rev 5462 : 8026251: New type profiling points: parameters to methods
Summary: x86 interpreter and c1 type profiling for parameters on method entries
Reviewed-by:

   1 /*
   2  * Copyright (c) 1997, 2013, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 #include "precompiled.hpp"
  26 #include "asm/macroAssembler.hpp"
  27 #include "interpreter/bytecodeHistogram.hpp"
  28 #include "interpreter/interpreter.hpp"
  29 #include "interpreter/interpreterGenerator.hpp"
  30 #include "interpreter/interpreterRuntime.hpp"
  31 #include "interpreter/templateTable.hpp"
  32 #include "oops/arrayOop.hpp"
  33 #include "oops/methodData.hpp"
  34 #include "oops/method.hpp"
  35 #include "oops/oop.inline.hpp"
  36 #include "prims/jvmtiExport.hpp"
  37 #include "prims/jvmtiThreadState.hpp"
  38 #include "runtime/arguments.hpp"
  39 #include "runtime/deoptimization.hpp"
  40 #include "runtime/frame.inline.hpp"
  41 #include "runtime/sharedRuntime.hpp"
  42 #include "runtime/stubRoutines.hpp"
  43 #include "runtime/synchronizer.hpp"
  44 #include "runtime/timer.hpp"
  45 #include "runtime/vframeArray.hpp"
  46 #include "utilities/debug.hpp"
  47 #include "utilities/macros.hpp"
  48 
  49 #define __ _masm->
  50 
  51 
  52 #ifndef CC_INTERP
  53 const int method_offset = frame::interpreter_frame_method_offset * wordSize;
  54 const int bci_offset    = frame::interpreter_frame_bcx_offset    * wordSize;
  55 const int locals_offset = frame::interpreter_frame_locals_offset * wordSize;
  56 
  57 //------------------------------------------------------------------------------------------------------------------------
  58 
  59 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() {
  60   address entry = __ pc();
  61 
  62   // Note: There should be a minimal interpreter frame set up when stack
  63   // overflow occurs since we check explicitly for it now.
  64   //
  65 #ifdef ASSERT
  66   { Label L;
  67     __ lea(rax, Address(rbp,
  68                 frame::interpreter_frame_monitor_block_top_offset * wordSize));
  69     __ cmpptr(rax, rsp);  // rax, = maximal rsp for current rbp,
  70                         //  (stack grows negative)
  71     __ jcc(Assembler::aboveEqual, L); // check if frame is complete
  72     __ stop ("interpreter frame not set up");
  73     __ bind(L);
  74   }
  75 #endif // ASSERT
  76   // Restore bcp under the assumption that the current frame is still
  77   // interpreted
  78   __ restore_bcp();
  79 
  80   // expression stack must be empty before entering the VM if an exception
  81   // happened
  82   __ empty_expression_stack();
  83   __ empty_FPU_stack();
  84   // throw exception
  85   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_StackOverflowError));
  86   return entry;
  87 }
  88 
  89 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler(const char* name) {
  90   address entry = __ pc();
  91   // expression stack must be empty before entering the VM if an exception happened
  92   __ empty_expression_stack();
  93   __ empty_FPU_stack();
  94   // setup parameters
  95   // ??? convention: expect aberrant index in register rbx,
  96   __ lea(rax, ExternalAddress((address)name));
  97   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ArrayIndexOutOfBoundsException), rax, rbx);
  98   return entry;
  99 }
 100 
 101 address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
 102   address entry = __ pc();
 103   // object is at TOS
 104   __ pop(rax);
 105   // expression stack must be empty before entering the VM if an exception
 106   // happened
 107   __ empty_expression_stack();
 108   __ empty_FPU_stack();
 109   __ call_VM(noreg,
 110              CAST_FROM_FN_PTR(address,
 111                               InterpreterRuntime::throw_ClassCastException),
 112              rax);
 113   return entry;
 114 }
 115 
 116 address TemplateInterpreterGenerator::generate_exception_handler_common(const char* name, const char* message, bool pass_oop) {
 117   assert(!pass_oop || message == NULL, "either oop or message but not both");
 118   address entry = __ pc();
 119   if (pass_oop) {
 120     // object is at TOS
 121     __ pop(rbx);
 122   }
 123   // expression stack must be empty before entering the VM if an exception happened
 124   __ empty_expression_stack();
 125   __ empty_FPU_stack();
 126   // setup parameters
 127   __ lea(rax, ExternalAddress((address)name));
 128   if (pass_oop) {
 129     __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_klass_exception), rax, rbx);
 130   } else {
 131     if (message != NULL) {
 132       __ lea(rbx, ExternalAddress((address)message));
 133     } else {
 134       __ movptr(rbx, NULL_WORD);
 135     }
 136     __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception), rax, rbx);
 137   }
 138   // throw exception
 139   __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
 140   return entry;
 141 }
 142 
 143 
 144 address TemplateInterpreterGenerator::generate_continuation_for(TosState state) {
 145   address entry = __ pc();
 146   // NULL last_sp until next java call
 147   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
 148   __ dispatch_next(state);
 149   return entry;
 150 }
 151 
 152 
 153 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step) {
 154   TosState incoming_state = state;
 155   address entry = __ pc();
 156 
 157 #ifdef COMPILER2
 158   // The FPU stack is clean if UseSSE >= 2 but must be cleaned in other cases
 159   if ((incoming_state == ftos && UseSSE < 1) || (incoming_state == dtos && UseSSE < 2)) {
 160     for (int i = 1; i < 8; i++) {
 161         __ ffree(i);
 162     }
 163   } else if (UseSSE < 2) {
 164     __ empty_FPU_stack();
 165   }
 166 #endif
 167   if ((incoming_state == ftos && UseSSE < 1) || (incoming_state == dtos && UseSSE < 2)) {
 168     __ MacroAssembler::verify_FPU(1, "generate_return_entry_for compiled");
 169   } else {
 170     __ MacroAssembler::verify_FPU(0, "generate_return_entry_for compiled");
 171   }
 172 
 173   // In SSE mode, interpreter returns FP results in xmm0 but they need
 174   // to end up back on the FPU so it can operate on them.
 175   if (incoming_state == ftos && UseSSE >= 1) {
 176     __ subptr(rsp, wordSize);
 177     __ movflt(Address(rsp, 0), xmm0);
 178     __ fld_s(Address(rsp, 0));
 179     __ addptr(rsp, wordSize);
 180   } else if (incoming_state == dtos && UseSSE >= 2) {
 181     __ subptr(rsp, 2*wordSize);
 182     __ movdbl(Address(rsp, 0), xmm0);
 183     __ fld_d(Address(rsp, 0));
 184     __ addptr(rsp, 2*wordSize);
 185   }
 186 
 187   __ MacroAssembler::verify_FPU(state == ftos || state == dtos ? 1 : 0, "generate_return_entry_for in interpreter");
 188 
 189   // Restore stack bottom in case i2c adjusted stack
 190   __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
 191   // and NULL it as marker that rsp is now tos until next java call
 192   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
 193 
 194   __ restore_bcp();
 195   __ restore_locals();
 196 
 197   if (incoming_state == atos) {
 198     Register mdp = rbx;
 199     Register tmp = rcx;
 200     __ profile_return_type(mdp, rax, tmp);
 201   }
 202 
 203   Label L_got_cache, L_giant_index;
 204   if (EnableInvokeDynamic) {
 205     __ cmpb(Address(rsi, 0), Bytecodes::_invokedynamic);
 206     __ jcc(Assembler::equal, L_giant_index);
 207   }
 208   __ get_cache_and_index_at_bcp(rbx, rcx, 1, sizeof(u2));
 209   __ bind(L_got_cache);
 210   __ movl(rbx, Address(rbx, rcx,
 211                     Address::times_ptr, ConstantPoolCache::base_offset() +
 212                     ConstantPoolCacheEntry::flags_offset()));
 213   __ andptr(rbx, 0xFF);
 214   __ lea(rsp, Address(rsp, rbx, Interpreter::stackElementScale()));
 215   __ dispatch_next(state, step);
 216 
 217   // out of the main line of code...
 218   if (EnableInvokeDynamic) {
 219     __ bind(L_giant_index);
 220     __ get_cache_and_index_at_bcp(rbx, rcx, 1, sizeof(u4));
 221     __ jmp(L_got_cache);
 222   }
 223 
 224   return entry;
 225 }
 226 
 227 
 228 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state, int step) {
 229   address entry = __ pc();
 230 
 231   // In SSE mode, FP results are in xmm0
 232   if (state == ftos && UseSSE > 0) {
 233     __ subptr(rsp, wordSize);
 234     __ movflt(Address(rsp, 0), xmm0);
 235     __ fld_s(Address(rsp, 0));
 236     __ addptr(rsp, wordSize);
 237   } else if (state == dtos && UseSSE >= 2) {
 238     __ subptr(rsp, 2*wordSize);
 239     __ movdbl(Address(rsp, 0), xmm0);
 240     __ fld_d(Address(rsp, 0));
 241     __ addptr(rsp, 2*wordSize);
 242   }
 243 
 244   __ MacroAssembler::verify_FPU(state == ftos || state == dtos ? 1 : 0, "generate_deopt_entry_for in interpreter");
 245 
 246   // The stack is not extended by deopt but we must NULL last_sp as this
 247   // entry is like a "return".
 248   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
 249   __ restore_bcp();
 250   __ restore_locals();
 251   // handle exceptions
 252   { Label L;
 253     const Register thread = rcx;
 254     __ get_thread(thread);
 255     __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
 256     __ jcc(Assembler::zero, L);
 257     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_pending_exception));
 258     __ should_not_reach_here();
 259     __ bind(L);
 260   }
 261   __ dispatch_next(state, step);
 262   return entry;
 263 }
 264 
 265 
 266 int AbstractInterpreter::BasicType_as_index(BasicType type) {
 267   int i = 0;
 268   switch (type) {
 269     case T_BOOLEAN: i = 0; break;
 270     case T_CHAR   : i = 1; break;
 271     case T_BYTE   : i = 2; break;
 272     case T_SHORT  : i = 3; break;
 273     case T_INT    : // fall through
 274     case T_LONG   : // fall through
 275     case T_VOID   : i = 4; break;
 276     case T_FLOAT  : i = 5; break;  // have to treat float and double separately for SSE
 277     case T_DOUBLE : i = 6; break;
 278     case T_OBJECT : // fall through
 279     case T_ARRAY  : i = 7; break;
 280     default       : ShouldNotReachHere();
 281   }
 282   assert(0 <= i && i < AbstractInterpreter::number_of_result_handlers, "index out of bounds");
 283   return i;
 284 }
 285 
 286 
 287 address TemplateInterpreterGenerator::generate_result_handler_for(BasicType type) {
 288   address entry = __ pc();
 289   switch (type) {
 290     case T_BOOLEAN: __ c2bool(rax);            break;
 291     case T_CHAR   : __ andptr(rax, 0xFFFF);    break;
 292     case T_BYTE   : __ sign_extend_byte (rax); break;
 293     case T_SHORT  : __ sign_extend_short(rax); break;
 294     case T_INT    : /* nothing to do */        break;
 295     case T_DOUBLE :
 296     case T_FLOAT  :
 297       { const Register t = InterpreterRuntime::SignatureHandlerGenerator::temp();
 298         __ pop(t);                            // remove return address first
 299         // Must return a result for interpreter or compiler. In SSE
 300         // mode, results are returned in xmm0 and the FPU stack must
 301         // be empty.
 302         if (type == T_FLOAT && UseSSE >= 1) {
 303           // Load ST0
 304           __ fld_d(Address(rsp, 0));
 305           // Store as float and empty fpu stack
 306           __ fstp_s(Address(rsp, 0));
 307           // and reload
 308           __ movflt(xmm0, Address(rsp, 0));
 309         } else if (type == T_DOUBLE && UseSSE >= 2 ) {
 310           __ movdbl(xmm0, Address(rsp, 0));
 311         } else {
 312           // restore ST0
 313           __ fld_d(Address(rsp, 0));
 314         }
 315         // and pop the temp
 316         __ addptr(rsp, 2 * wordSize);
 317         __ push(t);                           // restore return address
 318       }
 319       break;
 320     case T_OBJECT :
 321       // retrieve result from frame
 322       __ movptr(rax, Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize));
 323       // and verify it
 324       __ verify_oop(rax);
 325       break;
 326     default       : ShouldNotReachHere();
 327   }
 328   __ ret(0);                                   // return from result handler
 329   return entry;
 330 }
 331 
 332 address TemplateInterpreterGenerator::generate_safept_entry_for(TosState state, address runtime_entry) {
 333   address entry = __ pc();
 334   __ push(state);
 335   __ call_VM(noreg, runtime_entry);
 336   __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
 337   return entry;
 338 }
 339 
 340 
 341 // Helpers for commoning out cases in the various type of method entries.
 342 //
 343 
 344 // increment invocation count & check for overflow
 345 //
 346 // Note: checking for negative value instead of overflow
 347 //       so we have a 'sticky' overflow test
 348 //
 349 // rbx,: method
 350 // rcx: invocation counter
 351 //
 352 void InterpreterGenerator::generate_counter_incr(Label* overflow, Label* profile_method, Label* profile_method_continue) {
 353   Label done;
 354   // Note: In tiered we increment either counters in MethodCounters* or in MDO
 355   // depending if we're profiling or not.
 356   if (TieredCompilation) {
 357     int increment = InvocationCounter::count_increment;
 358     int mask = ((1 << Tier0InvokeNotifyFreqLog)  - 1) << InvocationCounter::count_shift;
 359     Label no_mdo;
 360     if (ProfileInterpreter) {
 361       // Are we profiling?
 362       __ movptr(rax, Address(rbx, Method::method_data_offset()));
 363       __ testptr(rax, rax);
 364       __ jccb(Assembler::zero, no_mdo);
 365       // Increment counter in the MDO
 366       const Address mdo_invocation_counter(rax, in_bytes(MethodData::invocation_counter_offset()) +
 367                                                 in_bytes(InvocationCounter::counter_offset()));
 368       __ increment_mask_and_jump(mdo_invocation_counter, increment, mask, rcx, false, Assembler::zero, overflow);
 369       __ jmp(done);
 370     }
 371     __ bind(no_mdo);
 372     // Increment counter in MethodCounters
 373     const Address invocation_counter(rax,
 374                   MethodCounters::invocation_counter_offset() +
 375                   InvocationCounter::counter_offset());
 376 
 377     __ get_method_counters(rbx, rax, done);
 378     __ increment_mask_and_jump(invocation_counter, increment, mask,
 379                                rcx, false, Assembler::zero, overflow);
 380     __ bind(done);
 381   } else {
 382     const Address backedge_counter  (rax,
 383                   MethodCounters::backedge_counter_offset() +
 384                   InvocationCounter::counter_offset());
 385     const Address invocation_counter(rax,
 386                   MethodCounters::invocation_counter_offset() +
 387                   InvocationCounter::counter_offset());
 388 
 389     __ get_method_counters(rbx, rax, done);
 390 
 391     if (ProfileInterpreter) {
 392       __ incrementl(Address(rax,
 393               MethodCounters::interpreter_invocation_counter_offset()));
 394     }
 395 
 396     // Update standard invocation counters
 397     __ movl(rcx, invocation_counter);
 398     __ incrementl(rcx, InvocationCounter::count_increment);
 399     __ movl(invocation_counter, rcx);             // save invocation count
 400 
 401     __ movl(rax, backedge_counter);               // load backedge counter
 402     __ andl(rax, InvocationCounter::count_mask_value);  // mask out the status bits
 403 
 404     __ addl(rcx, rax);                            // add both counters
 405 
 406     // profile_method is non-null only for interpreted method so
 407     // profile_method != NULL == !native_call
 408     // BytecodeInterpreter only calls for native so code is elided.
 409 
 410     if (ProfileInterpreter && profile_method != NULL) {
 411       // Test to see if we should create a method data oop
 412       __ cmp32(rcx,
 413                ExternalAddress((address)&InvocationCounter::InterpreterProfileLimit));
 414       __ jcc(Assembler::less, *profile_method_continue);
 415 
 416       // if no method data exists, go to profile_method
 417       __ test_method_data_pointer(rax, *profile_method);
 418     }
 419 
 420     __ cmp32(rcx,
 421              ExternalAddress((address)&InvocationCounter::InterpreterInvocationLimit));
 422     __ jcc(Assembler::aboveEqual, *overflow);
 423     __ bind(done);
 424   }
 425 }
 426 
 427 void InterpreterGenerator::generate_counter_overflow(Label* do_continue) {
 428 
 429   // Asm interpreter on entry
 430   // rdi - locals
 431   // rsi - bcp
 432   // rbx, - method
 433   // rdx - cpool
 434   // rbp, - interpreter frame
 435 
 436   // C++ interpreter on entry
 437   // rsi - new interpreter state pointer
 438   // rbp - interpreter frame pointer
 439   // rbx - method
 440 
 441   // On return (i.e. jump to entry_point) [ back to invocation of interpreter ]
 442   // rbx, - method
 443   // rcx - rcvr (assuming there is one)
 444   // top of stack return address of interpreter caller
 445   // rsp - sender_sp
 446 
 447   // C++ interpreter only
 448   // rsi - previous interpreter state pointer
 449 
 450   // InterpreterRuntime::frequency_counter_overflow takes one argument
 451   // indicating if the counter overflow occurs at a backwards branch (non-NULL bcp).
 452   // The call returns the address of the verified entry point for the method or NULL
 453   // if the compilation did not complete (either went background or bailed out).
 454   __ movptr(rax, (intptr_t)false);
 455   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), rax);
 456 
 457   __ movptr(rbx, Address(rbp, method_offset));   // restore Method*
 458 
 459   // Preserve invariant that rsi/rdi contain bcp/locals of sender frame
 460   // and jump to the interpreted entry.
 461   __ jmp(*do_continue, relocInfo::none);
 462 
 463 }
 464 
 465 void InterpreterGenerator::generate_stack_overflow_check(void) {
 466   // see if we've got enough room on the stack for locals plus overhead.
 467   // the expression stack grows down incrementally, so the normal guard
 468   // page mechanism will work for that.
 469   //
 470   // Registers live on entry:
 471   //
 472   // Asm interpreter
 473   // rdx: number of additional locals this frame needs (what we must check)
 474   // rbx,: Method*
 475 
 476   // destroyed on exit
 477   // rax,
 478 
 479   // NOTE:  since the additional locals are also always pushed (wasn't obvious in
 480   // generate_method_entry) so the guard should work for them too.
 481   //
 482 
 483   // monitor entry size: see picture of stack set (generate_method_entry) and frame_x86.hpp
 484   const int entry_size    = frame::interpreter_frame_monitor_size() * wordSize;
 485 
 486   // total overhead size: entry_size + (saved rbp, thru expr stack bottom).
 487   // be sure to change this if you add/subtract anything to/from the overhead area
 488   const int overhead_size = -(frame::interpreter_frame_initial_sp_offset*wordSize) + entry_size;
 489 
 490   const int page_size = os::vm_page_size();
 491 
 492   Label after_frame_check;
 493 
 494   // see if the frame is greater than one page in size. If so,
 495   // then we need to verify there is enough stack space remaining
 496   // for the additional locals.
 497   __ cmpl(rdx, (page_size - overhead_size)/Interpreter::stackElementSize);
 498   __ jcc(Assembler::belowEqual, after_frame_check);
 499 
 500   // compute rsp as if this were going to be the last frame on
 501   // the stack before the red zone
 502 
 503   Label after_frame_check_pop;
 504 
 505   __ push(rsi);
 506 
 507   const Register thread = rsi;
 508 
 509   __ get_thread(thread);
 510 
 511   const Address stack_base(thread, Thread::stack_base_offset());
 512   const Address stack_size(thread, Thread::stack_size_offset());
 513 
 514   // locals + overhead, in bytes
 515   __ lea(rax, Address(noreg, rdx, Interpreter::stackElementScale(), overhead_size));
 516 
 517 #ifdef ASSERT
 518   Label stack_base_okay, stack_size_okay;
 519   // verify that thread stack base is non-zero
 520   __ cmpptr(stack_base, (int32_t)NULL_WORD);
 521   __ jcc(Assembler::notEqual, stack_base_okay);
 522   __ stop("stack base is zero");
 523   __ bind(stack_base_okay);
 524   // verify that thread stack size is non-zero
 525   __ cmpptr(stack_size, 0);
 526   __ jcc(Assembler::notEqual, stack_size_okay);
 527   __ stop("stack size is zero");
 528   __ bind(stack_size_okay);
 529 #endif
 530 
 531   // Add stack base to locals and subtract stack size
 532   __ addptr(rax, stack_base);
 533   __ subptr(rax, stack_size);
 534 
 535   // Use the maximum number of pages we might bang.
 536   const int max_pages = StackShadowPages > (StackRedPages+StackYellowPages) ? StackShadowPages :
 537                                                                               (StackRedPages+StackYellowPages);
 538   __ addptr(rax, max_pages * page_size);
 539 
 540   // check against the current stack bottom
 541   __ cmpptr(rsp, rax);
 542   __ jcc(Assembler::above, after_frame_check_pop);
 543 
 544   __ pop(rsi);  // get saved bcp / (c++ prev state ).
 545 
 546   // Restore sender's sp as SP. This is necessary if the sender's
 547   // frame is an extended compiled frame (see gen_c2i_adapter())
 548   // and safer anyway in case of JSR292 adaptations.
 549 
 550   __ pop(rax); // return address must be moved if SP is changed
 551   __ mov(rsp, rsi);
 552   __ push(rax);
 553 
 554   // Note: the restored frame is not necessarily interpreted.
 555   // Use the shared runtime version of the StackOverflowError.
 556   assert(StubRoutines::throw_StackOverflowError_entry() != NULL, "stub not yet generated");
 557   __ jump(ExternalAddress(StubRoutines::throw_StackOverflowError_entry()));
 558   // all done with frame size check
 559   __ bind(after_frame_check_pop);
 560   __ pop(rsi);
 561 
 562   __ bind(after_frame_check);
 563 }
 564 
 565 // Allocate monitor and lock method (asm interpreter)
 566 // rbx, - Method*
 567 //
 568 void InterpreterGenerator::lock_method(void) {
 569   // synchronize method
 570   const Address access_flags      (rbx, Method::access_flags_offset());
 571   const Address monitor_block_top (rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
 572   const int entry_size            = frame::interpreter_frame_monitor_size() * wordSize;
 573 
 574   #ifdef ASSERT
 575     { Label L;
 576       __ movl(rax, access_flags);
 577       __ testl(rax, JVM_ACC_SYNCHRONIZED);
 578       __ jcc(Assembler::notZero, L);
 579       __ stop("method doesn't need synchronization");
 580       __ bind(L);
 581     }
 582   #endif // ASSERT
 583   // get synchronization object
 584   { Label done;
 585     const int mirror_offset = in_bytes(Klass::java_mirror_offset());
 586     __ movl(rax, access_flags);
 587     __ testl(rax, JVM_ACC_STATIC);
 588     __ movptr(rax, Address(rdi, Interpreter::local_offset_in_bytes(0)));  // get receiver (assume this is frequent case)
 589     __ jcc(Assembler::zero, done);
 590     __ movptr(rax, Address(rbx, Method::const_offset()));
 591     __ movptr(rax, Address(rax, ConstMethod::constants_offset()));
 592     __ movptr(rax, Address(rax, ConstantPool::pool_holder_offset_in_bytes()));
 593     __ movptr(rax, Address(rax, mirror_offset));
 594     __ bind(done);
 595   }
 596   // add space for monitor & lock
 597   __ subptr(rsp, entry_size);                                           // add space for a monitor entry
 598   __ movptr(monitor_block_top, rsp);                                    // set new monitor block top
 599   __ movptr(Address(rsp, BasicObjectLock::obj_offset_in_bytes()), rax); // store object
 600   __ mov(rdx, rsp);                                                    // object address
 601   __ lock_object(rdx);
 602 }
 603 
 604 //
 605 // Generate a fixed interpreter frame. This is identical setup for interpreted methods
 606 // and for native methods hence the shared code.
 607 
 608 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
 609   // initialize fixed part of activation frame
 610   __ push(rax);                                       // save return address
 611   __ enter();                                         // save old & set new rbp,
 612 
 613 
 614   __ push(rsi);                                       // set sender sp
 615   __ push((int32_t)NULL_WORD);                        // leave last_sp as null
 616   __ movptr(rsi, Address(rbx,Method::const_offset())); // get ConstMethod*
 617   __ lea(rsi, Address(rsi,ConstMethod::codes_offset())); // get codebase
 618   __ push(rbx);                                      // save Method*
 619   if (ProfileInterpreter) {
 620     Label method_data_continue;
 621     __ movptr(rdx, Address(rbx, in_bytes(Method::method_data_offset())));
 622     __ testptr(rdx, rdx);
 623     __ jcc(Assembler::zero, method_data_continue);
 624     __ addptr(rdx, in_bytes(MethodData::data_offset()));
 625     __ bind(method_data_continue);
 626     __ push(rdx);                                       // set the mdp (method data pointer)
 627   } else {
 628     __ push(0);
 629   }
 630 
 631   __ movptr(rdx, Address(rbx, Method::const_offset()));
 632   __ movptr(rdx, Address(rdx, ConstMethod::constants_offset()));
 633   __ movptr(rdx, Address(rdx, ConstantPool::cache_offset_in_bytes()));
 634   __ push(rdx);                                       // set constant pool cache
 635   __ push(rdi);                                       // set locals pointer
 636   if (native_call) {
 637     __ push(0);                                       // no bcp
 638   } else {
 639     __ push(rsi);                                     // set bcp
 640     }
 641   __ push(0);                                         // reserve word for pointer to expression stack bottom
 642   __ movptr(Address(rsp, 0), rsp);                    // set expression stack bottom
 643 }
 644 
 645 // End of helpers
 646 
 647 //
 648 // Various method entries
 649 //------------------------------------------------------------------------------------------------------------------------
 650 //
 651 //
 652 
 653 // Call an accessor method (assuming it is resolved, otherwise drop into vanilla (slow path) entry
 654 
 655 address InterpreterGenerator::generate_accessor_entry(void) {
 656 
 657   // rbx,: Method*
 658   // rcx: receiver (preserve for slow entry into asm interpreter)
 659 
 660   // rsi: senderSP must preserved for slow path, set SP to it on fast path
 661 
 662   address entry_point = __ pc();
 663   Label xreturn_path;
 664 
 665   // do fastpath for resolved accessor methods
 666   if (UseFastAccessorMethods) {
 667     Label slow_path;
 668     // If we need a safepoint check, generate full interpreter entry.
 669     ExternalAddress state(SafepointSynchronize::address_of_state());
 670     __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
 671              SafepointSynchronize::_not_synchronized);
 672 
 673     __ jcc(Assembler::notEqual, slow_path);
 674     // ASM/C++ Interpreter
 675     // Code: _aload_0, _(i|a)getfield, _(i|a)return or any rewrites thereof; parameter size = 1
 676     // Note: We can only use this code if the getfield has been resolved
 677     //       and if we don't have a null-pointer exception => check for
 678     //       these conditions first and use slow path if necessary.
 679     // rbx,: method
 680     // rcx: receiver
 681     __ movptr(rax, Address(rsp, wordSize));
 682 
 683     // check if local 0 != NULL and read field
 684     __ testptr(rax, rax);
 685     __ jcc(Assembler::zero, slow_path);
 686 
 687     // read first instruction word and extract bytecode @ 1 and index @ 2
 688     __ movptr(rdx, Address(rbx, Method::const_offset()));
 689     __ movptr(rdi, Address(rdx, ConstMethod::constants_offset()));
 690     __ movl(rdx, Address(rdx, ConstMethod::codes_offset()));
 691     // Shift codes right to get the index on the right.
 692     // The bytecode fetched looks like <index><0xb4><0x2a>
 693     __ shrl(rdx, 2*BitsPerByte);
 694     __ shll(rdx, exact_log2(in_words(ConstantPoolCacheEntry::size())));
 695     __ movptr(rdi, Address(rdi, ConstantPool::cache_offset_in_bytes()));
 696 
 697     // rax,: local 0
 698     // rbx,: method
 699     // rcx: receiver - do not destroy since it is needed for slow path!
 700     // rcx: scratch
 701     // rdx: constant pool cache index
 702     // rdi: constant pool cache
 703     // rsi: sender sp
 704 
 705     // check if getfield has been resolved and read constant pool cache entry
 706     // check the validity of the cache entry by testing whether _indices field
 707     // contains Bytecode::_getfield in b1 byte.
 708     assert(in_words(ConstantPoolCacheEntry::size()) == 4, "adjust shift below");
 709     __ movl(rcx,
 710             Address(rdi,
 711                     rdx,
 712                     Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset()));
 713     __ shrl(rcx, 2*BitsPerByte);
 714     __ andl(rcx, 0xFF);
 715     __ cmpl(rcx, Bytecodes::_getfield);
 716     __ jcc(Assembler::notEqual, slow_path);
 717 
 718     // Note: constant pool entry is not valid before bytecode is resolved
 719     __ movptr(rcx,
 720               Address(rdi,
 721                       rdx,
 722                       Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::f2_offset()));
 723     __ movl(rdx,
 724             Address(rdi,
 725                     rdx,
 726                     Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()));
 727 
 728     Label notByte, notShort, notChar;
 729     const Address field_address (rax, rcx, Address::times_1);
 730 
 731     // Need to differentiate between igetfield, agetfield, bgetfield etc.
 732     // because they are different sizes.
 733     // Use the type from the constant pool cache
 734     __ shrl(rdx, ConstantPoolCacheEntry::tos_state_shift);
 735     // Make sure we don't need to mask rdx after the above shift
 736     ConstantPoolCacheEntry::verify_tos_state_shift();
 737     __ cmpl(rdx, btos);
 738     __ jcc(Assembler::notEqual, notByte);
 739     __ load_signed_byte(rax, field_address);
 740     __ jmp(xreturn_path);
 741 
 742     __ bind(notByte);
 743     __ cmpl(rdx, stos);
 744     __ jcc(Assembler::notEqual, notShort);
 745     __ load_signed_short(rax, field_address);
 746     __ jmp(xreturn_path);
 747 
 748     __ bind(notShort);
 749     __ cmpl(rdx, ctos);
 750     __ jcc(Assembler::notEqual, notChar);
 751     __ load_unsigned_short(rax, field_address);
 752     __ jmp(xreturn_path);
 753 
 754     __ bind(notChar);
 755 #ifdef ASSERT
 756     Label okay;
 757     __ cmpl(rdx, atos);
 758     __ jcc(Assembler::equal, okay);
 759     __ cmpl(rdx, itos);
 760     __ jcc(Assembler::equal, okay);
 761     __ stop("what type is this?");
 762     __ bind(okay);
 763 #endif // ASSERT
 764     // All the rest are a 32 bit wordsize
 765     // This is ok for now. Since fast accessors should be going away
 766     __ movptr(rax, field_address);
 767 
 768     __ bind(xreturn_path);
 769 
 770     // _ireturn/_areturn
 771     __ pop(rdi);                               // get return address
 772     __ mov(rsp, rsi);                          // set sp to sender sp
 773     __ jmp(rdi);
 774 
 775     // generate a vanilla interpreter entry as the slow path
 776     __ bind(slow_path);
 777 
 778     (void) generate_normal_entry(false);
 779     return entry_point;
 780   }
 781   return NULL;
 782 
 783 }
 784 
 785 // Method entry for java.lang.ref.Reference.get.
 786 address InterpreterGenerator::generate_Reference_get_entry(void) {
 787 #if INCLUDE_ALL_GCS
 788   // Code: _aload_0, _getfield, _areturn
 789   // parameter size = 1
 790   //
 791   // The code that gets generated by this routine is split into 2 parts:
 792   //    1. The "intrinsified" code for G1 (or any SATB based GC),
 793   //    2. The slow path - which is an expansion of the regular method entry.
 794   //
 795   // Notes:-
 796   // * In the G1 code we do not check whether we need to block for
 797   //   a safepoint. If G1 is enabled then we must execute the specialized
 798   //   code for Reference.get (except when the Reference object is null)
 799   //   so that we can log the value in the referent field with an SATB
 800   //   update buffer.
 801   //   If the code for the getfield template is modified so that the
 802   //   G1 pre-barrier code is executed when the current method is
 803   //   Reference.get() then going through the normal method entry
 804   //   will be fine.
 805   // * The G1 code below can, however, check the receiver object (the instance
 806   //   of java.lang.Reference) and jump to the slow path if null. If the
 807   //   Reference object is null then we obviously cannot fetch the referent
 808   //   and so we don't need to call the G1 pre-barrier. Thus we can use the
 809   //   regular method entry code to generate the NPE.
 810   //
 811   // This code is based on generate_accessor_enty.
 812 
 813   // rbx,: Method*
 814   // rcx: receiver (preserve for slow entry into asm interpreter)
 815 
 816   // rsi: senderSP must preserved for slow path, set SP to it on fast path
 817 
 818   address entry = __ pc();
 819 
 820   const int referent_offset = java_lang_ref_Reference::referent_offset;
 821   guarantee(referent_offset > 0, "referent offset not initialized");
 822 
 823   if (UseG1GC) {
 824     Label slow_path;
 825 
 826     // Check if local 0 != NULL
 827     // If the receiver is null then it is OK to jump to the slow path.
 828     __ movptr(rax, Address(rsp, wordSize));
 829     __ testptr(rax, rax);
 830     __ jcc(Assembler::zero, slow_path);
 831 
 832     // rax: local 0 (must be preserved across the G1 barrier call)
 833     //
 834     // rbx: method (at this point it's scratch)
 835     // rcx: receiver (at this point it's scratch)
 836     // rdx: scratch
 837     // rdi: scratch
 838     //
 839     // rsi: sender sp
 840 
 841     // Preserve the sender sp in case the pre-barrier
 842     // calls the runtime
 843     __ push(rsi);
 844 
 845     // Load the value of the referent field.
 846     const Address field_address(rax, referent_offset);
 847     __ movptr(rax, field_address);
 848 
 849     // Generate the G1 pre-barrier code to log the value of
 850     // the referent field in an SATB buffer.
 851     __ get_thread(rcx);
 852     __ g1_write_barrier_pre(noreg /* obj */,
 853                             rax /* pre_val */,
 854                             rcx /* thread */,
 855                             rbx /* tmp */,
 856                             true /* tosca_save */,
 857                             true /* expand_call */);
 858 
 859     // _areturn
 860     __ pop(rsi);                // get sender sp
 861     __ pop(rdi);                // get return address
 862     __ mov(rsp, rsi);           // set sp to sender sp
 863     __ jmp(rdi);
 864 
 865     __ bind(slow_path);
 866     (void) generate_normal_entry(false);
 867 
 868     return entry;
 869   }
 870 #endif // INCLUDE_ALL_GCS
 871 
 872   // If G1 is not enabled then attempt to go through the accessor entry point
 873   // Reference.get is an accessor
 874   return generate_accessor_entry();
 875 }
 876 
 877 /**
 878  * Method entry for static native methods:
 879  *   int java.util.zip.CRC32.update(int crc, int b)
 880  */
 881 address InterpreterGenerator::generate_CRC32_update_entry() {
 882   if (UseCRC32Intrinsics) {
 883     address entry = __ pc();
 884 
 885     // rbx,: Method*
 886     // rsi: senderSP must preserved for slow path, set SP to it on fast path
 887     // rdx: scratch
 888     // rdi: scratch
 889 
 890     Label slow_path;
 891     // If we need a safepoint check, generate full interpreter entry.
 892     ExternalAddress state(SafepointSynchronize::address_of_state());
 893     __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
 894              SafepointSynchronize::_not_synchronized);
 895     __ jcc(Assembler::notEqual, slow_path);
 896 
 897     // We don't generate local frame and don't align stack because
 898     // we call stub code and there is no safepoint on this path.
 899 
 900     // Load parameters
 901     const Register crc = rax;  // crc
 902     const Register val = rdx;  // source java byte value
 903     const Register tbl = rdi;  // scratch
 904 
 905     // Arguments are reversed on java expression stack
 906     __ movl(val, Address(rsp,   wordSize)); // byte value
 907     __ movl(crc, Address(rsp, 2*wordSize)); // Initial CRC
 908 
 909     __ lea(tbl, ExternalAddress(StubRoutines::crc_table_addr()));
 910     __ notl(crc); // ~crc
 911     __ update_byte_crc32(crc, val, tbl);
 912     __ notl(crc); // ~crc
 913     // result in rax
 914 
 915     // _areturn
 916     __ pop(rdi);                // get return address
 917     __ mov(rsp, rsi);           // set sp to sender sp
 918     __ jmp(rdi);
 919 
 920     // generate a vanilla native entry as the slow path
 921     __ bind(slow_path);
 922 
 923     (void) generate_native_entry(false);
 924 
 925     return entry;
 926   }
 927   return generate_native_entry(false);
 928 }
 929 
 930 /**
 931  * Method entry for static native methods:
 932  *   int java.util.zip.CRC32.updateBytes(int crc, byte[] b, int off, int len)
 933  *   int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
 934  */
 935 address InterpreterGenerator::generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
 936   if (UseCRC32Intrinsics) {
 937     address entry = __ pc();
 938 
 939     // rbx,: Method*
 940     // rsi: senderSP must preserved for slow path, set SP to it on fast path
 941     // rdx: scratch
 942     // rdi: scratch
 943 
 944     Label slow_path;
 945     // If we need a safepoint check, generate full interpreter entry.
 946     ExternalAddress state(SafepointSynchronize::address_of_state());
 947     __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
 948              SafepointSynchronize::_not_synchronized);
 949     __ jcc(Assembler::notEqual, slow_path);
 950 
 951     // We don't generate local frame and don't align stack because
 952     // we call stub code and there is no safepoint on this path.
 953 
 954     // Load parameters
 955     const Register crc = rax;  // crc
 956     const Register buf = rdx;  // source java byte array address
 957     const Register len = rdi;  // length
 958 
 959     // Arguments are reversed on java expression stack
 960     __ movl(len,   Address(rsp,   wordSize)); // Length
 961     // Calculate address of start element
 962     if (kind == Interpreter::java_util_zip_CRC32_updateByteBuffer) {
 963       __ movptr(buf, Address(rsp, 3*wordSize)); // long buf
 964       __ addptr(buf, Address(rsp, 2*wordSize)); // + offset
 965       __ movl(crc,   Address(rsp, 5*wordSize)); // Initial CRC
 966     } else {
 967       __ movptr(buf, Address(rsp, 3*wordSize)); // byte[] array
 968       __ addptr(buf, arrayOopDesc::base_offset_in_bytes(T_BYTE)); // + header size
 969       __ addptr(buf, Address(rsp, 2*wordSize)); // + offset
 970       __ movl(crc,   Address(rsp, 4*wordSize)); // Initial CRC
 971     }
 972 
 973     __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, StubRoutines::updateBytesCRC32()), crc, buf, len);
 974     // result in rax
 975 
 976     // _areturn
 977     __ pop(rdi);                // get return address
 978     __ mov(rsp, rsi);           // set sp to sender sp
 979     __ jmp(rdi);
 980 
 981     // generate a vanilla native entry as the slow path
 982     __ bind(slow_path);
 983 
 984     (void) generate_native_entry(false);
 985 
 986     return entry;
 987   }
 988   return generate_native_entry(false);
 989 }
 990 
 991 //
 992 // Interpreter stub for calling a native method. (asm interpreter)
 993 // This sets up a somewhat different looking stack for calling the native method
 994 // than the typical interpreter frame setup.
 995 //
 996 
 997 address InterpreterGenerator::generate_native_entry(bool synchronized) {
 998   // determine code generation flags
 999   bool inc_counter  = UseCompiler || CountCompiledCalls;
1000 
1001   // rbx,: Method*
1002   // rsi: sender sp
1003   // rsi: previous interpreter state (C++ interpreter) must preserve
1004   address entry_point = __ pc();
1005 
1006   const Address constMethod       (rbx, Method::const_offset());
1007   const Address access_flags      (rbx, Method::access_flags_offset());
1008   const Address size_of_parameters(rcx, ConstMethod::size_of_parameters_offset());
1009 
1010   // get parameter size (always needed)
1011   __ movptr(rcx, constMethod);
1012   __ load_unsigned_short(rcx, size_of_parameters);
1013 
1014   // native calls don't need the stack size check since they have no expression stack
1015   // and the arguments are already on the stack and we only add a handful of words
1016   // to the stack
1017 
1018   // rbx,: Method*
1019   // rcx: size of parameters
1020   // rsi: sender sp
1021 
1022   __ pop(rax);                                       // get return address
1023   // for natives the size of locals is zero
1024 
1025   // compute beginning of parameters (rdi)
1026   __ lea(rdi, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize));
1027 
1028 
1029   // add 2 zero-initialized slots for native calls
1030   // NULL result handler
1031   __ push((int32_t)NULL_WORD);
1032   // NULL oop temp (mirror or jni oop result)
1033   __ push((int32_t)NULL_WORD);
1034 
1035   // initialize fixed part of activation frame
1036   generate_fixed_frame(true);
1037 
1038   // make sure method is native & not abstract
1039 #ifdef ASSERT
1040   __ movl(rax, access_flags);
1041   {
1042     Label L;
1043     __ testl(rax, JVM_ACC_NATIVE);
1044     __ jcc(Assembler::notZero, L);
1045     __ stop("tried to execute non-native method as native");
1046     __ bind(L);
1047   }
1048   { Label L;
1049     __ testl(rax, JVM_ACC_ABSTRACT);
1050     __ jcc(Assembler::zero, L);
1051     __ stop("tried to execute abstract method in interpreter");
1052     __ bind(L);
1053   }
1054 #endif
1055 
1056   // Since at this point in the method invocation the exception handler
1057   // would try to exit the monitor of synchronized methods which hasn't
1058   // been entered yet, we set the thread local variable
1059   // _do_not_unlock_if_synchronized to true. The remove_activation will
1060   // check this flag.
1061 
1062   __ get_thread(rax);
1063   const Address do_not_unlock_if_synchronized(rax,
1064         in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1065   __ movbool(do_not_unlock_if_synchronized, true);
1066 
1067   // increment invocation count & check for overflow
1068   Label invocation_counter_overflow;
1069   if (inc_counter) {
1070     generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
1071   }
1072 
1073   Label continue_after_compile;
1074   __ bind(continue_after_compile);
1075 
1076   bang_stack_shadow_pages(true);
1077 
1078   // reset the _do_not_unlock_if_synchronized flag
1079   __ get_thread(rax);
1080   __ movbool(do_not_unlock_if_synchronized, false);
1081 
1082   // check for synchronized methods
1083   // Must happen AFTER invocation_counter check and stack overflow check,
1084   // so method is not locked if overflows.
1085   //
1086   if (synchronized) {
1087     lock_method();
1088   } else {
1089     // no synchronization necessary
1090 #ifdef ASSERT
1091       { Label L;
1092         __ movl(rax, access_flags);
1093         __ testl(rax, JVM_ACC_SYNCHRONIZED);
1094         __ jcc(Assembler::zero, L);
1095         __ stop("method needs synchronization");
1096         __ bind(L);
1097       }
1098 #endif
1099   }
1100 
1101   // start execution
1102 #ifdef ASSERT
1103   { Label L;
1104     const Address monitor_block_top (rbp,
1105                  frame::interpreter_frame_monitor_block_top_offset * wordSize);
1106     __ movptr(rax, monitor_block_top);
1107     __ cmpptr(rax, rsp);
1108     __ jcc(Assembler::equal, L);
1109     __ stop("broken stack frame setup in interpreter");
1110     __ bind(L);
1111   }
1112 #endif
1113 
1114   // jvmti/dtrace support
1115   __ notify_method_entry();
1116 
1117   // work registers
1118   const Register method = rbx;
1119   const Register thread = rdi;
1120   const Register t      = rcx;
1121 
1122   // allocate space for parameters
1123   __ get_method(method);
1124   __ movptr(t, Address(method, Method::const_offset()));
1125   __ load_unsigned_short(t, Address(t, ConstMethod::size_of_parameters_offset()));
1126 
1127   __ shlptr(t, Interpreter::logStackElementSize);
1128   __ addptr(t, 2*wordSize);     // allocate two more slots for JNIEnv and possible mirror
1129   __ subptr(rsp, t);
1130   __ andptr(rsp, -(StackAlignmentInBytes)); // gcc needs 16 byte aligned stacks to do XMM intrinsics
1131 
1132   // get signature handler
1133   { Label L;
1134     __ movptr(t, Address(method, Method::signature_handler_offset()));
1135     __ testptr(t, t);
1136     __ jcc(Assembler::notZero, L);
1137     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), method);
1138     __ get_method(method);
1139     __ movptr(t, Address(method, Method::signature_handler_offset()));
1140     __ bind(L);
1141   }
1142 
1143   // call signature handler
1144   assert(InterpreterRuntime::SignatureHandlerGenerator::from() == rdi, "adjust this code");
1145   assert(InterpreterRuntime::SignatureHandlerGenerator::to  () == rsp, "adjust this code");
1146   assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == t  , "adjust this code");
1147   // The generated handlers do not touch RBX (the method oop).
1148   // However, large signatures cannot be cached and are generated
1149   // each time here.  The slow-path generator will blow RBX
1150   // sometime, so we must reload it after the call.
1151   __ call(t);
1152   __ get_method(method);        // slow path call blows RBX on DevStudio 5.0
1153 
1154   // result handler is in rax,
1155   // set result handler
1156   __ movptr(Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize), rax);
1157 
1158   // pass mirror handle if static call
1159   { Label L;
1160     const int mirror_offset = in_bytes(Klass::java_mirror_offset());
1161     __ movl(t, Address(method, Method::access_flags_offset()));
1162     __ testl(t, JVM_ACC_STATIC);
1163     __ jcc(Assembler::zero, L);
1164     // get mirror
1165     __ movptr(t, Address(method, Method:: const_offset()));
1166     __ movptr(t, Address(t, ConstMethod::constants_offset()));
1167     __ movptr(t, Address(t, ConstantPool::pool_holder_offset_in_bytes()));
1168     __ movptr(t, Address(t, mirror_offset));
1169     // copy mirror into activation frame
1170     __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize), t);
1171     // pass handle to mirror
1172     __ lea(t, Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize));
1173     __ movptr(Address(rsp, wordSize), t);
1174     __ bind(L);
1175   }
1176 
1177   // get native function entry point
1178   { Label L;
1179     __ movptr(rax, Address(method, Method::native_function_offset()));
1180     ExternalAddress unsatisfied(SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
1181     __ cmpptr(rax, unsatisfied.addr());
1182     __ jcc(Assembler::notEqual, L);
1183     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), method);
1184     __ get_method(method);
1185     __ movptr(rax, Address(method, Method::native_function_offset()));
1186     __ bind(L);
1187   }
1188 
1189   // pass JNIEnv
1190   __ get_thread(thread);
1191   __ lea(t, Address(thread, JavaThread::jni_environment_offset()));
1192   __ movptr(Address(rsp, 0), t);
1193 
1194   // set_last_Java_frame_before_call
1195   // It is enough that the pc()
1196   // points into the right code segment. It does not have to be the correct return pc.
1197   __ set_last_Java_frame(thread, noreg, rbp, __ pc());
1198 
1199   // change thread state
1200 #ifdef ASSERT
1201   { Label L;
1202     __ movl(t, Address(thread, JavaThread::thread_state_offset()));
1203     __ cmpl(t, _thread_in_Java);
1204     __ jcc(Assembler::equal, L);
1205     __ stop("Wrong thread state in native stub");
1206     __ bind(L);
1207   }
1208 #endif
1209 
1210   // Change state to native
1211   __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_native);
1212   __ call(rax);
1213 
1214   // result potentially in rdx:rax or ST0
1215 
1216   // Verify or restore cpu control state after JNI call
1217   __ restore_cpu_control_state_after_jni();
1218 
1219   // save potential result in ST(0) & rdx:rax
1220   // (if result handler is the T_FLOAT or T_DOUBLE handler, result must be in ST0 -
1221   // the check is necessary to avoid potential Intel FPU overflow problems by saving/restoring 'empty' FPU registers)
1222   // It is safe to do this push because state is _thread_in_native and return address will be found
1223   // via _last_native_pc and not via _last_jave_sp
1224 
1225   // NOTE: the order of theses push(es) is known to frame::interpreter_frame_result.
1226   // If the order changes or anything else is added to the stack the code in
1227   // interpreter_frame_result will have to be changed.
1228 
1229   { Label L;
1230     Label push_double;
1231     ExternalAddress float_handler(AbstractInterpreter::result_handler(T_FLOAT));
1232     ExternalAddress double_handler(AbstractInterpreter::result_handler(T_DOUBLE));
1233     __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize),
1234               float_handler.addr());
1235     __ jcc(Assembler::equal, push_double);
1236     __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize),
1237               double_handler.addr());
1238     __ jcc(Assembler::notEqual, L);
1239     __ bind(push_double);
1240     __ push(dtos);
1241     __ bind(L);
1242   }
1243   __ push(ltos);
1244 
1245   // change thread state
1246   __ get_thread(thread);
1247   __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_native_trans);
1248   if(os::is_MP()) {
1249     if (UseMembar) {
1250       // Force this write out before the read below
1251       __ membar(Assembler::Membar_mask_bits(
1252            Assembler::LoadLoad | Assembler::LoadStore |
1253            Assembler::StoreLoad | Assembler::StoreStore));
1254     } else {
1255       // Write serialization page so VM thread can do a pseudo remote membar.
1256       // We use the current thread pointer to calculate a thread specific
1257       // offset to write to within the page. This minimizes bus traffic
1258       // due to cache line collision.
1259       __ serialize_memory(thread, rcx);
1260     }
1261   }
1262 
1263   if (AlwaysRestoreFPU) {
1264     //  Make sure the control word is correct.
1265     __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
1266   }
1267 
1268   // check for safepoint operation in progress and/or pending suspend requests
1269   { Label Continue;
1270 
1271     __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
1272              SafepointSynchronize::_not_synchronized);
1273 
1274     Label L;
1275     __ jcc(Assembler::notEqual, L);
1276     __ cmpl(Address(thread, JavaThread::suspend_flags_offset()), 0);
1277     __ jcc(Assembler::equal, Continue);
1278     __ bind(L);
1279 
1280     // Don't use call_VM as it will see a possible pending exception and forward it
1281     // and never return here preventing us from clearing _last_native_pc down below.
1282     // Also can't use call_VM_leaf either as it will check to see if rsi & rdi are
1283     // preserved and correspond to the bcp/locals pointers. So we do a runtime call
1284     // by hand.
1285     //
1286     __ push(thread);
1287     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address,
1288                                             JavaThread::check_special_condition_for_native_trans)));
1289     __ increment(rsp, wordSize);
1290     __ get_thread(thread);
1291 
1292     __ bind(Continue);
1293   }
1294 
1295   // change thread state
1296   __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_Java);
1297 
1298   __ reset_last_Java_frame(thread, true, true);
1299 
1300   // reset handle block
1301   __ movptr(t, Address(thread, JavaThread::active_handles_offset()));
1302   __ movptr(Address(t, JNIHandleBlock::top_offset_in_bytes()), NULL_WORD);
1303 
1304   // If result was an oop then unbox and save it in the frame
1305   { Label L;
1306     Label no_oop, store_result;
1307     ExternalAddress handler(AbstractInterpreter::result_handler(T_OBJECT));
1308     __ cmpptr(Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize),
1309               handler.addr());
1310     __ jcc(Assembler::notEqual, no_oop);
1311     __ cmpptr(Address(rsp, 0), (int32_t)NULL_WORD);
1312     __ pop(ltos);
1313     __ testptr(rax, rax);
1314     __ jcc(Assembler::zero, store_result);
1315     // unbox
1316     __ movptr(rax, Address(rax, 0));
1317     __ bind(store_result);
1318     __ movptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset)*wordSize), rax);
1319     // keep stack depth as expected by pushing oop which will eventually be discarded
1320     __ push(ltos);
1321     __ bind(no_oop);
1322   }
1323 
1324   {
1325      Label no_reguard;
1326      __ cmpl(Address(thread, JavaThread::stack_guard_state_offset()), JavaThread::stack_guard_yellow_disabled);
1327      __ jcc(Assembler::notEqual, no_reguard);
1328 
1329      __ pusha();
1330      __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
1331      __ popa();
1332 
1333      __ bind(no_reguard);
1334    }
1335 
1336   // restore rsi to have legal interpreter frame,
1337   // i.e., bci == 0 <=> rsi == code_base()
1338   // Can't call_VM until bcp is within reasonable.
1339   __ get_method(method);      // method is junk from thread_in_native to now.
1340   __ movptr(rsi, Address(method,Method::const_offset()));   // get ConstMethod*
1341   __ lea(rsi, Address(rsi,ConstMethod::codes_offset()));    // get codebase
1342 
1343   // handle exceptions (exception handling will handle unlocking!)
1344   { Label L;
1345     __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
1346     __ jcc(Assembler::zero, L);
1347     // Note: At some point we may want to unify this with the code used in call_VM_base();
1348     //       i.e., we should use the StubRoutines::forward_exception code. For now this
1349     //       doesn't work here because the rsp is not correctly set at this point.
1350     __ MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_pending_exception));
1351     __ should_not_reach_here();
1352     __ bind(L);
1353   }
1354 
1355   // do unlocking if necessary
1356   { Label L;
1357     __ movl(t, Address(method, Method::access_flags_offset()));
1358     __ testl(t, JVM_ACC_SYNCHRONIZED);
1359     __ jcc(Assembler::zero, L);
1360     // the code below should be shared with interpreter macro assembler implementation
1361     { Label unlock;
1362       // BasicObjectLock will be first in list, since this is a synchronized method. However, need
1363       // to check that the object has not been unlocked by an explicit monitorexit bytecode.
1364       const Address monitor(rbp, frame::interpreter_frame_initial_sp_offset * wordSize - (int)sizeof(BasicObjectLock));
1365 
1366       __ lea(rdx, monitor);                   // address of first monitor
1367 
1368       __ movptr(t, Address(rdx, BasicObjectLock::obj_offset_in_bytes()));
1369       __ testptr(t, t);
1370       __ jcc(Assembler::notZero, unlock);
1371 
1372       // Entry already unlocked, need to throw exception
1373       __ MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
1374       __ should_not_reach_here();
1375 
1376       __ bind(unlock);
1377       __ unlock_object(rdx);
1378     }
1379     __ bind(L);
1380   }
1381 
1382   // jvmti/dtrace support
1383   // Note: This must happen _after_ handling/throwing any exceptions since
1384   //       the exception handler code notifies the runtime of method exits
1385   //       too. If this happens before, method entry/exit notifications are
1386   //       not properly paired (was bug - gri 11/22/99).
1387   __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI);
1388 
1389   // restore potential result in rdx:rax, call result handler to restore potential result in ST0 & handle result
1390   __ pop(ltos);
1391   __ movptr(t, Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize));
1392   __ call(t);
1393 
1394   // remove activation
1395   __ movptr(t, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); // get sender sp
1396   __ leave();                                // remove frame anchor
1397   __ pop(rdi);                               // get return address
1398   __ mov(rsp, t);                            // set sp to sender sp
1399   __ jmp(rdi);
1400 
1401   if (inc_counter) {
1402     // Handle overflow of counter and compile method
1403     __ bind(invocation_counter_overflow);
1404     generate_counter_overflow(&continue_after_compile);
1405   }
1406 
1407   return entry_point;
1408 }
1409 
1410 //
1411 // Generic interpreted method entry to (asm) interpreter
1412 //
1413 address InterpreterGenerator::generate_normal_entry(bool synchronized) {
1414   // determine code generation flags
1415   bool inc_counter  = UseCompiler || CountCompiledCalls;
1416 
1417   // rbx,: Method*
1418   // rsi: sender sp
1419   address entry_point = __ pc();
1420 
1421   const Address constMethod       (rbx, Method::const_offset());
1422   const Address access_flags      (rbx, Method::access_flags_offset());
1423   const Address size_of_parameters(rdx, ConstMethod::size_of_parameters_offset());
1424   const Address size_of_locals    (rdx, ConstMethod::size_of_locals_offset());
1425 
1426   // get parameter size (always needed)
1427   __ movptr(rdx, constMethod);
1428   __ load_unsigned_short(rcx, size_of_parameters);
1429 
1430   // rbx,: Method*
1431   // rcx: size of parameters
1432 
1433   // rsi: sender_sp (could differ from sp+wordSize if we were called via c2i )
1434 
1435   __ load_unsigned_short(rdx, size_of_locals);       // get size of locals in words
1436   __ subl(rdx, rcx);                                // rdx = no. of additional locals
1437 
1438   // see if we've got enough room on the stack for locals plus overhead.
1439   generate_stack_overflow_check();
1440 
1441   // get return address
1442   __ pop(rax);
1443 
1444   // compute beginning of parameters (rdi)
1445   __ lea(rdi, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize));
1446 
1447   // rdx - # of additional locals
1448   // allocate space for locals
1449   // explicitly initialize locals
1450   {
1451     Label exit, loop;
1452     __ testl(rdx, rdx);
1453     __ jcc(Assembler::lessEqual, exit);               // do nothing if rdx <= 0
1454     __ bind(loop);
1455     __ push((int32_t)NULL_WORD);                      // initialize local variables
1456     __ decrement(rdx);                                // until everything initialized
1457     __ jcc(Assembler::greater, loop);
1458     __ bind(exit);
1459   }
1460 
1461   // initialize fixed part of activation frame
1462   generate_fixed_frame(false);
1463 
1464   // make sure method is not native & not abstract
1465 #ifdef ASSERT
1466   __ movl(rax, access_flags);
1467   {
1468     Label L;
1469     __ testl(rax, JVM_ACC_NATIVE);
1470     __ jcc(Assembler::zero, L);
1471     __ stop("tried to execute native method as non-native");
1472     __ bind(L);
1473   }
1474   { Label L;
1475     __ testl(rax, JVM_ACC_ABSTRACT);
1476     __ jcc(Assembler::zero, L);
1477     __ stop("tried to execute abstract method in interpreter");
1478     __ bind(L);
1479   }
1480 #endif
1481 
1482   // Since at this point in the method invocation the exception handler
1483   // would try to exit the monitor of synchronized methods which hasn't
1484   // been entered yet, we set the thread local variable
1485   // _do_not_unlock_if_synchronized to true. The remove_activation will
1486   // check this flag.
1487 
1488   __ get_thread(rax);
1489   const Address do_not_unlock_if_synchronized(rax,
1490         in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1491   __ movbool(do_not_unlock_if_synchronized, true);
1492 
1493   __ profile_parameters_type(rax, rcx, rdx);
1494   // increment invocation count & check for overflow
1495   Label invocation_counter_overflow;
1496   Label profile_method;
1497   Label profile_method_continue;
1498   if (inc_counter) {
1499     generate_counter_incr(&invocation_counter_overflow, &profile_method, &profile_method_continue);
1500     if (ProfileInterpreter) {
1501       __ bind(profile_method_continue);
1502     }
1503   }
1504   Label continue_after_compile;
1505   __ bind(continue_after_compile);
1506 
1507   bang_stack_shadow_pages(false);
1508 
1509   // reset the _do_not_unlock_if_synchronized flag
1510   __ get_thread(rax);
1511   __ movbool(do_not_unlock_if_synchronized, false);
1512 
1513   // check for synchronized methods
1514   // Must happen AFTER invocation_counter check and stack overflow check,
1515   // so method is not locked if overflows.
1516   //
1517   if (synchronized) {
1518     // Allocate monitor and lock method
1519     lock_method();
1520   } else {
1521     // no synchronization necessary
1522 #ifdef ASSERT
1523       { Label L;
1524         __ movl(rax, access_flags);
1525         __ testl(rax, JVM_ACC_SYNCHRONIZED);
1526         __ jcc(Assembler::zero, L);
1527         __ stop("method needs synchronization");
1528         __ bind(L);
1529       }
1530 #endif
1531   }
1532 
1533   // start execution
1534 #ifdef ASSERT
1535   { Label L;
1536      const Address monitor_block_top (rbp,
1537                  frame::interpreter_frame_monitor_block_top_offset * wordSize);
1538     __ movptr(rax, monitor_block_top);
1539     __ cmpptr(rax, rsp);
1540     __ jcc(Assembler::equal, L);
1541     __ stop("broken stack frame setup in interpreter");
1542     __ bind(L);
1543   }
1544 #endif
1545 
1546   // jvmti support
1547   __ notify_method_entry();
1548 
1549   __ dispatch_next(vtos);
1550 
1551   // invocation counter overflow
1552   if (inc_counter) {
1553     if (ProfileInterpreter) {
1554       // We have decided to profile this method in the interpreter
1555       __ bind(profile_method);
1556       __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1557       __ set_method_data_pointer_for_bcp();
1558       __ get_method(rbx);
1559       __ jmp(profile_method_continue);
1560     }
1561     // Handle overflow of counter and compile method
1562     __ bind(invocation_counter_overflow);
1563     generate_counter_overflow(&continue_after_compile);
1564   }
1565 
1566   return entry_point;
1567 }
1568 
1569 //------------------------------------------------------------------------------------------------------------------------
1570 // Entry points
1571 //
1572 // Here we generate the various kind of entries into the interpreter.
1573 // The two main entry type are generic bytecode methods and native call method.
1574 // These both come in synchronized and non-synchronized versions but the
1575 // frame layout they create is very similar. The other method entry
1576 // types are really just special purpose entries that are really entry
1577 // and interpretation all in one. These are for trivial methods like
1578 // accessor, empty, or special math methods.
1579 //
1580 // When control flow reaches any of the entry types for the interpreter
1581 // the following holds ->
1582 //
1583 // Arguments:
1584 //
1585 // rbx,: Method*
1586 // rcx: receiver
1587 //
1588 //
1589 // Stack layout immediately at entry
1590 //
1591 // [ return address     ] <--- rsp
1592 // [ parameter n        ]
1593 //   ...
1594 // [ parameter 1        ]
1595 // [ expression stack   ] (caller's java expression stack)
1596 
1597 // Assuming that we don't go to one of the trivial specialized
1598 // entries the stack will look like below when we are ready to execute
1599 // the first bytecode (or call the native routine). The register usage
1600 // will be as the template based interpreter expects (see interpreter_x86.hpp).
1601 //
1602 // local variables follow incoming parameters immediately; i.e.
1603 // the return address is moved to the end of the locals).
1604 //
1605 // [ monitor entry      ] <--- rsp
1606 //   ...
1607 // [ monitor entry      ]
1608 // [ expr. stack bottom ]
1609 // [ saved rsi          ]
1610 // [ current rdi        ]
1611 // [ Method*            ]
1612 // [ saved rbp,          ] <--- rbp,
1613 // [ return address     ]
1614 // [ local variable m   ]
1615 //   ...
1616 // [ local variable 1   ]
1617 // [ parameter n        ]
1618 //   ...
1619 // [ parameter 1        ] <--- rdi
1620 
1621 address AbstractInterpreterGenerator::generate_method_entry(AbstractInterpreter::MethodKind kind) {
1622   // determine code generation flags
1623   bool synchronized = false;
1624   address entry_point = NULL;
1625   InterpreterGenerator* ig_this = (InterpreterGenerator*)this;
1626 
1627   switch (kind) {
1628     case Interpreter::zerolocals             :                                                       break;
1629     case Interpreter::zerolocals_synchronized: synchronized = true;                                  break;
1630     case Interpreter::native                 : entry_point = ig_this->generate_native_entry(false);  break;
1631     case Interpreter::native_synchronized    : entry_point = ig_this->generate_native_entry(true);   break;
1632     case Interpreter::empty                  : entry_point = ig_this->generate_empty_entry();        break;
1633     case Interpreter::accessor               : entry_point = ig_this->generate_accessor_entry();     break;
1634     case Interpreter::abstract               : entry_point = ig_this->generate_abstract_entry();     break;
1635 
1636     case Interpreter::java_lang_math_sin     : // fall thru
1637     case Interpreter::java_lang_math_cos     : // fall thru
1638     case Interpreter::java_lang_math_tan     : // fall thru
1639     case Interpreter::java_lang_math_abs     : // fall thru
1640     case Interpreter::java_lang_math_log     : // fall thru
1641     case Interpreter::java_lang_math_log10   : // fall thru
1642     case Interpreter::java_lang_math_sqrt    : // fall thru
1643     case Interpreter::java_lang_math_pow     : // fall thru
1644     case Interpreter::java_lang_math_exp     : entry_point = ig_this->generate_math_entry(kind);      break;
1645     case Interpreter::java_lang_ref_reference_get
1646                                              : entry_point = ig_this->generate_Reference_get_entry(); break;
1647     case Interpreter::java_util_zip_CRC32_update
1648                                              : entry_point = ig_this->generate_CRC32_update_entry();  break;
1649     case Interpreter::java_util_zip_CRC32_updateBytes
1650                                              : // fall thru
1651     case Interpreter::java_util_zip_CRC32_updateByteBuffer
1652                                              : entry_point = ig_this->generate_CRC32_updateBytes_entry(kind); break;
1653     default:
1654       fatal(err_msg("unexpected method kind: %d", kind));
1655       break;
1656   }
1657 
1658   if (entry_point) return entry_point;
1659 
1660   return ig_this->generate_normal_entry(synchronized);
1661 
1662 }
1663 
1664 // These should never be compiled since the interpreter will prefer
1665 // the compiled version to the intrinsic version.
1666 bool AbstractInterpreter::can_be_compiled(methodHandle m) {
1667   switch (method_kind(m)) {
1668     case Interpreter::java_lang_math_sin     : // fall thru
1669     case Interpreter::java_lang_math_cos     : // fall thru
1670     case Interpreter::java_lang_math_tan     : // fall thru
1671     case Interpreter::java_lang_math_abs     : // fall thru
1672     case Interpreter::java_lang_math_log     : // fall thru
1673     case Interpreter::java_lang_math_log10   : // fall thru
1674     case Interpreter::java_lang_math_sqrt    : // fall thru
1675     case Interpreter::java_lang_math_pow     : // fall thru
1676     case Interpreter::java_lang_math_exp     :
1677       return false;
1678     default:
1679       return true;
1680   }
1681 }
1682 
1683 // How much stack a method activation needs in words.
1684 int AbstractInterpreter::size_top_interpreter_activation(Method* method) {
1685 
1686   const int stub_code = 4;  // see generate_call_stub
1687   // Save space for one monitor to get into the interpreted method in case
1688   // the method is synchronized
1689   int monitor_size    = method->is_synchronized() ?
1690                                 1*frame::interpreter_frame_monitor_size() : 0;
1691 
1692   // total overhead size: entry_size + (saved rbp, thru expr stack bottom).
1693   // be sure to change this if you add/subtract anything to/from the overhead area
1694   const int overhead_size = -frame::interpreter_frame_initial_sp_offset;
1695 
1696   const int method_stack = (method->max_locals() + method->max_stack()) *
1697                            Interpreter::stackElementWords;
1698   return overhead_size + method_stack + stub_code;
1699 }
1700 
1701 // asm based interpreter deoptimization helpers
1702 
1703 int AbstractInterpreter::layout_activation(Method* method,
1704                                            int tempcount,
1705                                            int popframe_extra_args,
1706                                            int moncount,
1707                                            int caller_actual_parameters,
1708                                            int callee_param_count,
1709                                            int callee_locals,
1710                                            frame* caller,
1711                                            frame* interpreter_frame,
1712                                            bool is_top_frame,
1713                                            bool is_bottom_frame) {
1714   // Note: This calculation must exactly parallel the frame setup
1715   // in AbstractInterpreterGenerator::generate_method_entry.
1716   // If interpreter_frame!=NULL, set up the method, locals, and monitors.
1717   // The frame interpreter_frame, if not NULL, is guaranteed to be the right size,
1718   // as determined by a previous call to this method.
1719   // It is also guaranteed to be walkable even though it is in a skeletal state
1720   // NOTE: return size is in words not bytes
1721 
1722   // fixed size of an interpreter frame:
1723   int max_locals = method->max_locals() * Interpreter::stackElementWords;
1724   int extra_locals = (method->max_locals() - method->size_of_parameters()) *
1725                      Interpreter::stackElementWords;
1726 
1727   int overhead = frame::sender_sp_offset - frame::interpreter_frame_initial_sp_offset;
1728 
1729   // Our locals were accounted for by the caller (or last_frame_adjust on the transistion)
1730   // Since the callee parameters already account for the callee's params we only need to account for
1731   // the extra locals.
1732 
1733 
1734   int size = overhead +
1735          ((callee_locals - callee_param_count)*Interpreter::stackElementWords) +
1736          (moncount*frame::interpreter_frame_monitor_size()) +
1737          tempcount*Interpreter::stackElementWords + popframe_extra_args;
1738 
1739   if (interpreter_frame != NULL) {
1740 #ifdef ASSERT
1741     if (!EnableInvokeDynamic)
1742       // @@@ FIXME: Should we correct interpreter_frame_sender_sp in the calling sequences?
1743       // Probably, since deoptimization doesn't work yet.
1744       assert(caller->unextended_sp() == interpreter_frame->interpreter_frame_sender_sp(), "Frame not properly walkable");
1745     assert(caller->sp() == interpreter_frame->sender_sp(), "Frame not properly walkable(2)");
1746 #endif
1747 
1748     interpreter_frame->interpreter_frame_set_method(method);
1749     // NOTE the difference in using sender_sp and interpreter_frame_sender_sp
1750     // interpreter_frame_sender_sp is the original sp of the caller (the unextended_sp)
1751     // and sender_sp is fp+8
1752     intptr_t* locals = interpreter_frame->sender_sp() + max_locals - 1;
1753 
1754 #ifdef ASSERT
1755     if (caller->is_interpreted_frame()) {
1756       assert(locals < caller->fp() + frame::interpreter_frame_initial_sp_offset, "bad placement");
1757     }
1758 #endif
1759 
1760     interpreter_frame->interpreter_frame_set_locals(locals);
1761     BasicObjectLock* montop = interpreter_frame->interpreter_frame_monitor_begin();
1762     BasicObjectLock* monbot = montop - moncount;
1763     interpreter_frame->interpreter_frame_set_monitor_end(monbot);
1764 
1765     // Set last_sp
1766     intptr_t*  rsp = (intptr_t*) monbot  -
1767                      tempcount*Interpreter::stackElementWords -
1768                      popframe_extra_args;
1769     interpreter_frame->interpreter_frame_set_last_sp(rsp);
1770 
1771     // All frames but the initial (oldest) interpreter frame we fill in have a
1772     // value for sender_sp that allows walking the stack but isn't
1773     // truly correct. Correct the value here.
1774 
1775     if (extra_locals != 0 &&
1776         interpreter_frame->sender_sp() == interpreter_frame->interpreter_frame_sender_sp() ) {
1777       interpreter_frame->set_interpreter_frame_sender_sp(caller->sp() + extra_locals);
1778     }
1779     *interpreter_frame->interpreter_frame_cache_addr() =
1780       method->constants()->cache();
1781   }
1782   return size;
1783 }
1784 
1785 
1786 //------------------------------------------------------------------------------------------------------------------------
1787 // Exceptions
1788 
1789 void TemplateInterpreterGenerator::generate_throw_exception() {
1790   // Entry point in previous activation (i.e., if the caller was interpreted)
1791   Interpreter::_rethrow_exception_entry = __ pc();
1792   const Register thread = rcx;
1793 
1794   // Restore sp to interpreter_frame_last_sp even though we are going
1795   // to empty the expression stack for the exception processing.
1796   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
1797   // rax,: exception
1798   // rdx: return address/pc that threw exception
1799   __ restore_bcp();                              // rsi points to call/send
1800   __ restore_locals();
1801 
1802   // Entry point for exceptions thrown within interpreter code
1803   Interpreter::_throw_exception_entry = __ pc();
1804   // expression stack is undefined here
1805   // rax,: exception
1806   // rsi: exception bcp
1807   __ verify_oop(rax);
1808 
1809   // expression stack must be empty before entering the VM in case of an exception
1810   __ empty_expression_stack();
1811   __ empty_FPU_stack();
1812   // find exception handler address and preserve exception oop
1813   __ call_VM(rdx, CAST_FROM_FN_PTR(address, InterpreterRuntime::exception_handler_for_exception), rax);
1814   // rax,: exception handler entry point
1815   // rdx: preserved exception oop
1816   // rsi: bcp for exception handler
1817   __ push_ptr(rdx);                              // push exception which is now the only value on the stack
1818   __ jmp(rax);                                   // jump to exception handler (may be _remove_activation_entry!)
1819 
1820   // If the exception is not handled in the current frame the frame is removed and
1821   // the exception is rethrown (i.e. exception continuation is _rethrow_exception).
1822   //
1823   // Note: At this point the bci is still the bxi for the instruction which caused
1824   //       the exception and the expression stack is empty. Thus, for any VM calls
1825   //       at this point, GC will find a legal oop map (with empty expression stack).
1826 
1827   // In current activation
1828   // tos: exception
1829   // rsi: exception bcp
1830 
1831   //
1832   // JVMTI PopFrame support
1833   //
1834 
1835    Interpreter::_remove_activation_preserving_args_entry = __ pc();
1836   __ empty_expression_stack();
1837   __ empty_FPU_stack();
1838   // Set the popframe_processing bit in pending_popframe_condition indicating that we are
1839   // currently handling popframe, so that call_VMs that may happen later do not trigger new
1840   // popframe handling cycles.
1841   __ get_thread(thread);
1842   __ movl(rdx, Address(thread, JavaThread::popframe_condition_offset()));
1843   __ orl(rdx, JavaThread::popframe_processing_bit);
1844   __ movl(Address(thread, JavaThread::popframe_condition_offset()), rdx);
1845 
1846   {
1847     // Check to see whether we are returning to a deoptimized frame.
1848     // (The PopFrame call ensures that the caller of the popped frame is
1849     // either interpreted or compiled and deoptimizes it if compiled.)
1850     // In this case, we can't call dispatch_next() after the frame is
1851     // popped, but instead must save the incoming arguments and restore
1852     // them after deoptimization has occurred.
1853     //
1854     // Note that we don't compare the return PC against the
1855     // deoptimization blob's unpack entry because of the presence of
1856     // adapter frames in C2.
1857     Label caller_not_deoptimized;
1858     __ movptr(rdx, Address(rbp, frame::return_addr_offset * wordSize));
1859     __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::interpreter_contains), rdx);
1860     __ testl(rax, rax);
1861     __ jcc(Assembler::notZero, caller_not_deoptimized);
1862 
1863     // Compute size of arguments for saving when returning to deoptimized caller
1864     __ get_method(rax);
1865     __ movptr(rax, Address(rax, Method::const_offset()));
1866     __ load_unsigned_short(rax, Address(rax, ConstMethod::size_of_parameters_offset()));
1867     __ shlptr(rax, Interpreter::logStackElementSize);
1868     __ restore_locals();
1869     __ subptr(rdi, rax);
1870     __ addptr(rdi, wordSize);
1871     // Save these arguments
1872     __ get_thread(thread);
1873     __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, Deoptimization::popframe_preserve_args), thread, rax, rdi);
1874 
1875     __ remove_activation(vtos, rdx,
1876                          /* throw_monitor_exception */ false,
1877                          /* install_monitor_exception */ false,
1878                          /* notify_jvmdi */ false);
1879 
1880     // Inform deoptimization that it is responsible for restoring these arguments
1881     __ get_thread(thread);
1882     __ movl(Address(thread, JavaThread::popframe_condition_offset()), JavaThread::popframe_force_deopt_reexecution_bit);
1883 
1884     // Continue in deoptimization handler
1885     __ jmp(rdx);
1886 
1887     __ bind(caller_not_deoptimized);
1888   }
1889 
1890   __ remove_activation(vtos, rdx,
1891                        /* throw_monitor_exception */ false,
1892                        /* install_monitor_exception */ false,
1893                        /* notify_jvmdi */ false);
1894 
1895   // Finish with popframe handling
1896   // A previous I2C followed by a deoptimization might have moved the
1897   // outgoing arguments further up the stack. PopFrame expects the
1898   // mutations to those outgoing arguments to be preserved and other
1899   // constraints basically require this frame to look exactly as
1900   // though it had previously invoked an interpreted activation with
1901   // no space between the top of the expression stack (current
1902   // last_sp) and the top of stack. Rather than force deopt to
1903   // maintain this kind of invariant all the time we call a small
1904   // fixup routine to move the mutated arguments onto the top of our
1905   // expression stack if necessary.
1906   __ mov(rax, rsp);
1907   __ movptr(rbx, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1908   __ get_thread(thread);
1909   // PC must point into interpreter here
1910   __ set_last_Java_frame(thread, noreg, rbp, __ pc());
1911   __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), thread, rax, rbx);
1912   __ get_thread(thread);
1913   __ reset_last_Java_frame(thread, true, true);
1914   // Restore the last_sp and null it out
1915   __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1916   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
1917 
1918   __ restore_bcp();
1919   __ restore_locals();
1920   // The method data pointer was incremented already during
1921   // call profiling. We have to restore the mdp for the current bcp.
1922   if (ProfileInterpreter) {
1923     __ set_method_data_pointer_for_bcp();
1924   }
1925 
1926   // Clear the popframe condition flag
1927   __ get_thread(thread);
1928   __ movl(Address(thread, JavaThread::popframe_condition_offset()), JavaThread::popframe_inactive);
1929 
1930 #if INCLUDE_JVMTI
1931   if (EnableInvokeDynamic) {
1932     Label L_done;
1933     const Register local0 = rdi;
1934 
1935     __ cmpb(Address(rsi, 0), Bytecodes::_invokestatic);
1936     __ jcc(Assembler::notEqual, L_done);
1937 
1938     // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call.
1939     // Detect such a case in the InterpreterRuntime function and return the member name argument, or NULL.
1940 
1941     __ get_method(rdx);
1942     __ movptr(rax, Address(local0, 0));
1943     __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), rax, rdx, rsi);
1944 
1945     __ testptr(rax, rax);
1946     __ jcc(Assembler::zero, L_done);
1947 
1948     __ movptr(Address(rbx, 0), rax);
1949     __ bind(L_done);
1950   }
1951 #endif // INCLUDE_JVMTI
1952 
1953   __ dispatch_next(vtos);
1954   // end of PopFrame support
1955 
1956   Interpreter::_remove_activation_entry = __ pc();
1957 
1958   // preserve exception over this code sequence
1959   __ pop_ptr(rax);
1960   __ get_thread(thread);
1961   __ movptr(Address(thread, JavaThread::vm_result_offset()), rax);
1962   // remove the activation (without doing throws on illegalMonitorExceptions)
1963   __ remove_activation(vtos, rdx, false, true, false);
1964   // restore exception
1965   __ get_thread(thread);
1966   __ get_vm_result(rax, thread);
1967 
1968   // Inbetween activations - previous activation type unknown yet
1969   // compute continuation point - the continuation point expects
1970   // the following registers set up:
1971   //
1972   // rax: exception
1973   // rdx: return address/pc that threw exception
1974   // rsp: expression stack of caller
1975   // rbp: rbp, of caller
1976   __ push(rax);                                  // save exception
1977   __ push(rdx);                                  // save return address
1978   __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), thread, rdx);
1979   __ mov(rbx, rax);                              // save exception handler
1980   __ pop(rdx);                                   // restore return address
1981   __ pop(rax);                                   // restore exception
1982   // Note that an "issuing PC" is actually the next PC after the call
1983   __ jmp(rbx);                                   // jump to exception handler of caller
1984 }
1985 
1986 
1987 //
1988 // JVMTI ForceEarlyReturn support
1989 //
1990 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
1991   address entry = __ pc();
1992   const Register thread = rcx;
1993 
1994   __ restore_bcp();
1995   __ restore_locals();
1996   __ empty_expression_stack();
1997   __ empty_FPU_stack();
1998   __ load_earlyret_value(state);
1999 
2000   __ get_thread(thread);
2001   __ movptr(rcx, Address(thread, JavaThread::jvmti_thread_state_offset()));
2002   const Address cond_addr(rcx, JvmtiThreadState::earlyret_state_offset());
2003 
2004   // Clear the earlyret state
2005   __ movl(cond_addr, JvmtiThreadState::earlyret_inactive);
2006 
2007   __ remove_activation(state, rsi,
2008                        false, /* throw_monitor_exception */
2009                        false, /* install_monitor_exception */
2010                        true); /* notify_jvmdi */
2011   __ jmp(rsi);
2012   return entry;
2013 } // end of ForceEarlyReturn support
2014 
2015 
2016 //------------------------------------------------------------------------------------------------------------------------
2017 // Helper for vtos entry point generation
2018 
2019 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) {
2020   assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
2021   Label L;
2022   fep = __ pc(); __ push(ftos); __ jmp(L);
2023   dep = __ pc(); __ push(dtos); __ jmp(L);
2024   lep = __ pc(); __ push(ltos); __ jmp(L);
2025   aep = __ pc(); __ push(atos); __ jmp(L);
2026   bep = cep = sep =             // fall through
2027   iep = __ pc(); __ push(itos); // fall through
2028   vep = __ pc(); __ bind(L);    // fall through
2029   generate_and_dispatch(t);
2030 }
2031 
2032 //------------------------------------------------------------------------------------------------------------------------
2033 // Generation of individual instructions
2034 
2035 // helpers for generate_and_dispatch
2036 
2037 
2038 
2039 InterpreterGenerator::InterpreterGenerator(StubQueue* code)
2040  : TemplateInterpreterGenerator(code) {
2041    generate_all(); // down here so it can be "virtual"
2042 }
2043 
2044 //------------------------------------------------------------------------------------------------------------------------
2045 
2046 // Non-product code
2047 #ifndef PRODUCT
2048 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
2049   address entry = __ pc();
2050 
2051   // prepare expression stack
2052   __ pop(rcx);          // pop return address so expression stack is 'pure'
2053   __ push(state);       // save tosca
2054 
2055   // pass tosca registers as arguments & call tracer
2056   __ call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::trace_bytecode), rcx, rax, rdx);
2057   __ mov(rcx, rax);     // make sure return address is not destroyed by pop(state)
2058   __ pop(state);        // restore tosca
2059 
2060   // return
2061   __ jmp(rcx);
2062 
2063   return entry;
2064 }
2065 
2066 
2067 void TemplateInterpreterGenerator::count_bytecode() {
2068   __ incrementl(ExternalAddress((address) &BytecodeCounter::_counter_value));
2069 }
2070 
2071 
2072 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
2073   __ incrementl(ExternalAddress((address) &BytecodeHistogram::_counters[t->bytecode()]));
2074 }
2075 
2076 
2077 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
2078   __ mov32(ExternalAddress((address) &BytecodePairHistogram::_index), rbx);
2079   __ shrl(rbx, BytecodePairHistogram::log2_number_of_codes);
2080   __ orl(rbx, ((int)t->bytecode()) << BytecodePairHistogram::log2_number_of_codes);
2081   ExternalAddress table((address) BytecodePairHistogram::_counters);
2082   Address index(noreg, rbx, Address::times_4);
2083   __ incrementl(ArrayAddress(table, index));
2084 }
2085 
2086 
2087 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
2088   // Call a little run-time stub to avoid blow-up for each bytecode.
2089   // The run-time runtime saves the right registers, depending on
2090   // the tosca in-state for the given template.
2091   assert(Interpreter::trace_code(t->tos_in()) != NULL,
2092          "entry must have been generated");
2093   __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in())));
2094 }
2095 
2096 
2097 void TemplateInterpreterGenerator::stop_interpreter_at() {
2098   Label L;
2099   __ cmp32(ExternalAddress((address) &BytecodeCounter::_counter_value),
2100            StopInterpreterAt);
2101   __ jcc(Assembler::notEqual, L);
2102   __ int3();
2103   __ bind(L);
2104 }
2105 #endif // !PRODUCT
2106 #endif // CC_INTERP
--- EOF ---