1 /*
   2  * Copyright 1997-2009 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 // Arguments are: required type at TOS+4, failing object (or NULL) at TOS.
  96 address TemplateInterpreterGenerator::generate_WrongMethodType_handler() {
  97   address entry = __ pc();
  98 
  99   __ pop(rbx);                  // actual failing object is at TOS
 100   __ pop(rax);                  // required type is at TOS+4
 101 
 102   __ verify_oop(rbx);
 103   __ verify_oop(rax);
 104 
 105   // Various method handle types use interpreter registers as temps.
 106   __ restore_bcp();
 107   __ restore_locals();
 108 
 109   // Expression stack must be empty before entering the VM for an exception.
 110   __ empty_expression_stack();
 111   __ empty_FPU_stack();
 112   __ call_VM(noreg,
 113              CAST_FROM_FN_PTR(address,
 114                               InterpreterRuntime::throw_WrongMethodTypeException),
 115              // pass required type, failing object (or NULL)
 116              rax, rbx);
 117   return entry;
 118 }
 119 
 120 
 121 address TemplateInterpreterGenerator::generate_exception_handler_common(const char* name, const char* message, bool pass_oop) {
 122   assert(!pass_oop || message == NULL, "either oop or message but not both");
 123   address entry = __ pc();
 124   if (pass_oop) {
 125     // object is at TOS
 126     __ pop(rbx);
 127   }
 128   // expression stack must be empty before entering the VM if an exception happened
 129   __ empty_expression_stack();
 130   __ empty_FPU_stack();
 131   // setup parameters
 132   __ lea(rax, ExternalAddress((address)name));
 133   if (pass_oop) {
 134     __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_klass_exception), rax, rbx);
 135   } else {
 136     if (message != NULL) {
 137       __ lea(rbx, ExternalAddress((address)message));
 138     } else {
 139       __ movptr(rbx, NULL_WORD);
 140     }
 141     __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception), rax, rbx);
 142   }
 143   // throw exception
 144   __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
 145   return entry;
 146 }
 147 
 148 
 149 address TemplateInterpreterGenerator::generate_continuation_for(TosState state) {
 150   address entry = __ pc();
 151   // NULL last_sp until next java call
 152   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
 153   __ dispatch_next(state);
 154   return entry;
 155 }
 156 
 157 
 158 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step) {
 159   TosState incoming_state = state;
 160 
 161   Label interpreter_entry;
 162   address compiled_entry = __ pc();
 163 
 164 #ifdef COMPILER2
 165   // The FPU stack is clean if UseSSE >= 2 but must be cleaned in other cases
 166   if ((incoming_state == ftos && UseSSE < 1) || (incoming_state == dtos && UseSSE < 2)) {
 167     for (int i = 1; i < 8; i++) {
 168         __ ffree(i);
 169     }
 170   } else if (UseSSE < 2) {
 171     __ empty_FPU_stack();
 172   }
 173 #endif
 174   if ((incoming_state == ftos && UseSSE < 1) || (incoming_state == dtos && UseSSE < 2)) {
 175     __ MacroAssembler::verify_FPU(1, "generate_return_entry_for compiled");
 176   } else {
 177     __ MacroAssembler::verify_FPU(0, "generate_return_entry_for compiled");
 178   }
 179 
 180   __ jmp(interpreter_entry, relocInfo::none);
 181   // emit a sentinel we can test for when converting an interpreter
 182   // entry point to a compiled entry point.
 183   __ a_long(Interpreter::return_sentinel);
 184   __ a_long((int)compiled_entry);
 185   address entry = __ pc();
 186   __ bind(interpreter_entry);
 187 
 188   // In SSE mode, interpreter returns FP results in xmm0 but they need
 189   // to end up back on the FPU so it can operate on them.
 190   if (incoming_state == ftos && UseSSE >= 1) {
 191     __ subptr(rsp, wordSize);
 192     __ movflt(Address(rsp, 0), xmm0);
 193     __ fld_s(Address(rsp, 0));
 194     __ addptr(rsp, wordSize);
 195   } else if (incoming_state == dtos && UseSSE >= 2) {
 196     __ subptr(rsp, 2*wordSize);
 197     __ movdbl(Address(rsp, 0), xmm0);
 198     __ fld_d(Address(rsp, 0));
 199     __ addptr(rsp, 2*wordSize);
 200   }
 201 
 202   __ MacroAssembler::verify_FPU(state == ftos || state == dtos ? 1 : 0, "generate_return_entry_for in interpreter");
 203 
 204   // Restore stack bottom in case i2c adjusted stack
 205   __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
 206   // and NULL it as marker that rsp is now tos until next java call
 207   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
 208 
 209   __ restore_bcp();
 210   __ restore_locals();
 211 
 212   Label L_got_cache, L_giant_index;
 213   if (EnableInvokeDynamic) {
 214     __ cmpb(Address(rsi, 0), Bytecodes::_invokedynamic);
 215     __ jcc(Assembler::equal, L_giant_index);
 216   }
 217   __ get_cache_and_index_at_bcp(rbx, rcx, 1, false);
 218   __ bind(L_got_cache);
 219   __ movl(rbx, Address(rbx, rcx,
 220                     Address::times_ptr, constantPoolCacheOopDesc::base_offset() +
 221                     ConstantPoolCacheEntry::flags_offset()));
 222   __ andptr(rbx, 0xFF);
 223   __ lea(rsp, Address(rsp, rbx, Interpreter::stackElementScale()));
 224   __ dispatch_next(state, step);
 225 
 226   // out of the main line of code...
 227   if (EnableInvokeDynamic) {
 228     __ bind(L_giant_index);
 229     __ get_cache_and_index_at_bcp(rbx, rcx, 1, true);
 230     __ jmp(L_got_cache);
 231   }
 232 
 233   return entry;
 234 }
 235 
 236 
 237 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state, int step) {
 238   address entry = __ pc();
 239 
 240   // In SSE mode, FP results are in xmm0
 241   if (state == ftos && UseSSE > 0) {
 242     __ subptr(rsp, wordSize);
 243     __ movflt(Address(rsp, 0), xmm0);
 244     __ fld_s(Address(rsp, 0));
 245     __ addptr(rsp, wordSize);
 246   } else if (state == dtos && UseSSE >= 2) {
 247     __ subptr(rsp, 2*wordSize);
 248     __ movdbl(Address(rsp, 0), xmm0);
 249     __ fld_d(Address(rsp, 0));
 250     __ addptr(rsp, 2*wordSize);
 251   }
 252 
 253   __ MacroAssembler::verify_FPU(state == ftos || state == dtos ? 1 : 0, "generate_deopt_entry_for in interpreter");
 254 
 255   // The stack is not extended by deopt but we must NULL last_sp as this
 256   // entry is like a "return".
 257   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
 258   __ restore_bcp();
 259   __ restore_locals();
 260   // handle exceptions
 261   { Label L;
 262     const Register thread = rcx;
 263     __ get_thread(thread);
 264     __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
 265     __ jcc(Assembler::zero, L);
 266     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_pending_exception));
 267     __ should_not_reach_here();
 268     __ bind(L);
 269   }
 270   __ dispatch_next(state, step);
 271   return entry;
 272 }
 273 
 274 
 275 int AbstractInterpreter::BasicType_as_index(BasicType type) {
 276   int i = 0;
 277   switch (type) {
 278     case T_BOOLEAN: i = 0; break;
 279     case T_CHAR   : i = 1; break;
 280     case T_BYTE   : i = 2; break;
 281     case T_SHORT  : i = 3; break;
 282     case T_INT    : // fall through
 283     case T_LONG   : // fall through
 284     case T_VOID   : i = 4; break;
 285     case T_FLOAT  : i = 5; break;  // have to treat float and double separately for SSE
 286     case T_DOUBLE : i = 6; break;
 287     case T_OBJECT : // fall through
 288     case T_ARRAY  : i = 7; break;
 289     default       : ShouldNotReachHere();
 290   }
 291   assert(0 <= i && i < AbstractInterpreter::number_of_result_handlers, "index out of bounds");
 292   return i;
 293 }
 294 
 295 
 296 address TemplateInterpreterGenerator::generate_result_handler_for(BasicType type) {
 297   address entry = __ pc();
 298   switch (type) {
 299     case T_BOOLEAN: __ c2bool(rax);            break;
 300     case T_CHAR   : __ andptr(rax, 0xFFFF);    break;
 301     case T_BYTE   : __ sign_extend_byte (rax); break;
 302     case T_SHORT  : __ sign_extend_short(rax); break;
 303     case T_INT    : /* nothing to do */        break;
 304     case T_DOUBLE :
 305     case T_FLOAT  :
 306       { const Register t = InterpreterRuntime::SignatureHandlerGenerator::temp();
 307         __ pop(t);                            // remove return address first
 308         __ pop_dtos_to_rsp();
 309         // Must return a result for interpreter or compiler. In SSE
 310         // mode, results are returned in xmm0 and the FPU stack must
 311         // be empty.
 312         if (type == T_FLOAT && UseSSE >= 1) {
 313           // Load ST0
 314           __ fld_d(Address(rsp, 0));
 315           // Store as float and empty fpu stack
 316           __ fstp_s(Address(rsp, 0));
 317           // and reload
 318           __ movflt(xmm0, Address(rsp, 0));
 319         } else if (type == T_DOUBLE && UseSSE >= 2 ) {
 320           __ movdbl(xmm0, Address(rsp, 0));
 321         } else {
 322           // restore ST0
 323           __ fld_d(Address(rsp, 0));
 324         }
 325         // and pop the temp
 326         __ addptr(rsp, 2 * wordSize);
 327         __ push(t);                           // restore return address
 328       }
 329       break;
 330     case T_OBJECT :
 331       // retrieve result from frame
 332       __ movptr(rax, Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize));
 333       // and verify it
 334       __ verify_oop(rax);
 335       break;
 336     default       : ShouldNotReachHere();
 337   }
 338   __ ret(0);                                   // return from result handler
 339   return entry;
 340 }
 341 
 342 address TemplateInterpreterGenerator::generate_safept_entry_for(TosState state, address runtime_entry) {
 343   address entry = __ pc();
 344   __ push(state);
 345   __ call_VM(noreg, runtime_entry);
 346   __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
 347   return entry;
 348 }
 349 
 350 
 351 // Helpers for commoning out cases in the various type of method entries.
 352 //
 353 
 354 // increment invocation count & check for overflow
 355 //
 356 // Note: checking for negative value instead of overflow
 357 //       so we have a 'sticky' overflow test
 358 //
 359 // rbx,: method
 360 // rcx: invocation counter
 361 //
 362 void InterpreterGenerator::generate_counter_incr(Label* overflow, Label* profile_method, Label* profile_method_continue) {
 363 
 364   const Address invocation_counter(rbx, methodOopDesc::invocation_counter_offset() + InvocationCounter::counter_offset());
 365   const Address backedge_counter  (rbx, methodOopDesc::backedge_counter_offset() + InvocationCounter::counter_offset());
 366 
 367   if (ProfileInterpreter) { // %%% Merge this into methodDataOop
 368     __ incrementl(Address(rbx,methodOopDesc::interpreter_invocation_counter_offset()));
 369   }
 370   // Update standard invocation counters
 371   __ movl(rax, backedge_counter);               // load backedge counter
 372 
 373   __ incrementl(rcx, InvocationCounter::count_increment);
 374   __ andl(rax, InvocationCounter::count_mask_value);  // mask out the status bits
 375 
 376   __ movl(invocation_counter, rcx);             // save invocation count
 377   __ addl(rcx, rax);                            // add both counters
 378 
 379   // profile_method is non-null only for interpreted method so
 380   // profile_method != NULL == !native_call
 381   // BytecodeInterpreter only calls for native so code is elided.
 382 
 383   if (ProfileInterpreter && profile_method != NULL) {
 384     // Test to see if we should create a method data oop
 385     __ cmp32(rcx,
 386              ExternalAddress((address)&InvocationCounter::InterpreterProfileLimit));
 387     __ jcc(Assembler::less, *profile_method_continue);
 388 
 389     // if no method data exists, go to profile_method
 390     __ test_method_data_pointer(rax, *profile_method);
 391   }
 392 
 393   __ cmp32(rcx,
 394            ExternalAddress((address)&InvocationCounter::InterpreterInvocationLimit));
 395   __ jcc(Assembler::aboveEqual, *overflow);
 396 
 397 }
 398 
 399 void InterpreterGenerator::generate_counter_overflow(Label* do_continue) {
 400 
 401   // Asm interpreter on entry
 402   // rdi - locals
 403   // rsi - bcp
 404   // rbx, - method
 405   // rdx - cpool
 406   // rbp, - interpreter frame
 407 
 408   // C++ interpreter on entry
 409   // rsi - new interpreter state pointer
 410   // rbp - interpreter frame pointer
 411   // rbx - method
 412 
 413   // On return (i.e. jump to entry_point) [ back to invocation of interpreter ]
 414   // rbx, - method
 415   // rcx - rcvr (assuming there is one)
 416   // top of stack return address of interpreter caller
 417   // rsp - sender_sp
 418 
 419   // C++ interpreter only
 420   // rsi - previous interpreter state pointer
 421 
 422   const Address size_of_parameters(rbx, methodOopDesc::size_of_parameters_offset());
 423 
 424   // InterpreterRuntime::frequency_counter_overflow takes one argument
 425   // indicating if the counter overflow occurs at a backwards branch (non-NULL bcp).
 426   // The call returns the address of the verified entry point for the method or NULL
 427   // if the compilation did not complete (either went background or bailed out).
 428   __ movptr(rax, (intptr_t)false);
 429   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), rax);
 430 
 431   __ movptr(rbx, Address(rbp, method_offset));   // restore methodOop
 432 
 433   // Preserve invariant that rsi/rdi contain bcp/locals of sender frame
 434   // and jump to the interpreted entry.
 435   __ jmp(*do_continue, relocInfo::none);
 436 
 437 }
 438 
 439 void InterpreterGenerator::generate_stack_overflow_check(void) {
 440   // see if we've got enough room on the stack for locals plus overhead.
 441   // the expression stack grows down incrementally, so the normal guard
 442   // page mechanism will work for that.
 443   //
 444   // Registers live on entry:
 445   //
 446   // Asm interpreter
 447   // rdx: number of additional locals this frame needs (what we must check)
 448   // rbx,: methodOop
 449 
 450   // destroyed on exit
 451   // rax,
 452 
 453   // NOTE:  since the additional locals are also always pushed (wasn't obvious in
 454   // generate_method_entry) so the guard should work for them too.
 455   //
 456 
 457   // monitor entry size: see picture of stack set (generate_method_entry) and frame_x86.hpp
 458   const int entry_size    = frame::interpreter_frame_monitor_size() * wordSize;
 459 
 460   // total overhead size: entry_size + (saved rbp, thru expr stack bottom).
 461   // be sure to change this if you add/subtract anything to/from the overhead area
 462   const int overhead_size = -(frame::interpreter_frame_initial_sp_offset*wordSize) + entry_size;
 463 
 464   const int page_size = os::vm_page_size();
 465 
 466   Label after_frame_check;
 467 
 468   // see if the frame is greater than one page in size. If so,
 469   // then we need to verify there is enough stack space remaining
 470   // for the additional locals.
 471   __ cmpl(rdx, (page_size - overhead_size)/Interpreter::stackElementSize());
 472   __ jcc(Assembler::belowEqual, after_frame_check);
 473 
 474   // compute rsp as if this were going to be the last frame on
 475   // the stack before the red zone
 476 
 477   Label after_frame_check_pop;
 478 
 479   __ push(rsi);
 480 
 481   const Register thread = rsi;
 482 
 483   __ get_thread(thread);
 484 
 485   const Address stack_base(thread, Thread::stack_base_offset());
 486   const Address stack_size(thread, Thread::stack_size_offset());
 487 
 488   // locals + overhead, in bytes
 489   __ lea(rax, Address(noreg, rdx, Interpreter::stackElementScale(), overhead_size));
 490 
 491 #ifdef ASSERT
 492   Label stack_base_okay, stack_size_okay;
 493   // verify that thread stack base is non-zero
 494   __ cmpptr(stack_base, (int32_t)NULL_WORD);
 495   __ jcc(Assembler::notEqual, stack_base_okay);
 496   __ stop("stack base is zero");
 497   __ bind(stack_base_okay);
 498   // verify that thread stack size is non-zero
 499   __ cmpptr(stack_size, 0);
 500   __ jcc(Assembler::notEqual, stack_size_okay);
 501   __ stop("stack size is zero");
 502   __ bind(stack_size_okay);
 503 #endif
 504 
 505   // Add stack base to locals and subtract stack size
 506   __ addptr(rax, stack_base);
 507   __ subptr(rax, stack_size);
 508 
 509   // Use the maximum number of pages we might bang.
 510   const int max_pages = StackShadowPages > (StackRedPages+StackYellowPages) ? StackShadowPages :
 511                                                                               (StackRedPages+StackYellowPages);
 512   __ addptr(rax, max_pages * page_size);
 513 
 514   // check against the current stack bottom
 515   __ cmpptr(rsp, rax);
 516   __ jcc(Assembler::above, after_frame_check_pop);
 517 
 518   __ pop(rsi);  // get saved bcp / (c++ prev state ).
 519 
 520   __ pop(rax);  // get return address
 521   __ jump(ExternalAddress(Interpreter::throw_StackOverflowError_entry()));
 522 
 523   // all done with frame size check
 524   __ bind(after_frame_check_pop);
 525   __ pop(rsi);
 526 
 527   __ bind(after_frame_check);
 528 }
 529 
 530 // Allocate monitor and lock method (asm interpreter)
 531 // rbx, - methodOop
 532 //
 533 void InterpreterGenerator::lock_method(void) {
 534   // synchronize method
 535   const Address access_flags      (rbx, methodOopDesc::access_flags_offset());
 536   const Address monitor_block_top (rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
 537   const int entry_size            = frame::interpreter_frame_monitor_size() * wordSize;
 538 
 539   #ifdef ASSERT
 540     { Label L;
 541       __ movl(rax, access_flags);
 542       __ testl(rax, JVM_ACC_SYNCHRONIZED);
 543       __ jcc(Assembler::notZero, L);
 544       __ stop("method doesn't need synchronization");
 545       __ bind(L);
 546     }
 547   #endif // ASSERT
 548   // get synchronization object
 549   { Label done;
 550     const int mirror_offset = klassOopDesc::klass_part_offset_in_bytes() + Klass::java_mirror_offset_in_bytes();
 551     __ movl(rax, access_flags);
 552     __ testl(rax, JVM_ACC_STATIC);
 553     __ movptr(rax, Address(rdi, Interpreter::local_offset_in_bytes(0)));  // get receiver (assume this is frequent case)
 554     __ jcc(Assembler::zero, done);
 555     __ movptr(rax, Address(rbx, methodOopDesc::constants_offset()));
 556     __ movptr(rax, Address(rax, constantPoolOopDesc::pool_holder_offset_in_bytes()));
 557     __ movptr(rax, Address(rax, mirror_offset));
 558     __ bind(done);
 559   }
 560   // add space for monitor & lock
 561   __ subptr(rsp, entry_size);                                           // add space for a monitor entry
 562   __ movptr(monitor_block_top, rsp);                                    // set new monitor block top
 563   __ movptr(Address(rsp, BasicObjectLock::obj_offset_in_bytes()), rax); // store object
 564   __ mov(rdx, rsp);                                                    // object address
 565   __ lock_object(rdx);
 566 }
 567 
 568 //
 569 // Generate a fixed interpreter frame. This is identical setup for interpreted methods
 570 // and for native methods hence the shared code.
 571 
 572 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
 573   // initialize fixed part of activation frame
 574   __ push(rax);                                       // save return address
 575   __ enter();                                         // save old & set new rbp,
 576 
 577 
 578   __ push(rsi);                                       // set sender sp
 579   __ push((int32_t)NULL_WORD);                        // leave last_sp as null
 580   __ movptr(rsi, Address(rbx,methodOopDesc::const_offset())); // get constMethodOop
 581   __ lea(rsi, Address(rsi,constMethodOopDesc::codes_offset())); // get codebase
 582   __ push(rbx);                                      // save methodOop
 583   if (ProfileInterpreter) {
 584     Label method_data_continue;
 585     __ movptr(rdx, Address(rbx, in_bytes(methodOopDesc::method_data_offset())));
 586     __ testptr(rdx, rdx);
 587     __ jcc(Assembler::zero, method_data_continue);
 588     __ addptr(rdx, in_bytes(methodDataOopDesc::data_offset()));
 589     __ bind(method_data_continue);
 590     __ push(rdx);                                       // set the mdp (method data pointer)
 591   } else {
 592     __ push(0);
 593   }
 594 
 595   __ movptr(rdx, Address(rbx, methodOopDesc::constants_offset()));
 596   __ movptr(rdx, Address(rdx, constantPoolOopDesc::cache_offset_in_bytes()));
 597   __ push(rdx);                                       // set constant pool cache
 598   __ push(rdi);                                       // set locals pointer
 599   if (native_call) {
 600     __ push(0);                                       // no bcp
 601   } else {
 602     __ push(rsi);                                     // set bcp
 603     }
 604   __ push(0);                                         // reserve word for pointer to expression stack bottom
 605   __ movptr(Address(rsp, 0), rsp);                    // set expression stack bottom
 606 }
 607 
 608 // End of helpers
 609 
 610 //
 611 // Various method entries
 612 //------------------------------------------------------------------------------------------------------------------------
 613 //
 614 //
 615 
 616 // Call an accessor method (assuming it is resolved, otherwise drop into vanilla (slow path) entry
 617 
 618 address InterpreterGenerator::generate_accessor_entry(void) {
 619 
 620   // rbx,: methodOop
 621   // rcx: receiver (preserve for slow entry into asm interpreter)
 622 
 623   // rsi: senderSP must preserved for slow path, set SP to it on fast path
 624 
 625   address entry_point = __ pc();
 626   Label xreturn_path;
 627 
 628   // do fastpath for resolved accessor methods
 629   if (UseFastAccessorMethods) {
 630     Label slow_path;
 631     // If we need a safepoint check, generate full interpreter entry.
 632     ExternalAddress state(SafepointSynchronize::address_of_state());
 633     __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
 634              SafepointSynchronize::_not_synchronized);
 635 
 636     __ jcc(Assembler::notEqual, slow_path);
 637     // ASM/C++ Interpreter
 638     // Code: _aload_0, _(i|a)getfield, _(i|a)return or any rewrites thereof; parameter size = 1
 639     // Note: We can only use this code if the getfield has been resolved
 640     //       and if we don't have a null-pointer exception => check for
 641     //       these conditions first and use slow path if necessary.
 642     // rbx,: method
 643     // rcx: receiver
 644     __ movptr(rax, Address(rsp, wordSize));
 645 
 646     // check if local 0 != NULL and read field
 647     __ testptr(rax, rax);
 648     __ jcc(Assembler::zero, slow_path);
 649 
 650     __ movptr(rdi, Address(rbx, methodOopDesc::constants_offset()));
 651     // read first instruction word and extract bytecode @ 1 and index @ 2
 652     __ movptr(rdx, Address(rbx, methodOopDesc::const_offset()));
 653     __ movl(rdx, Address(rdx, constMethodOopDesc::codes_offset()));
 654     // Shift codes right to get the index on the right.
 655     // The bytecode fetched looks like <index><0xb4><0x2a>
 656     __ shrl(rdx, 2*BitsPerByte);
 657     __ shll(rdx, exact_log2(in_words(ConstantPoolCacheEntry::size())));
 658     __ movptr(rdi, Address(rdi, constantPoolOopDesc::cache_offset_in_bytes()));
 659 
 660     // rax,: local 0
 661     // rbx,: method
 662     // rcx: receiver - do not destroy since it is needed for slow path!
 663     // rcx: scratch
 664     // rdx: constant pool cache index
 665     // rdi: constant pool cache
 666     // rsi: sender sp
 667 
 668     // check if getfield has been resolved and read constant pool cache entry
 669     // check the validity of the cache entry by testing whether _indices field
 670     // contains Bytecode::_getfield in b1 byte.
 671     assert(in_words(ConstantPoolCacheEntry::size()) == 4, "adjust shift below");
 672     __ movl(rcx,
 673             Address(rdi,
 674                     rdx,
 675                     Address::times_ptr, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::indices_offset()));
 676     __ shrl(rcx, 2*BitsPerByte);
 677     __ andl(rcx, 0xFF);
 678     __ cmpl(rcx, Bytecodes::_getfield);
 679     __ jcc(Assembler::notEqual, slow_path);
 680 
 681     // Note: constant pool entry is not valid before bytecode is resolved
 682     __ movptr(rcx,
 683               Address(rdi,
 684                       rdx,
 685                       Address::times_ptr, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::f2_offset()));
 686     __ movl(rdx,
 687             Address(rdi,
 688                     rdx,
 689                     Address::times_ptr, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::flags_offset()));
 690 
 691     Label notByte, notShort, notChar;
 692     const Address field_address (rax, rcx, Address::times_1);
 693 
 694     // Need to differentiate between igetfield, agetfield, bgetfield etc.
 695     // because they are different sizes.
 696     // Use the type from the constant pool cache
 697     __ shrl(rdx, ConstantPoolCacheEntry::tosBits);
 698     // Make sure we don't need to mask rdx for tosBits after the above shift
 699     ConstantPoolCacheEntry::verify_tosBits();
 700     __ cmpl(rdx, btos);
 701     __ jcc(Assembler::notEqual, notByte);
 702     __ load_signed_byte(rax, field_address);
 703     __ jmp(xreturn_path);
 704 
 705     __ bind(notByte);
 706     __ cmpl(rdx, stos);
 707     __ jcc(Assembler::notEqual, notShort);
 708     __ load_signed_short(rax, field_address);
 709     __ jmp(xreturn_path);
 710 
 711     __ bind(notShort);
 712     __ cmpl(rdx, ctos);
 713     __ jcc(Assembler::notEqual, notChar);
 714     __ load_unsigned_short(rax, field_address);
 715     __ jmp(xreturn_path);
 716 
 717     __ bind(notChar);
 718 #ifdef ASSERT
 719     Label okay;
 720     __ cmpl(rdx, atos);
 721     __ jcc(Assembler::equal, okay);
 722     __ cmpl(rdx, itos);
 723     __ jcc(Assembler::equal, okay);
 724     __ stop("what type is this?");
 725     __ bind(okay);
 726 #endif // ASSERT
 727     // All the rest are a 32 bit wordsize
 728     // This is ok for now. Since fast accessors should be going away
 729     __ movptr(rax, field_address);
 730 
 731     __ bind(xreturn_path);
 732 
 733     // _ireturn/_areturn
 734     __ pop(rdi);                               // get return address
 735     __ mov(rsp, rsi);                          // set sp to sender sp
 736     __ jmp(rdi);
 737 
 738     // generate a vanilla interpreter entry as the slow path
 739     __ bind(slow_path);
 740 
 741     (void) generate_normal_entry(false);
 742     return entry_point;
 743   }
 744   return NULL;
 745 
 746 }
 747 
 748 //
 749 // Interpreter stub for calling a native method. (asm interpreter)
 750 // This sets up a somewhat different looking stack for calling the native method
 751 // than the typical interpreter frame setup.
 752 //
 753 
 754 address InterpreterGenerator::generate_native_entry(bool synchronized) {
 755   // determine code generation flags
 756   bool inc_counter  = UseCompiler || CountCompiledCalls;
 757 
 758   // rbx,: methodOop
 759   // rsi: sender sp
 760   // rsi: previous interpreter state (C++ interpreter) must preserve
 761   address entry_point = __ pc();
 762 
 763 
 764   const Address size_of_parameters(rbx, methodOopDesc::size_of_parameters_offset());
 765   const Address invocation_counter(rbx, methodOopDesc::invocation_counter_offset() + InvocationCounter::counter_offset());
 766   const Address access_flags      (rbx, methodOopDesc::access_flags_offset());
 767 
 768   // get parameter size (always needed)
 769   __ load_unsigned_short(rcx, size_of_parameters);
 770 
 771   // native calls don't need the stack size check since they have no expression stack
 772   // and the arguments are already on the stack and we only add a handful of words
 773   // to the stack
 774 
 775   // rbx,: methodOop
 776   // rcx: size of parameters
 777   // rsi: sender sp
 778 
 779   __ pop(rax);                                       // get return address
 780   // for natives the size of locals is zero
 781 
 782   // compute beginning of parameters (rdi)
 783   __ lea(rdi, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize));
 784 
 785 
 786   // add 2 zero-initialized slots for native calls
 787   // NULL result handler
 788   __ push((int32_t)NULL_WORD);
 789   // NULL oop temp (mirror or jni oop result)
 790   __ push((int32_t)NULL_WORD);
 791 
 792   if (inc_counter) __ movl(rcx, invocation_counter);  // (pre-)fetch invocation count
 793   // initialize fixed part of activation frame
 794 
 795   generate_fixed_frame(true);
 796 
 797   // make sure method is native & not abstract
 798 #ifdef ASSERT
 799   __ movl(rax, access_flags);
 800   {
 801     Label L;
 802     __ testl(rax, JVM_ACC_NATIVE);
 803     __ jcc(Assembler::notZero, L);
 804     __ stop("tried to execute non-native method as native");
 805     __ bind(L);
 806   }
 807   { Label L;
 808     __ testl(rax, JVM_ACC_ABSTRACT);
 809     __ jcc(Assembler::zero, L);
 810     __ stop("tried to execute abstract method in interpreter");
 811     __ bind(L);
 812   }
 813 #endif
 814 
 815   // Since at this point in the method invocation the exception handler
 816   // would try to exit the monitor of synchronized methods which hasn't
 817   // been entered yet, we set the thread local variable
 818   // _do_not_unlock_if_synchronized to true. The remove_activation will
 819   // check this flag.
 820 
 821   __ get_thread(rax);
 822   const Address do_not_unlock_if_synchronized(rax,
 823         in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
 824   __ movbool(do_not_unlock_if_synchronized, true);
 825 
 826   // increment invocation count & check for overflow
 827   Label invocation_counter_overflow;
 828   if (inc_counter) {
 829     generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
 830   }
 831 
 832   Label continue_after_compile;
 833   __ bind(continue_after_compile);
 834 
 835   bang_stack_shadow_pages(true);
 836 
 837   // reset the _do_not_unlock_if_synchronized flag
 838   __ get_thread(rax);
 839   __ movbool(do_not_unlock_if_synchronized, false);
 840 
 841   // check for synchronized methods
 842   // Must happen AFTER invocation_counter check and stack overflow check,
 843   // so method is not locked if overflows.
 844   //
 845   if (synchronized) {
 846     lock_method();
 847   } else {
 848     // no synchronization necessary
 849 #ifdef ASSERT
 850       { Label L;
 851         __ movl(rax, access_flags);
 852         __ testl(rax, JVM_ACC_SYNCHRONIZED);
 853         __ jcc(Assembler::zero, L);
 854         __ stop("method needs synchronization");
 855         __ bind(L);
 856       }
 857 #endif
 858   }
 859 
 860   // start execution
 861 #ifdef ASSERT
 862   { Label L;
 863     const Address monitor_block_top (rbp,
 864                  frame::interpreter_frame_monitor_block_top_offset * wordSize);
 865     __ movptr(rax, monitor_block_top);
 866     __ cmpptr(rax, rsp);
 867     __ jcc(Assembler::equal, L);
 868     __ stop("broken stack frame setup in interpreter");
 869     __ bind(L);
 870   }
 871 #endif
 872 
 873   // jvmti/dtrace support
 874   __ notify_method_entry();
 875 
 876   // work registers
 877   const Register method = rbx;
 878   const Register thread = rdi;
 879   const Register t      = rcx;
 880 
 881   // allocate space for parameters
 882   __ get_method(method);
 883   __ verify_oop(method);
 884   __ load_unsigned_short(t, Address(method, methodOopDesc::size_of_parameters_offset()));
 885   __ shlptr(t, Interpreter::logStackElementSize());
 886   __ addptr(t, 2*wordSize);     // allocate two more slots for JNIEnv and possible mirror
 887   __ subptr(rsp, t);
 888   __ andptr(rsp, -(StackAlignmentInBytes)); // gcc needs 16 byte aligned stacks to do XMM intrinsics
 889 
 890   // get signature handler
 891   { Label L;
 892     __ movptr(t, Address(method, methodOopDesc::signature_handler_offset()));
 893     __ testptr(t, t);
 894     __ jcc(Assembler::notZero, L);
 895     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), method);
 896     __ get_method(method);
 897     __ movptr(t, Address(method, methodOopDesc::signature_handler_offset()));
 898     __ bind(L);
 899   }
 900 
 901   // call signature handler
 902   assert(InterpreterRuntime::SignatureHandlerGenerator::from() == rdi, "adjust this code");
 903   assert(InterpreterRuntime::SignatureHandlerGenerator::to  () == rsp, "adjust this code");
 904   assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == t  , "adjust this code");
 905   // The generated handlers do not touch RBX (the method oop).
 906   // However, large signatures cannot be cached and are generated
 907   // each time here.  The slow-path generator will blow RBX
 908   // sometime, so we must reload it after the call.
 909   __ call(t);
 910   __ get_method(method);        // slow path call blows RBX on DevStudio 5.0
 911 
 912   // result handler is in rax,
 913   // set result handler
 914   __ movptr(Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize), rax);
 915 
 916   // pass mirror handle if static call
 917   { Label L;
 918     const int mirror_offset = klassOopDesc::klass_part_offset_in_bytes() + Klass::java_mirror_offset_in_bytes();
 919     __ movl(t, Address(method, methodOopDesc::access_flags_offset()));
 920     __ testl(t, JVM_ACC_STATIC);
 921     __ jcc(Assembler::zero, L);
 922     // get mirror
 923     __ movptr(t, Address(method, methodOopDesc:: constants_offset()));
 924     __ movptr(t, Address(t, constantPoolOopDesc::pool_holder_offset_in_bytes()));
 925     __ movptr(t, Address(t, mirror_offset));
 926     // copy mirror into activation frame
 927     __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize), t);
 928     // pass handle to mirror
 929     __ lea(t, Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize));
 930     __ movptr(Address(rsp, wordSize), t);
 931     __ bind(L);
 932   }
 933 
 934   // get native function entry point
 935   { Label L;
 936     __ movptr(rax, Address(method, methodOopDesc::native_function_offset()));
 937     ExternalAddress unsatisfied(SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
 938     __ cmpptr(rax, unsatisfied.addr());
 939     __ jcc(Assembler::notEqual, L);
 940     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), method);
 941     __ get_method(method);
 942     __ verify_oop(method);
 943     __ movptr(rax, Address(method, methodOopDesc::native_function_offset()));
 944     __ bind(L);
 945   }
 946 
 947   // pass JNIEnv
 948   __ get_thread(thread);
 949   __ lea(t, Address(thread, JavaThread::jni_environment_offset()));
 950   __ movptr(Address(rsp, 0), t);
 951 
 952   // set_last_Java_frame_before_call
 953   // It is enough that the pc()
 954   // points into the right code segment. It does not have to be the correct return pc.
 955   __ set_last_Java_frame(thread, noreg, rbp, __ pc());
 956 
 957   // change thread state
 958 #ifdef ASSERT
 959   { Label L;
 960     __ movl(t, Address(thread, JavaThread::thread_state_offset()));
 961     __ cmpl(t, _thread_in_Java);
 962     __ jcc(Assembler::equal, L);
 963     __ stop("Wrong thread state in native stub");
 964     __ bind(L);
 965   }
 966 #endif
 967 
 968   // Change state to native
 969   __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_native);
 970   __ call(rax);
 971 
 972   // result potentially in rdx:rax or ST0
 973 
 974   // Either restore the MXCSR register after returning from the JNI Call
 975   // or verify that it wasn't changed.
 976   if (VM_Version::supports_sse()) {
 977     if (RestoreMXCSROnJNICalls) {
 978       __ ldmxcsr(ExternalAddress(StubRoutines::addr_mxcsr_std()));
 979     }
 980     else if (CheckJNICalls ) {
 981       __ call(RuntimeAddress(StubRoutines::x86::verify_mxcsr_entry()));
 982     }
 983   }
 984 
 985   // Either restore the x87 floating pointer control word after returning
 986   // from the JNI call or verify that it wasn't changed.
 987   if (CheckJNICalls) {
 988     __ call(RuntimeAddress(StubRoutines::x86::verify_fpu_cntrl_wrd_entry()));
 989   }
 990 
 991   // save potential result in ST(0) & rdx:rax
 992   // (if result handler is the T_FLOAT or T_DOUBLE handler, result must be in ST0 -
 993   // the check is necessary to avoid potential Intel FPU overflow problems by saving/restoring 'empty' FPU registers)
 994   // It is safe to do this push because state is _thread_in_native and return address will be found
 995   // via _last_native_pc and not via _last_jave_sp
 996 
 997   // NOTE: the order of theses push(es) is known to frame::interpreter_frame_result.
 998   // If the order changes or anything else is added to the stack the code in
 999   // interpreter_frame_result will have to be changed.
1000 
1001   { Label L;
1002     Label push_double;
1003     ExternalAddress float_handler(AbstractInterpreter::result_handler(T_FLOAT));
1004     ExternalAddress double_handler(AbstractInterpreter::result_handler(T_DOUBLE));
1005     __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize),
1006               float_handler.addr());
1007     __ jcc(Assembler::equal, push_double);
1008     __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize),
1009               double_handler.addr());
1010     __ jcc(Assembler::notEqual, L);
1011     __ bind(push_double);
1012     __ push(dtos);
1013     __ bind(L);
1014   }
1015   __ push(ltos);
1016 
1017   // change thread state
1018   __ get_thread(thread);
1019   __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_native_trans);
1020   if(os::is_MP()) {
1021     if (UseMembar) {
1022       // Force this write out before the read below
1023       __ membar(Assembler::Membar_mask_bits(
1024            Assembler::LoadLoad | Assembler::LoadStore |
1025            Assembler::StoreLoad | Assembler::StoreStore));
1026     } else {
1027       // Write serialization page so VM thread can do a pseudo remote membar.
1028       // We use the current thread pointer to calculate a thread specific
1029       // offset to write to within the page. This minimizes bus traffic
1030       // due to cache line collision.
1031       __ serialize_memory(thread, rcx);
1032     }
1033   }
1034 
1035   if (AlwaysRestoreFPU) {
1036     //  Make sure the control word is correct.
1037     __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
1038   }
1039 
1040   // check for safepoint operation in progress and/or pending suspend requests
1041   { Label Continue;
1042 
1043     __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
1044              SafepointSynchronize::_not_synchronized);
1045 
1046     Label L;
1047     __ jcc(Assembler::notEqual, L);
1048     __ cmpl(Address(thread, JavaThread::suspend_flags_offset()), 0);
1049     __ jcc(Assembler::equal, Continue);
1050     __ bind(L);
1051 
1052     // Don't use call_VM as it will see a possible pending exception and forward it
1053     // and never return here preventing us from clearing _last_native_pc down below.
1054     // Also can't use call_VM_leaf either as it will check to see if rsi & rdi are
1055     // preserved and correspond to the bcp/locals pointers. So we do a runtime call
1056     // by hand.
1057     //
1058     __ push(thread);
1059     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address,
1060                                             JavaThread::check_special_condition_for_native_trans)));
1061     __ increment(rsp, wordSize);
1062     __ get_thread(thread);
1063 
1064     __ bind(Continue);
1065   }
1066 
1067   // change thread state
1068   __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_Java);
1069 
1070   __ reset_last_Java_frame(thread, true, true);
1071 
1072   // reset handle block
1073   __ movptr(t, Address(thread, JavaThread::active_handles_offset()));
1074   __ movptr(Address(t, JNIHandleBlock::top_offset_in_bytes()), NULL_WORD);
1075 
1076   // If result was an oop then unbox and save it in the frame
1077   { Label L;
1078     Label no_oop, store_result;
1079     ExternalAddress handler(AbstractInterpreter::result_handler(T_OBJECT));
1080     __ cmpptr(Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize),
1081               handler.addr());
1082     __ jcc(Assembler::notEqual, no_oop);
1083     __ cmpptr(Address(rsp, 0), (int32_t)NULL_WORD);
1084     __ pop(ltos);
1085     __ testptr(rax, rax);
1086     __ jcc(Assembler::zero, store_result);
1087     // unbox
1088     __ movptr(rax, Address(rax, 0));
1089     __ bind(store_result);
1090     __ movptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset)*wordSize), rax);
1091     // keep stack depth as expected by pushing oop which will eventually be discarded
1092     __ push(ltos);
1093     __ bind(no_oop);
1094   }
1095 
1096   {
1097      Label no_reguard;
1098      __ cmpl(Address(thread, JavaThread::stack_guard_state_offset()), JavaThread::stack_guard_yellow_disabled);
1099      __ jcc(Assembler::notEqual, no_reguard);
1100 
1101      __ pusha();
1102      __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
1103      __ popa();
1104 
1105      __ bind(no_reguard);
1106    }
1107 
1108   // restore rsi to have legal interpreter frame,
1109   // i.e., bci == 0 <=> rsi == code_base()
1110   // Can't call_VM until bcp is within reasonable.
1111   __ get_method(method);      // method is junk from thread_in_native to now.
1112   __ verify_oop(method);
1113   __ movptr(rsi, Address(method,methodOopDesc::const_offset()));   // get constMethodOop
1114   __ lea(rsi, Address(rsi,constMethodOopDesc::codes_offset()));    // get codebase
1115 
1116   // handle exceptions (exception handling will handle unlocking!)
1117   { Label L;
1118     __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
1119     __ jcc(Assembler::zero, L);
1120     // Note: At some point we may want to unify this with the code used in call_VM_base();
1121     //       i.e., we should use the StubRoutines::forward_exception code. For now this
1122     //       doesn't work here because the rsp is not correctly set at this point.
1123     __ MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_pending_exception));
1124     __ should_not_reach_here();
1125     __ bind(L);
1126   }
1127 
1128   // do unlocking if necessary
1129   { Label L;
1130     __ movl(t, Address(method, methodOopDesc::access_flags_offset()));
1131     __ testl(t, JVM_ACC_SYNCHRONIZED);
1132     __ jcc(Assembler::zero, L);
1133     // the code below should be shared with interpreter macro assembler implementation
1134     { Label unlock;
1135       // BasicObjectLock will be first in list, since this is a synchronized method. However, need
1136       // to check that the object has not been unlocked by an explicit monitorexit bytecode.
1137       const Address monitor(rbp, frame::interpreter_frame_initial_sp_offset * wordSize - (int)sizeof(BasicObjectLock));
1138 
1139       __ lea(rdx, monitor);                   // address of first monitor
1140 
1141       __ movptr(t, Address(rdx, BasicObjectLock::obj_offset_in_bytes()));
1142       __ testptr(t, t);
1143       __ jcc(Assembler::notZero, unlock);
1144 
1145       // Entry already unlocked, need to throw exception
1146       __ MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
1147       __ should_not_reach_here();
1148 
1149       __ bind(unlock);
1150       __ unlock_object(rdx);
1151     }
1152     __ bind(L);
1153   }
1154 
1155   // jvmti/dtrace support
1156   // Note: This must happen _after_ handling/throwing any exceptions since
1157   //       the exception handler code notifies the runtime of method exits
1158   //       too. If this happens before, method entry/exit notifications are
1159   //       not properly paired (was bug - gri 11/22/99).
1160   __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI);
1161 
1162   // restore potential result in rdx:rax, call result handler to restore potential result in ST0 & handle result
1163   __ pop(ltos);
1164   __ movptr(t, Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize));
1165   __ call(t);
1166 
1167   // remove activation
1168   __ movptr(t, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); // get sender sp
1169   __ leave();                                // remove frame anchor
1170   __ pop(rdi);                               // get return address
1171   __ mov(rsp, t);                            // set sp to sender sp
1172   __ jmp(rdi);
1173 
1174   if (inc_counter) {
1175     // Handle overflow of counter and compile method
1176     __ bind(invocation_counter_overflow);
1177     generate_counter_overflow(&continue_after_compile);
1178   }
1179 
1180   return entry_point;
1181 }
1182 
1183 //
1184 // Generic interpreted method entry to (asm) interpreter
1185 //
1186 address InterpreterGenerator::generate_normal_entry(bool synchronized) {
1187   // determine code generation flags
1188   bool inc_counter  = UseCompiler || CountCompiledCalls;
1189 
1190   // rbx,: methodOop
1191   // rsi: sender sp
1192   address entry_point = __ pc();
1193 
1194 
1195   const Address size_of_parameters(rbx, methodOopDesc::size_of_parameters_offset());
1196   const Address size_of_locals    (rbx, methodOopDesc::size_of_locals_offset());
1197   const Address invocation_counter(rbx, methodOopDesc::invocation_counter_offset() + InvocationCounter::counter_offset());
1198   const Address access_flags      (rbx, methodOopDesc::access_flags_offset());
1199 
1200   // get parameter size (always needed)
1201   __ load_unsigned_short(rcx, size_of_parameters);
1202 
1203   // rbx,: methodOop
1204   // rcx: size of parameters
1205 
1206   // rsi: sender_sp (could differ from sp+wordSize if we were called via c2i )
1207 
1208   __ load_unsigned_short(rdx, size_of_locals);       // get size of locals in words
1209   __ subl(rdx, rcx);                                // rdx = no. of additional locals
1210 
1211   // see if we've got enough room on the stack for locals plus overhead.
1212   generate_stack_overflow_check();
1213 
1214   // get return address
1215   __ pop(rax);
1216 
1217   // compute beginning of parameters (rdi)
1218   __ lea(rdi, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize));
1219 
1220   // rdx - # of additional locals
1221   // allocate space for locals
1222   // explicitly initialize locals
1223   {
1224     Label exit, loop;
1225     __ testl(rdx, rdx);
1226     __ jcc(Assembler::lessEqual, exit);               // do nothing if rdx <= 0
1227     __ bind(loop);
1228     if (TaggedStackInterpreter) {
1229       __ push((int32_t)NULL_WORD);                    // push tag
1230     }
1231     __ push((int32_t)NULL_WORD);                      // initialize local variables
1232     __ decrement(rdx);                                // until everything initialized
1233     __ jcc(Assembler::greater, loop);
1234     __ bind(exit);
1235   }
1236 
1237   if (inc_counter) __ movl(rcx, invocation_counter);  // (pre-)fetch invocation count
1238   // initialize fixed part of activation frame
1239   generate_fixed_frame(false);
1240 
1241   // make sure method is not native & not abstract
1242 #ifdef ASSERT
1243   __ movl(rax, access_flags);
1244   {
1245     Label L;
1246     __ testl(rax, JVM_ACC_NATIVE);
1247     __ jcc(Assembler::zero, L);
1248     __ stop("tried to execute native method as non-native");
1249     __ bind(L);
1250   }
1251   { Label L;
1252     __ testl(rax, JVM_ACC_ABSTRACT);
1253     __ jcc(Assembler::zero, L);
1254     __ stop("tried to execute abstract method in interpreter");
1255     __ bind(L);
1256   }
1257 #endif
1258 
1259   // Since at this point in the method invocation the exception handler
1260   // would try to exit the monitor of synchronized methods which hasn't
1261   // been entered yet, we set the thread local variable
1262   // _do_not_unlock_if_synchronized to true. The remove_activation will
1263   // check this flag.
1264 
1265   __ get_thread(rax);
1266   const Address do_not_unlock_if_synchronized(rax,
1267         in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1268   __ movbool(do_not_unlock_if_synchronized, true);
1269 
1270   // increment invocation count & check for overflow
1271   Label invocation_counter_overflow;
1272   Label profile_method;
1273   Label profile_method_continue;
1274   if (inc_counter) {
1275     generate_counter_incr(&invocation_counter_overflow, &profile_method, &profile_method_continue);
1276     if (ProfileInterpreter) {
1277       __ bind(profile_method_continue);
1278     }
1279   }
1280   Label continue_after_compile;
1281   __ bind(continue_after_compile);
1282 
1283   bang_stack_shadow_pages(false);
1284 
1285   // reset the _do_not_unlock_if_synchronized flag
1286   __ get_thread(rax);
1287   __ movbool(do_not_unlock_if_synchronized, false);
1288 
1289   // check for synchronized methods
1290   // Must happen AFTER invocation_counter check and stack overflow check,
1291   // so method is not locked if overflows.
1292   //
1293   if (synchronized) {
1294     // Allocate monitor and lock method
1295     lock_method();
1296   } else {
1297     // no synchronization necessary
1298 #ifdef ASSERT
1299       { Label L;
1300         __ movl(rax, access_flags);
1301         __ testl(rax, JVM_ACC_SYNCHRONIZED);
1302         __ jcc(Assembler::zero, L);
1303         __ stop("method needs synchronization");
1304         __ bind(L);
1305       }
1306 #endif
1307   }
1308 
1309   // start execution
1310 #ifdef ASSERT
1311   { Label L;
1312      const Address monitor_block_top (rbp,
1313                  frame::interpreter_frame_monitor_block_top_offset * wordSize);
1314     __ movptr(rax, monitor_block_top);
1315     __ cmpptr(rax, rsp);
1316     __ jcc(Assembler::equal, L);
1317     __ stop("broken stack frame setup in interpreter");
1318     __ bind(L);
1319   }
1320 #endif
1321 
1322   // jvmti support
1323   __ notify_method_entry();
1324 
1325   __ dispatch_next(vtos);
1326 
1327   // invocation counter overflow
1328   if (inc_counter) {
1329     if (ProfileInterpreter) {
1330       // We have decided to profile this method in the interpreter
1331       __ bind(profile_method);
1332 
1333       __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method), rsi, true);
1334 
1335       __ movptr(rbx, Address(rbp, method_offset));   // restore methodOop
1336       __ movptr(rax, Address(rbx, in_bytes(methodOopDesc::method_data_offset())));
1337       __ movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), rax);
1338       __ test_method_data_pointer(rax, profile_method_continue);
1339       __ addptr(rax, in_bytes(methodDataOopDesc::data_offset()));
1340       __ movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), rax);
1341       __ jmp(profile_method_continue);
1342     }
1343     // Handle overflow of counter and compile method
1344     __ bind(invocation_counter_overflow);
1345     generate_counter_overflow(&continue_after_compile);
1346   }
1347 
1348   return entry_point;
1349 }
1350 
1351 //------------------------------------------------------------------------------------------------------------------------
1352 // Entry points
1353 //
1354 // Here we generate the various kind of entries into the interpreter.
1355 // The two main entry type are generic bytecode methods and native call method.
1356 // These both come in synchronized and non-synchronized versions but the
1357 // frame layout they create is very similar. The other method entry
1358 // types are really just special purpose entries that are really entry
1359 // and interpretation all in one. These are for trivial methods like
1360 // accessor, empty, or special math methods.
1361 //
1362 // When control flow reaches any of the entry types for the interpreter
1363 // the following holds ->
1364 //
1365 // Arguments:
1366 //
1367 // rbx,: methodOop
1368 // rcx: receiver
1369 //
1370 //
1371 // Stack layout immediately at entry
1372 //
1373 // [ return address     ] <--- rsp
1374 // [ parameter n        ]
1375 //   ...
1376 // [ parameter 1        ]
1377 // [ expression stack   ] (caller's java expression stack)
1378 
1379 // Assuming that we don't go to one of the trivial specialized
1380 // entries the stack will look like below when we are ready to execute
1381 // the first bytecode (or call the native routine). The register usage
1382 // will be as the template based interpreter expects (see interpreter_x86.hpp).
1383 //
1384 // local variables follow incoming parameters immediately; i.e.
1385 // the return address is moved to the end of the locals).
1386 //
1387 // [ monitor entry      ] <--- rsp
1388 //   ...
1389 // [ monitor entry      ]
1390 // [ expr. stack bottom ]
1391 // [ saved rsi          ]
1392 // [ current rdi        ]
1393 // [ methodOop          ]
1394 // [ saved rbp,          ] <--- rbp,
1395 // [ return address     ]
1396 // [ local variable m   ]
1397 //   ...
1398 // [ local variable 1   ]
1399 // [ parameter n        ]
1400 //   ...
1401 // [ parameter 1        ] <--- rdi
1402 
1403 address AbstractInterpreterGenerator::generate_method_entry(AbstractInterpreter::MethodKind kind) {
1404   // determine code generation flags
1405   bool synchronized = false;
1406   address entry_point = NULL;
1407 
1408   switch (kind) {
1409     case Interpreter::zerolocals             :                                                                             break;
1410     case Interpreter::zerolocals_synchronized: synchronized = true;                                                        break;
1411     case Interpreter::native                 : entry_point = ((InterpreterGenerator*)this)->generate_native_entry(false);  break;
1412     case Interpreter::native_synchronized    : entry_point = ((InterpreterGenerator*)this)->generate_native_entry(true);   break;
1413     case Interpreter::empty                  : entry_point = ((InterpreterGenerator*)this)->generate_empty_entry();        break;
1414     case Interpreter::accessor               : entry_point = ((InterpreterGenerator*)this)->generate_accessor_entry();     break;
1415     case Interpreter::abstract               : entry_point = ((InterpreterGenerator*)this)->generate_abstract_entry();     break;
1416     case Interpreter::method_handle          : entry_point = ((InterpreterGenerator*)this)->generate_method_handle_entry(); break;
1417 
1418     case Interpreter::java_lang_math_sin     : // fall thru
1419     case Interpreter::java_lang_math_cos     : // fall thru
1420     case Interpreter::java_lang_math_tan     : // fall thru
1421     case Interpreter::java_lang_math_abs     : // fall thru
1422     case Interpreter::java_lang_math_log     : // fall thru
1423     case Interpreter::java_lang_math_log10   : // fall thru
1424     case Interpreter::java_lang_math_sqrt    : entry_point = ((InterpreterGenerator*)this)->generate_math_entry(kind);     break;
1425     default                                  : ShouldNotReachHere();                                                       break;
1426   }
1427 
1428   if (entry_point) return entry_point;
1429 
1430   return ((InterpreterGenerator*)this)->generate_normal_entry(synchronized);
1431 
1432 }
1433 
1434 // How much stack a method activation needs in words.
1435 int AbstractInterpreter::size_top_interpreter_activation(methodOop method) {
1436 
1437   const int stub_code = 4;  // see generate_call_stub
1438   // Save space for one monitor to get into the interpreted method in case
1439   // the method is synchronized
1440   int monitor_size    = method->is_synchronized() ?
1441                                 1*frame::interpreter_frame_monitor_size() : 0;
1442 
1443   // total overhead size: entry_size + (saved rbp, thru expr stack bottom).
1444   // be sure to change this if you add/subtract anything to/from the overhead area
1445   const int overhead_size = -frame::interpreter_frame_initial_sp_offset;
1446 
1447   const int extra_stack = methodOopDesc::extra_stack_entries();
1448   const int method_stack = (method->max_locals() + method->max_stack() + extra_stack) *
1449                            Interpreter::stackElementWords();
1450   return overhead_size + method_stack + stub_code;
1451 }
1452 
1453 // asm based interpreter deoptimization helpers
1454 
1455 int AbstractInterpreter::layout_activation(methodOop method,
1456                                            int tempcount,
1457                                            int popframe_extra_args,
1458                                            int moncount,
1459                                            int callee_param_count,
1460                                            int callee_locals,
1461                                            frame* caller,
1462                                            frame* interpreter_frame,
1463                                            bool is_top_frame) {
1464   // Note: This calculation must exactly parallel the frame setup
1465   // in AbstractInterpreterGenerator::generate_method_entry.
1466   // If interpreter_frame!=NULL, set up the method, locals, and monitors.
1467   // The frame interpreter_frame, if not NULL, is guaranteed to be the right size,
1468   // as determined by a previous call to this method.
1469   // It is also guaranteed to be walkable even though it is in a skeletal state
1470   // NOTE: return size is in words not bytes
1471 
1472   // fixed size of an interpreter frame:
1473   int max_locals = method->max_locals() * Interpreter::stackElementWords();
1474   int extra_locals = (method->max_locals() - method->size_of_parameters()) *
1475                      Interpreter::stackElementWords();
1476 
1477   int overhead = frame::sender_sp_offset - frame::interpreter_frame_initial_sp_offset;
1478 
1479   // Our locals were accounted for by the caller (or last_frame_adjust on the transistion)
1480   // Since the callee parameters already account for the callee's params we only need to account for
1481   // the extra locals.
1482 
1483 
1484   int size = overhead +
1485          ((callee_locals - callee_param_count)*Interpreter::stackElementWords()) +
1486          (moncount*frame::interpreter_frame_monitor_size()) +
1487          tempcount*Interpreter::stackElementWords() + popframe_extra_args;
1488 
1489   if (interpreter_frame != NULL) {
1490 #ifdef ASSERT
1491     if (!EnableMethodHandles)
1492       // @@@ FIXME: Should we correct interpreter_frame_sender_sp in the calling sequences?
1493       // Probably, since deoptimization doesn't work yet.
1494       assert(caller->unextended_sp() == interpreter_frame->interpreter_frame_sender_sp(), "Frame not properly walkable");
1495     assert(caller->sp() == interpreter_frame->sender_sp(), "Frame not properly walkable(2)");
1496 #endif
1497 
1498     interpreter_frame->interpreter_frame_set_method(method);
1499     // NOTE the difference in using sender_sp and interpreter_frame_sender_sp
1500     // interpreter_frame_sender_sp is the original sp of the caller (the unextended_sp)
1501     // and sender_sp is fp+8
1502     intptr_t* locals = interpreter_frame->sender_sp() + max_locals - 1;
1503 
1504     interpreter_frame->interpreter_frame_set_locals(locals);
1505     BasicObjectLock* montop = interpreter_frame->interpreter_frame_monitor_begin();
1506     BasicObjectLock* monbot = montop - moncount;
1507     interpreter_frame->interpreter_frame_set_monitor_end(monbot);
1508 
1509     // Set last_sp
1510     intptr_t*  rsp = (intptr_t*) monbot  -
1511                      tempcount*Interpreter::stackElementWords() -
1512                      popframe_extra_args;
1513     interpreter_frame->interpreter_frame_set_last_sp(rsp);
1514 
1515     // All frames but the initial (oldest) interpreter frame we fill in have a
1516     // value for sender_sp that allows walking the stack but isn't
1517     // truly correct. Correct the value here.
1518 
1519     if (extra_locals != 0 &&
1520         interpreter_frame->sender_sp() == interpreter_frame->interpreter_frame_sender_sp() ) {
1521       interpreter_frame->set_interpreter_frame_sender_sp(caller->sp() + extra_locals);
1522     }
1523     *interpreter_frame->interpreter_frame_cache_addr() =
1524       method->constants()->cache();
1525   }
1526   return size;
1527 }
1528 
1529 
1530 //------------------------------------------------------------------------------------------------------------------------
1531 // Exceptions
1532 
1533 void TemplateInterpreterGenerator::generate_throw_exception() {
1534   // Entry point in previous activation (i.e., if the caller was interpreted)
1535   Interpreter::_rethrow_exception_entry = __ pc();
1536 
1537   // Restore sp to interpreter_frame_last_sp even though we are going
1538   // to empty the expression stack for the exception processing.
1539   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
1540   // rax,: exception
1541   // rdx: return address/pc that threw exception
1542   __ restore_bcp();                              // rsi points to call/send
1543   __ restore_locals();
1544 
1545   // Entry point for exceptions thrown within interpreter code
1546   Interpreter::_throw_exception_entry = __ pc();
1547   // expression stack is undefined here
1548   // rax,: exception
1549   // rsi: exception bcp
1550   __ verify_oop(rax);
1551 
1552   // expression stack must be empty before entering the VM in case of an exception
1553   __ empty_expression_stack();
1554   __ empty_FPU_stack();
1555   // find exception handler address and preserve exception oop
1556   __ call_VM(rdx, CAST_FROM_FN_PTR(address, InterpreterRuntime::exception_handler_for_exception), rax);
1557   // rax,: exception handler entry point
1558   // rdx: preserved exception oop
1559   // rsi: bcp for exception handler
1560   __ push_ptr(rdx);                              // push exception which is now the only value on the stack
1561   __ jmp(rax);                                   // jump to exception handler (may be _remove_activation_entry!)
1562 
1563   // If the exception is not handled in the current frame the frame is removed and
1564   // the exception is rethrown (i.e. exception continuation is _rethrow_exception).
1565   //
1566   // Note: At this point the bci is still the bxi for the instruction which caused
1567   //       the exception and the expression stack is empty. Thus, for any VM calls
1568   //       at this point, GC will find a legal oop map (with empty expression stack).
1569 
1570   // In current activation
1571   // tos: exception
1572   // rsi: exception bcp
1573 
1574   //
1575   // JVMTI PopFrame support
1576   //
1577 
1578    Interpreter::_remove_activation_preserving_args_entry = __ pc();
1579   __ empty_expression_stack();
1580   __ empty_FPU_stack();
1581   // Set the popframe_processing bit in pending_popframe_condition indicating that we are
1582   // currently handling popframe, so that call_VMs that may happen later do not trigger new
1583   // popframe handling cycles.
1584   __ get_thread(rcx);
1585   __ movl(rdx, Address(rcx, JavaThread::popframe_condition_offset()));
1586   __ orl(rdx, JavaThread::popframe_processing_bit);
1587   __ movl(Address(rcx, JavaThread::popframe_condition_offset()), rdx);
1588 
1589   {
1590     // Check to see whether we are returning to a deoptimized frame.
1591     // (The PopFrame call ensures that the caller of the popped frame is
1592     // either interpreted or compiled and deoptimizes it if compiled.)
1593     // In this case, we can't call dispatch_next() after the frame is
1594     // popped, but instead must save the incoming arguments and restore
1595     // them after deoptimization has occurred.
1596     //
1597     // Note that we don't compare the return PC against the
1598     // deoptimization blob's unpack entry because of the presence of
1599     // adapter frames in C2.
1600     Label caller_not_deoptimized;
1601     __ movptr(rdx, Address(rbp, frame::return_addr_offset * wordSize));
1602     __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::interpreter_contains), rdx);
1603     __ testl(rax, rax);
1604     __ jcc(Assembler::notZero, caller_not_deoptimized);
1605 
1606     // Compute size of arguments for saving when returning to deoptimized caller
1607     __ get_method(rax);
1608     __ verify_oop(rax);
1609     __ load_unsigned_short(rax, Address(rax, in_bytes(methodOopDesc::size_of_parameters_offset())));
1610     __ shlptr(rax, Interpreter::logStackElementSize());
1611     __ restore_locals();
1612     __ subptr(rdi, rax);
1613     __ addptr(rdi, wordSize);
1614     // Save these arguments
1615     __ get_thread(rcx);
1616     __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, Deoptimization::popframe_preserve_args), rcx, rax, rdi);
1617 
1618     __ remove_activation(vtos, rdx,
1619                          /* throw_monitor_exception */ false,
1620                          /* install_monitor_exception */ false,
1621                          /* notify_jvmdi */ false);
1622 
1623     // Inform deoptimization that it is responsible for restoring these arguments
1624     __ get_thread(rcx);
1625     __ movl(Address(rcx, JavaThread::popframe_condition_offset()), JavaThread::popframe_force_deopt_reexecution_bit);
1626 
1627     // Continue in deoptimization handler
1628     __ jmp(rdx);
1629 
1630     __ bind(caller_not_deoptimized);
1631   }
1632 
1633   __ remove_activation(vtos, rdx,
1634                        /* throw_monitor_exception */ false,
1635                        /* install_monitor_exception */ false,
1636                        /* notify_jvmdi */ false);
1637 
1638   // Finish with popframe handling
1639   // A previous I2C followed by a deoptimization might have moved the
1640   // outgoing arguments further up the stack. PopFrame expects the
1641   // mutations to those outgoing arguments to be preserved and other
1642   // constraints basically require this frame to look exactly as
1643   // though it had previously invoked an interpreted activation with
1644   // no space between the top of the expression stack (current
1645   // last_sp) and the top of stack. Rather than force deopt to
1646   // maintain this kind of invariant all the time we call a small
1647   // fixup routine to move the mutated arguments onto the top of our
1648   // expression stack if necessary.
1649   __ mov(rax, rsp);
1650   __ movptr(rbx, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1651   __ get_thread(rcx);
1652   // PC must point into interpreter here
1653   __ set_last_Java_frame(rcx, noreg, rbp, __ pc());
1654   __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), rcx, rax, rbx);
1655   __ get_thread(rcx);
1656   __ reset_last_Java_frame(rcx, true, true);
1657   // Restore the last_sp and null it out
1658   __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1659   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
1660 
1661   __ restore_bcp();
1662   __ restore_locals();
1663   // The method data pointer was incremented already during
1664   // call profiling. We have to restore the mdp for the current bcp.
1665   if (ProfileInterpreter) {
1666     __ set_method_data_pointer_for_bcp();
1667   }
1668 
1669   // Clear the popframe condition flag
1670   __ get_thread(rcx);
1671   __ movl(Address(rcx, JavaThread::popframe_condition_offset()), JavaThread::popframe_inactive);
1672 
1673   __ dispatch_next(vtos);
1674   // end of PopFrame support
1675 
1676   Interpreter::_remove_activation_entry = __ pc();
1677 
1678   // preserve exception over this code sequence
1679   __ pop_ptr(rax);
1680   __ get_thread(rcx);
1681   __ movptr(Address(rcx, JavaThread::vm_result_offset()), rax);
1682   // remove the activation (without doing throws on illegalMonitorExceptions)
1683   __ remove_activation(vtos, rdx, false, true, false);
1684   // restore exception
1685   __ get_thread(rcx);
1686   __ movptr(rax, Address(rcx, JavaThread::vm_result_offset()));
1687   __ movptr(Address(rcx, JavaThread::vm_result_offset()), NULL_WORD);
1688   __ verify_oop(rax);
1689 
1690   // Inbetween activations - previous activation type unknown yet
1691   // compute continuation point - the continuation point expects
1692   // the following registers set up:
1693   //
1694   // rax,: exception
1695   // rdx: return address/pc that threw exception
1696   // rsp: expression stack of caller
1697   // rbp,: rbp, of caller
1698   __ push(rax);                                  // save exception
1699   __ push(rdx);                                  // save return address
1700   __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), rdx);
1701   __ mov(rbx, rax);                              // save exception handler
1702   __ pop(rdx);                                   // restore return address
1703   __ pop(rax);                                   // restore exception
1704   // Note that an "issuing PC" is actually the next PC after the call
1705   __ jmp(rbx);                                   // jump to exception handler of caller
1706 }
1707 
1708 
1709 //
1710 // JVMTI ForceEarlyReturn support
1711 //
1712 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
1713   address entry = __ pc();
1714 
1715   __ restore_bcp();
1716   __ restore_locals();
1717   __ empty_expression_stack();
1718   __ empty_FPU_stack();
1719   __ load_earlyret_value(state);
1720 
1721   __ get_thread(rcx);
1722   __ movptr(rcx, Address(rcx, JavaThread::jvmti_thread_state_offset()));
1723   const Address cond_addr(rcx, JvmtiThreadState::earlyret_state_offset());
1724 
1725   // Clear the earlyret state
1726   __ movl(cond_addr, JvmtiThreadState::earlyret_inactive);
1727 
1728   __ remove_activation(state, rsi,
1729                        false, /* throw_monitor_exception */
1730                        false, /* install_monitor_exception */
1731                        true); /* notify_jvmdi */
1732   __ jmp(rsi);
1733   return entry;
1734 } // end of ForceEarlyReturn support
1735 
1736 
1737 //------------------------------------------------------------------------------------------------------------------------
1738 // Helper for vtos entry point generation
1739 
1740 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) {
1741   assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
1742   Label L;
1743   fep = __ pc(); __ push(ftos); __ jmp(L);
1744   dep = __ pc(); __ push(dtos); __ jmp(L);
1745   lep = __ pc(); __ push(ltos); __ jmp(L);
1746   aep = __ pc(); __ push(atos); __ jmp(L);
1747   bep = cep = sep =             // fall through
1748   iep = __ pc(); __ push(itos); // fall through
1749   vep = __ pc(); __ bind(L);    // fall through
1750   generate_and_dispatch(t);
1751 }
1752 
1753 //------------------------------------------------------------------------------------------------------------------------
1754 // Generation of individual instructions
1755 
1756 // helpers for generate_and_dispatch
1757 
1758 
1759 
1760 InterpreterGenerator::InterpreterGenerator(StubQueue* code)
1761  : TemplateInterpreterGenerator(code) {
1762    generate_all(); // down here so it can be "virtual"
1763 }
1764 
1765 //------------------------------------------------------------------------------------------------------------------------
1766 
1767 // Non-product code
1768 #ifndef PRODUCT
1769 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
1770   address entry = __ pc();
1771 
1772   // prepare expression stack
1773   __ pop(rcx);          // pop return address so expression stack is 'pure'
1774   __ push(state);       // save tosca
1775 
1776   // pass tosca registers as arguments & call tracer
1777   __ call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::trace_bytecode), rcx, rax, rdx);
1778   __ mov(rcx, rax);     // make sure return address is not destroyed by pop(state)
1779   __ pop(state);        // restore tosca
1780 
1781   // return
1782   __ jmp(rcx);
1783 
1784   return entry;
1785 }
1786 
1787 
1788 void TemplateInterpreterGenerator::count_bytecode() {
1789   __ incrementl(ExternalAddress((address) &BytecodeCounter::_counter_value));
1790 }
1791 
1792 
1793 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
1794   __ incrementl(ExternalAddress((address) &BytecodeHistogram::_counters[t->bytecode()]));
1795 }
1796 
1797 
1798 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
1799   __ mov32(ExternalAddress((address) &BytecodePairHistogram::_index), rbx);
1800   __ shrl(rbx, BytecodePairHistogram::log2_number_of_codes);
1801   __ orl(rbx, ((int)t->bytecode()) << BytecodePairHistogram::log2_number_of_codes);
1802   ExternalAddress table((address) BytecodePairHistogram::_counters);
1803   Address index(noreg, rbx, Address::times_4);
1804   __ incrementl(ArrayAddress(table, index));
1805 }
1806 
1807 
1808 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
1809   // Call a little run-time stub to avoid blow-up for each bytecode.
1810   // The run-time runtime saves the right registers, depending on
1811   // the tosca in-state for the given template.
1812   assert(Interpreter::trace_code(t->tos_in()) != NULL,
1813          "entry must have been generated");
1814   __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in())));
1815 }
1816 
1817 
1818 void TemplateInterpreterGenerator::stop_interpreter_at() {
1819   Label L;
1820   __ cmp32(ExternalAddress((address) &BytecodeCounter::_counter_value),
1821            StopInterpreterAt);
1822   __ jcc(Assembler::notEqual, L);
1823   __ int3();
1824   __ bind(L);
1825 }
1826 #endif // !PRODUCT
1827 #endif // CC_INTERP