate_native_entry(bool synchronized) {
 992   // determine code generation flags
 993   bool inc_counter  = UseCompiler || CountCompiledCalls;
 994 
 995   // rbx,: Method*
 996   // rsi: sender sp
 997   // rsi: previous interpreter state (C++ interpreter) must preserve
 998   address entry_point = __ pc();
 999 
1000   const Address constMethod       (rbx, Method::const_offset());
1001   const Address access_flags      (rbx, Method::access_flags_offset());
1002   const Address size_of_parameters(rcx, ConstMethod::size_of_parameters_offset());
1003 
1004   // get parameter size (always needed)
1005   __ movptr(rcx, constMethod);
1006   __ load_unsigned_short(rcx, size_of_parameters);
1007 
1008   // native calls don't need the stack size check since they have no expression stack
1009   // and the arguments are already on the stack and we only add a handful of words
1010   // to the stack
1011 
1012   // rbx,: Method*
1013   // rcx: size of parameters
1014   // rsi: sender sp
1015 
1016   __ pop(rax);                                       // get return address
1017   // for natives the size of locals is zero
1018 
1019   // compute beginning of parameters (rdi)
1020   __ lea(rdi, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize));
1021 
1022 
1023   // add 2 zero-initialized slots for native calls
1024   // NULL result handler
1025   __ push((int32_t)NULL_WORD);
1026   // NULL oop temp (mirror or jni oop result)
1027   __ push((int32_t)NULL_WORD);
1028 
1029   // initialize fixed part of activation frame
1030   generate_fixed_frame(true);
1031 
1032   // make sure method is native & not abstract
1033 #ifdef ASSERT
1034   __ movl(rax, access_flags);
1035   {
1036     Label L;
1037     __ testl(rax, JVM_ACC_NATIVE);
1038     __ jcc(Assembler::notZero, L);
1039     __ stop("tried to execute non-native method as native");
1040     __ bind(L);
1041   }
1042   { Label L;
1043     __ testl(rax, JVM_ACC_ABSTRACT);
1044     __ jcc(Assembler::zero, L);
1045     __ stop("tried to execute abstract method in interpreter");
1046     __ bind(L);
1047   }
1048 #endif
1049 
1050   // Since at this point in the method invocation the exception handler
1051   // would try to exit the monitor of synchronized methods which hasn't
1052   // been entered yet, we set the thread local variable
1053   // _do_not_unlock_if_synchronized to true. The remove_activation will
1054   // check this flag.
1055 
1056   __ get_thread(rax);
1057   const Address do_not_unlock_if_synchronized(rax,
1058         in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1059   __ movbool(do_not_unlock_if_synchronized, true);
1060 
1061   // increment invocation count & check for overflow
1062   Label invocation_counter_overflow;
1063   if (inc_counter) {
1064     generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
1065   }
1066 
1067   Label continue_after_compile;
1068   __ bind(continue_after_compile);
1069 
1070   bang_stack_shadow_pages(true);
1071 
1072   // reset the _do_not_unlock_if_synchronized flag
1073   __ get_thread(rax);
1074   __ movbool(do_not_unlock_if_synchronized, false);
1075 
1076   // check for synchronized methods
1077   // Must happen AFTER invocation_counter check and stack overflow check,
1078   // so method is not locked if overflows.
1079   //
1080   if (synchronized) {
1081     lock_method();
1082   } else {
1083     // no synchronization necessary
1084 #ifdef ASSERT
1085       { Label L;
1086         __ movl(rax, access_flags);
1087         __ testl(rax, JVM_ACC_SYNCHRONIZED);
1088         __ jcc(Assembler::zero, L);
1089         __ stop("method needs synchronization");
1090         __ bind(L);
1091       }
1092 #endif
1093   }
1094 
1095   // start execution
1096 #ifdef ASSERT
1097   { Label L;
1098     const Address monitor_block_top (rbp,
1099                  frame::interpreter_frame_monitor_block_top_offset * wordSize);
1100     __ movptr(rax, monitor_block_top);
1101     __ cmpptr(rax, rsp);
1102     __ jcc(Assembler::equal, L);
1103     __ stop("broken stack frame setup in interpreter");
1104     __ bind(L);
1105   }
1106 #endif
1107 
1108   // jvmti/dtrace support
1109   __ notify_method_entry();
1110 
1111   // work registers
1112   const Register method = rbx;
1113   const Register thread = rdi;
1114   const Register t      = rcx;
1115 
1116   // allocate space for parameters
1117   __ get_method(method);
1118   __ movptr(t, Address(method, Method::const_offset()));
1119   __ load_unsigned_short(t, Address(t, ConstMethod::size_of_parameters_offset()));
1120 
1121   __ shlptr(t, Interpreter::logStackElementSize);
1122   __ addptr(t, 2*wordSize);     // allocate two more slots for JNIEnv and possible mirror
1123   __ subptr(rsp, t);
1124   __ andptr(rsp, -(StackAlignmentInBytes)); // gcc needs 16 byte aligned stacks to do XMM intrinsics
1125 
1126   // get signature handler
1127   { Label L;
1128     __ movptr(t, Address(method, Method::signature_handler_offset()));
1129     __ testptr(t, t);
1130     __ jcc(Assembler::notZero, L);
1131     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), method);
1132     __ get_method(method);
1133     __ movptr(t, Address(method, Method::signature_handler_offset()));
1134     __ bind(L);
1135   }
1136 
1137   // call signature handler
1138   assert(InterpreterRuntime::SignatureHandlerGenerator::from() == rdi, "adjust this code");
1139   assert(InterpreterRuntime::SignatureHandlerGenerator::to  () == rsp, "adjust this code");
1140   assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == t  , "adjust this code");
1141   // The generated handlers do not touch RBX (the method oop).
1142   // However, large signatures cannot be cached and are generated
1143   // each time here.  The slow-path generator will blow RBX
1144   // sometime, so we must reload it after the call.
1145   __ call(t);
1146   __ get_method(method);        // slow path call blows RBX on DevStudio 5.0
1147 
1148   // result handler is in rax,
1149   // set result handler
1150   __ movptr(Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize), rax);
1151 
1152   // pass mirror handle if static call
1153   { Label L;
1154     const int mirror_offset = in_bytes(Klass::java_mirror_offset());
1155     __ movl(t, Address(method, Method::access_flags_offset()));
1156     __ testl(t, JVM_ACC_STATIC);
1157     __ jcc(Assembler::zero, L);
1158     // get mirror
1159     __ movptr(t, Address(method, Method:: const_offset()));
1160     __ movptr(t, Address(t, ConstMethod::constants_offset()));
1161     __ movptr(t, Address(t, ConstantPool::pool_holder_offset_in_bytes()));
1162     __ movptr(t, Address(t, mirror_offset));
1163     // copy mirror into activation frame
1164     __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize), t);
1165     // pass handle to mirror
1166     __ lea(t, Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize));
1167     __ movptr(Address(rsp, wordSize), t);
1168     __ bind(L);
1169   }
1170 
1171   // get native function entry point
1172   { Label L;
1173     __ movptr(rax, Address(method, Method::native_function_offset()));
1174     ExternalAddress unsatisfied(SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
1175     __ cmpptr(rax, unsatisfied.addr());
1176     __ jcc(Assembler::notEqual, L);
1177     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), method);
1178     __ get_method(method);
1179     __ movptr(rax, Address(method, Method::native_function_offset()));
1180     __ bind(L);
1181   }
1182 
1183   // pass JNIEnv
1184   __ get_thread(thread);
1185   __ lea(t, Address(thread, JavaThread::jni_environment_offset()));
1186   __ movptr(Address(rsp, 0), t);
1187 
1188   // set_last_Java_frame_before_call
1189   // It is enough that the pc()
1190   // points into the right code segment. It does not have to be the correct return pc.
1191   __ set_last_Java_frame(thread, noreg, rbp, __ pc());
1192 
1193   // change thread state
1194 #ifdef ASSERT
1195   { Label L;
1196     __ movl(t, Address(thread, JavaThread::thread_state_offset()));
1197     __ cmpl(t, _thread_in_Java);
1198     __ jcc(Assembler::equal, L);
1199     __ stop("Wrong thread state in native stub");
1200     __ bind(L);
1201   }
1202 #endif
1203 
1204   // Change state to native
1205   __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_native);
1206   __ call(rax);
1207 
1208   // result potentially in rdx:rax or ST0
1209 
1210   // Verify or restore cpu control state after JNI call
1211   __ restore_cpu_control_state_after_jni();
1212 
1213   // save potential result in ST(0) & rdx:rax
1214   // (if result handler is the T_FLOAT or T_DOUBLE handler, result must be in ST0 -
1215   // the check is necessary to avoid potential Intel FPU overflow problems by saving/restoring 'empty' FPU registers)
1216   // It is safe to do this push because state is _thread_in_native and return address will be found
1217   // via _last_native_pc and not via _last_jave_sp
1218 
1219   // NOTE: the order of theses push(es) is known to frame::interpreter_frame_result.
1220   // If the order changes or anything else is added to the stack the code in
1221   // interpreter_frame_result will have to be changed.
1222 
1223   { Label L;
1224     Label push_double;
1225     ExternalAddress float_handler(AbstractInterpreter::result_handler(T_FLOAT));
1226     ExternalAddress double_handler(AbstractInterpreter::result_handler(T_DOUBLE));
1227     __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize),
1228               float_handler.addr());
1229     __ jcc(Assembler::equal, push_double);
1230     __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize),
1231               double_handler.addr());
1232     __ jcc(Assembler::notEqual, L);
1233     __ bind(push_double);
1234     __ push(dtos);
1235     __ bind(L);
1236   }
1237   __ push(ltos);
1238 
1239   // change thread state
1240   __ get_thread(thread);
1241   __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_native_trans);
1242   if(os::is_MP()) {
1243     if (UseMembar) {
1244       // Force this write out before the read below
1245       __ membar(Assembler::Membar_mask_bits(
1246            Assembler::LoadLoad | Assembler::LoadStore |
1247            Assembler::StoreLoad | Assembler::StoreStore));
1248     } else {
1249       // Write serialization page so VM thread can do a pseudo remote membar.
1250       // We use the current thread pointer to calculate a thread specific
1251       // offset to write to within the page. This minimizes bus traffic
1252       // due to cache line collision.
1253       __ serialize_memory(thread, rcx);
1254     }
1255   }
1256 
1257   if (AlwaysRestoreFPU) {
1258     //  Make sure the control word is correct.
1259     __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
1260   }
1261 
1262   // check for safepoint operation in progress and/or pending suspend requests
1263   { Label Continue;
1264 
1265     __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
1266              SafepointSynchronize::_not_synchronized);
1267 
1268     Label L;
1269     __ jcc(Assembler::notEqual, L);
1270     __ cmpl(Address(thread, JavaThread::suspend_flags_offset()), 0);
1271     __ jcc(Assembler::equal, Continue);
1272     __ bind(L);
1273 
1274     // Don't use call_VM as it will see a possible pending exception and forward it
1275     // and never return here preventing us from clearing _last_native_pc down below.
1276     // Also can't use call_VM_leaf either as it will check to see if rsi & rdi are
1277     // preserved and correspond to the bcp/locals pointers. So we do a runtime call
1278     // by hand.
1279     //
1280     __ push(thread);
1281     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address,
1282                                             JavaThread::check_special_condition_for_native_trans)));
1283     __ increment(rsp, wordSize);
1284     __ get_thread(thread);
1285 
1286     __ bind(Continue);
1287   }
1288 
1289   // change thread state
1290   __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_Java);
1291 
1292   __ reset_last_Java_frame(thread, true);
1293 
1294   // reset handle block
1295   __ movptr(t, Address(thread, JavaThread::active_handles_offset()));
1296   __ movl(Address(t, JNIHandleBlock::top_offset_in_bytes()), NULL_WORD);
1297 
1298   // If result was an oop then unbox and save it in the frame
1299   {
1300     Label no_oop;
1301     ExternalAddress handler(AbstractInterpreter::result_handler(T_OBJECT));
1302     __ cmpptr(Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize),
1303               handler.addr());
1304     __ jcc(Assembler::notEqual, no_oop);
1305     __ cmpptr(Address(rsp, 0), (int32_t)NULL_WORD);
1306     __ pop(ltos);
1307     // Unbox oop result, e.g. JNIHandles::resolve value.
1308     __ resolve_jobject(rax /* value */,
1309                        thread /* thread */,
1310                        t /* tmp */);
1311     __ movptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset)*wordSize), rax);
1312     // keep stack depth as expected by pushing oop which will eventually be discarded
1313     __ push(ltos);
1314     __ bind(no_oop);
1315   }
1316 
1317   {
1318      Label no_reguard;
1319      __ cmpl(Address(thread, JavaThread::stack_guard_state_offset()), JavaThread::stack_guard_yellow_disabled);
1320      __ jcc(Assembler::notEqual, no_reguard);
1321 
1322      __ pusha();
1323      __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
1324      __ popa();
1325 
1326      __ bind(no_reguard);
1327    }
1328 
1329   // restore rsi to have legal interpreter frame,
1330   // i.e., bci == 0 <=> rsi == code_base()
1331   // Can't call_VM until bcp is within reasonable.
1332   __ get_method(method);      // method is junk from thread_in_native to now.
1333   __ movptr(rsi, Address(method,Method::const_offset()));   // get ConstMethod*
1334   __ lea(rsi, Address(rsi,ConstMethod::codes_offset()));    // get codebase
1335 
1336   // handle exceptions (exception handling will handle unlocking!)
1337   { Label L;
1338     __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
1339     __ jcc(Assembler::zero, L);
1340     // Note: At some point we may want to unify this with the code used in call_VM_base();
1341     //       i.e., we should use the StubRoutines::forward_exception code. For now this
1342     //       doesn't work here because the rsp is not correctly set at this point.
1343     __ MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_pending_exception));
1344     __ should_not_reach_here();
1345     __ bind(L);
1346   }
1347 
1348   // do unlocking if necessary
1349   { Label L;
1350     __ movl(t, Address(method, Method::access_flags_offset()));
1351     __ testl(t, JVM_ACC_SYNCHRONIZED);
1352     __ jcc(Assembler::zero, L);
1353     // the code below should be shared with interpreter macro assembler implementation
1354     { Label unlock;
1355       // BasicObjectLock will be first in list, since this is a synchronized method. However, need
1356       // to check that the object has not been unlocked by an explicit monitorexit bytecode.
1357       const Address monitor(rbp, frame::interpreter_frame_initial_sp_offset * wordSize - (int)sizeof(BasicObjectLock));
1358 
1359       __ lea(rdx, monitor);                   // address of first monitor
1360 
1361       __ movptr(t, Address(rdx, BasicObjectLock::obj_offset_in_bytes()));
1362       __ testptr(t, t);
1363       __ jcc(Assembler::notZero, unlock);
1364 
1365       // Entry already unlocked, need to throw exception
1366       __ MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
1367       __ should_not_reach_here();
1368 
1369       __ bind(unlock);
1370       __ unlock_object(rdx);
1371     }
1372     __ bind(L);
1373   }
1374 
1375   // jvmti/dtrace support
1376   // Note: This must happen _after_ handling/throwing any exceptions since
1377   //       the exception handler code notifies the runtime of method exits
1378   //       too. If this happens before, method entry/exit notifications are
1379   //       not properly paired (was bug - gri 11/22/99).
1380   __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI);
1381 
1382   // restore potential result in rdx:rax, call result handler to restore potential result in ST0 & handle result
1383   __ pop(ltos);
1384   __ movptr(t, Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize));
1385   __ call(t);
1386 
1387   // remove activation
1388   __ movptr(t, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); // get sender sp
1389   __ leave();                                // remove frame anchor
1390   __ pop(rdi);                               // get return address
1391   __ mov(rsp, t);                            // set sp to sender sp
1392   __ jmp(rdi);
1393 
1394   if (inc_counter) {
1395     // Handle overflow of counter and compile method
1396     __ bind(invocation_counter_overflow);
1397     generate_counter_overflow(&continue_after_compile);
1398   }
1399 
1400   return entry_point;
1401 }
1402 
1403 //
1404 // Generic interpreted method entry to (asm) interpreter
1405 //
1406 address InterpreterGenerator::generate_normal_entry(bool synchronized) {
1407   // determine code generation flags
1408   bool inc_counter  = UseCompiler || CountCompiledCalls;
1409 
1410   // rbx,: Method*
1411   // rsi: sender sp
1412   address entry_point = __ pc();
1413 
1414   const Address constMethod       (rbx, Method::const_offset());
1415   const Address access_flags      (rbx, Method::access_flags_offset());
1416   const Address size_of_parameters(rdx, ConstMethod::size_of_parameters_offset());
1417   const Address size_of_locals    (rdx, ConstMethod::size_of_locals_offset());
1418 
1419   // get parameter size (always needed)
1420   __ movptr(rdx, constMethod);
1421   __ load_unsigned_short(rcx, size_of_parameters);
1422 
1423   // rbx,: Method*
1424   // rcx: size of parameters
1425 
1426   // rsi: sender_sp (could differ from sp+wordSize if we were called via c2i )
1427 
1428   __ load_unsigned_short(rdx, size_of_locals);       // get size of locals in words
1429   __ subl(rdx, rcx);                                // rdx = no. of additional locals
1430 
1431   // see if we've got enough room on the stack for locals plus overhead.
1432   generate_stack_overflow_check();
1433 
1434   // get return address
1435   __ pop(rax);
1436 
1437   // compute beginning of parameters (rdi)
1438   __ lea(rdi, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize));
1439 
1440   // rdx - # of additional locals
1441   // allocate space for locals
1442   // explicitly initialize locals
1443   {
1444     Label exit, loop;
1445     __ testl(rdx, rdx);
1446     __ jcc(Assembler::lessEqual, exit);               // do nothing if rdx <= 0
1447     __ bind(loop);
1448     __ push((int32_t)NULL_WORD);                      // initialize local variables
1449     __ decrement(rdx);                                // until everything initialized
1450     __ jcc(Assembler::greater, loop);
1451     __ bind(exit);
1452   }
1453 
1454   // initialize fixed part of activation frame
1455   generate_fixed_frame(false);
1456 
1457   // make sure method is not native & not abstract
1458 #ifdef ASSERT
1459   __ movl(rax, access_flags);
1460   {
1461     Label L;
1462     __ testl(rax, JVM_ACC_NATIVE);
1463     __ jcc(Assembler::zero, L);
1464     __ stop("tried to execute native method as non-native");
1465     __ bind(L);
1466   }
1467   { Label L;
1468     __ testl(rax, JVM_ACC_ABSTRACT);
1469     __ jcc(Assembler::zero, L);
1470     __ stop("tried to execute abstract method in interpreter");
1471     __ bind(L);
1472   }
1473 #endif
1474 
1475   // Since at this point in the method invocation the exception handler
1476   // would try to exit the monitor of synchronized methods which hasn't
1477   // been entered yet, we set the thread local variable
1478   // _do_not_unlock_if_synchronized to true. The remove_activation will
1479   // check this flag.
1480 
1481   __ get_thread(rax);
1482   const Address do_not_unlock_if_synchronized(rax,
1483         in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1484   __ movbool(do_not_unlock_if_synchronized, true);
1485 
1486   __ profile_parameters_type(rax, rcx, rdx);
1487   // increment invocation count & check for overflow
1488   Label invocation_counter_overflow;
1489   Label profile_method;
1490   Label profile_method_continue;
1491   if (inc_counter) {
1492     generate_counter_incr(&invocation_counter_overflow, &profile_method, &profile_method_continue);
1493     if (ProfileInterpreter) {
1494       __ bind(profile_method_continue);
1495     }
1496   }
1497   Label continue_after_compile;
1498   __ bind(continue_after_compile);
1499 
1500   bang_stack_shadow_pages(false);
1501 
1502   // reset the _do_not_unlock_if_synchronized flag
1503   __ get_thread(rax);
1504   __ movbool(do_not_unlock_if_synchronized, false);
1505 
1506   // check for synchronized methods
1507   // Must happen AFTER invocation_counter check and stack overflow check,
1508   // so method is not locked if overflows.
1509   //
1510   if (synchronized) {
1511     // Allocate monitor and lock method
1512     lock_method();
1513   } else {
1514     // no synchronization necessary
1515 #ifdef ASSERT
1516       { Label L;
1517         __ movl(rax, access_flags);
1518         __ testl(rax, JVM_ACC_SYNCHRONIZED);
1519         __ jcc(Assembler::zero, L);
1520         __ stop("method needs synchronization");
1521         __ bind(L);
1522       }
1523 #endif
1524   }
1525 
1526   // start execution
1527 #ifdef ASSERT
1528   { Label L;
1529      const Address monitor_block_top (rbp,
1530                  frame::interpreter_frame_monitor_block_top_offset * wordSize);
1531     __ movptr(rax, monitor_block_top);
1532     __ cmpptr(rax, rsp);
1533     __ jcc(Assembler::equal, L);
1534     __ stop("broken stack frame setup in interpreter");
1535     __ bind(L);
1536   }
1537 #endif
1538 
1539   // jvmti support
1540   __ notify_method_entry();
1541 
1542   __ dispatch_next(vtos);
1543 
1544   // invocation counter overflow
1545   if (inc_counter) {
1546     if (ProfileInterpreter) {
1547       // We have decided to profile this method in the interpreter
1548       __ bind(profile_method);
1549       __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1550       __ set_method_data_pointer_for_bcp();
1551       __ get_method(rbx);
1552       __ jmp(profile_method_continue);
1553     }
1554     // Handle overflow of counter and compile method
1555     __ bind(invocation_counter_overflow);
1556     generate_counter_overflow(&continue_after_compile);
1557   }
1558 
1559   return entry_point;
1560 }
1561 
1562 //------------------------------------------------------------------------------------------------------------------------
1563 // Entry points
1564 //
1565 // Here we generate the various kind of entries into the interpreter.
1566 // The two main entry type are generic bytecode methods and native call method.
1567 // These both come in synchronized and non-synchronized versions but the
1568 // frame layout they create is very similar. The other method entry
1569 // types are really just special purpose entries that are really entry
1570 // and interpretation all in one. These are for trivial methods like
1571 // accessor, empty, or special math methods.
1572 //
1573 // When control flow reaches any of the entry types for the interpreter
1574 // the following holds ->
1575 //
1576 // Arguments:
1577 //
1578 // rbx,: Method*
1579 // rcx: receiver
1580 //
1581 //
1582 // Stack layout immediately at entry
1583 //
1584 // [ return address     ] <--- rsp
1585 // [ parameter n        ]
1586 //   ...
1587 // [ parameter 1        ]
1588 // [ expression stack   ] (caller's java expression stack)
1589 
1590 // Assuming that we don't go to one of the trivial specialized
1591 // entries the stack will look like below when we are ready to execute
1592 // the first bytecode (or call the native routine). The register usage
1593 // will be as the template based interpreter expects (see interpreter_x86.hpp).
1594 //
1595 // local variables follow incoming parameters immediately; i.e.
1596 // the return address is moved to the end of the locals).
1597 //
1598 // [ monitor entry      ] <--- rsp
1599 //   ...
1600 // [ monitor entry      ]
1601 // [ expr. stack bottom ]
1602 // [ saved rsi          ]
1603 // [ current rdi        ]
1604 // [ Method*            ]
1605 // [ saved rbp,          ] <--- rbp,
1606 // [ return address     ]
1607 // [ local variable m   ]
1608 //   ...
1609 // [ local variable 1   ]
1610 // [ parameter n        ]
1611 //   ...
1612 // [ parameter 1        ] <--- rdi
1613 
1614 address AbstractInterpreterGenerator::generate_method_entry(AbstractInterpreter::MethodKind kind) {
1615   // determine code generation flags
1616   bool synchronized = false;
1617   address entry_point = NULL;
1618   InterpreterGenerator* ig_this = (InterpreterGenerator*)this;
1619 
1620   switch (kind) {
1621     case Interpreter::zerolocals             :                                                       break;
1622     case Interpreter::zerolocals_synchronized: synchronized = true;                                  break;
1623     case Interpreter::native                 : entry_point = ig_this->generate_native_entry(false);  break;
1624     case Interpreter::native_synchronized    : entry_point = ig_this->generate_native_entry(true);   break;
1625     case Interpreter::empty                  : entry_point = ig_this->generate_empty_entry();        break;
1626     case Interpreter::accessor               : entry_point = ig_this->generate_accessor_entry();     break;
1627     case Interpreter::abstract               : entry_point = ig_this->generate_abstract_entry();     break;
1628 
1629     case Interpreter::java_lang_math_sin     : // fall thru
1630     case Interpreter::java_lang_math_cos     : // fall thru
1631     case Interpreter::java_lang_math_tan     : // fall thru
1632     case Interpreter::java_lang_math_abs     : // fall thru
1633     case Interpreter::java_lang_math_log     : // fall thru
1634     case Interpreter::java_lang_math_log10   : // fall thru
1635     case Interpreter::java_lang_math_sqrt    : // fall thru
1636     case Interpreter::java_lang_math_pow     : // fall thru
1637     case Interpreter::java_lang_math_exp     : entry_point = ig_this->generate_math_entry(kind);      break;
1638     case Interpreter::java_lang_ref_reference_get
1639                                              : entry_point = ig_this->generate_Reference_get_entry(); break;
1640     case Interpreter::java_util_zip_CRC32_update
1641                                              : entry_point = ig_this->generate_CRC32_update_entry();  break;
1642     case Interpreter::java_util_zip_CRC32_updateBytes
1643                                              : // fall thru
1644     case Interpreter::java_util_zip_CRC32_updateByteBuffer
1645                                              : entry_point = ig_this->generate_CRC32_updateBytes_entry(kind); break;
1646     default:
1647       fatal(err_msg("unexpected method kind: %d", kind));
1648       break;
1649   }
1650 
1651   if (entry_point) return entry_point;
1652 
1653   return ig_this->generate_normal_entry(synchronized);
1654 
1655 }
1656 
1657 // These should never be compiled since the interpreter will prefer
1658 // the compiled version to the intrinsic version.
1659 bool AbstractInterpreter::can_be_compiled(methodHandle m) {
1660   switch (method_kind(m)) {
1661     case Interpreter::java_lang_math_sin     : // fall thru
1662     case Interpreter::java_lang_math_cos     : // fall thru
1663     case Interpreter::java_lang_math_tan     : // fall thru
1664     case Interpreter::java_lang_math_abs     : // fall thru
1665     case Interpreter::java_lang_math_log     : // fall thru
1666     case Interpreter::java_lang_math_log10   : // fall thru
1667     case Interpreter::java_lang_math_sqrt    : // fall thru
1668     case Interpreter::java_lang_math_pow     : // fall thru
1669     case Interpreter::java_lang_math_exp     :
1670       return false;
1671     default:
1672       return true;
1673   }
1674 }
1675 
1676 // How much stack a method activation needs in words.
1677 int AbstractInterpreter::size_top_interpreter_activation(Method* method) {
1678 
1679   const int stub_code = 4;  // see generate_call_stub
1680   // Save space for one monitor to get into the interpreted method in case
1681   // the method is synchronized
1682   int monitor_size    = method->is_synchronized() ?
1683                                 1*frame::interpreter_frame_monitor_size() : 0;
1684 
1685   // total overhead size: entry_size + (saved rbp, thru expr stack bottom).
1686   // be sure to change this if you add/subtract anything to/from the overhead area
1687   const int overhead_size = -frame::interpreter_frame_initial_sp_offset;
1688 
1689   const int method_stack = (method->max_locals() + method->max_stack()) *
1690                            Interpreter::stackElementWords;
1691   return overhead_size + method_stack + stub_code;
1692 }
1693 
1694 //------------------------------------------------------------------------------------------------------------------------
1695 // Exceptions
1696 
1697 void TemplateInterpreterGenerator::generate_throw_exception() {
1698   // Entry point in previous activation (i.e., if the caller was interpreted)
1699   Interpreter::_rethrow_exception_entry = __ pc();
1700   const Register thread = rcx;
1701 
1702   // Restore sp to interpreter_frame_last_sp even though we are going
1703   // to empty the expression stack for the exception processing.
1704   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
1705   // rax,: exception
1706   // rdx: return address/pc that threw exception
1707   __ restore_bcp();                              // rsi points to call/send
1708   __ restore_locals();
1709 
1710   // Entry point for exceptions thrown within interpreter code
1711   Interpreter::_throw_exception_entry = __ pc();
1712   // expression stack is undefined here
1713   // rax,: exception
1714   // rsi: exception bcp
1715   __ verify_oop(rax);
1716 
1717   // expression stack must be empty before entering the VM in case of an exception
1718   __ empty_expression_stack();
1719   __ empty_FPU_stack();
1720   // find exception handler address and preserve exception oop
1721   __ call_VM(rdx, CAST_FROM_FN_PTR(address, InterpreterRuntime::exception_handler_for_exception), rax);
1722   // rax,: exception handler entry point
1723   // rdx: preserved exception oop
1724   // rsi: bcp for exception handler
1725   __ push_ptr(rdx);                              // push exception which is now the only value on the stack
1726   __ jmp(rax);                                   // jump to exception handler (may be _remove_activation_entry!)
1727 
1728   // If the exception is not handled in the current frame the frame is removed and
1729   // the exception is rethrown (i.e. exception continuation is _rethrow_exception).
1730   //
1731   // Note: At this point the bci is still the bxi for the instruction which caused
1732   //       the exception and the expression stack is empty. Thus, for any VM calls
1733   //       at this point, GC will find a legal oop map (with empty expression stack).
1734 
1735   // In current activation
1736   // tos: exception
1737   // rsi: exception bcp
1738 
1739   //
1740   // JVMTI PopFrame support
1741   //
1742 
1743    Interpreter::_remove_activation_preserving_args_entry = __ pc();
1744   __ empty_expression_stack();
1745   __ empty_FPU_stack();
1746   // Set the popframe_processing bit in pending_popframe_condition indicating that we are
1747   // currently handling popframe, so that call_VMs that may happen later do not trigger new
1748   // popframe handling cycles.
1749   __ get_thread(thread);
1750   __ movl(rdx, Address(thread, JavaThread::popframe_condition_offset()));
1751   __ orl(rdx, JavaThread::popframe_processing_bit);
1752   __ movl(Address(thread, JavaThread::popframe_condition_offset()), rdx);
1753 
1754   {
1755     // Check to see whether we are returning to a deoptimized frame.
1756     // (The PopFrame call ensures that the caller of the popped frame is
1757     // either interpreted or compiled and deoptimizes it if compiled.)
1758     // In this case, we can't call dispatch_next() after the frame is
1759     // popped, but instead must save the incoming arguments and restore
1760     // them after deoptimization has occurred.
1761     //
1762     // Note that we don't compare the return PC against the
1763     // deoptimization blob's unpack entry because of the presence of
1764     // adapter frames in C2.
1765     Label caller_not_deoptimized;
1766     __ movptr(rdx, Address(rbp, frame::return_addr_offset * wordSize));
1767     __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::interpreter_contains), rdx);
1768     __ testl(rax, rax);
1769     __ jcc(Assembler::notZero, caller_not_deoptimized);
1770 
1771     // Compute size of arguments for saving when returning to deoptimized caller
1772     __ get_method(rax);
1773     __ movptr(rax, Address(rax, Method::const_offset()));
1774     __ load_unsigned_short(rax, Address(rax, ConstMethod::size_of_parameters_offset()));
1775     __ shlptr(rax, Interpreter::logStackElementSize);
1776     __ restore_locals();
1777     __ subptr(rdi, rax);
1778     __ addptr(rdi, wordSize);
1779     // Save these arguments
1780     __ get_thread(thread);
1781     __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, Deoptimization::popframe_preserve_args), thread, rax, rdi);
1782 
1783     __ remove_activation(vtos, rdx,
1784                          /* throw_monitor_exception */ false,
1785                          /* install_monitor_exception */ false,
1786                          /* notify_jvmdi */ false);
1787 
1788     // Inform deoptimization that it is responsible for restoring these arguments
1789     __ get_thread(thread);
1790     __ movl(Address(thread, JavaThread::popframe_condition_offset()), JavaThread::popframe_force_deopt_reexecution_bit);
1791 
1792     // Continue in deoptimization handler
1793     __ jmp(rdx);
1794 
1795     __ bind(caller_not_deoptimized);
1796   }
1797 
1798   __ remove_activation(vtos, rdx,
1799                        /* throw_monitor_exception */ false,
1800                        /* install_monitor_exception */ false,
1801                        /* notify_jvmdi */ false);
1802 
1803   // Finish with popframe handling
1804   // A previous I2C followed by a deoptimization might have moved the
1805   // outgoing arguments further up the stack. PopFrame expects the
1806   // mutations to those outgoing arguments to be preserved and other
1807   // constraints basically require this frame to look exactly as
1808   // though it had previously invoked an interpreted activation with
1809   // no space between the top of the expression stack (current
1810   // last_sp) and the top of stack. Rather than force deopt to
1811   // maintain this kind of invariant all the time we call a small
1812   // fixup routine to move the mutated arguments onto the top of our
1813   // expression stack if necessary.
1814   __ mov(rax, rsp);
1815   __ movptr(rbx, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1816   __ get_thread(thread);
1817   // PC must point into interpreter here
1818   __ set_last_Java_frame(thread, noreg, rbp, __ pc());
1819   __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), thread, rax, rbx);
1820   __ get_thread(thread);
1821   __ reset_last_Java_frame(thread, true);
1822   // Restore the last_sp and null it out
1823   __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1824   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
1825 
1826   __ restore_bcp();
1827   __ restore_locals();
1828   // The method data pointer was incremented already during
1829   // call profiling. We have to restore the mdp for the current bcp.
1830   if (ProfileInterpreter) {
1831     __ set_method_data_pointer_for_bcp();
1832   }
1833 
1834   // Clear the popframe condition flag
1835   __ get_thread(thread);
1836   __ movl(Address(thread, JavaThread::popframe_condition_offset()), JavaThread::popframe_inactive);
1837 
1838 #if INCLUDE_JVMTI
1839   if (EnableInvokeDynamic) {
1840     Label L_done;
1841     const Register local0 = rdi;
1842 
1843     __ cmpb(Address(rsi, 0), Bytecodes::_invokestatic);
1844     __ jcc(Assembler::notEqual, L_done);
1845 
1846     // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call.
1847     // Detect such a case in the InterpreterRuntime function and return the member name argument, or NULL.
1848 
1849     __ get_method(rdx);
1850     __ movptr(rax, Address(local0, 0));
1851     __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), rax, rdx, rsi);
1852 
1853     __ testptr(rax, rax);
1854     __ jcc(Assembler::zero, L_done);
1855 
1856     __ movptr(Address(rbx, 0), rax);
1857     __ bind(L_done);
1858   }
1859 #endif // INCLUDE_JVMTI
1860 
1861   __ dispatch_next(vtos);
1862   // end of PopFrame support
1863 
1864   Interpreter::_remove_activation_entry = __ pc();
1865 
1866   // preserve exception over this code sequence
1867   __ pop_ptr(rax);
1868   __ get_thread(thread);
1869   __ movptr(Address(thread, JavaThread::vm_result_offset()), rax);
1870   // remove the activation (without doing throws on illegalMonitorExceptions)
1871   __ remove_activation(vtos, rdx, false, true, false);
1872   // restore exception
1873   __ get_thread(thread);
1874   __ get_vm_result(rax, thread);
1875 
1876   // Inbetween activations - previous activation type unknown yet
1877   // compute continuation point - the continuation point expects
1878   // the following registers set up:
1879   //
1880   // rax: exception
1881   // rdx: return address/pc that threw exception
1882   // rsp: expression stack of caller
1883   // rbp: rbp, of caller
1884   __ push(rax);                                  // save exception
1885   __ push(rdx);                                  // save return address
1886   __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), thread, rdx);
1887   __ mov(rbx, rax);                              // save exception handler
1888   __ pop(rdx);                                   // restore return address
1889   __ pop(rax);                                   // restore exception
1890   // Note that an "issuing PC" is actually the next PC after the call
1891   __ jmp(rbx);                                   // jump to exception handler of caller
1892 }
1893 
1894 
1895 //
1896 // JVMTI ForceEarlyReturn support
1897 //
1898 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
1899   address entry = __ pc();
1900   const Register thread = rcx;
1901 
1902   __ restore_bcp();
1903   __ restore_locals();
1904   __ empty_expression_stack();
1905   __ empty_FPU_stack();
1906   __ load_earlyret_value(state);
1907 
1908   __ get_thread(thread);
1909   __ movptr(rcx, Address(thread, JavaThread::jvmti_thread_state_offset()));
1910   const Address cond_addr(rcx, JvmtiThreadState::earlyret_state_offset());
1911 
1912   // Clear the earlyret state
1913   __ movl(cond_addr, JvmtiThreadState::earlyret_inactive);
1914 
1915   __ remove_activation(state, rsi,
1916                        false, /* throw_monitor_exception */
1917                        false, /* install_monitor_exception */
1918                        true); /* notify_jvmdi */
1919   __ jmp(rsi);
1920   return entry;
1921 } // end of ForceEarlyReturn support
1922 
1923 
1924 //------------------------------------------------------------------------------------------------------------------------
1925 // Helper for vtos entry point generation
1926 
1927 void TemplateInterpreterGenerator::set_vtos_entry_points (Template* t, address& bep, address& cep, address& sep, address& aep, address& iep, address& lep, address& fep, address& dep, address& vep) {
1928   assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
1929   Label L;
1930   fep = __ pc(); __ push(ftos); __ jmp(L);
1931   dep = __ pc(); __ push(dtos); __ jmp(L);
1932   lep = __ pc(); __ push(ltos); __ jmp(L);
1933   aep = __ pc(); __ push(atos); __ jmp(L);
1934   bep = cep = sep =             // fall through
1935   iep = __ pc(); __ push(itos); // fall through
1936   vep = __ pc(); __ bind(L);    // fall through
1937   generate_and_dispatch(t);
1938 }
1939 
1940 //------------------------------------------------------------------------------------------------------------------------
1941 // Generation of individual instructions
1942 
1943 // helpers for generate_and_dispatch
1944 
1945 
1946 
1947 InterpreterGenerator::InterpreterGenerator(StubQueue* code)
1948  : TemplateInterpreterGenerator(code) {
1949    generate_all(); // down here so it can be "virtual"
1950 }
1951 
1952 //------------------------------------------------------------------------------------------------------------------------
1953 
1954 // Non-product code
1955 #ifndef PRODUCT
1956 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
1957   address entry = __ pc();
1958 
1959   // prepare expression stack
1960   __ pop(rcx);          // pop return address so expression stack is 'pure'
1961   __ push(state);       // save tosca
1962 
1963   // pass tosca registers as arguments & call tracer
1964   __ call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::trace_bytecode), rcx, rax, rdx);
1965   __ mov(rcx, rax);     // make sure return address is not destroyed by pop(state)
1966   __ pop(state);        // restore tosca
1967 
1968   // return
1969   __ jmp(rcx);
1970 
1971   return entry;
1972 }
1973 
1974 
1975 void TemplateInterpreterGenerator::count_bytecode() {
1976   __ incrementl(ExternalAddress((address) &BytecodeCounter::_counter_value));
1977 }
1978 
1979 
1980 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
1981   __ incrementl(ExternalAddress((address) &BytecodeHistogram::_counters[t->bytecode()]));
1982 }
1983 
1984 
1985 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
1986   __ mov32(ExternalAddress((address) &BytecodePairHistogram::_index), rbx);
1987   __ shrl(rbx, BytecodePairHistogram::log2_number_of_codes);
1988   __ orl(rbx, ((int)t->bytecode()) << BytecodePairHistogram::log2_number_of_codes);
1989   ExternalAddress table((address) BytecodePairHistogram::_counters);
1990   Address index(noreg, rbx, Address::times_4);
1991   __ incrementl(ArrayAddress(table, index));
1992 }
1993 
1994 
1995 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
1996   // Call a little run-time stub to avoid blow-up for each bytecode.
1997   // The run-time runtime saves the right registers, depending on
1998   // the tosca in-state for the given template.
1999   assert(Interpreter::trace_code(t->tos_in()) != NULL,
2000          "entry must have been generated");
2001   __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in())));
2002 }
2003 
2004 
2005 void TemplateInterpreterGenerator::stop_interpreter_at() {
2006   Label L;
2007   __ cmp32(ExternalAddress((address) &BytecodeCounter::_counter_value),
2008            StopInterpreterAt);
2009   __ jcc(Assembler::notEqual, L);
2010   __ int3();
2011   __ bind(L);
2012 }
2013 #endif // !PRODUCT
2014 #endif // CC_INTERP