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 
 750 #if INCLUDE_ALL_GCS
 751     if (UseShenandoahGC) {
 752       Label notObj;
 753 
 754       // Needs GC barriers
 755       __ cmpl(rdx, atos);
 756       __ jcc(Assembler::notEqual, notObj);
 757       __ load_heap_oop(rax, field_address);
 758       __ jmp(xreturn_path);
 759 
 760       __ bind(notObj);
 761     }
 762 #endif
 763 
 764 #ifdef ASSERT
 765     Label okay;
 766     __ cmpl(rdx, atos);
 767     __ jcc(Assembler::equal, okay);
 768     __ cmpl(rdx, itos);
 769     __ jcc(Assembler::equal, okay);
 770     __ stop("what type is this?");
 771     __ bind(okay);
 772 #endif // ASSERT
 773     // All the rest are a 32 bit wordsize
 774     // This is ok for now. Since fast accessors should be going away
 775     __ movptr(rax, field_address);
 776 
 777     __ bind(xreturn_path);
 778 
 779     // _ireturn/_areturn
 780     __ pop(rdi);                               // get return address
 781     __ mov(rsp, rsi);                          // set sp to sender sp
 782     __ jmp(rdi);
 783 
 784     // generate a vanilla interpreter entry as the slow path
 785     __ bind(slow_path);
 786 
 787     (void) generate_normal_entry(false);
 788     return entry_point;
 789   }
 790   return NULL;
 791 
 792 }
 793 
 794 // Method entry for java.lang.ref.Reference.get.
 795 address InterpreterGenerator::generate_Reference_get_entry(void) {
 796 #if INCLUDE_ALL_GCS
 797   // Code: _aload_0, _getfield, _areturn
 798   // parameter size = 1
 799   //
 800   // The code that gets generated by this routine is split into 2 parts:
 801   //    1. The "intrinsified" code for G1 (or any SATB based GC),
 802   //    2. The slow path - which is an expansion of the regular method entry.
 803   //
 804   // Notes:-
 805   // * In the G1 code we do not check whether we need to block for
 806   //   a safepoint. If G1 is enabled then we must execute the specialized
 807   //   code for Reference.get (except when the Reference object is null)
 808   //   so that we can log the value in the referent field with an SATB
 809   //   update buffer.
 810   //   If the code for the getfield template is modified so that the
 811   //   G1 pre-barrier code is executed when the current method is
 812   //   Reference.get() then going through the normal method entry
 813   //   will be fine.
 814   // * The G1 code below can, however, check the receiver object (the instance
 815   //   of java.lang.Reference) and jump to the slow path if null. If the
 816   //   Reference object is null then we obviously cannot fetch the referent
 817   //   and so we don't need to call the G1 pre-barrier. Thus we can use the
 818   //   regular method entry code to generate the NPE.
 819   //
 820   // This code is based on generate_accessor_enty.
 821 
 822   // rbx,: Method*
 823   // rcx: receiver (preserve for slow entry into asm interpreter)
 824 
 825   // rsi: senderSP must preserved for slow path, set SP to it on fast path
 826 
 827   address entry = __ pc();
 828 
 829   const int referent_offset = java_lang_ref_Reference::referent_offset;
 830   guarantee(referent_offset > 0, "referent offset not initialized");
 831 
 832   if (UseG1GC || UseShenandoahGC) {
 833     Label slow_path;
 834 
 835     // Check if local 0 != NULL
 836     // If the receiver is null then it is OK to jump to the slow path.
 837     __ movptr(rax, Address(rsp, wordSize));
 838     __ testptr(rax, rax);
 839     __ jcc(Assembler::zero, slow_path);
 840 
 841     // rax: local 0 (must be preserved across the G1 barrier call)
 842     //
 843     // rbx: method (at this point it's scratch)
 844     // rcx: receiver (at this point it's scratch)
 845     // rdx: scratch
 846     // rdi: scratch
 847     //
 848     // rsi: sender sp
 849 
 850     // Preserve the sender sp in case the pre-barrier
 851     // calls the runtime
 852     __ push(rsi);
 853 
 854     // Load the value of the referent field.
 855     const Address field_address(rax, referent_offset);
 856 #if INCLUDE_ALL_GCS
 857     if (UseShenandoahGC) {
 858       // Needs GC barriers
 859       __ load_heap_oop(rax, field_address);
 860     } else
 861 #endif
 862     __ movptr(rax, field_address);
 863 
 864     // Generate the G1 pre-barrier code to log the value of
 865     // the referent field in an SATB buffer.
 866     if (!UseShenandoahGC || ShenandoahKeepAliveBarrier) {
 867     __ get_thread(rcx);
 868     __ g1_write_barrier_pre(noreg /* obj */,
 869                             rax /* pre_val */,
 870                             rcx /* thread */,
 871                             rbx /* tmp */,
 872                             true /* tosca_save */,
 873                             true /* expand_call */);
 874     }
 875 
 876     // _areturn
 877     __ pop(rsi);                // get sender sp
 878     __ pop(rdi);                // get return address
 879     __ mov(rsp, rsi);           // set sp to sender sp
 880     __ jmp(rdi);
 881 
 882     __ bind(slow_path);
 883     (void) generate_normal_entry(false);
 884 
 885     return entry;
 886   }
 887 #endif // INCLUDE_ALL_GCS
 888 
 889   // If G1 is not enabled then attempt to go through the accessor entry point
 890   // Reference.get is an accessor
 891   return generate_accessor_entry();
 892 }
 893 
 894 /**
 895  * Method entry for static native methods:
 896  *   int java.util.zip.CRC32.update(int crc, int b)
 897  */
 898 address InterpreterGenerator::generate_CRC32_update_entry() {
 899   if (UseCRC32Intrinsics) {
 900     address entry = __ pc();
 901 
 902     // rbx,: Method*
 903     // rsi: senderSP must preserved for slow path, set SP to it on fast path
 904     // rdx: scratch
 905     // rdi: scratch
 906 
 907     Label slow_path;
 908     // If we need a safepoint check, generate full interpreter entry.
 909     ExternalAddress state(SafepointSynchronize::address_of_state());
 910     __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
 911              SafepointSynchronize::_not_synchronized);
 912     __ jcc(Assembler::notEqual, slow_path);
 913 
 914     // We don't generate local frame and don't align stack because
 915     // we call stub code and there is no safepoint on this path.
 916 
 917     // Load parameters
 918     const Register crc = rax;  // crc
 919     const Register val = rdx;  // source java byte value
 920     const Register tbl = rdi;  // scratch
 921 
 922     // Arguments are reversed on java expression stack
 923     __ movl(val, Address(rsp,   wordSize)); // byte value
 924     __ movl(crc, Address(rsp, 2*wordSize)); // Initial CRC
 925 
 926     __ lea(tbl, ExternalAddress(StubRoutines::crc_table_addr()));
 927     __ notl(crc); // ~crc
 928     __ update_byte_crc32(crc, val, tbl);
 929     __ notl(crc); // ~crc
 930     // result in rax
 931 
 932     // _areturn
 933     __ pop(rdi);                // get return address
 934     __ mov(rsp, rsi);           // set sp to sender sp
 935     __ jmp(rdi);
 936 
 937     // generate a vanilla native entry as the slow path
 938     __ bind(slow_path);
 939 
 940     (void) generate_native_entry(false);
 941 
 942     return entry;
 943   }
 944   return generate_native_entry(false);
 945 }
 946 
 947 /**
 948  * Method entry for static native methods:
 949  *   int java.util.zip.CRC32.updateBytes(int crc, byte[] b, int off, int len)
 950  *   int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
 951  */
 952 address InterpreterGenerator::generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
 953   if (UseCRC32Intrinsics) {
 954     address entry = __ pc();
 955 
 956     // rbx,: Method*
 957     // rsi: senderSP must preserved for slow path, set SP to it on fast path
 958     // rdx: scratch
 959     // rdi: scratch
 960 
 961     Label slow_path;
 962     // If we need a safepoint check, generate full interpreter entry.
 963     ExternalAddress state(SafepointSynchronize::address_of_state());
 964     __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
 965              SafepointSynchronize::_not_synchronized);
 966     __ jcc(Assembler::notEqual, slow_path);
 967 
 968     // We don't generate local frame and don't align stack because
 969     // we call stub code and there is no safepoint on this path.
 970 
 971     // Load parameters
 972     const Register crc = rax;  // crc
 973     const Register buf = rdx;  // source java byte array address
 974     const Register len = rdi;  // length
 975 
 976     // Arguments are reversed on java expression stack
 977     __ movl(len,   Address(rsp,   wordSize)); // Length
 978     // Calculate address of start element
 979     if (kind == Interpreter::java_util_zip_CRC32_updateByteBuffer) {
 980       __ movptr(buf, Address(rsp, 3*wordSize)); // long buf
 981       __ addptr(buf, Address(rsp, 2*wordSize)); // + offset
 982       __ movl(crc,   Address(rsp, 5*wordSize)); // Initial CRC
 983     } else {
 984       __ movptr(buf, Address(rsp, 3*wordSize)); // byte[] array
 985       __ addptr(buf, arrayOopDesc::base_offset_in_bytes(T_BYTE)); // + header size
 986       __ addptr(buf, Address(rsp, 2*wordSize)); // + offset
 987       __ movl(crc,   Address(rsp, 4*wordSize)); // Initial CRC
 988     }
 989 
 990     __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, StubRoutines::updateBytesCRC32()), crc, buf, len);
 991     // result in rax
 992 
 993     // _areturn
 994     __ pop(rdi);                // get return address
 995     __ mov(rsp, rsi);           // set sp to sender sp
 996     __ jmp(rdi);
 997 
 998     // generate a vanilla native entry as the slow path
 999     __ bind(slow_path);
1000 
1001     (void) generate_native_entry(false);
1002 
1003     return entry;
1004   }
1005   return generate_native_entry(false);
1006 }
1007 
1008 //
1009 // Interpreter stub for calling a native method. (asm interpreter)
1010 // This sets up a somewhat different looking stack for calling the native method
1011 // than the typical interpreter frame setup.
1012 //
1013 
1014 address InterpreterGenerator::generate_native_entry(bool synchronized) {
1015   // determine code generation flags
1016   bool inc_counter  = UseCompiler || CountCompiledCalls;
1017 
1018   // rbx,: Method*
1019   // rsi: sender sp
1020   // rsi: previous interpreter state (C++ interpreter) must preserve
1021   address entry_point = __ pc();
1022 
1023   const Address constMethod       (rbx, Method::const_offset());
1024   const Address access_flags      (rbx, Method::access_flags_offset());
1025   const Address size_of_parameters(rcx, ConstMethod::size_of_parameters_offset());
1026 
1027   // get parameter size (always needed)
1028   __ movptr(rcx, constMethod);
1029   __ load_unsigned_short(rcx, size_of_parameters);
1030 
1031   // native calls don't need the stack size check since they have no expression stack
1032   // and the arguments are already on the stack and we only add a handful of words
1033   // to the stack
1034 
1035   // rbx,: Method*
1036   // rcx: size of parameters
1037   // rsi: sender sp
1038 
1039   __ pop(rax);                                       // get return address
1040   // for natives the size of locals is zero
1041 
1042   // compute beginning of parameters (rdi)
1043   __ lea(rdi, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize));
1044 
1045 
1046   // add 2 zero-initialized slots for native calls
1047   // NULL result handler
1048   __ push((int32_t)NULL_WORD);
1049   // NULL oop temp (mirror or jni oop result)
1050   __ push((int32_t)NULL_WORD);
1051 
1052   // initialize fixed part of activation frame
1053   generate_fixed_frame(true);
1054 
1055   // make sure method is native & not abstract
1056 #ifdef ASSERT
1057   __ movl(rax, access_flags);
1058   {
1059     Label L;
1060     __ testl(rax, JVM_ACC_NATIVE);
1061     __ jcc(Assembler::notZero, L);
1062     __ stop("tried to execute non-native method as native");
1063     __ bind(L);
1064   }
1065   { Label L;
1066     __ testl(rax, JVM_ACC_ABSTRACT);
1067     __ jcc(Assembler::zero, L);
1068     __ stop("tried to execute abstract method in interpreter");
1069     __ bind(L);
1070   }
1071 #endif
1072 
1073   // Since at this point in the method invocation the exception handler
1074   // would try to exit the monitor of synchronized methods which hasn't
1075   // been entered yet, we set the thread local variable
1076   // _do_not_unlock_if_synchronized to true. The remove_activation will
1077   // check this flag.
1078 
1079   __ get_thread(rax);
1080   const Address do_not_unlock_if_synchronized(rax,
1081         in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1082   __ movbool(do_not_unlock_if_synchronized, true);
1083 
1084   // increment invocation count & check for overflow
1085   Label invocation_counter_overflow;
1086   if (inc_counter) {
1087     generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
1088   }
1089 
1090   Label continue_after_compile;
1091   __ bind(continue_after_compile);
1092 
1093   bang_stack_shadow_pages(true);
1094 
1095   // reset the _do_not_unlock_if_synchronized flag
1096   __ get_thread(rax);
1097   __ movbool(do_not_unlock_if_synchronized, false);
1098 
1099   // check for synchronized methods
1100   // Must happen AFTER invocation_counter check and stack overflow check,
1101   // so method is not locked if overflows.
1102   //
1103   if (synchronized) {
1104     lock_method();
1105   } else {
1106     // no synchronization necessary
1107 #ifdef ASSERT
1108       { Label L;
1109         __ movl(rax, access_flags);
1110         __ testl(rax, JVM_ACC_SYNCHRONIZED);
1111         __ jcc(Assembler::zero, L);
1112         __ stop("method needs synchronization");
1113         __ bind(L);
1114       }
1115 #endif
1116   }
1117 
1118   // start execution
1119 #ifdef ASSERT
1120   { Label L;
1121     const Address monitor_block_top (rbp,
1122                  frame::interpreter_frame_monitor_block_top_offset * wordSize);
1123     __ movptr(rax, monitor_block_top);
1124     __ cmpptr(rax, rsp);
1125     __ jcc(Assembler::equal, L);
1126     __ stop("broken stack frame setup in interpreter");
1127     __ bind(L);
1128   }
1129 #endif
1130 
1131   // jvmti/dtrace support
1132   __ notify_method_entry();
1133 
1134   // work registers
1135   const Register method = rbx;
1136   const Register thread = rdi;
1137   const Register t      = rcx;
1138 
1139   // allocate space for parameters
1140   __ get_method(method);
1141   __ movptr(t, Address(method, Method::const_offset()));
1142   __ load_unsigned_short(t, Address(t, ConstMethod::size_of_parameters_offset()));
1143 
1144   __ shlptr(t, Interpreter::logStackElementSize);
1145   __ addptr(t, 2*wordSize);     // allocate two more slots for JNIEnv and possible mirror
1146   __ subptr(rsp, t);
1147   __ andptr(rsp, -(StackAlignmentInBytes)); // gcc needs 16 byte aligned stacks to do XMM intrinsics
1148 
1149   // get signature handler
1150   { Label L;
1151     __ movptr(t, Address(method, Method::signature_handler_offset()));
1152     __ testptr(t, t);
1153     __ jcc(Assembler::notZero, L);
1154     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), method);
1155     __ get_method(method);
1156     __ movptr(t, Address(method, Method::signature_handler_offset()));
1157     __ bind(L);
1158   }
1159 
1160   // call signature handler
1161   assert(InterpreterRuntime::SignatureHandlerGenerator::from() == rdi, "adjust this code");
1162   assert(InterpreterRuntime::SignatureHandlerGenerator::to  () == rsp, "adjust this code");
1163   assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == t  , "adjust this code");
1164   // The generated handlers do not touch RBX (the method oop).
1165   // However, large signatures cannot be cached and are generated
1166   // each time here.  The slow-path generator will blow RBX
1167   // sometime, so we must reload it after the call.
1168   __ call(t);
1169   __ get_method(method);        // slow path call blows RBX on DevStudio 5.0
1170 
1171   // result handler is in rax,
1172   // set result handler
1173   __ movptr(Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize), rax);
1174 
1175   // pass mirror handle if static call
1176   { Label L;
1177     const int mirror_offset = in_bytes(Klass::java_mirror_offset());
1178     __ movl(t, Address(method, Method::access_flags_offset()));
1179     __ testl(t, JVM_ACC_STATIC);
1180     __ jcc(Assembler::zero, L);
1181     // get mirror
1182     __ movptr(t, Address(method, Method:: const_offset()));
1183     __ movptr(t, Address(t, ConstMethod::constants_offset()));
1184     __ movptr(t, Address(t, ConstantPool::pool_holder_offset_in_bytes()));
1185     __ movptr(t, Address(t, mirror_offset));
1186     // copy mirror into activation frame
1187     __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize), t);
1188     // pass handle to mirror
1189     __ lea(t, Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize));
1190     __ movptr(Address(rsp, wordSize), t);
1191     __ bind(L);
1192   }
1193 
1194   // get native function entry point
1195   { Label L;
1196     __ movptr(rax, Address(method, Method::native_function_offset()));
1197     ExternalAddress unsatisfied(SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
1198     __ cmpptr(rax, unsatisfied.addr());
1199     __ jcc(Assembler::notEqual, L);
1200     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), method);
1201     __ get_method(method);
1202     __ movptr(rax, Address(method, Method::native_function_offset()));
1203     __ bind(L);
1204   }
1205 
1206   // pass JNIEnv
1207   __ get_thread(thread);
1208   __ lea(t, Address(thread, JavaThread::jni_environment_offset()));
1209   __ movptr(Address(rsp, 0), t);
1210 
1211   // set_last_Java_frame_before_call
1212   // It is enough that the pc()
1213   // points into the right code segment. It does not have to be the correct return pc.
1214   __ set_last_Java_frame(thread, noreg, rbp, __ pc());
1215 
1216   // change thread state
1217 #ifdef ASSERT
1218   { Label L;
1219     __ movl(t, Address(thread, JavaThread::thread_state_offset()));
1220     __ cmpl(t, _thread_in_Java);
1221     __ jcc(Assembler::equal, L);
1222     __ stop("Wrong thread state in native stub");
1223     __ bind(L);
1224   }
1225 #endif
1226 
1227   // Change state to native
1228   __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_native);
1229   __ call(rax);
1230 
1231   // result potentially in rdx:rax or ST0
1232 
1233   // Verify or restore cpu control state after JNI call
1234   __ restore_cpu_control_state_after_jni();
1235 
1236   // save potential result in ST(0) & rdx:rax
1237   // (if result handler is the T_FLOAT or T_DOUBLE handler, result must be in ST0 -
1238   // the check is necessary to avoid potential Intel FPU overflow problems by saving/restoring 'empty' FPU registers)
1239   // It is safe to do this push because state is _thread_in_native and return address will be found
1240   // via _last_native_pc and not via _last_jave_sp
1241 
1242   // NOTE: the order of theses push(es) is known to frame::interpreter_frame_result.
1243   // If the order changes or anything else is added to the stack the code in
1244   // interpreter_frame_result will have to be changed.
1245 
1246   { Label L;
1247     Label push_double;
1248     ExternalAddress float_handler(AbstractInterpreter::result_handler(T_FLOAT));
1249     ExternalAddress double_handler(AbstractInterpreter::result_handler(T_DOUBLE));
1250     __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize),
1251               float_handler.addr());
1252     __ jcc(Assembler::equal, push_double);
1253     __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize),
1254               double_handler.addr());
1255     __ jcc(Assembler::notEqual, L);
1256     __ bind(push_double);
1257     __ push(dtos);
1258     __ bind(L);
1259   }
1260   __ push(ltos);
1261 
1262   // change thread state
1263   __ get_thread(thread);
1264   __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_native_trans);
1265   if(os::is_MP()) {
1266     if (UseMembar) {
1267       // Force this write out before the read below
1268       __ membar(Assembler::Membar_mask_bits(
1269            Assembler::LoadLoad | Assembler::LoadStore |
1270            Assembler::StoreLoad | Assembler::StoreStore));
1271     } else {
1272       // Write serialization page so VM thread can do a pseudo remote membar.
1273       // We use the current thread pointer to calculate a thread specific
1274       // offset to write to within the page. This minimizes bus traffic
1275       // due to cache line collision.
1276       __ serialize_memory(thread, rcx);
1277     }
1278   }
1279 
1280   if (AlwaysRestoreFPU) {
1281     //  Make sure the control word is correct.
1282     __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
1283   }
1284 
1285   // check for safepoint operation in progress and/or pending suspend requests
1286   { Label Continue;
1287 
1288     __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
1289              SafepointSynchronize::_not_synchronized);
1290 
1291     Label L;
1292     __ jcc(Assembler::notEqual, L);
1293     __ cmpl(Address(thread, JavaThread::suspend_flags_offset()), 0);
1294     __ jcc(Assembler::equal, Continue);
1295     __ bind(L);
1296 
1297     // Don't use call_VM as it will see a possible pending exception and forward it
1298     // and never return here preventing us from clearing _last_native_pc down below.
1299     // Also can't use call_VM_leaf either as it will check to see if rsi & rdi are
1300     // preserved and correspond to the bcp/locals pointers. So we do a runtime call
1301     // by hand.
1302     //
1303     __ push(thread);
1304     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address,
1305                                             JavaThread::check_special_condition_for_native_trans)));
1306     __ increment(rsp, wordSize);
1307     __ get_thread(thread);
1308 
1309     __ bind(Continue);
1310   }
1311 
1312   // change thread state
1313   __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_Java);
1314 
1315   __ reset_last_Java_frame(thread, true);
1316 
1317   // reset handle block
1318   __ movptr(t, Address(thread, JavaThread::active_handles_offset()));
1319   __ movl(Address(t, JNIHandleBlock::top_offset_in_bytes()), NULL_WORD);
1320 
1321   // If result was an oop then unbox and save it in the frame
1322   {
1323     Label no_oop;
1324     ExternalAddress handler(AbstractInterpreter::result_handler(T_OBJECT));
1325     __ cmpptr(Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize),
1326               handler.addr());
1327     __ jcc(Assembler::notEqual, no_oop);
1328     __ cmpptr(Address(rsp, 0), (int32_t)NULL_WORD);
1329     __ pop(ltos);
1330     // Unbox oop result, e.g. JNIHandles::resolve value.
1331     __ resolve_jobject(rax /* value */,
1332                        thread /* thread */,
1333                        t /* tmp */);
1334     __ movptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset)*wordSize), rax);
1335     // keep stack depth as expected by pushing oop which will eventually be discarded
1336     __ push(ltos);
1337     __ bind(no_oop);
1338   }
1339 
1340   {
1341      Label no_reguard;
1342      __ cmpl(Address(thread, JavaThread::stack_guard_state_offset()), JavaThread::stack_guard_yellow_disabled);
1343      __ jcc(Assembler::notEqual, no_reguard);
1344 
1345      __ pusha();
1346      __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
1347      __ popa();
1348 
1349      __ bind(no_reguard);
1350    }
1351 
1352   // restore rsi to have legal interpreter frame,
1353   // i.e., bci == 0 <=> rsi == code_base()
1354   // Can't call_VM until bcp is within reasonable.
1355   __ get_method(method);      // method is junk from thread_in_native to now.
1356   __ movptr(rsi, Address(method,Method::const_offset()));   // get ConstMethod*
1357   __ lea(rsi, Address(rsi,ConstMethod::codes_offset()));    // get codebase
1358 
1359   // handle exceptions (exception handling will handle unlocking!)
1360   { Label L;
1361     __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
1362     __ jcc(Assembler::zero, L);
1363     // Note: At some point we may want to unify this with the code used in call_VM_base();
1364     //       i.e., we should use the StubRoutines::forward_exception code. For now this
1365     //       doesn't work here because the rsp is not correctly set at this point.
1366     __ MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_pending_exception));
1367     __ should_not_reach_here();
1368     __ bind(L);
1369   }
1370 
1371   // do unlocking if necessary
1372   { Label L;
1373     __ movl(t, Address(method, Method::access_flags_offset()));
1374     __ testl(t, JVM_ACC_SYNCHRONIZED);
1375     __ jcc(Assembler::zero, L);
1376     // the code below should be shared with interpreter macro assembler implementation
1377     { Label unlock;
1378       // BasicObjectLock will be first in list, since this is a synchronized method. However, need
1379       // to check that the object has not been unlocked by an explicit monitorexit bytecode.
1380       const Address monitor(rbp, frame::interpreter_frame_initial_sp_offset * wordSize - (int)sizeof(BasicObjectLock));
1381 
1382       __ lea(rdx, monitor);                   // address of first monitor
1383 
1384       __ movptr(t, Address(rdx, BasicObjectLock::obj_offset_in_bytes()));
1385       __ testptr(t, t);
1386       __ jcc(Assembler::notZero, unlock);
1387 
1388       // Entry already unlocked, need to throw exception
1389       __ MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
1390       __ should_not_reach_here();
1391 
1392       __ bind(unlock);
1393       __ unlock_object(rdx);
1394     }
1395     __ bind(L);
1396   }
1397 
1398   // jvmti/dtrace support
1399   // Note: This must happen _after_ handling/throwing any exceptions since
1400   //       the exception handler code notifies the runtime of method exits
1401   //       too. If this happens before, method entry/exit notifications are
1402   //       not properly paired (was bug - gri 11/22/99).
1403   __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI);
1404 
1405   // restore potential result in rdx:rax, call result handler to restore potential result in ST0 & handle result
1406   __ pop(ltos);
1407   __ movptr(t, Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize));
1408   __ call(t);
1409 
1410   // remove activation
1411   __ movptr(t, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); // get sender sp
1412   __ leave();                                // remove frame anchor
1413   __ pop(rdi);                               // get return address
1414   __ mov(rsp, t);                            // set sp to sender sp
1415   __ jmp(rdi);
1416 
1417   if (inc_counter) {
1418     // Handle overflow of counter and compile method
1419     __ bind(invocation_counter_overflow);
1420     generate_counter_overflow(&continue_after_compile);
1421   }
1422 
1423   return entry_point;
1424 }
1425 
1426 //
1427 // Generic interpreted method entry to (asm) interpreter
1428 //
1429 address InterpreterGenerator::generate_normal_entry(bool synchronized) {
1430   // determine code generation flags
1431   bool inc_counter  = UseCompiler || CountCompiledCalls;
1432 
1433   // rbx,: Method*
1434   // rsi: sender sp
1435   address entry_point = __ pc();
1436 
1437   const Address constMethod       (rbx, Method::const_offset());
1438   const Address access_flags      (rbx, Method::access_flags_offset());
1439   const Address size_of_parameters(rdx, ConstMethod::size_of_parameters_offset());
1440   const Address size_of_locals    (rdx, ConstMethod::size_of_locals_offset());
1441 
1442   // get parameter size (always needed)
1443   __ movptr(rdx, constMethod);
1444   __ load_unsigned_short(rcx, size_of_parameters);
1445 
1446   // rbx,: Method*
1447   // rcx: size of parameters
1448 
1449   // rsi: sender_sp (could differ from sp+wordSize if we were called via c2i )
1450 
1451   __ load_unsigned_short(rdx, size_of_locals);       // get size of locals in words
1452   __ subl(rdx, rcx);                                // rdx = no. of additional locals
1453 
1454   // see if we've got enough room on the stack for locals plus overhead.
1455   generate_stack_overflow_check();
1456 
1457   // get return address
1458   __ pop(rax);
1459 
1460   // compute beginning of parameters (rdi)
1461   __ lea(rdi, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize));
1462 
1463   // rdx - # of additional locals
1464   // allocate space for locals
1465   // explicitly initialize locals
1466   {
1467     Label exit, loop;
1468     __ testl(rdx, rdx);
1469     __ jcc(Assembler::lessEqual, exit);               // do nothing if rdx <= 0
1470     __ bind(loop);
1471     __ push((int32_t)NULL_WORD);                      // initialize local variables
1472     __ decrement(rdx);                                // until everything initialized
1473     __ jcc(Assembler::greater, loop);
1474     __ bind(exit);
1475   }
1476 
1477   // initialize fixed part of activation frame
1478   generate_fixed_frame(false);
1479 
1480   // make sure method is not native & not abstract
1481 #ifdef ASSERT
1482   __ movl(rax, access_flags);
1483   {
1484     Label L;
1485     __ testl(rax, JVM_ACC_NATIVE);
1486     __ jcc(Assembler::zero, L);
1487     __ stop("tried to execute native method as non-native");
1488     __ bind(L);
1489   }
1490   { Label L;
1491     __ testl(rax, JVM_ACC_ABSTRACT);
1492     __ jcc(Assembler::zero, L);
1493     __ stop("tried to execute abstract method in interpreter");
1494     __ bind(L);
1495   }
1496 #endif
1497 
1498   // Since at this point in the method invocation the exception handler
1499   // would try to exit the monitor of synchronized methods which hasn't
1500   // been entered yet, we set the thread local variable
1501   // _do_not_unlock_if_synchronized to true. The remove_activation will
1502   // check this flag.
1503 
1504   __ get_thread(rax);
1505   const Address do_not_unlock_if_synchronized(rax,
1506         in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1507   __ movbool(do_not_unlock_if_synchronized, true);
1508 
1509   __ profile_parameters_type(rax, rcx, rdx);
1510   // increment invocation count & check for overflow
1511   Label invocation_counter_overflow;
1512   Label profile_method;
1513   Label profile_method_continue;
1514   if (inc_counter) {
1515     generate_counter_incr(&invocation_counter_overflow, &profile_method, &profile_method_continue);
1516     if (ProfileInterpreter) {
1517       __ bind(profile_method_continue);
1518     }
1519   }
1520   Label continue_after_compile;
1521   __ bind(continue_after_compile);
1522 
1523   bang_stack_shadow_pages(false);
1524 
1525   // reset the _do_not_unlock_if_synchronized flag
1526   __ get_thread(rax);
1527   __ movbool(do_not_unlock_if_synchronized, false);
1528 
1529   // check for synchronized methods
1530   // Must happen AFTER invocation_counter check and stack overflow check,
1531   // so method is not locked if overflows.
1532   //
1533   if (synchronized) {
1534     // Allocate monitor and lock method
1535     lock_method();
1536   } else {
1537     // no synchronization necessary
1538 #ifdef ASSERT
1539       { Label L;
1540         __ movl(rax, access_flags);
1541         __ testl(rax, JVM_ACC_SYNCHRONIZED);
1542         __ jcc(Assembler::zero, L);
1543         __ stop("method needs synchronization");
1544         __ bind(L);
1545       }
1546 #endif
1547   }
1548 
1549   // start execution
1550 #ifdef ASSERT
1551   { Label L;
1552      const Address monitor_block_top (rbp,
1553                  frame::interpreter_frame_monitor_block_top_offset * wordSize);
1554     __ movptr(rax, monitor_block_top);
1555     __ cmpptr(rax, rsp);
1556     __ jcc(Assembler::equal, L);
1557     __ stop("broken stack frame setup in interpreter");
1558     __ bind(L);
1559   }
1560 #endif
1561 
1562   // jvmti support
1563   __ notify_method_entry();
1564 
1565   __ dispatch_next(vtos);
1566 
1567   // invocation counter overflow
1568   if (inc_counter) {
1569     if (ProfileInterpreter) {
1570       // We have decided to profile this method in the interpreter
1571       __ bind(profile_method);
1572       __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1573       __ set_method_data_pointer_for_bcp();
1574       __ get_method(rbx);
1575       __ jmp(profile_method_continue);
1576     }
1577     // Handle overflow of counter and compile method
1578     __ bind(invocation_counter_overflow);
1579     generate_counter_overflow(&continue_after_compile);
1580   }
1581 
1582   return entry_point;
1583 }
1584 
1585 //------------------------------------------------------------------------------------------------------------------------
1586 // Entry points
1587 //
1588 // Here we generate the various kind of entries into the interpreter.
1589 // The two main entry type are generic bytecode methods and native call method.
1590 // These both come in synchronized and non-synchronized versions but the
1591 // frame layout they create is very similar. The other method entry
1592 // types are really just special purpose entries that are really entry
1593 // and interpretation all in one. These are for trivial methods like
1594 // accessor, empty, or special math methods.
1595 //
1596 // When control flow reaches any of the entry types for the interpreter
1597 // the following holds ->
1598 //
1599 // Arguments:
1600 //
1601 // rbx,: Method*
1602 // rcx: receiver
1603 //
1604 //
1605 // Stack layout immediately at entry
1606 //
1607 // [ return address     ] <--- rsp
1608 // [ parameter n        ]
1609 //   ...
1610 // [ parameter 1        ]
1611 // [ expression stack   ] (caller's java expression stack)
1612 
1613 // Assuming that we don't go to one of the trivial specialized
1614 // entries the stack will look like below when we are ready to execute
1615 // the first bytecode (or call the native routine). The register usage
1616 // will be as the template based interpreter expects (see interpreter_x86.hpp).
1617 //
1618 // local variables follow incoming parameters immediately; i.e.
1619 // the return address is moved to the end of the locals).
1620 //
1621 // [ monitor entry      ] <--- rsp
1622 //   ...
1623 // [ monitor entry      ]
1624 // [ expr. stack bottom ]
1625 // [ saved rsi          ]
1626 // [ current rdi        ]
1627 // [ Method*            ]
1628 // [ saved rbp,          ] <--- rbp,
1629 // [ return address     ]
1630 // [ local variable m   ]
1631 //   ...
1632 // [ local variable 1   ]
1633 // [ parameter n        ]
1634 //   ...
1635 // [ parameter 1        ] <--- rdi
1636 
1637 address AbstractInterpreterGenerator::generate_method_entry(AbstractInterpreter::MethodKind kind) {
1638   // determine code generation flags
1639   bool synchronized = false;
1640   address entry_point = NULL;
1641   InterpreterGenerator* ig_this = (InterpreterGenerator*)this;
1642 
1643   switch (kind) {
1644     case Interpreter::zerolocals             :                                                       break;
1645     case Interpreter::zerolocals_synchronized: synchronized = true;                                  break;
1646     case Interpreter::native                 : entry_point = ig_this->generate_native_entry(false);  break;
1647     case Interpreter::native_synchronized    : entry_point = ig_this->generate_native_entry(true);   break;
1648     case Interpreter::empty                  : entry_point = ig_this->generate_empty_entry();        break;
1649     case Interpreter::accessor               : entry_point = ig_this->generate_accessor_entry();     break;
1650     case Interpreter::abstract               : entry_point = ig_this->generate_abstract_entry();     break;
1651 
1652     case Interpreter::java_lang_math_sin     : // fall thru
1653     case Interpreter::java_lang_math_cos     : // fall thru
1654     case Interpreter::java_lang_math_tan     : // fall thru
1655     case Interpreter::java_lang_math_abs     : // fall thru
1656     case Interpreter::java_lang_math_log     : // fall thru
1657     case Interpreter::java_lang_math_log10   : // fall thru
1658     case Interpreter::java_lang_math_sqrt    : // fall thru
1659     case Interpreter::java_lang_math_pow     : // fall thru
1660     case Interpreter::java_lang_math_exp     : entry_point = ig_this->generate_math_entry(kind);      break;
1661     case Interpreter::java_lang_ref_reference_get
1662                                              : entry_point = ig_this->generate_Reference_get_entry(); break;
1663     case Interpreter::java_util_zip_CRC32_update
1664                                              : entry_point = ig_this->generate_CRC32_update_entry();  break;
1665     case Interpreter::java_util_zip_CRC32_updateBytes
1666                                              : // fall thru
1667     case Interpreter::java_util_zip_CRC32_updateByteBuffer
1668                                              : entry_point = ig_this->generate_CRC32_updateBytes_entry(kind); break;
1669     default:
1670       fatal(err_msg("unexpected method kind: %d", kind));
1671       break;
1672   }
1673 
1674   if (entry_point) return entry_point;
1675 
1676   return ig_this->generate_normal_entry(synchronized);
1677 
1678 }
1679 
1680 // These should never be compiled since the interpreter will prefer
1681 // the compiled version to the intrinsic version.
1682 bool AbstractInterpreter::can_be_compiled(methodHandle m) {
1683   switch (method_kind(m)) {
1684     case Interpreter::java_lang_math_sin     : // fall thru
1685     case Interpreter::java_lang_math_cos     : // fall thru
1686     case Interpreter::java_lang_math_tan     : // fall thru
1687     case Interpreter::java_lang_math_abs     : // fall thru
1688     case Interpreter::java_lang_math_log     : // fall thru
1689     case Interpreter::java_lang_math_log10   : // fall thru
1690     case Interpreter::java_lang_math_sqrt    : // fall thru
1691     case Interpreter::java_lang_math_pow     : // fall thru
1692     case Interpreter::java_lang_math_exp     :
1693       return false;
1694     default:
1695       return true;
1696   }
1697 }
1698 
1699 // How much stack a method activation needs in words.
1700 int AbstractInterpreter::size_top_interpreter_activation(Method* method) {
1701 
1702   const int stub_code = 4;  // see generate_call_stub
1703   // Save space for one monitor to get into the interpreted method in case
1704   // the method is synchronized
1705   int monitor_size    = method->is_synchronized() ?
1706                                 1*frame::interpreter_frame_monitor_size() : 0;
1707 
1708   // total overhead size: entry_size + (saved rbp, thru expr stack bottom).
1709   // be sure to change this if you add/subtract anything to/from the overhead area
1710   const int overhead_size = -frame::interpreter_frame_initial_sp_offset;
1711 
1712   const int method_stack = (method->max_locals() + method->max_stack()) *
1713                            Interpreter::stackElementWords;
1714   return overhead_size + method_stack + stub_code;
1715 }
1716 
1717 //------------------------------------------------------------------------------------------------------------------------
1718 // Exceptions
1719 
1720 void TemplateInterpreterGenerator::generate_throw_exception() {
1721   // Entry point in previous activation (i.e., if the caller was interpreted)
1722   Interpreter::_rethrow_exception_entry = __ pc();
1723   const Register thread = rcx;
1724 
1725   // Restore sp to interpreter_frame_last_sp even though we are going
1726   // to empty the expression stack for the exception processing.
1727   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
1728   // rax,: exception
1729   // rdx: return address/pc that threw exception
1730   __ restore_bcp();                              // rsi points to call/send
1731   __ restore_locals();
1732 
1733   // Entry point for exceptions thrown within interpreter code
1734   Interpreter::_throw_exception_entry = __ pc();
1735   // expression stack is undefined here
1736   // rax,: exception
1737   // rsi: exception bcp
1738   __ verify_oop(rax);
1739 
1740   // expression stack must be empty before entering the VM in case of an exception
1741   __ empty_expression_stack();
1742   __ empty_FPU_stack();
1743   // find exception handler address and preserve exception oop
1744   __ call_VM(rdx, CAST_FROM_FN_PTR(address, InterpreterRuntime::exception_handler_for_exception), rax);
1745   // rax,: exception handler entry point
1746   // rdx: preserved exception oop
1747   // rsi: bcp for exception handler
1748   __ push_ptr(rdx);                              // push exception which is now the only value on the stack
1749   __ jmp(rax);                                   // jump to exception handler (may be _remove_activation_entry!)
1750 
1751   // If the exception is not handled in the current frame the frame is removed and
1752   // the exception is rethrown (i.e. exception continuation is _rethrow_exception).
1753   //
1754   // Note: At this point the bci is still the bxi for the instruction which caused
1755   //       the exception and the expression stack is empty. Thus, for any VM calls
1756   //       at this point, GC will find a legal oop map (with empty expression stack).
1757 
1758   // In current activation
1759   // tos: exception
1760   // rsi: exception bcp
1761 
1762   //
1763   // JVMTI PopFrame support
1764   //
1765 
1766    Interpreter::_remove_activation_preserving_args_entry = __ pc();
1767   __ empty_expression_stack();
1768   __ empty_FPU_stack();
1769   // Set the popframe_processing bit in pending_popframe_condition indicating that we are
1770   // currently handling popframe, so that call_VMs that may happen later do not trigger new
1771   // popframe handling cycles.
1772   __ get_thread(thread);
1773   __ movl(rdx, Address(thread, JavaThread::popframe_condition_offset()));
1774   __ orl(rdx, JavaThread::popframe_processing_bit);
1775   __ movl(Address(thread, JavaThread::popframe_condition_offset()), rdx);
1776 
1777   {
1778     // Check to see whether we are returning to a deoptimized frame.
1779     // (The PopFrame call ensures that the caller of the popped frame is
1780     // either interpreted or compiled and deoptimizes it if compiled.)
1781     // In this case, we can't call dispatch_next() after the frame is
1782     // popped, but instead must save the incoming arguments and restore
1783     // them after deoptimization has occurred.
1784     //
1785     // Note that we don't compare the return PC against the
1786     // deoptimization blob's unpack entry because of the presence of
1787     // adapter frames in C2.
1788     Label caller_not_deoptimized;
1789     __ movptr(rdx, Address(rbp, frame::return_addr_offset * wordSize));
1790     __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::interpreter_contains), rdx);
1791     __ testl(rax, rax);
1792     __ jcc(Assembler::notZero, caller_not_deoptimized);
1793 
1794     // Compute size of arguments for saving when returning to deoptimized caller
1795     __ get_method(rax);
1796     __ movptr(rax, Address(rax, Method::const_offset()));
1797     __ load_unsigned_short(rax, Address(rax, ConstMethod::size_of_parameters_offset()));
1798     __ shlptr(rax, Interpreter::logStackElementSize);
1799     __ restore_locals();
1800     __ subptr(rdi, rax);
1801     __ addptr(rdi, wordSize);
1802     // Save these arguments
1803     __ get_thread(thread);
1804     __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, Deoptimization::popframe_preserve_args), thread, rax, rdi);
1805 
1806     __ remove_activation(vtos, rdx,
1807                          /* throw_monitor_exception */ false,
1808                          /* install_monitor_exception */ false,
1809                          /* notify_jvmdi */ false);
1810 
1811     // Inform deoptimization that it is responsible for restoring these arguments
1812     __ get_thread(thread);
1813     __ movl(Address(thread, JavaThread::popframe_condition_offset()), JavaThread::popframe_force_deopt_reexecution_bit);
1814 
1815     // Continue in deoptimization handler
1816     __ jmp(rdx);
1817 
1818     __ bind(caller_not_deoptimized);
1819   }
1820 
1821   __ remove_activation(vtos, rdx,
1822                        /* throw_monitor_exception */ false,
1823                        /* install_monitor_exception */ false,
1824                        /* notify_jvmdi */ false);
1825 
1826   // Finish with popframe handling
1827   // A previous I2C followed by a deoptimization might have moved the
1828   // outgoing arguments further up the stack. PopFrame expects the
1829   // mutations to those outgoing arguments to be preserved and other
1830   // constraints basically require this frame to look exactly as
1831   // though it had previously invoked an interpreted activation with
1832   // no space between the top of the expression stack (current
1833   // last_sp) and the top of stack. Rather than force deopt to
1834   // maintain this kind of invariant all the time we call a small
1835   // fixup routine to move the mutated arguments onto the top of our
1836   // expression stack if necessary.
1837   __ mov(rax, rsp);
1838   __ movptr(rbx, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1839   __ get_thread(thread);
1840   // PC must point into interpreter here
1841   __ set_last_Java_frame(thread, noreg, rbp, __ pc());
1842   __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), thread, rax, rbx);
1843   __ get_thread(thread);
1844   __ reset_last_Java_frame(thread, true);
1845   // Restore the last_sp and null it out
1846   __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1847   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
1848 
1849   __ restore_bcp();
1850   __ restore_locals();
1851   // The method data pointer was incremented already during
1852   // call profiling. We have to restore the mdp for the current bcp.
1853   if (ProfileInterpreter) {
1854     __ set_method_data_pointer_for_bcp();
1855   }
1856 
1857   // Clear the popframe condition flag
1858   __ get_thread(thread);
1859   __ movl(Address(thread, JavaThread::popframe_condition_offset()), JavaThread::popframe_inactive);
1860 
1861 #if INCLUDE_JVMTI
1862   if (EnableInvokeDynamic) {
1863     Label L_done;
1864     const Register local0 = rdi;
1865 
1866     __ cmpb(Address(rsi, 0), Bytecodes::_invokestatic);
1867     __ jcc(Assembler::notEqual, L_done);
1868 
1869     // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call.
1870     // Detect such a case in the InterpreterRuntime function and return the member name argument, or NULL.
1871 
1872     __ get_method(rdx);
1873     __ movptr(rax, Address(local0, 0));
1874     __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), rax, rdx, rsi);
1875 
1876     __ testptr(rax, rax);
1877     __ jcc(Assembler::zero, L_done);
1878 
1879     __ movptr(Address(rbx, 0), rax);
1880     __ bind(L_done);
1881   }
1882 #endif // INCLUDE_JVMTI
1883 
1884   __ dispatch_next(vtos);
1885   // end of PopFrame support
1886 
1887   Interpreter::_remove_activation_entry = __ pc();
1888 
1889   // preserve exception over this code sequence
1890   __ pop_ptr(rax);
1891   __ get_thread(thread);
1892   __ movptr(Address(thread, JavaThread::vm_result_offset()), rax);
1893   // remove the activation (without doing throws on illegalMonitorExceptions)
1894   __ remove_activation(vtos, rdx, false, true, false);
1895   // restore exception
1896   __ get_thread(thread);
1897   __ get_vm_result(rax, thread);
1898 
1899   // Inbetween activations - previous activation type unknown yet
1900   // compute continuation point - the continuation point expects
1901   // the following registers set up:
1902   //
1903   // rax: exception
1904   // rdx: return address/pc that threw exception
1905   // rsp: expression stack of caller
1906   // rbp: rbp, of caller
1907   __ push(rax);                                  // save exception
1908   __ push(rdx);                                  // save return address
1909   __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), thread, rdx);
1910   __ mov(rbx, rax);                              // save exception handler
1911   __ pop(rdx);                                   // restore return address
1912   __ pop(rax);                                   // restore exception
1913   // Note that an "issuing PC" is actually the next PC after the call
1914   __ jmp(rbx);                                   // jump to exception handler of caller
1915 }
1916 
1917 
1918 //
1919 // JVMTI ForceEarlyReturn support
1920 //
1921 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
1922   address entry = __ pc();
1923   const Register thread = rcx;
1924 
1925   __ restore_bcp();
1926   __ restore_locals();
1927   __ empty_expression_stack();
1928   __ empty_FPU_stack();
1929   __ load_earlyret_value(state);
1930 
1931   __ get_thread(thread);
1932   __ movptr(rcx, Address(thread, JavaThread::jvmti_thread_state_offset()));
1933   const Address cond_addr(rcx, JvmtiThreadState::earlyret_state_offset());
1934 
1935   // Clear the earlyret state
1936   __ movl(cond_addr, JvmtiThreadState::earlyret_inactive);
1937 
1938   __ remove_activation(state, rsi,
1939                        false, /* throw_monitor_exception */
1940                        false, /* install_monitor_exception */
1941                        true); /* notify_jvmdi */
1942   __ jmp(rsi);
1943   return entry;
1944 } // end of ForceEarlyReturn support
1945 
1946 
1947 //------------------------------------------------------------------------------------------------------------------------
1948 // Helper for vtos entry point generation
1949 
1950 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) {
1951   assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
1952   Label L;
1953   fep = __ pc(); __ push(ftos); __ jmp(L);
1954   dep = __ pc(); __ push(dtos); __ jmp(L);
1955   lep = __ pc(); __ push(ltos); __ jmp(L);
1956   aep = __ pc(); __ push(atos); __ jmp(L);
1957   bep = cep = sep =             // fall through
1958   iep = __ pc(); __ push(itos); // fall through
1959   vep = __ pc(); __ bind(L);    // fall through
1960   generate_and_dispatch(t);
1961 }
1962 
1963 //------------------------------------------------------------------------------------------------------------------------
1964 // Generation of individual instructions
1965 
1966 // helpers for generate_and_dispatch
1967 
1968 
1969 
1970 InterpreterGenerator::InterpreterGenerator(StubQueue* code)
1971  : TemplateInterpreterGenerator(code) {
1972    generate_all(); // down here so it can be "virtual"
1973 }
1974 
1975 //------------------------------------------------------------------------------------------------------------------------
1976 
1977 // Non-product code
1978 #ifndef PRODUCT
1979 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
1980   address entry = __ pc();
1981 
1982   // prepare expression stack
1983   __ pop(rcx);          // pop return address so expression stack is 'pure'
1984   __ push(state);       // save tosca
1985 
1986   // pass tosca registers as arguments & call tracer
1987   __ call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::trace_bytecode), rcx, rax, rdx);
1988   __ mov(rcx, rax);     // make sure return address is not destroyed by pop(state)
1989   __ pop(state);        // restore tosca
1990 
1991   // return
1992   __ jmp(rcx);
1993 
1994   return entry;
1995 }
1996 
1997 
1998 void TemplateInterpreterGenerator::count_bytecode() {
1999   __ incrementl(ExternalAddress((address) &BytecodeCounter::_counter_value));
2000 }
2001 
2002 
2003 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
2004   __ incrementl(ExternalAddress((address) &BytecodeHistogram::_counters[t->bytecode()]));
2005 }
2006 
2007 
2008 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
2009   __ mov32(ExternalAddress((address) &BytecodePairHistogram::_index), rbx);
2010   __ shrl(rbx, BytecodePairHistogram::log2_number_of_codes);
2011   __ orl(rbx, ((int)t->bytecode()) << BytecodePairHistogram::log2_number_of_codes);
2012   ExternalAddress table((address) BytecodePairHistogram::_counters);
2013   Address index(noreg, rbx, Address::times_4);
2014   __ incrementl(ArrayAddress(table, index));
2015 }
2016 
2017 
2018 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
2019   // Call a little run-time stub to avoid blow-up for each bytecode.
2020   // The run-time runtime saves the right registers, depending on
2021   // the tosca in-state for the given template.
2022   assert(Interpreter::trace_code(t->tos_in()) != NULL,
2023          "entry must have been generated");
2024   __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in())));
2025 }
2026 
2027 
2028 void TemplateInterpreterGenerator::stop_interpreter_at() {
2029   Label L;
2030   __ cmp32(ExternalAddress((address) &BytecodeCounter::_counter_value),
2031            StopInterpreterAt);
2032   __ jcc(Assembler::notEqual, L);
2033   __ int3();
2034   __ bind(L);
2035 }
2036 #endif // !PRODUCT
2037 #endif // CC_INTERP