1 /*
   2  * Copyright 1997-2008 Sun Microsystems, Inc.  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 Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  20  * CA 95054 USA or visit www.sun.com if you need additional information or
  21  * have any questions.
  22  *
  23  */
  24 
  25 #include "incls/_precompiled.incl"
  26 #include "incls/_templateInterpreter_x86_32.cpp.incl"
  27 
  28 #define __ _masm->
  29 
  30 
  31 #ifndef CC_INTERP
  32 const int method_offset = frame::interpreter_frame_method_offset * wordSize;
  33 const int bci_offset    = frame::interpreter_frame_bcx_offset    * wordSize;
  34 const int locals_offset = frame::interpreter_frame_locals_offset * wordSize;
  35 
  36 //------------------------------------------------------------------------------------------------------------------------
  37 
  38 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() {
  39   address entry = __ pc();
  40 
  41   // Note: There should be a minimal interpreter frame set up when stack
  42   // overflow occurs since we check explicitly for it now.
  43   //
  44 #ifdef ASSERT
  45   { Label L;
  46     __ lea(rax, Address(rbp,
  47                 frame::interpreter_frame_monitor_block_top_offset * wordSize));
  48     __ cmpptr(rax, rsp);  // rax, = maximal rsp for current rbp,
  49                         //  (stack grows negative)
  50     __ jcc(Assembler::aboveEqual, L); // check if frame is complete
  51     __ stop ("interpreter frame not set up");
  52     __ bind(L);
  53   }
  54 #endif // ASSERT
  55   // Restore bcp under the assumption that the current frame is still
  56   // interpreted
  57   __ restore_bcp();
  58 
  59   // expression stack must be empty before entering the VM if an exception
  60   // happened
  61   __ empty_expression_stack();
  62   __ empty_FPU_stack();
  63   // throw exception
  64   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_StackOverflowError));
  65   return entry;
  66 }
  67 
  68 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler(const char* name) {
  69   address entry = __ pc();
  70   // expression stack must be empty before entering the VM if an exception happened
  71   __ empty_expression_stack();
  72   __ empty_FPU_stack();
  73   // setup parameters
  74   // ??? convention: expect aberrant index in register rbx,
  75   __ lea(rax, ExternalAddress((address)name));
  76   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ArrayIndexOutOfBoundsException), rax, rbx);
  77   return entry;
  78 }
  79 
  80 address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
  81   address entry = __ pc();
  82   // object is at TOS
  83   __ pop(rax);
  84   // expression stack must be empty before entering the VM if an exception
  85   // happened
  86   __ empty_expression_stack();
  87   __ empty_FPU_stack();
  88   __ call_VM(noreg,
  89              CAST_FROM_FN_PTR(address,
  90                               InterpreterRuntime::throw_ClassCastException),
  91              rax);
  92   return entry;
  93 }
  94 
  95 #ifdef ASSERT
  96 address last_WrongMethodType_caller;
  97 #endif //ASSERT
  98 
  99 // Arguments are: required type at TOS+8, failing object (or NULL) at TOS+4.
 100 // pc at TOS (just for debugging)
 101 address TemplateInterpreterGenerator::generate_WrongMethodType_handler() {
 102   address entry = __ pc();
 103 
 104   __ pop(rcx);                  // address raising the error (for debug)
 105 #ifdef ASSERT
 106   __ lea(rax, ExternalAddress((address) &last_WrongMethodType_caller));
 107   __ movptr(Address(rax, 0), rcx);
 108 #endif //ASSERT
 109 
 110   __ pop(rbx);                  // actual failing object is at TOS
 111   __ pop(rax);                  // required type is at TOS+4
 112 
 113   __ verify_oop(rbx);
 114   __ verify_oop(rax);
 115 
 116   // Various method handle types use interpreter registers as temps.
 117   __ restore_bcp();
 118   __ restore_locals();
 119 
 120   // Expression stack must be empty before entering the VM for an exception.
 121   __ empty_expression_stack();
 122   __ empty_FPU_stack();
 123   __ call_VM(noreg,
 124              CAST_FROM_FN_PTR(address,
 125                               InterpreterRuntime::throw_WrongMethodTypeException),
 126              // pass required type, failing object (or NULL)
 127              rax, rbx);
 128   return entry;
 129 }
 130 
 131 
 132 address TemplateInterpreterGenerator::generate_exception_handler_common(const char* name, const char* message, bool pass_oop) {
 133   assert(!pass_oop || message == NULL, "either oop or message but not both");
 134   address entry = __ pc();
 135   if (pass_oop) {
 136     // object is at TOS
 137     __ pop(rbx);
 138   }
 139   // expression stack must be empty before entering the VM if an exception happened
 140   __ empty_expression_stack();
 141   __ empty_FPU_stack();
 142   // setup parameters
 143   __ lea(rax, ExternalAddress((address)name));
 144   if (pass_oop) {
 145     __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_klass_exception), rax, rbx);
 146   } else {
 147     if (message != NULL) {
 148       __ lea(rbx, ExternalAddress((address)message));
 149     } else {
 150       __ movptr(rbx, (int32_t)NULL_WORD);
 151     }
 152     __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception), rax, rbx);
 153   }
 154   // throw exception
 155   __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
 156   return entry;
 157 }
 158 
 159 
 160 address TemplateInterpreterGenerator::generate_continuation_for(TosState state) {
 161   address entry = __ pc();
 162   // NULL last_sp until next java call
 163   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
 164   __ dispatch_next(state);
 165   return entry;
 166 }
 167 
 168 
 169 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step) {
 170   Label interpreter_entry;
 171   address compiled_entry = __ pc();
 172 
 173 #ifdef COMPILER2
 174   // The FPU stack is clean if UseSSE >= 2 but must be cleaned in other cases
 175   if ((state == ftos && UseSSE < 1) || (state == dtos && UseSSE < 2)) {
 176     for (int i = 1; i < 8; i++) {
 177         __ ffree(i);
 178     }
 179   } else if (UseSSE < 2) {
 180     __ empty_FPU_stack();
 181   }
 182 #endif
 183   if ((state == ftos && UseSSE < 1) || (state == dtos && UseSSE < 2)) {
 184     __ MacroAssembler::verify_FPU(1, "generate_return_entry_for compiled");
 185   } else {
 186     __ MacroAssembler::verify_FPU(0, "generate_return_entry_for compiled");
 187   }
 188 
 189   __ jmp(interpreter_entry, relocInfo::none);
 190   // emit a sentinel we can test for when converting an interpreter
 191   // entry point to a compiled entry point.
 192   __ a_long(Interpreter::return_sentinel);
 193   __ a_long((int)compiled_entry);
 194   address entry = __ pc();
 195   __ bind(interpreter_entry);
 196 
 197   // In SSE mode, interpreter returns FP results in xmm0 but they need
 198   // to end up back on the FPU so it can operate on them.
 199   if (state == ftos && UseSSE >= 1) {
 200     __ subptr(rsp, wordSize);
 201     __ movflt(Address(rsp, 0), xmm0);
 202     __ fld_s(Address(rsp, 0));
 203     __ addptr(rsp, wordSize);
 204   } else if (state == dtos && UseSSE >= 2) {
 205     __ subptr(rsp, 2*wordSize);
 206     __ movdbl(Address(rsp, 0), xmm0);
 207     __ fld_d(Address(rsp, 0));
 208     __ addptr(rsp, 2*wordSize);
 209   }
 210 
 211   __ MacroAssembler::verify_FPU(state == ftos || state == dtos ? 1 : 0, "generate_return_entry_for in interpreter");
 212 
 213   // Restore stack bottom in case i2c adjusted stack
 214   __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
 215   // and NULL it as marker that rsp is now tos until next java call
 216   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
 217 
 218   __ restore_bcp();
 219   __ restore_locals();
 220   __ get_cache_and_index_at_bcp(rbx, rcx, 1);
 221   __ movl(rbx, Address(rbx, rcx,
 222                     Address::times_ptr, constantPoolCacheOopDesc::base_offset() +
 223                     ConstantPoolCacheEntry::flags_offset()));
 224   __ andptr(rbx, 0xFF);
 225   __ lea(rsp, Address(rsp, rbx, Interpreter::stackElementScale()));
 226   __ dispatch_next(state, step);
 227   return entry;
 228 }
 229 
 230 
 231 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state, int step) {
 232   address entry = __ pc();
 233 
 234   // In SSE mode, FP results are in xmm0
 235   if (state == ftos && UseSSE > 0) {
 236     __ subptr(rsp, wordSize);
 237     __ movflt(Address(rsp, 0), xmm0);
 238     __ fld_s(Address(rsp, 0));
 239     __ addptr(rsp, wordSize);
 240   } else if (state == dtos && UseSSE >= 2) {
 241     __ subptr(rsp, 2*wordSize);
 242     __ movdbl(Address(rsp, 0), xmm0);
 243     __ fld_d(Address(rsp, 0));
 244     __ addptr(rsp, 2*wordSize);
 245   }
 246 
 247   __ MacroAssembler::verify_FPU(state == ftos || state == dtos ? 1 : 0, "generate_deopt_entry_for in interpreter");
 248 
 249   // The stack is not extended by deopt but we must NULL last_sp as this
 250   // entry is like a "return".
 251   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
 252   __ restore_bcp();
 253   __ restore_locals();
 254   // handle exceptions
 255   { Label L;
 256     const Register thread = rcx;
 257     __ get_thread(thread);
 258     __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
 259     __ jcc(Assembler::zero, L);
 260     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_pending_exception));
 261     __ should_not_reach_here();
 262     __ bind(L);
 263   }
 264   __ dispatch_next(state, step);
 265   return entry;
 266 }
 267 
 268 
 269 int AbstractInterpreter::BasicType_as_index(BasicType type) {
 270   int i = 0;
 271   switch (type) {
 272     case T_BOOLEAN: i = 0; break;
 273     case T_CHAR   : i = 1; break;
 274     case T_BYTE   : i = 2; break;
 275     case T_SHORT  : i = 3; break;
 276     case T_INT    : // fall through
 277     case T_LONG   : // fall through
 278     case T_VOID   : i = 4; break;
 279     case T_FLOAT  : i = 5; break;  // have to treat float and double separately for SSE
 280     case T_DOUBLE : i = 6; break;
 281     case T_OBJECT : // fall through
 282     case T_ARRAY  : i = 7; break;
 283     default       : ShouldNotReachHere();
 284   }
 285   assert(0 <= i && i < AbstractInterpreter::number_of_result_handlers, "index out of bounds");
 286   return i;
 287 }
 288 
 289 
 290 address TemplateInterpreterGenerator::generate_result_handler_for(BasicType type) {
 291   address entry = __ pc();
 292   switch (type) {
 293     case T_BOOLEAN: __ c2bool(rax);            break;
 294     case T_CHAR   : __ andptr(rax, 0xFFFF);    break;
 295     case T_BYTE   : __ sign_extend_byte (rax); break;
 296     case T_SHORT  : __ sign_extend_short(rax); break;
 297     case T_INT    : /* nothing to do */        break;
 298     case T_DOUBLE :
 299     case T_FLOAT  :
 300       { const Register t = InterpreterRuntime::SignatureHandlerGenerator::temp();
 301         __ pop(t);                            // remove return address first
 302         __ pop_dtos_to_rsp();
 303         // Must return a result for interpreter or compiler. In SSE
 304         // mode, results are returned in xmm0 and the FPU stack must
 305         // be empty.
 306         if (type == T_FLOAT && UseSSE >= 1) {
 307           // Load ST0
 308           __ fld_d(Address(rsp, 0));
 309           // Store as float and empty fpu stack
 310           __ fstp_s(Address(rsp, 0));
 311           // and reload
 312           __ movflt(xmm0, Address(rsp, 0));
 313         } else if (type == T_DOUBLE && UseSSE >= 2 ) {
 314           __ movdbl(xmm0, Address(rsp, 0));
 315         } else {
 316           // restore ST0
 317           __ fld_d(Address(rsp, 0));
 318         }
 319         // and pop the temp
 320         __ addptr(rsp, 2 * wordSize);
 321         __ push(t);                           // restore return address
 322       }
 323       break;
 324     case T_OBJECT :
 325       // retrieve result from frame
 326       __ movptr(rax, Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize));
 327       // and verify it
 328       __ verify_oop(rax);
 329       break;
 330     default       : ShouldNotReachHere();
 331   }
 332   __ ret(0);                                   // return from result handler
 333   return entry;
 334 }
 335 
 336 address TemplateInterpreterGenerator::generate_safept_entry_for(TosState state, address runtime_entry) {
 337   address entry = __ pc();
 338   __ push(state);
 339   __ call_VM(noreg, runtime_entry);
 340   __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
 341   return entry;
 342 }
 343 
 344 
 345 // Helpers for commoning out cases in the various type of method entries.
 346 //
 347 
 348 // increment invocation count & check for overflow
 349 //
 350 // Note: checking for negative value instead of overflow
 351 //       so we have a 'sticky' overflow test
 352 //
 353 // rbx,: method
 354 // rcx: invocation counter
 355 //
 356 void InterpreterGenerator::generate_counter_incr(Label* overflow, Label* profile_method, Label* profile_method_continue) {
 357 
 358   const Address invocation_counter(rbx, methodOopDesc::invocation_counter_offset() + InvocationCounter::counter_offset());
 359   const Address backedge_counter  (rbx, methodOopDesc::backedge_counter_offset() + InvocationCounter::counter_offset());
 360 
 361   if (ProfileInterpreter) { // %%% Merge this into methodDataOop
 362     __ incrementl(Address(rbx,methodOopDesc::interpreter_invocation_counter_offset()));
 363   }
 364   // Update standard invocation counters
 365   __ movl(rax, backedge_counter);               // load backedge counter
 366 
 367   __ incrementl(rcx, InvocationCounter::count_increment);
 368   __ andl(rax, InvocationCounter::count_mask_value);  // mask out the status bits
 369 
 370   __ movl(invocation_counter, rcx);             // save invocation count
 371   __ addl(rcx, rax);                            // add both counters
 372 
 373   // profile_method is non-null only for interpreted method so
 374   // profile_method != NULL == !native_call
 375   // BytecodeInterpreter only calls for native so code is elided.
 376 
 377   if (ProfileInterpreter && profile_method != NULL) {
 378     // Test to see if we should create a method data oop
 379     __ cmp32(rcx,
 380              ExternalAddress((address)&InvocationCounter::InterpreterProfileLimit));
 381     __ jcc(Assembler::less, *profile_method_continue);
 382 
 383     // if no method data exists, go to profile_method
 384     __ test_method_data_pointer(rax, *profile_method);
 385   }
 386 
 387   __ cmp32(rcx,
 388            ExternalAddress((address)&InvocationCounter::InterpreterInvocationLimit));
 389   __ jcc(Assembler::aboveEqual, *overflow);
 390 
 391 }
 392 
 393 void InterpreterGenerator::generate_counter_overflow(Label* do_continue) {
 394 
 395   // Asm interpreter on entry
 396   // rdi - locals
 397   // rsi - bcp
 398   // rbx, - method
 399   // rdx - cpool
 400   // rbp, - interpreter frame
 401 
 402   // C++ interpreter on entry
 403   // rsi - new interpreter state pointer
 404   // rbp - interpreter frame pointer
 405   // rbx - method
 406 
 407   // On return (i.e. jump to entry_point) [ back to invocation of interpreter ]
 408   // rbx, - method
 409   // rcx - rcvr (assuming there is one)
 410   // top of stack return address of interpreter caller
 411   // rsp - sender_sp
 412 
 413   // C++ interpreter only
 414   // rsi - previous interpreter state pointer
 415 
 416   const Address size_of_parameters(rbx, methodOopDesc::size_of_parameters_offset());
 417 
 418   // InterpreterRuntime::frequency_counter_overflow takes one argument
 419   // indicating if the counter overflow occurs at a backwards branch (non-NULL bcp).
 420   // The call returns the address of the verified entry point for the method or NULL
 421   // if the compilation did not complete (either went background or bailed out).
 422   __ movptr(rax, (int32_t)false);
 423   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), rax);
 424 
 425   __ movptr(rbx, Address(rbp, method_offset));   // restore methodOop
 426 
 427   // Preserve invariant that rsi/rdi contain bcp/locals of sender frame
 428   // and jump to the interpreted entry.
 429   __ jmp(*do_continue, relocInfo::none);
 430 
 431 }
 432 
 433 void InterpreterGenerator::generate_stack_overflow_check(void) {
 434   // see if we've got enough room on the stack for locals plus overhead.
 435   // the expression stack grows down incrementally, so the normal guard
 436   // page mechanism will work for that.
 437   //
 438   // Registers live on entry:
 439   //
 440   // Asm interpreter
 441   // rdx: number of additional locals this frame needs (what we must check)
 442   // rbx,: methodOop
 443 
 444   // destroyed on exit
 445   // rax,
 446 
 447   // NOTE:  since the additional locals are also always pushed (wasn't obvious in
 448   // generate_method_entry) so the guard should work for them too.
 449   //
 450 
 451   // monitor entry size: see picture of stack set (generate_method_entry) and frame_x86.hpp
 452   const int entry_size    = frame::interpreter_frame_monitor_size() * wordSize;
 453 
 454   // total overhead size: entry_size + (saved rbp, thru expr stack bottom).
 455   // be sure to change this if you add/subtract anything to/from the overhead area
 456   const int overhead_size = -(frame::interpreter_frame_initial_sp_offset*wordSize) + entry_size;
 457 
 458   const int page_size = os::vm_page_size();
 459 
 460   Label after_frame_check;
 461 
 462   // see if the frame is greater than one page in size. If so,
 463   // then we need to verify there is enough stack space remaining
 464   // for the additional locals.
 465   __ cmpl(rdx, (page_size - overhead_size)/Interpreter::stackElementSize());
 466   __ jcc(Assembler::belowEqual, after_frame_check);
 467 
 468   // compute rsp as if this were going to be the last frame on
 469   // the stack before the red zone
 470 
 471   Label after_frame_check_pop;
 472 
 473   __ push(rsi);
 474 
 475   const Register thread = rsi;
 476 
 477   __ get_thread(thread);
 478 
 479   const Address stack_base(thread, Thread::stack_base_offset());
 480   const Address stack_size(thread, Thread::stack_size_offset());
 481 
 482   // locals + overhead, in bytes
 483   __ lea(rax, Address(noreg, rdx, Interpreter::stackElementScale(), overhead_size));
 484 
 485 #ifdef ASSERT
 486   Label stack_base_okay, stack_size_okay;
 487   // verify that thread stack base is non-zero
 488   __ cmpptr(stack_base, (int32_t)NULL_WORD);
 489   __ jcc(Assembler::notEqual, stack_base_okay);
 490   __ stop("stack base is zero");
 491   __ bind(stack_base_okay);
 492   // verify that thread stack size is non-zero
 493   __ cmpptr(stack_size, 0);
 494   __ jcc(Assembler::notEqual, stack_size_okay);
 495   __ stop("stack size is zero");
 496   __ bind(stack_size_okay);
 497 #endif
 498 
 499   // Add stack base to locals and subtract stack size
 500   __ addptr(rax, stack_base);
 501   __ subptr(rax, stack_size);
 502 
 503   // Use the maximum number of pages we might bang.
 504   const int max_pages = StackShadowPages > (StackRedPages+StackYellowPages) ? StackShadowPages :
 505                                                                               (StackRedPages+StackYellowPages);
 506   __ addptr(rax, max_pages * page_size);
 507 
 508   // check against the current stack bottom
 509   __ cmpptr(rsp, rax);
 510   __ jcc(Assembler::above, after_frame_check_pop);
 511 
 512   __ pop(rsi);  // get saved bcp / (c++ prev state ).
 513 
 514   __ pop(rax);  // get return address
 515   __ jump(ExternalAddress(Interpreter::throw_StackOverflowError_entry()));
 516 
 517   // all done with frame size check
 518   __ bind(after_frame_check_pop);
 519   __ pop(rsi);
 520 
 521   __ bind(after_frame_check);
 522 }
 523 
 524 // Allocate monitor and lock method (asm interpreter)
 525 // rbx, - methodOop
 526 //
 527 void InterpreterGenerator::lock_method(void) {
 528   // synchronize method
 529   const Address access_flags      (rbx, methodOopDesc::access_flags_offset());
 530   const Address monitor_block_top (rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
 531   const int entry_size            = frame::interpreter_frame_monitor_size() * wordSize;
 532 
 533   #ifdef ASSERT
 534     { Label L;
 535       __ movl(rax, access_flags);
 536       __ testl(rax, JVM_ACC_SYNCHRONIZED);
 537       __ jcc(Assembler::notZero, L);
 538       __ stop("method doesn't need synchronization");
 539       __ bind(L);
 540     }
 541   #endif // ASSERT
 542   // get synchronization object
 543   { Label done;
 544     const int mirror_offset = klassOopDesc::klass_part_offset_in_bytes() + Klass::java_mirror_offset_in_bytes();
 545     __ movl(rax, access_flags);
 546     __ testl(rax, JVM_ACC_STATIC);
 547     __ movptr(rax, Address(rdi, Interpreter::local_offset_in_bytes(0)));  // get receiver (assume this is frequent case)
 548     __ jcc(Assembler::zero, done);
 549     __ movptr(rax, Address(rbx, methodOopDesc::constants_offset()));
 550     __ movptr(rax, Address(rax, constantPoolOopDesc::pool_holder_offset_in_bytes()));
 551     __ movptr(rax, Address(rax, mirror_offset));
 552     __ bind(done);
 553   }
 554   // add space for monitor & lock
 555   __ subptr(rsp, entry_size);                                           // add space for a monitor entry
 556   __ movptr(monitor_block_top, rsp);                                    // set new monitor block top
 557   __ movptr(Address(rsp, BasicObjectLock::obj_offset_in_bytes()), rax); // store object
 558   __ mov(rdx, rsp);                                                    // object address
 559   __ lock_object(rdx);
 560 }
 561 
 562 //
 563 // Generate a fixed interpreter frame. This is identical setup for interpreted methods
 564 // and for native methods hence the shared code.
 565 
 566 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
 567   // initialize fixed part of activation frame
 568   __ push(rax);                                       // save return address
 569   __ enter();                                         // save old & set new rbp,
 570 
 571 
 572   __ push(rsi);                                       // set sender sp
 573   __ push((int32_t)NULL_WORD);                        // leave last_sp as null
 574   __ movptr(rsi, Address(rbx,methodOopDesc::const_offset())); // get constMethodOop
 575   __ lea(rsi, Address(rsi,constMethodOopDesc::codes_offset())); // get codebase
 576   __ push(rbx);                                      // save methodOop
 577   if (ProfileInterpreter) {
 578     Label method_data_continue;
 579     __ movptr(rdx, Address(rbx, in_bytes(methodOopDesc::method_data_offset())));
 580     __ testptr(rdx, rdx);
 581     __ jcc(Assembler::zero, method_data_continue);
 582     __ addptr(rdx, in_bytes(methodDataOopDesc::data_offset()));
 583     __ bind(method_data_continue);
 584     __ push(rdx);                                       // set the mdp (method data pointer)
 585   } else {
 586     __ push(0);
 587   }
 588 
 589   __ movptr(rdx, Address(rbx, methodOopDesc::constants_offset()));
 590   __ movptr(rdx, Address(rdx, constantPoolOopDesc::cache_offset_in_bytes()));
 591   __ push(rdx);                                       // set constant pool cache
 592   __ push(rdi);                                       // set locals pointer
 593   if (native_call) {
 594     __ push(0);                                       // no bcp
 595   } else {
 596     __ push(rsi);                                     // set bcp
 597     }
 598   __ push(0);                                         // reserve word for pointer to expression stack bottom
 599   __ movptr(Address(rsp, 0), rsp);                    // set expression stack bottom
 600 }
 601 
 602 // End of helpers
 603 
 604 //
 605 // Various method entries
 606 //------------------------------------------------------------------------------------------------------------------------
 607 //
 608 //
 609 
 610 // Call an accessor method (assuming it is resolved, otherwise drop into vanilla (slow path) entry
 611 
 612 address InterpreterGenerator::generate_accessor_entry(void) {
 613 
 614   // rbx,: methodOop
 615   // rcx: receiver (preserve for slow entry into asm interpreter)
 616 
 617   // rsi: senderSP must preserved for slow path, set SP to it on fast path
 618 
 619   address entry_point = __ pc();
 620   Label xreturn_path;
 621 
 622   // do fastpath for resolved accessor methods
 623   if (UseFastAccessorMethods) {
 624     Label slow_path;
 625     // If we need a safepoint check, generate full interpreter entry.
 626     ExternalAddress state(SafepointSynchronize::address_of_state());
 627     __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
 628              SafepointSynchronize::_not_synchronized);
 629 
 630     __ jcc(Assembler::notEqual, slow_path);
 631     // ASM/C++ Interpreter
 632     // Code: _aload_0, _(i|a)getfield, _(i|a)return or any rewrites thereof; parameter size = 1
 633     // Note: We can only use this code if the getfield has been resolved
 634     //       and if we don't have a null-pointer exception => check for
 635     //       these conditions first and use slow path if necessary.
 636     // rbx,: method
 637     // rcx: receiver
 638     __ movptr(rax, Address(rsp, wordSize));
 639 
 640     // check if local 0 != NULL and read field
 641     __ testptr(rax, rax);
 642     __ jcc(Assembler::zero, slow_path);
 643 
 644     __ movptr(rdi, Address(rbx, methodOopDesc::constants_offset()));
 645     // read first instruction word and extract bytecode @ 1 and index @ 2
 646     __ movptr(rdx, Address(rbx, methodOopDesc::const_offset()));
 647     __ movl(rdx, Address(rdx, constMethodOopDesc::codes_offset()));
 648     // Shift codes right to get the index on the right.
 649     // The bytecode fetched looks like <index><0xb4><0x2a>
 650     __ shrl(rdx, 2*BitsPerByte);
 651     __ shll(rdx, exact_log2(in_words(ConstantPoolCacheEntry::size())));
 652     __ movptr(rdi, Address(rdi, constantPoolOopDesc::cache_offset_in_bytes()));
 653 
 654     // rax,: local 0
 655     // rbx,: method
 656     // rcx: receiver - do not destroy since it is needed for slow path!
 657     // rcx: scratch
 658     // rdx: constant pool cache index
 659     // rdi: constant pool cache
 660     // rsi: sender sp
 661 
 662     // check if getfield has been resolved and read constant pool cache entry
 663     // check the validity of the cache entry by testing whether _indices field
 664     // contains Bytecode::_getfield in b1 byte.
 665     assert(in_words(ConstantPoolCacheEntry::size()) == 4, "adjust shift below");
 666     __ movl(rcx,
 667             Address(rdi,
 668                     rdx,
 669                     Address::times_ptr, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::indices_offset()));
 670     __ shrl(rcx, 2*BitsPerByte);
 671     __ andl(rcx, 0xFF);
 672     __ cmpl(rcx, Bytecodes::_getfield);
 673     __ jcc(Assembler::notEqual, slow_path);
 674 
 675     // Note: constant pool entry is not valid before bytecode is resolved
 676     __ movptr(rcx,
 677               Address(rdi,
 678                       rdx,
 679                       Address::times_ptr, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::f2_offset()));
 680     __ movl(rdx,
 681             Address(rdi,
 682                     rdx,
 683                     Address::times_ptr, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::flags_offset()));
 684 
 685     Label notByte, notShort, notChar;
 686     const Address field_address (rax, rcx, Address::times_1);
 687 
 688     // Need to differentiate between igetfield, agetfield, bgetfield etc.
 689     // because they are different sizes.
 690     // Use the type from the constant pool cache
 691     __ shrl(rdx, ConstantPoolCacheEntry::tosBits);
 692     // Make sure we don't need to mask rdx for tosBits after the above shift
 693     ConstantPoolCacheEntry::verify_tosBits();
 694     __ cmpl(rdx, btos);
 695     __ jcc(Assembler::notEqual, notByte);
 696     __ load_signed_byte(rax, field_address);
 697     __ jmp(xreturn_path);
 698 
 699     __ bind(notByte);
 700     __ cmpl(rdx, stos);
 701     __ jcc(Assembler::notEqual, notShort);
 702     __ load_signed_word(rax, field_address);
 703     __ jmp(xreturn_path);
 704 
 705     __ bind(notShort);
 706     __ cmpl(rdx, ctos);
 707     __ jcc(Assembler::notEqual, notChar);
 708     __ load_unsigned_word(rax, field_address);
 709     __ jmp(xreturn_path);
 710 
 711     __ bind(notChar);
 712 #ifdef ASSERT
 713     Label okay;
 714     __ cmpl(rdx, atos);
 715     __ jcc(Assembler::equal, okay);
 716     __ cmpl(rdx, itos);
 717     __ jcc(Assembler::equal, okay);
 718     __ stop("what type is this?");
 719     __ bind(okay);
 720 #endif // ASSERT
 721     // All the rest are a 32 bit wordsize
 722     // This is ok for now. Since fast accessors should be going away
 723     __ movptr(rax, field_address);
 724 
 725     __ bind(xreturn_path);
 726 
 727     // _ireturn/_areturn
 728     __ pop(rdi);                               // get return address
 729     __ mov(rsp, rsi);                          // set sp to sender sp
 730     __ jmp(rdi);
 731 
 732     // generate a vanilla interpreter entry as the slow path
 733     __ bind(slow_path);
 734 
 735     (void) generate_normal_entry(false);
 736     return entry_point;
 737   }
 738   return NULL;
 739 
 740 }
 741 
 742 //
 743 // Interpreter stub for calling a native method. (asm interpreter)
 744 // This sets up a somewhat different looking stack for calling the native method
 745 // than the typical interpreter frame setup.
 746 //
 747 
 748 address InterpreterGenerator::generate_native_entry(bool synchronized) {
 749   // determine code generation flags
 750   bool inc_counter  = UseCompiler || CountCompiledCalls;
 751 
 752   // rbx,: methodOop
 753   // rsi: sender sp
 754   // rsi: previous interpreter state (C++ interpreter) must preserve
 755   address entry_point = __ pc();
 756 
 757 
 758   const Address size_of_parameters(rbx, methodOopDesc::size_of_parameters_offset());
 759   const Address invocation_counter(rbx, methodOopDesc::invocation_counter_offset() + InvocationCounter::counter_offset());
 760   const Address access_flags      (rbx, methodOopDesc::access_flags_offset());
 761 
 762   // get parameter size (always needed)
 763   __ load_unsigned_word(rcx, size_of_parameters);
 764 
 765   // native calls don't need the stack size check since they have no expression stack
 766   // and the arguments are already on the stack and we only add a handful of words
 767   // to the stack
 768 
 769   // rbx,: methodOop
 770   // rcx: size of parameters
 771   // rsi: sender sp
 772 
 773   __ pop(rax);                                       // get return address
 774   // for natives the size of locals is zero
 775 
 776   // compute beginning of parameters (rdi)
 777   __ lea(rdi, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize));
 778 
 779 
 780   // add 2 zero-initialized slots for native calls
 781   // NULL result handler
 782   __ push((int32_t)NULL_WORD);
 783   // NULL oop temp (mirror or jni oop result)
 784   __ push((int32_t)NULL_WORD);
 785 
 786   if (inc_counter) __ movl(rcx, invocation_counter);  // (pre-)fetch invocation count
 787   // initialize fixed part of activation frame
 788 
 789   generate_fixed_frame(true);
 790 
 791   // make sure method is native & not abstract
 792 #ifdef ASSERT
 793   __ movl(rax, access_flags);
 794   {
 795     Label L;
 796     __ testl(rax, JVM_ACC_NATIVE);
 797     __ jcc(Assembler::notZero, L);
 798     __ stop("tried to execute non-native method as native");
 799     __ bind(L);
 800   }
 801   { Label L;
 802     __ testl(rax, JVM_ACC_ABSTRACT);
 803     __ jcc(Assembler::zero, L);
 804     __ stop("tried to execute abstract method in interpreter");
 805     __ bind(L);
 806   }
 807 #endif
 808 
 809   // Since at this point in the method invocation the exception handler
 810   // would try to exit the monitor of synchronized methods which hasn't
 811   // been entered yet, we set the thread local variable
 812   // _do_not_unlock_if_synchronized to true. The remove_activation will
 813   // check this flag.
 814 
 815   __ get_thread(rax);
 816   const Address do_not_unlock_if_synchronized(rax,
 817         in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
 818   __ movbool(do_not_unlock_if_synchronized, true);
 819 
 820   // increment invocation count & check for overflow
 821   Label invocation_counter_overflow;
 822   if (inc_counter) {
 823     generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
 824   }
 825 
 826   Label continue_after_compile;
 827   __ bind(continue_after_compile);
 828 
 829   bang_stack_shadow_pages(true);
 830 
 831   // reset the _do_not_unlock_if_synchronized flag
 832   __ get_thread(rax);
 833   __ movbool(do_not_unlock_if_synchronized, false);
 834 
 835   // check for synchronized methods
 836   // Must happen AFTER invocation_counter check and stack overflow check,
 837   // so method is not locked if overflows.
 838   //
 839   if (synchronized) {
 840     lock_method();
 841   } else {
 842     // no synchronization necessary
 843 #ifdef ASSERT
 844       { Label L;
 845         __ movl(rax, access_flags);
 846         __ testl(rax, JVM_ACC_SYNCHRONIZED);
 847         __ jcc(Assembler::zero, L);
 848         __ stop("method needs synchronization");
 849         __ bind(L);
 850       }
 851 #endif
 852   }
 853 
 854   // start execution
 855 #ifdef ASSERT
 856   { Label L;
 857     const Address monitor_block_top (rbp,
 858                  frame::interpreter_frame_monitor_block_top_offset * wordSize);
 859     __ movptr(rax, monitor_block_top);
 860     __ cmpptr(rax, rsp);
 861     __ jcc(Assembler::equal, L);
 862     __ stop("broken stack frame setup in interpreter");
 863     __ bind(L);
 864   }
 865 #endif
 866 
 867   // jvmti/dtrace support
 868   __ notify_method_entry();
 869 
 870   // work registers
 871   const Register method = rbx;
 872   const Register thread = rdi;
 873   const Register t      = rcx;
 874 
 875   // allocate space for parameters
 876   __ get_method(method);
 877   __ verify_oop(method);
 878   __ load_unsigned_word(t, Address(method, methodOopDesc::size_of_parameters_offset()));
 879   __ shlptr(t, Interpreter::logStackElementSize());
 880   __ addptr(t, 2*wordSize);     // allocate two more slots for JNIEnv and possible mirror
 881   __ subptr(rsp, t);
 882   __ andptr(rsp, -(StackAlignmentInBytes)); // gcc needs 16 byte aligned stacks to do XMM intrinsics
 883 
 884   // get signature handler
 885   { Label L;
 886     __ movptr(t, Address(method, methodOopDesc::signature_handler_offset()));
 887     __ testptr(t, t);
 888     __ jcc(Assembler::notZero, L);
 889     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), method);
 890     __ get_method(method);
 891     __ movptr(t, Address(method, methodOopDesc::signature_handler_offset()));
 892     __ bind(L);
 893   }
 894 
 895   // call signature handler
 896   assert(InterpreterRuntime::SignatureHandlerGenerator::from() == rdi, "adjust this code");
 897   assert(InterpreterRuntime::SignatureHandlerGenerator::to  () == rsp, "adjust this code");
 898   assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == t  , "adjust this code");
 899   // The generated handlers do not touch RBX (the method oop).
 900   // However, large signatures cannot be cached and are generated
 901   // each time here.  The slow-path generator will blow RBX
 902   // sometime, so we must reload it after the call.
 903   __ call(t);
 904   __ get_method(method);        // slow path call blows RBX on DevStudio 5.0
 905 
 906   // result handler is in rax,
 907   // set result handler
 908   __ movptr(Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize), rax);
 909 
 910   // pass mirror handle if static call
 911   { Label L;
 912     const int mirror_offset = klassOopDesc::klass_part_offset_in_bytes() + Klass::java_mirror_offset_in_bytes();
 913     __ movl(t, Address(method, methodOopDesc::access_flags_offset()));
 914     __ testl(t, JVM_ACC_STATIC);
 915     __ jcc(Assembler::zero, L);
 916     // get mirror
 917     __ movptr(t, Address(method, methodOopDesc:: constants_offset()));
 918     __ movptr(t, Address(t, constantPoolOopDesc::pool_holder_offset_in_bytes()));
 919     __ movptr(t, Address(t, mirror_offset));
 920     // copy mirror into activation frame
 921     __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize), t);
 922     // pass handle to mirror
 923     __ lea(t, Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize));
 924     __ movptr(Address(rsp, wordSize), t);
 925     __ bind(L);
 926   }
 927 
 928   // get native function entry point
 929   { Label L;
 930     __ movptr(rax, Address(method, methodOopDesc::native_function_offset()));
 931     ExternalAddress unsatisfied(SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
 932     __ cmpptr(rax, unsatisfied.addr());
 933     __ jcc(Assembler::notEqual, L);
 934     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), method);
 935     __ get_method(method);
 936     __ verify_oop(method);
 937     __ movptr(rax, Address(method, methodOopDesc::native_function_offset()));
 938     __ bind(L);
 939   }
 940 
 941   // pass JNIEnv
 942   __ get_thread(thread);
 943   __ lea(t, Address(thread, JavaThread::jni_environment_offset()));
 944   __ movptr(Address(rsp, 0), t);
 945 
 946   // set_last_Java_frame_before_call
 947   // It is enough that the pc()
 948   // points into the right code segment. It does not have to be the correct return pc.
 949   __ set_last_Java_frame(thread, noreg, rbp, __ pc());
 950 
 951   // change thread state
 952 #ifdef ASSERT
 953   { Label L;
 954     __ movl(t, Address(thread, JavaThread::thread_state_offset()));
 955     __ cmpl(t, _thread_in_Java);
 956     __ jcc(Assembler::equal, L);
 957     __ stop("Wrong thread state in native stub");
 958     __ bind(L);
 959   }
 960 #endif
 961 
 962   // Change state to native
 963   __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_native);
 964   __ call(rax);
 965 
 966   // result potentially in rdx:rax or ST0
 967 
 968   // Either restore the MXCSR register after returning from the JNI Call
 969   // or verify that it wasn't changed.
 970   if (VM_Version::supports_sse()) {
 971     if (RestoreMXCSROnJNICalls) {
 972       __ ldmxcsr(ExternalAddress(StubRoutines::addr_mxcsr_std()));
 973     }
 974     else if (CheckJNICalls ) {
 975       __ call(RuntimeAddress(StubRoutines::x86::verify_mxcsr_entry()));
 976     }
 977   }
 978 
 979   // Either restore the x87 floating pointer control word after returning
 980   // from the JNI call or verify that it wasn't changed.
 981   if (CheckJNICalls) {
 982     __ call(RuntimeAddress(StubRoutines::x86::verify_fpu_cntrl_wrd_entry()));
 983   }
 984 
 985   // save potential result in ST(0) & rdx:rax
 986   // (if result handler is the T_FLOAT or T_DOUBLE handler, result must be in ST0 -
 987   // the check is necessary to avoid potential Intel FPU overflow problems by saving/restoring 'empty' FPU registers)
 988   // It is safe to do this push because state is _thread_in_native and return address will be found
 989   // via _last_native_pc and not via _last_jave_sp
 990 
 991   // NOTE: the order of theses push(es) is known to frame::interpreter_frame_result.
 992   // If the order changes or anything else is added to the stack the code in
 993   // interpreter_frame_result will have to be changed.
 994 
 995   { Label L;
 996     Label push_double;
 997     ExternalAddress float_handler(AbstractInterpreter::result_handler(T_FLOAT));
 998     ExternalAddress double_handler(AbstractInterpreter::result_handler(T_DOUBLE));
 999     __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize),
1000               float_handler.addr());
1001     __ jcc(Assembler::equal, push_double);
1002     __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize),
1003               double_handler.addr());
1004     __ jcc(Assembler::notEqual, L);
1005     __ bind(push_double);
1006     __ push(dtos);
1007     __ bind(L);
1008   }
1009   __ push(ltos);
1010 
1011   // change thread state
1012   __ get_thread(thread);
1013   __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_native_trans);
1014   if(os::is_MP()) {
1015     if (UseMembar) {
1016       // Force this write out before the read below
1017       __ membar(Assembler::Membar_mask_bits(
1018            Assembler::LoadLoad | Assembler::LoadStore |
1019            Assembler::StoreLoad | Assembler::StoreStore));
1020     } else {
1021       // Write serialization page so VM thread can do a pseudo remote membar.
1022       // We use the current thread pointer to calculate a thread specific
1023       // offset to write to within the page. This minimizes bus traffic
1024       // due to cache line collision.
1025       __ serialize_memory(thread, rcx);
1026     }
1027   }
1028 
1029   if (AlwaysRestoreFPU) {
1030     //  Make sure the control word is correct.
1031     __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
1032   }
1033 
1034   // check for safepoint operation in progress and/or pending suspend requests
1035   { Label Continue;
1036 
1037     __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
1038              SafepointSynchronize::_not_synchronized);
1039 
1040     Label L;
1041     __ jcc(Assembler::notEqual, L);
1042     __ cmpl(Address(thread, JavaThread::suspend_flags_offset()), 0);
1043     __ jcc(Assembler::equal, Continue);
1044     __ bind(L);
1045 
1046     // Don't use call_VM as it will see a possible pending exception and forward it
1047     // and never return here preventing us from clearing _last_native_pc down below.
1048     // Also can't use call_VM_leaf either as it will check to see if rsi & rdi are
1049     // preserved and correspond to the bcp/locals pointers. So we do a runtime call
1050     // by hand.
1051     //
1052     __ push(thread);
1053     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address,
1054                                             JavaThread::check_special_condition_for_native_trans)));
1055     __ increment(rsp, wordSize);
1056     __ get_thread(thread);
1057 
1058     __ bind(Continue);
1059   }
1060 
1061   // change thread state
1062   __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_Java);
1063 
1064   __ reset_last_Java_frame(thread, true, true);
1065 
1066   // reset handle block
1067   __ movptr(t, Address(thread, JavaThread::active_handles_offset()));
1068   __ movptr(Address(t, JNIHandleBlock::top_offset_in_bytes()), (int32_t)NULL_WORD);
1069 
1070   // If result was an oop then unbox and save it in the frame
1071   { Label L;
1072     Label no_oop, store_result;
1073     ExternalAddress handler(AbstractInterpreter::result_handler(T_OBJECT));
1074     __ cmpptr(Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize),
1075               handler.addr());
1076     __ jcc(Assembler::notEqual, no_oop);
1077     __ cmpptr(Address(rsp, 0), (int32_t)NULL_WORD);
1078     __ pop(ltos);
1079     __ testptr(rax, rax);
1080     __ jcc(Assembler::zero, store_result);
1081     // unbox
1082     __ movptr(rax, Address(rax, 0));
1083     __ bind(store_result);
1084     __ movptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset)*wordSize), rax);
1085     // keep stack depth as expected by pushing oop which will eventually be discarded
1086     __ push(ltos);
1087     __ bind(no_oop);
1088   }
1089 
1090   {
1091      Label no_reguard;
1092      __ cmpl(Address(thread, JavaThread::stack_guard_state_offset()), JavaThread::stack_guard_yellow_disabled);
1093      __ jcc(Assembler::notEqual, no_reguard);
1094 
1095      __ pusha();
1096      __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
1097      __ popa();
1098 
1099      __ bind(no_reguard);
1100    }
1101 
1102   // restore rsi to have legal interpreter frame,
1103   // i.e., bci == 0 <=> rsi == code_base()
1104   // Can't call_VM until bcp is within reasonable.
1105   __ get_method(method);      // method is junk from thread_in_native to now.
1106   __ verify_oop(method);
1107   __ movptr(rsi, Address(method,methodOopDesc::const_offset()));   // get constMethodOop
1108   __ lea(rsi, Address(rsi,constMethodOopDesc::codes_offset()));    // get codebase
1109 
1110   // handle exceptions (exception handling will handle unlocking!)
1111   { Label L;
1112     __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
1113     __ jcc(Assembler::zero, L);
1114     // Note: At some point we may want to unify this with the code used in call_VM_base();
1115     //       i.e., we should use the StubRoutines::forward_exception code. For now this
1116     //       doesn't work here because the rsp is not correctly set at this point.
1117     __ MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_pending_exception));
1118     __ should_not_reach_here();
1119     __ bind(L);
1120   }
1121 
1122   // do unlocking if necessary
1123   { Label L;
1124     __ movl(t, Address(method, methodOopDesc::access_flags_offset()));
1125     __ testl(t, JVM_ACC_SYNCHRONIZED);
1126     __ jcc(Assembler::zero, L);
1127     // the code below should be shared with interpreter macro assembler implementation
1128     { Label unlock;
1129       // BasicObjectLock will be first in list, since this is a synchronized method. However, need
1130       // to check that the object has not been unlocked by an explicit monitorexit bytecode.
1131       const Address monitor(rbp, frame::interpreter_frame_initial_sp_offset * wordSize - (int)sizeof(BasicObjectLock));
1132 
1133       __ lea(rdx, monitor);                   // address of first monitor
1134 
1135       __ movptr(t, Address(rdx, BasicObjectLock::obj_offset_in_bytes()));
1136       __ testptr(t, t);
1137       __ jcc(Assembler::notZero, unlock);
1138 
1139       // Entry already unlocked, need to throw exception
1140       __ MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
1141       __ should_not_reach_here();
1142 
1143       __ bind(unlock);
1144       __ unlock_object(rdx);
1145     }
1146     __ bind(L);
1147   }
1148 
1149   // jvmti/dtrace support
1150   // Note: This must happen _after_ handling/throwing any exceptions since
1151   //       the exception handler code notifies the runtime of method exits
1152   //       too. If this happens before, method entry/exit notifications are
1153   //       not properly paired (was bug - gri 11/22/99).
1154   __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI);
1155 
1156   // restore potential result in rdx:rax, call result handler to restore potential result in ST0 & handle result
1157   __ pop(ltos);
1158   __ movptr(t, Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize));
1159   __ call(t);
1160 
1161   // remove activation
1162   __ movptr(t, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); // get sender sp
1163   __ leave();                                // remove frame anchor
1164   __ pop(rdi);                               // get return address
1165   __ mov(rsp, t);                            // set sp to sender sp
1166   __ jmp(rdi);
1167 
1168   if (inc_counter) {
1169     // Handle overflow of counter and compile method
1170     __ bind(invocation_counter_overflow);
1171     generate_counter_overflow(&continue_after_compile);
1172   }
1173 
1174   return entry_point;
1175 }
1176 
1177 //
1178 // Generic interpreted method entry to (asm) interpreter
1179 //
1180 address InterpreterGenerator::generate_normal_entry(bool synchronized) {
1181   // determine code generation flags
1182   bool inc_counter  = UseCompiler || CountCompiledCalls;
1183 
1184   // rbx,: methodOop
1185   // rsi: sender sp
1186   address entry_point = __ pc();
1187 
1188 
1189   const Address size_of_parameters(rbx, methodOopDesc::size_of_parameters_offset());
1190   const Address size_of_locals    (rbx, methodOopDesc::size_of_locals_offset());
1191   const Address invocation_counter(rbx, methodOopDesc::invocation_counter_offset() + InvocationCounter::counter_offset());
1192   const Address access_flags      (rbx, methodOopDesc::access_flags_offset());
1193 
1194   // get parameter size (always needed)
1195   __ load_unsigned_word(rcx, size_of_parameters);
1196 
1197   // rbx,: methodOop
1198   // rcx: size of parameters
1199 
1200   // rsi: sender_sp (could differ from sp+wordSize if we were called via c2i )
1201 
1202   __ load_unsigned_word(rdx, size_of_locals);       // get size of locals in words
1203   __ subl(rdx, rcx);                                // rdx = no. of additional locals
1204 
1205   // see if we've got enough room on the stack for locals plus overhead.
1206   generate_stack_overflow_check();
1207 
1208   // get return address
1209   __ pop(rax);
1210 
1211   // compute beginning of parameters (rdi)
1212   __ lea(rdi, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize));
1213 
1214   // rdx - # of additional locals
1215   // allocate space for locals
1216   // explicitly initialize locals
1217   {
1218     Label exit, loop;
1219     __ testl(rdx, rdx);
1220     __ jcc(Assembler::lessEqual, exit);               // do nothing if rdx <= 0
1221     __ bind(loop);
1222     if (TaggedStackInterpreter) {
1223       __ push((int32_t)NULL_WORD);                    // push tag
1224     }
1225     __ push((int32_t)NULL_WORD);                      // initialize local variables
1226     __ decrement(rdx);                                // until everything initialized
1227     __ jcc(Assembler::greater, loop);
1228     __ bind(exit);
1229   }
1230 
1231   if (inc_counter) __ movl(rcx, invocation_counter);  // (pre-)fetch invocation count
1232   // initialize fixed part of activation frame
1233   generate_fixed_frame(false);
1234 
1235   // make sure method is not native & not abstract
1236 #ifdef ASSERT
1237   __ movl(rax, access_flags);
1238   {
1239     Label L;
1240     __ testl(rax, JVM_ACC_NATIVE);
1241     __ jcc(Assembler::zero, L);
1242     __ stop("tried to execute native method as non-native");
1243     __ bind(L);
1244   }
1245   { Label L;
1246     __ testl(rax, JVM_ACC_ABSTRACT);
1247     __ jcc(Assembler::zero, L);
1248     __ stop("tried to execute abstract method in interpreter");
1249     __ bind(L);
1250   }
1251 #endif
1252 
1253   // Since at this point in the method invocation the exception handler
1254   // would try to exit the monitor of synchronized methods which hasn't
1255   // been entered yet, we set the thread local variable
1256   // _do_not_unlock_if_synchronized to true. The remove_activation will
1257   // check this flag.
1258 
1259   __ get_thread(rax);
1260   const Address do_not_unlock_if_synchronized(rax,
1261         in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1262   __ movbool(do_not_unlock_if_synchronized, true);
1263 
1264   // increment invocation count & check for overflow
1265   Label invocation_counter_overflow;
1266   Label profile_method;
1267   Label profile_method_continue;
1268   if (inc_counter) {
1269     generate_counter_incr(&invocation_counter_overflow, &profile_method, &profile_method_continue);
1270     if (ProfileInterpreter) {
1271       __ bind(profile_method_continue);
1272     }
1273   }
1274   Label continue_after_compile;
1275   __ bind(continue_after_compile);
1276 
1277   bang_stack_shadow_pages(false);
1278 
1279   // reset the _do_not_unlock_if_synchronized flag
1280   __ get_thread(rax);
1281   __ movbool(do_not_unlock_if_synchronized, false);
1282 
1283   // check for synchronized methods
1284   // Must happen AFTER invocation_counter check and stack overflow check,
1285   // so method is not locked if overflows.
1286   //
1287   if (synchronized) {
1288     // Allocate monitor and lock method
1289     lock_method();
1290   } else {
1291     // no synchronization necessary
1292 #ifdef ASSERT
1293       { Label L;
1294         __ movl(rax, access_flags);
1295         __ testl(rax, JVM_ACC_SYNCHRONIZED);
1296         __ jcc(Assembler::zero, L);
1297         __ stop("method needs synchronization");
1298         __ bind(L);
1299       }
1300 #endif
1301   }
1302 
1303   // start execution
1304 #ifdef ASSERT
1305   { Label L;
1306      const Address monitor_block_top (rbp,
1307                  frame::interpreter_frame_monitor_block_top_offset * wordSize);
1308     __ movptr(rax, monitor_block_top);
1309     __ cmpptr(rax, rsp);
1310     __ jcc(Assembler::equal, L);
1311     __ stop("broken stack frame setup in interpreter");
1312     __ bind(L);
1313   }
1314 #endif
1315 
1316   // jvmti support
1317   __ notify_method_entry();
1318 
1319   __ dispatch_next(vtos);
1320 
1321   // invocation counter overflow
1322   if (inc_counter) {
1323     if (ProfileInterpreter) {
1324       // We have decided to profile this method in the interpreter
1325       __ bind(profile_method);
1326 
1327       __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method), rsi, true);
1328 
1329       __ movptr(rbx, Address(rbp, method_offset));   // restore methodOop
1330       __ movptr(rax, Address(rbx, in_bytes(methodOopDesc::method_data_offset())));
1331       __ movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), rax);
1332       __ test_method_data_pointer(rax, profile_method_continue);
1333       __ addptr(rax, in_bytes(methodDataOopDesc::data_offset()));
1334       __ movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), rax);
1335       __ jmp(profile_method_continue);
1336     }
1337     // Handle overflow of counter and compile method
1338     __ bind(invocation_counter_overflow);
1339     generate_counter_overflow(&continue_after_compile);
1340   }
1341 
1342   return entry_point;
1343 }
1344 
1345 //------------------------------------------------------------------------------------------------------------------------
1346 // Entry points
1347 //
1348 // Here we generate the various kind of entries into the interpreter.
1349 // The two main entry type are generic bytecode methods and native call method.
1350 // These both come in synchronized and non-synchronized versions but the
1351 // frame layout they create is very similar. The other method entry
1352 // types are really just special purpose entries that are really entry
1353 // and interpretation all in one. These are for trivial methods like
1354 // accessor, empty, or special math methods.
1355 //
1356 // When control flow reaches any of the entry types for the interpreter
1357 // the following holds ->
1358 //
1359 // Arguments:
1360 //
1361 // rbx,: methodOop
1362 // rcx: receiver
1363 //
1364 //
1365 // Stack layout immediately at entry
1366 //
1367 // [ return address     ] <--- rsp
1368 // [ parameter n        ]
1369 //   ...
1370 // [ parameter 1        ]
1371 // [ expression stack   ] (caller's java expression stack)
1372 
1373 // Assuming that we don't go to one of the trivial specialized
1374 // entries the stack will look like below when we are ready to execute
1375 // the first bytecode (or call the native routine). The register usage
1376 // will be as the template based interpreter expects (see interpreter_x86.hpp).
1377 //
1378 // local variables follow incoming parameters immediately; i.e.
1379 // the return address is moved to the end of the locals).
1380 //
1381 // [ monitor entry      ] <--- rsp
1382 //   ...
1383 // [ monitor entry      ]
1384 // [ expr. stack bottom ]
1385 // [ saved rsi          ]
1386 // [ current rdi        ]
1387 // [ methodOop          ]
1388 // [ saved rbp,          ] <--- rbp,
1389 // [ return address     ]
1390 // [ local variable m   ]
1391 //   ...
1392 // [ local variable 1   ]
1393 // [ parameter n        ]
1394 //   ...
1395 // [ parameter 1        ] <--- rdi
1396 
1397 address AbstractInterpreterGenerator::generate_method_entry(AbstractInterpreter::MethodKind kind) {
1398   // determine code generation flags
1399   bool synchronized = false;
1400   address entry_point = NULL;
1401 
1402   switch (kind) {
1403     case Interpreter::zerolocals             :                                                                             break;
1404     case Interpreter::zerolocals_synchronized: synchronized = true;                                                        break;
1405     case Interpreter::native                 : entry_point = ((InterpreterGenerator*)this)->generate_native_entry(false);  break;
1406     case Interpreter::native_synchronized    : entry_point = ((InterpreterGenerator*)this)->generate_native_entry(true);   break;
1407     case Interpreter::empty                  : entry_point = ((InterpreterGenerator*)this)->generate_empty_entry();        break;
1408     case Interpreter::accessor               : entry_point = ((InterpreterGenerator*)this)->generate_accessor_entry();     break;
1409     case Interpreter::abstract               : entry_point = ((InterpreterGenerator*)this)->generate_abstract_entry();     break;
1410     case Interpreter::method_handle          : entry_point = ((InterpreterGenerator*)this)->generate_method_handle_entry(); break;
1411 
1412     case Interpreter::java_lang_math_sin     : // fall thru
1413     case Interpreter::java_lang_math_cos     : // fall thru
1414     case Interpreter::java_lang_math_tan     : // fall thru
1415     case Interpreter::java_lang_math_abs     : // fall thru
1416     case Interpreter::java_lang_math_log     : // fall thru
1417     case Interpreter::java_lang_math_log10   : // fall thru
1418     case Interpreter::java_lang_math_sqrt    : entry_point = ((InterpreterGenerator*)this)->generate_math_entry(kind);     break;
1419     default                                  : ShouldNotReachHere();                                                       break;
1420   }
1421 
1422   if (entry_point) return entry_point;
1423 
1424   return ((InterpreterGenerator*)this)->generate_normal_entry(synchronized);
1425 
1426 }
1427 
1428 // How much stack a method activation needs in words.
1429 int AbstractInterpreter::size_top_interpreter_activation(methodOop method) {
1430 
1431   const int stub_code = 4;  // see generate_call_stub
1432   // Save space for one monitor to get into the interpreted method in case
1433   // the method is synchronized
1434   int monitor_size    = method->is_synchronized() ?
1435                                 1*frame::interpreter_frame_monitor_size() : 0;
1436 
1437   // total overhead size: entry_size + (saved rbp, thru expr stack bottom).
1438   // be sure to change this if you add/subtract anything to/from the overhead area
1439   const int overhead_size = -frame::interpreter_frame_initial_sp_offset;
1440 
1441   const int extra_stack = methodOopDesc::extra_stack();
1442   const int method_stack = (method->max_locals() + method->max_stack() + extra_stack) *
1443                            Interpreter::stackElementWords();
1444   return overhead_size + method_stack + extra_stack + stub_code;
1445 }
1446 
1447 // asm based interpreter deoptimization helpers
1448 
1449 int AbstractInterpreter::layout_activation(methodOop method,
1450                                            int tempcount,
1451                                            int popframe_extra_args,
1452                                            int moncount,
1453                                            int callee_param_count,
1454                                            int callee_locals,
1455                                            frame* caller,
1456                                            frame* interpreter_frame,
1457                                            bool is_top_frame) {
1458   // Note: This calculation must exactly parallel the frame setup
1459   // in AbstractInterpreterGenerator::generate_method_entry.
1460   // If interpreter_frame!=NULL, set up the method, locals, and monitors.
1461   // The frame interpreter_frame, if not NULL, is guaranteed to be the right size,
1462   // as determined by a previous call to this method.
1463   // It is also guaranteed to be walkable even though it is in a skeletal state
1464   // NOTE: return size is in words not bytes
1465 
1466   // fixed size of an interpreter frame:
1467   int max_locals = method->max_locals() * Interpreter::stackElementWords();
1468   int extra_locals = (method->max_locals() - method->size_of_parameters()) *
1469                      Interpreter::stackElementWords();
1470 
1471   int overhead = frame::sender_sp_offset - frame::interpreter_frame_initial_sp_offset;
1472 
1473   // Our locals were accounted for by the caller (or last_frame_adjust on the transistion)
1474   // Since the callee parameters already account for the callee's params we only need to account for
1475   // the extra locals.
1476 
1477 
1478   int size = overhead +
1479          ((callee_locals - callee_param_count)*Interpreter::stackElementWords()) +
1480          (moncount*frame::interpreter_frame_monitor_size()) +
1481          tempcount*Interpreter::stackElementWords() + popframe_extra_args;
1482 
1483   if (interpreter_frame != NULL) {
1484 #ifdef ASSERT
1485     assert(caller->unextended_sp() == interpreter_frame->interpreter_frame_sender_sp(), "Frame not properly walkable");
1486     assert(caller->sp() == interpreter_frame->sender_sp(), "Frame not properly walkable(2)");
1487 #endif
1488 
1489     interpreter_frame->interpreter_frame_set_method(method);
1490     // NOTE the difference in using sender_sp and interpreter_frame_sender_sp
1491     // interpreter_frame_sender_sp is the original sp of the caller (the unextended_sp)
1492     // and sender_sp is fp+8
1493     intptr_t* locals = interpreter_frame->sender_sp() + max_locals - 1;
1494 
1495     interpreter_frame->interpreter_frame_set_locals(locals);
1496     BasicObjectLock* montop = interpreter_frame->interpreter_frame_monitor_begin();
1497     BasicObjectLock* monbot = montop - moncount;
1498     interpreter_frame->interpreter_frame_set_monitor_end(monbot);
1499 
1500     // Set last_sp
1501     intptr_t*  rsp = (intptr_t*) monbot  -
1502                      tempcount*Interpreter::stackElementWords() -
1503                      popframe_extra_args;
1504     interpreter_frame->interpreter_frame_set_last_sp(rsp);
1505 
1506     // All frames but the initial (oldest) interpreter frame we fill in have a
1507     // value for sender_sp that allows walking the stack but isn't
1508     // truly correct. Correct the value here.
1509 
1510     if (extra_locals != 0 &&
1511         interpreter_frame->sender_sp() == interpreter_frame->interpreter_frame_sender_sp() ) {
1512       interpreter_frame->set_interpreter_frame_sender_sp(caller->sp() + extra_locals);
1513     }
1514     *interpreter_frame->interpreter_frame_cache_addr() =
1515       method->constants()->cache();
1516   }
1517   return size;
1518 }
1519 
1520 
1521 //------------------------------------------------------------------------------------------------------------------------
1522 // Exceptions
1523 
1524 void TemplateInterpreterGenerator::generate_throw_exception() {
1525   // Entry point in previous activation (i.e., if the caller was interpreted)
1526   Interpreter::_rethrow_exception_entry = __ pc();
1527 
1528   // Restore sp to interpreter_frame_last_sp even though we are going
1529   // to empty the expression stack for the exception processing.
1530   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
1531   // rax,: exception
1532   // rdx: return address/pc that threw exception
1533   __ restore_bcp();                              // rsi points to call/send
1534   __ restore_locals();
1535 
1536   // Entry point for exceptions thrown within interpreter code
1537   Interpreter::_throw_exception_entry = __ pc();
1538   // expression stack is undefined here
1539   // rax,: exception
1540   // rsi: exception bcp
1541   __ verify_oop(rax);
1542 
1543   // expression stack must be empty before entering the VM in case of an exception
1544   __ empty_expression_stack();
1545   __ empty_FPU_stack();
1546   // find exception handler address and preserve exception oop
1547   __ call_VM(rdx, CAST_FROM_FN_PTR(address, InterpreterRuntime::exception_handler_for_exception), rax);
1548   // rax,: exception handler entry point
1549   // rdx: preserved exception oop
1550   // rsi: bcp for exception handler
1551   __ push_ptr(rdx);                              // push exception which is now the only value on the stack
1552   __ jmp(rax);                                   // jump to exception handler (may be _remove_activation_entry!)
1553 
1554   // If the exception is not handled in the current frame the frame is removed and
1555   // the exception is rethrown (i.e. exception continuation is _rethrow_exception).
1556   //
1557   // Note: At this point the bci is still the bxi for the instruction which caused
1558   //       the exception and the expression stack is empty. Thus, for any VM calls
1559   //       at this point, GC will find a legal oop map (with empty expression stack).
1560 
1561   // In current activation
1562   // tos: exception
1563   // rsi: exception bcp
1564 
1565   //
1566   // JVMTI PopFrame support
1567   //
1568 
1569    Interpreter::_remove_activation_preserving_args_entry = __ pc();
1570   __ empty_expression_stack();
1571   __ empty_FPU_stack();
1572   // Set the popframe_processing bit in pending_popframe_condition indicating that we are
1573   // currently handling popframe, so that call_VMs that may happen later do not trigger new
1574   // popframe handling cycles.
1575   __ get_thread(rcx);
1576   __ movl(rdx, Address(rcx, JavaThread::popframe_condition_offset()));
1577   __ orl(rdx, JavaThread::popframe_processing_bit);
1578   __ movl(Address(rcx, JavaThread::popframe_condition_offset()), rdx);
1579 
1580   {
1581     // Check to see whether we are returning to a deoptimized frame.
1582     // (The PopFrame call ensures that the caller of the popped frame is
1583     // either interpreted or compiled and deoptimizes it if compiled.)
1584     // In this case, we can't call dispatch_next() after the frame is
1585     // popped, but instead must save the incoming arguments and restore
1586     // them after deoptimization has occurred.
1587     //
1588     // Note that we don't compare the return PC against the
1589     // deoptimization blob's unpack entry because of the presence of
1590     // adapter frames in C2.
1591     Label caller_not_deoptimized;
1592     __ movptr(rdx, Address(rbp, frame::return_addr_offset * wordSize));
1593     __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::interpreter_contains), rdx);
1594     __ testl(rax, rax);
1595     __ jcc(Assembler::notZero, caller_not_deoptimized);
1596 
1597     // Compute size of arguments for saving when returning to deoptimized caller
1598     __ get_method(rax);
1599     __ verify_oop(rax);
1600     __ load_unsigned_word(rax, Address(rax, in_bytes(methodOopDesc::size_of_parameters_offset())));
1601     __ shlptr(rax, Interpreter::logStackElementSize());
1602     __ restore_locals();
1603     __ subptr(rdi, rax);
1604     __ addptr(rdi, wordSize);
1605     // Save these arguments
1606     __ get_thread(rcx);
1607     __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, Deoptimization::popframe_preserve_args), rcx, rax, rdi);
1608 
1609     __ remove_activation(vtos, rdx,
1610                          /* throw_monitor_exception */ false,
1611                          /* install_monitor_exception */ false,
1612                          /* notify_jvmdi */ false);
1613 
1614     // Inform deoptimization that it is responsible for restoring these arguments
1615     __ get_thread(rcx);
1616     __ movl(Address(rcx, JavaThread::popframe_condition_offset()), JavaThread::popframe_force_deopt_reexecution_bit);
1617 
1618     // Continue in deoptimization handler
1619     __ jmp(rdx);
1620 
1621     __ bind(caller_not_deoptimized);
1622   }
1623 
1624   __ remove_activation(vtos, rdx,
1625                        /* throw_monitor_exception */ false,
1626                        /* install_monitor_exception */ false,
1627                        /* notify_jvmdi */ false);
1628 
1629   // Finish with popframe handling
1630   // A previous I2C followed by a deoptimization might have moved the
1631   // outgoing arguments further up the stack. PopFrame expects the
1632   // mutations to those outgoing arguments to be preserved and other
1633   // constraints basically require this frame to look exactly as
1634   // though it had previously invoked an interpreted activation with
1635   // no space between the top of the expression stack (current
1636   // last_sp) and the top of stack. Rather than force deopt to
1637   // maintain this kind of invariant all the time we call a small
1638   // fixup routine to move the mutated arguments onto the top of our
1639   // expression stack if necessary.
1640   __ mov(rax, rsp);
1641   __ movptr(rbx, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1642   __ get_thread(rcx);
1643   // PC must point into interpreter here
1644   __ set_last_Java_frame(rcx, noreg, rbp, __ pc());
1645   __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), rcx, rax, rbx);
1646   __ get_thread(rcx);
1647   __ reset_last_Java_frame(rcx, true, true);
1648   // Restore the last_sp and null it out
1649   __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1650   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
1651 
1652   __ restore_bcp();
1653   __ restore_locals();
1654   // The method data pointer was incremented already during
1655   // call profiling. We have to restore the mdp for the current bcp.
1656   if (ProfileInterpreter) {
1657     __ set_method_data_pointer_for_bcp();
1658   }
1659 
1660   // Clear the popframe condition flag
1661   __ get_thread(rcx);
1662   __ movl(Address(rcx, JavaThread::popframe_condition_offset()), JavaThread::popframe_inactive);
1663 
1664   __ dispatch_next(vtos);
1665   // end of PopFrame support
1666 
1667   Interpreter::_remove_activation_entry = __ pc();
1668 
1669   // preserve exception over this code sequence
1670   __ pop_ptr(rax);
1671   __ get_thread(rcx);
1672   __ movptr(Address(rcx, JavaThread::vm_result_offset()), rax);
1673   // remove the activation (without doing throws on illegalMonitorExceptions)
1674   __ remove_activation(vtos, rdx, false, true, false);
1675   // restore exception
1676   __ get_thread(rcx);
1677   __ movptr(rax, Address(rcx, JavaThread::vm_result_offset()));
1678   __ movptr(Address(rcx, JavaThread::vm_result_offset()), (int32_t)NULL_WORD);
1679   __ verify_oop(rax);
1680 
1681   // Inbetween activations - previous activation type unknown yet
1682   // compute continuation point - the continuation point expects
1683   // the following registers set up:
1684   //
1685   // rax,: exception
1686   // rdx: return address/pc that threw exception
1687   // rsp: expression stack of caller
1688   // rbp,: rbp, of caller
1689   __ push(rax);                                  // save exception
1690   __ push(rdx);                                  // save return address
1691   __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), rdx);
1692   __ mov(rbx, rax);                              // save exception handler
1693   __ pop(rdx);                                   // restore return address
1694   __ pop(rax);                                   // restore exception
1695   // Note that an "issuing PC" is actually the next PC after the call
1696   __ jmp(rbx);                                   // jump to exception handler of caller
1697 }
1698 
1699 
1700 //
1701 // JVMTI ForceEarlyReturn support
1702 //
1703 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
1704   address entry = __ pc();
1705 
1706   __ restore_bcp();
1707   __ restore_locals();
1708   __ empty_expression_stack();
1709   __ empty_FPU_stack();
1710   __ load_earlyret_value(state);
1711 
1712   __ get_thread(rcx);
1713   __ movptr(rcx, Address(rcx, JavaThread::jvmti_thread_state_offset()));
1714   const Address cond_addr(rcx, JvmtiThreadState::earlyret_state_offset());
1715 
1716   // Clear the earlyret state
1717   __ movl(cond_addr, JvmtiThreadState::earlyret_inactive);
1718 
1719   __ remove_activation(state, rsi,
1720                        false, /* throw_monitor_exception */
1721                        false, /* install_monitor_exception */
1722                        true); /* notify_jvmdi */
1723   __ jmp(rsi);
1724   return entry;
1725 } // end of ForceEarlyReturn support
1726 
1727 
1728 //------------------------------------------------------------------------------------------------------------------------
1729 // Helper for vtos entry point generation
1730 
1731 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) {
1732   assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
1733   Label L;
1734   fep = __ pc(); __ push(ftos); __ jmp(L);
1735   dep = __ pc(); __ push(dtos); __ jmp(L);
1736   lep = __ pc(); __ push(ltos); __ jmp(L);
1737   aep = __ pc(); __ push(atos); __ jmp(L);
1738   bep = cep = sep =             // fall through
1739   iep = __ pc(); __ push(itos); // fall through
1740   vep = __ pc(); __ bind(L);    // fall through
1741   generate_and_dispatch(t);
1742 }
1743 
1744 //------------------------------------------------------------------------------------------------------------------------
1745 // Generation of individual instructions
1746 
1747 // helpers for generate_and_dispatch
1748 
1749 
1750 
1751 InterpreterGenerator::InterpreterGenerator(StubQueue* code)
1752  : TemplateInterpreterGenerator(code) {
1753    generate_all(); // down here so it can be "virtual"
1754 }
1755 
1756 //------------------------------------------------------------------------------------------------------------------------
1757 
1758 // Non-product code
1759 #ifndef PRODUCT
1760 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
1761   address entry = __ pc();
1762 
1763   // prepare expression stack
1764   __ pop(rcx);          // pop return address so expression stack is 'pure'
1765   __ push(state);       // save tosca
1766 
1767   // pass tosca registers as arguments & call tracer
1768   __ call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::trace_bytecode), rcx, rax, rdx);
1769   __ mov(rcx, rax);     // make sure return address is not destroyed by pop(state)
1770   __ pop(state);        // restore tosca
1771 
1772   // return
1773   __ jmp(rcx);
1774 
1775   return entry;
1776 }
1777 
1778 
1779 void TemplateInterpreterGenerator::count_bytecode() {
1780   __ incrementl(ExternalAddress((address) &BytecodeCounter::_counter_value));
1781 }
1782 
1783 
1784 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
1785   __ incrementl(ExternalAddress((address) &BytecodeHistogram::_counters[t->bytecode()]));
1786 }
1787 
1788 
1789 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
1790   __ mov32(ExternalAddress((address) &BytecodePairHistogram::_index), rbx);
1791   __ shrl(rbx, BytecodePairHistogram::log2_number_of_codes);
1792   __ orl(rbx, ((int)t->bytecode()) << BytecodePairHistogram::log2_number_of_codes);
1793   ExternalAddress table((address) BytecodePairHistogram::_counters);
1794   Address index(noreg, rbx, Address::times_4);
1795   __ incrementl(ArrayAddress(table, index));
1796 }
1797 
1798 
1799 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
1800   // Call a little run-time stub to avoid blow-up for each bytecode.
1801   // The run-time runtime saves the right registers, depending on
1802   // the tosca in-state for the given template.
1803   assert(Interpreter::trace_code(t->tos_in()) != NULL,
1804          "entry must have been generated");
1805   __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in())));
1806 }
1807 
1808 
1809 void TemplateInterpreterGenerator::stop_interpreter_at() {
1810   Label L;
1811   __ cmp32(ExternalAddress((address) &BytecodeCounter::_counter_value),
1812            StopInterpreterAt);
1813   __ jcc(Assembler::notEqual, L);
1814   __ int3();
1815   __ bind(L);
1816 }
1817 #endif // !PRODUCT
1818 #endif // CC_INTERP