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