1 /*
   2  * Copyright (c) 2003, 2011, 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 "interp_masm_x86_64.hpp"
  27 #include "interpreter/interpreter.hpp"
  28 #include "interpreter/interpreterRuntime.hpp"
  29 #include "oops/arrayOop.hpp"
  30 #include "oops/markOop.hpp"
  31 #include "oops/methodDataOop.hpp"
  32 #include "oops/methodOop.hpp"
  33 #include "prims/jvmtiExport.hpp"
  34 #include "prims/jvmtiRedefineClassesTrace.hpp"
  35 #include "prims/jvmtiThreadState.hpp"
  36 #include "runtime/basicLock.hpp"
  37 #include "runtime/biasedLocking.hpp"
  38 #include "runtime/sharedRuntime.hpp"
  39 #ifdef TARGET_OS_FAMILY_linux
  40 # include "thread_linux.inline.hpp"
  41 #endif
  42 #ifdef TARGET_OS_FAMILY_solaris
  43 # include "thread_solaris.inline.hpp"
  44 #endif
  45 #ifdef TARGET_OS_FAMILY_windows
  46 # include "thread_windows.inline.hpp"
  47 #endif
  48 
  49 
  50 // Implementation of InterpreterMacroAssembler
  51 
  52 #ifdef CC_INTERP
  53 void InterpreterMacroAssembler::get_method(Register reg) {
  54   movptr(reg, Address(rbp, -((int)sizeof(BytecodeInterpreter) + 2 * wordSize)));
  55   movptr(reg, Address(reg, byte_offset_of(BytecodeInterpreter, _method)));
  56 }
  57 #endif // CC_INTERP
  58 
  59 #ifndef CC_INTERP
  60 
  61 void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point,
  62                                                   int number_of_arguments) {
  63   // interpreter specific
  64   //
  65   // Note: No need to save/restore bcp & locals (r13 & r14) pointer
  66   //       since these are callee saved registers and no blocking/
  67   //       GC can happen in leaf calls.
  68   // Further Note: DO NOT save/restore bcp/locals. If a caller has
  69   // already saved them so that it can use esi/edi as temporaries
  70   // then a save/restore here will DESTROY the copy the caller
  71   // saved! There used to be a save_bcp() that only happened in
  72   // the ASSERT path (no restore_bcp). Which caused bizarre failures
  73   // when jvm built with ASSERTs.
  74 #ifdef ASSERT
  75   {
  76     Label L;
  77     cmpptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
  78     jcc(Assembler::equal, L);
  79     stop("InterpreterMacroAssembler::call_VM_leaf_base:"
  80          " last_sp != NULL");
  81     bind(L);
  82   }
  83 #endif
  84   // super call
  85   MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments);
  86   // interpreter specific
  87   // Used to ASSERT that r13/r14 were equal to frame's bcp/locals
  88   // but since they may not have been saved (and we don't want to
  89   // save thme here (see note above) the assert is invalid.
  90 }
  91 
  92 void InterpreterMacroAssembler::call_VM_base(Register oop_result,
  93                                              Register java_thread,
  94                                              Register last_java_sp,
  95                                              address  entry_point,
  96                                              int      number_of_arguments,
  97                                              bool     check_exceptions) {
  98   // interpreter specific
  99   //
 100   // Note: Could avoid restoring locals ptr (callee saved) - however doesn't
 101   //       really make a difference for these runtime calls, since they are
 102   //       slow anyway. Btw., bcp must be saved/restored since it may change
 103   //       due to GC.
 104   // assert(java_thread == noreg , "not expecting a precomputed java thread");
 105   save_bcp();
 106 #ifdef ASSERT
 107   {
 108     Label L;
 109     cmpptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
 110     jcc(Assembler::equal, L);
 111     stop("InterpreterMacroAssembler::call_VM_leaf_base:"
 112          " last_sp != NULL");
 113     bind(L);
 114   }
 115 #endif /* ASSERT */
 116   // super call
 117   MacroAssembler::call_VM_base(oop_result, noreg, last_java_sp,
 118                                entry_point, number_of_arguments,
 119                                check_exceptions);
 120   // interpreter specific
 121   restore_bcp();
 122   restore_locals();
 123 }
 124 
 125 
 126 void InterpreterMacroAssembler::check_and_handle_popframe(Register java_thread) {
 127   if (JvmtiExport::can_pop_frame()) {
 128     Label L;
 129     // Initiate popframe handling only if it is not already being
 130     // processed.  If the flag has the popframe_processing bit set, it
 131     // means that this code is called *during* popframe handling - we
 132     // don't want to reenter.
 133     // This method is only called just after the call into the vm in
 134     // call_VM_base, so the arg registers are available.
 135     movl(c_rarg0, Address(r15_thread, JavaThread::popframe_condition_offset()));
 136     testl(c_rarg0, JavaThread::popframe_pending_bit);
 137     jcc(Assembler::zero, L);
 138     testl(c_rarg0, JavaThread::popframe_processing_bit);
 139     jcc(Assembler::notZero, L);
 140     // Call Interpreter::remove_activation_preserving_args_entry() to get the
 141     // address of the same-named entrypoint in the generated interpreter code.
 142     call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry));
 143     jmp(rax);
 144     bind(L);
 145   }
 146 }
 147 
 148 
 149 void InterpreterMacroAssembler::load_earlyret_value(TosState state) {
 150   movptr(rcx, Address(r15_thread, JavaThread::jvmti_thread_state_offset()));
 151   const Address tos_addr(rcx, JvmtiThreadState::earlyret_tos_offset());
 152   const Address oop_addr(rcx, JvmtiThreadState::earlyret_oop_offset());
 153   const Address val_addr(rcx, JvmtiThreadState::earlyret_value_offset());
 154   switch (state) {
 155     case atos: movptr(rax, oop_addr);
 156                movptr(oop_addr, (int32_t)NULL_WORD);
 157                verify_oop(rax, state);              break;
 158     case ltos: movptr(rax, val_addr);                 break;
 159     case btos:                                   // fall through
 160     case ctos:                                   // fall through
 161     case stos:                                   // fall through
 162     case itos: movl(rax, val_addr);                 break;
 163     case ftos: movflt(xmm0, val_addr);              break;
 164     case dtos: movdbl(xmm0, val_addr);              break;
 165     case vtos: /* nothing to do */                  break;
 166     default  : ShouldNotReachHere();
 167   }
 168   // Clean up tos value in the thread object
 169   movl(tos_addr,  (int) ilgl);
 170   movl(val_addr,  (int32_t) NULL_WORD);
 171 }
 172 
 173 
 174 void InterpreterMacroAssembler::check_and_handle_earlyret(Register java_thread) {
 175   if (JvmtiExport::can_force_early_return()) {
 176     Label L;
 177     movptr(c_rarg0, Address(r15_thread, JavaThread::jvmti_thread_state_offset()));
 178     testptr(c_rarg0, c_rarg0);
 179     jcc(Assembler::zero, L); // if (thread->jvmti_thread_state() == NULL) exit;
 180 
 181     // Initiate earlyret handling only if it is not already being processed.
 182     // If the flag has the earlyret_processing bit set, it means that this code
 183     // is called *during* earlyret handling - we don't want to reenter.
 184     movl(c_rarg0, Address(c_rarg0, JvmtiThreadState::earlyret_state_offset()));
 185     cmpl(c_rarg0, JvmtiThreadState::earlyret_pending);
 186     jcc(Assembler::notEqual, L);
 187 
 188     // Call Interpreter::remove_activation_early_entry() to get the address of the
 189     // same-named entrypoint in the generated interpreter code.
 190     movptr(c_rarg0, Address(r15_thread, JavaThread::jvmti_thread_state_offset()));
 191     movl(c_rarg0, Address(c_rarg0, JvmtiThreadState::earlyret_tos_offset()));
 192     call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), c_rarg0);
 193     jmp(rax);
 194     bind(L);
 195   }
 196 }
 197 
 198 
 199 void InterpreterMacroAssembler::get_unsigned_2_byte_index_at_bcp(
 200   Register reg,
 201   int bcp_offset) {
 202   assert(bcp_offset >= 0, "bcp is still pointing to start of bytecode");
 203   movl(reg, Address(r13, bcp_offset));
 204   bswapl(reg);
 205   shrl(reg, 16);
 206 }
 207 
 208 
 209 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register index,
 210                                                        int bcp_offset,
 211                                                        size_t index_size) {
 212   assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
 213   if (index_size == sizeof(u2)) {
 214     load_unsigned_short(index, Address(r13, bcp_offset));
 215   } else if (index_size == sizeof(u4)) {
 216     assert(EnableInvokeDynamic, "giant index used only for JSR 292");
 217     movl(index, Address(r13, bcp_offset));
 218     // Check if the secondary index definition is still ~x, otherwise
 219     // we have to change the following assembler code to calculate the
 220     // plain index.
 221     assert(constantPoolCacheOopDesc::decode_secondary_index(~123) == 123, "else change next line");
 222     notl(index);  // convert to plain index
 223   } else if (index_size == sizeof(u1)) {
 224     assert(EnableInvokeDynamic, "tiny index used only for JSR 292");
 225     load_unsigned_byte(index, Address(r13, bcp_offset));
 226   } else {
 227     ShouldNotReachHere();
 228   }
 229 }
 230 
 231 
 232 void InterpreterMacroAssembler::get_cache_and_index_at_bcp(Register cache,
 233                                                            Register index,
 234                                                            int bcp_offset,
 235                                                            size_t index_size) {
 236   assert(cache != index, "must use different registers");
 237   get_cache_index_at_bcp(index, bcp_offset, index_size);
 238   movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
 239   assert(sizeof(ConstantPoolCacheEntry) == 4 * wordSize, "adjust code below");
 240   // convert from field index to ConstantPoolCacheEntry index
 241   shll(index, 2);
 242 }
 243 
 244 
 245 void InterpreterMacroAssembler::get_cache_entry_pointer_at_bcp(Register cache,
 246                                                                Register tmp,
 247                                                                int bcp_offset,
 248                                                                size_t index_size) {
 249   assert(cache != tmp, "must use different register");
 250   get_cache_index_at_bcp(tmp, bcp_offset, index_size);
 251   assert(sizeof(ConstantPoolCacheEntry) == 4 * wordSize, "adjust code below");
 252   // convert from field index to ConstantPoolCacheEntry index
 253   // and from word offset to byte offset
 254   shll(tmp, 2 + LogBytesPerWord);
 255   movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
 256   // skip past the header
 257   addptr(cache, in_bytes(constantPoolCacheOopDesc::base_offset()));
 258   addptr(cache, tmp);  // construct pointer to cache entry
 259 }
 260 
 261 
 262 // Generate a subtype check: branch to ok_is_subtype if sub_klass is a
 263 // subtype of super_klass.
 264 //
 265 // Args:
 266 //      rax: superklass
 267 //      Rsub_klass: subklass
 268 //
 269 // Kills:
 270 //      rcx, rdi
 271 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass,
 272                                                   Label& ok_is_subtype) {
 273   assert(Rsub_klass != rax, "rax holds superklass");
 274   assert(Rsub_klass != r14, "r14 holds locals");
 275   assert(Rsub_klass != r13, "r13 holds bcp");
 276   assert(Rsub_klass != rcx, "rcx holds 2ndary super array length");
 277   assert(Rsub_klass != rdi, "rdi holds 2ndary super array scan ptr");
 278 
 279   // Profile the not-null value's klass.
 280   profile_typecheck(rcx, Rsub_klass, rdi); // blows rcx, reloads rdi
 281 
 282   // Do the check.
 283   check_klass_subtype(Rsub_klass, rax, rcx, ok_is_subtype); // blows rcx
 284 
 285   // Profile the failure of the check.
 286   profile_typecheck_failed(rcx); // blows rcx
 287 }
 288 
 289 
 290 
 291 // Java Expression Stack
 292 
 293 void InterpreterMacroAssembler::pop_ptr(Register r) {
 294   pop(r);
 295 }
 296 
 297 void InterpreterMacroAssembler::pop_i(Register r) {
 298   // XXX can't use pop currently, upper half non clean
 299   movl(r, Address(rsp, 0));
 300   addptr(rsp, wordSize);
 301 }
 302 
 303 void InterpreterMacroAssembler::pop_l(Register r) {
 304   movq(r, Address(rsp, 0));
 305   addptr(rsp, 2 * Interpreter::stackElementSize);
 306 }
 307 
 308 void InterpreterMacroAssembler::pop_f(XMMRegister r) {
 309   movflt(r, Address(rsp, 0));
 310   addptr(rsp, wordSize);
 311 }
 312 
 313 void InterpreterMacroAssembler::pop_d(XMMRegister r) {
 314   movdbl(r, Address(rsp, 0));
 315   addptr(rsp, 2 * Interpreter::stackElementSize);
 316 }
 317 
 318 void InterpreterMacroAssembler::push_ptr(Register r) {
 319   push(r);
 320 }
 321 
 322 void InterpreterMacroAssembler::push_i(Register r) {
 323   push(r);
 324 }
 325 
 326 void InterpreterMacroAssembler::push_l(Register r) {
 327   subptr(rsp, 2 * wordSize);
 328   movq(Address(rsp, 0), r);
 329 }
 330 
 331 void InterpreterMacroAssembler::push_f(XMMRegister r) {
 332   subptr(rsp, wordSize);
 333   movflt(Address(rsp, 0), r);
 334 }
 335 
 336 void InterpreterMacroAssembler::push_d(XMMRegister r) {
 337   subptr(rsp, 2 * wordSize);
 338   movdbl(Address(rsp, 0), r);
 339 }
 340 
 341 void InterpreterMacroAssembler::pop(TosState state) {
 342   switch (state) {
 343   case atos: pop_ptr();                 break;
 344   case btos:
 345   case ctos:
 346   case stos:
 347   case itos: pop_i();                   break;
 348   case ltos: pop_l();                   break;
 349   case ftos: pop_f();                   break;
 350   case dtos: pop_d();                   break;
 351   case vtos: /* nothing to do */        break;
 352   default:   ShouldNotReachHere();
 353   }
 354   verify_oop(rax, state);
 355 }
 356 
 357 void InterpreterMacroAssembler::push(TosState state) {
 358   verify_oop(rax, state);
 359   switch (state) {
 360   case atos: push_ptr();                break;
 361   case btos:
 362   case ctos:
 363   case stos:
 364   case itos: push_i();                  break;
 365   case ltos: push_l();                  break;
 366   case ftos: push_f();                  break;
 367   case dtos: push_d();                  break;
 368   case vtos: /* nothing to do */        break;
 369   default  : ShouldNotReachHere();
 370   }
 371 }
 372 
 373 
 374 // Helpers for swap and dup
 375 void InterpreterMacroAssembler::load_ptr(int n, Register val) {
 376   movptr(val, Address(rsp, Interpreter::expr_offset_in_bytes(n)));
 377 }
 378 
 379 void InterpreterMacroAssembler::store_ptr(int n, Register val) {
 380   movptr(Address(rsp, Interpreter::expr_offset_in_bytes(n)), val);
 381 }
 382 
 383 
 384 void InterpreterMacroAssembler::prepare_to_jump_from_interpreted() {
 385   // set sender sp
 386   lea(r13, Address(rsp, wordSize));
 387   // record last_sp
 388   movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), r13);
 389 }
 390 
 391 
 392 // Jump to from_interpreted entry of a call unless single stepping is possible
 393 // in this thread in which case we must call the i2i entry
 394 void InterpreterMacroAssembler::jump_from_interpreted(Register method, Register temp) {
 395   prepare_to_jump_from_interpreted();
 396 
 397   if (JvmtiExport::can_post_interpreter_events()) {
 398     Label run_compiled_code;
 399     // JVMTI events, such as single-stepping, are implemented partly by avoiding running
 400     // compiled code in threads for which the event is enabled.  Check here for
 401     // interp_only_mode if these events CAN be enabled.
 402     // interp_only is an int, on little endian it is sufficient to test the byte only
 403     // Is a cmpl faster?
 404     cmpb(Address(r15_thread, JavaThread::interp_only_mode_offset()), 0);
 405     jccb(Assembler::zero, run_compiled_code);
 406     jmp(Address(method, methodOopDesc::interpreter_entry_offset()));
 407     bind(run_compiled_code);
 408   }
 409 
 410   jmp(Address(method, methodOopDesc::from_interpreted_offset()));
 411 
 412 }
 413 
 414 
 415 // The following two routines provide a hook so that an implementation
 416 // can schedule the dispatch in two parts.  amd64 does not do this.
 417 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int step) {
 418   // Nothing amd64 specific to be done here
 419 }
 420 
 421 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int step) {
 422   dispatch_next(state, step);
 423 }
 424 
 425 void InterpreterMacroAssembler::dispatch_base(TosState state,
 426                                               address* table,
 427                                               bool verifyoop) {
 428   verify_FPU(1, state);
 429   if (VerifyActivationFrameSize) {
 430     Label L;
 431     mov(rcx, rbp);
 432     subptr(rcx, rsp);
 433     int32_t min_frame_size =
 434       (frame::link_offset - frame::interpreter_frame_initial_sp_offset) *
 435       wordSize;
 436     cmpptr(rcx, (int32_t)min_frame_size);
 437     jcc(Assembler::greaterEqual, L);
 438     stop("broken stack frame");
 439     bind(L);
 440   }
 441   if (verifyoop) {
 442     verify_oop(rax, state);
 443   }
 444   lea(rscratch1, ExternalAddress((address)table));
 445   jmp(Address(rscratch1, rbx, Address::times_8));
 446 }
 447 
 448 void InterpreterMacroAssembler::dispatch_only(TosState state) {
 449   dispatch_base(state, Interpreter::dispatch_table(state));
 450 }
 451 
 452 void InterpreterMacroAssembler::dispatch_only_normal(TosState state) {
 453   dispatch_base(state, Interpreter::normal_table(state));
 454 }
 455 
 456 void InterpreterMacroAssembler::dispatch_only_noverify(TosState state) {
 457   dispatch_base(state, Interpreter::normal_table(state), false);
 458 }
 459 
 460 
 461 void InterpreterMacroAssembler::dispatch_next(TosState state, int step) {
 462   // load next bytecode (load before advancing r13 to prevent AGI)
 463   load_unsigned_byte(rbx, Address(r13, step));
 464   // advance r13
 465   increment(r13, step);
 466   dispatch_base(state, Interpreter::dispatch_table(state));
 467 }
 468 
 469 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) {
 470   // load current bytecode
 471   load_unsigned_byte(rbx, Address(r13, 0));
 472   dispatch_base(state, table);
 473 }
 474 
 475 // remove activation
 476 //
 477 // Unlock the receiver if this is a synchronized method.
 478 // Unlock any Java monitors from syncronized blocks.
 479 // Remove the activation from the stack.
 480 //
 481 // If there are locked Java monitors
 482 //    If throw_monitor_exception
 483 //       throws IllegalMonitorStateException
 484 //    Else if install_monitor_exception
 485 //       installs IllegalMonitorStateException
 486 //    Else
 487 //       no error processing
 488 void InterpreterMacroAssembler::remove_activation(
 489         TosState state,
 490         Register ret_addr,
 491         bool throw_monitor_exception,
 492         bool install_monitor_exception,
 493         bool notify_jvmdi) {
 494   // Note: Registers rdx xmm0 may be in use for the
 495   // result check if synchronized method
 496   Label unlocked, unlock, no_unlock;
 497 
 498   // get the value of _do_not_unlock_if_synchronized into rdx
 499   const Address do_not_unlock_if_synchronized(r15_thread,
 500     in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
 501   movbool(rdx, do_not_unlock_if_synchronized);
 502   movbool(do_not_unlock_if_synchronized, false); // reset the flag
 503 
 504  // get method access flags
 505   movptr(rbx, Address(rbp, frame::interpreter_frame_method_offset * wordSize));
 506   movl(rcx, Address(rbx, methodOopDesc::access_flags_offset()));
 507   testl(rcx, JVM_ACC_SYNCHRONIZED);
 508   jcc(Assembler::zero, unlocked);
 509 
 510   // Don't unlock anything if the _do_not_unlock_if_synchronized flag
 511   // is set.
 512   testbool(rdx);
 513   jcc(Assembler::notZero, no_unlock);
 514 
 515   // unlock monitor
 516   push(state); // save result
 517 
 518   // BasicObjectLock will be first in list, since this is a
 519   // synchronized method. However, need to check that the object has
 520   // not been unlocked by an explicit monitorexit bytecode.
 521   const Address monitor(rbp, frame::interpreter_frame_initial_sp_offset *
 522                         wordSize - (int) sizeof(BasicObjectLock));
 523   // We use c_rarg1 so that if we go slow path it will be the correct
 524   // register for unlock_object to pass to VM directly
 525   lea(c_rarg1, monitor); // address of first monitor
 526 
 527   movptr(rax, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
 528   testptr(rax, rax);
 529   jcc(Assembler::notZero, unlock);
 530 
 531   pop(state);
 532   if (throw_monitor_exception) {
 533     // Entry already unlocked, need to throw exception
 534     call_VM(noreg, CAST_FROM_FN_PTR(address,
 535                    InterpreterRuntime::throw_illegal_monitor_state_exception));
 536     should_not_reach_here();
 537   } else {
 538     // Monitor already unlocked during a stack unroll. If requested,
 539     // install an illegal_monitor_state_exception.  Continue with
 540     // stack unrolling.
 541     if (install_monitor_exception) {
 542       call_VM(noreg, CAST_FROM_FN_PTR(address,
 543                      InterpreterRuntime::new_illegal_monitor_state_exception));
 544     }
 545     jmp(unlocked);
 546   }
 547 
 548   bind(unlock);
 549   unlock_object(c_rarg1);
 550   pop(state);
 551 
 552   // Check that for block-structured locking (i.e., that all locked
 553   // objects has been unlocked)
 554   bind(unlocked);
 555 
 556   // rax: Might contain return value
 557 
 558   // Check that all monitors are unlocked
 559   {
 560     Label loop, exception, entry, restart;
 561     const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
 562     const Address monitor_block_top(
 563         rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
 564     const Address monitor_block_bot(
 565         rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
 566 
 567     bind(restart);
 568     // We use c_rarg1 so that if we go slow path it will be the correct
 569     // register for unlock_object to pass to VM directly
 570     movptr(c_rarg1, monitor_block_top); // points to current entry, starting
 571                                   // with top-most entry
 572     lea(rbx, monitor_block_bot);  // points to word before bottom of
 573                                   // monitor block
 574     jmp(entry);
 575 
 576     // Entry already locked, need to throw exception
 577     bind(exception);
 578 
 579     if (throw_monitor_exception) {
 580       // Throw exception
 581       MacroAssembler::call_VM(noreg,
 582                               CAST_FROM_FN_PTR(address, InterpreterRuntime::
 583                                    throw_illegal_monitor_state_exception));
 584       should_not_reach_here();
 585     } else {
 586       // Stack unrolling. Unlock object and install illegal_monitor_exception.
 587       // Unlock does not block, so don't have to worry about the frame.
 588       // We don't have to preserve c_rarg1 since we are going to throw an exception.
 589 
 590       push(state);
 591       unlock_object(c_rarg1);
 592       pop(state);
 593 
 594       if (install_monitor_exception) {
 595         call_VM(noreg, CAST_FROM_FN_PTR(address,
 596                                         InterpreterRuntime::
 597                                         new_illegal_monitor_state_exception));
 598       }
 599 
 600       jmp(restart);
 601     }
 602 
 603     bind(loop);
 604     // check if current entry is used
 605     cmpptr(Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()), (int32_t) NULL);
 606     jcc(Assembler::notEqual, exception);
 607 
 608     addptr(c_rarg1, entry_size); // otherwise advance to next entry
 609     bind(entry);
 610     cmpptr(c_rarg1, rbx); // check if bottom reached
 611     jcc(Assembler::notEqual, loop); // if not at bottom then check this entry
 612   }
 613 
 614   bind(no_unlock);
 615 
 616   // jvmti support
 617   if (notify_jvmdi) {
 618     notify_method_exit(state, NotifyJVMTI);    // preserve TOSCA
 619   } else {
 620     notify_method_exit(state, SkipNotifyJVMTI); // preserve TOSCA
 621   }
 622 
 623   // remove activation
 624   // get sender sp
 625   movptr(rbx,
 626          Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize));
 627   leave();                           // remove frame anchor
 628   pop(ret_addr);                     // get return address
 629   mov(rsp, rbx);                     // set sp to sender sp
 630 }
 631 
 632 #endif // C_INTERP
 633 
 634 // Lock object
 635 //
 636 // Args:
 637 //      c_rarg1: BasicObjectLock to be used for locking
 638 //
 639 // Kills:
 640 //      rax
 641 //      c_rarg0, c_rarg1, c_rarg2, c_rarg3, .. (param regs)
 642 //      rscratch1, rscratch2 (scratch regs)
 643 void InterpreterMacroAssembler::lock_object(Register lock_reg) {
 644   assert(lock_reg == c_rarg1, "The argument is only for looks. It must be c_rarg1");
 645 
 646   if (UseHeavyMonitors) {
 647     call_VM(noreg,
 648             CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
 649             lock_reg);
 650   } else {
 651     Label done;
 652 
 653     const Register swap_reg = rax; // Must use rax for cmpxchg instruction
 654     const Register obj_reg = c_rarg3; // Will contain the oop
 655 
 656     const int obj_offset = BasicObjectLock::obj_offset_in_bytes();
 657     const int lock_offset = BasicObjectLock::lock_offset_in_bytes ();
 658     const int mark_offset = lock_offset +
 659                             BasicLock::displaced_header_offset_in_bytes();
 660 
 661     Label slow_case;
 662 
 663     // Load object pointer into obj_reg %c_rarg3
 664     movptr(obj_reg, Address(lock_reg, obj_offset));
 665 
 666     if (UseBiasedLocking) {
 667       biased_locking_enter(lock_reg, obj_reg, swap_reg, rscratch1, false, done, &slow_case);
 668     }
 669 
 670     // Load immediate 1 into swap_reg %rax
 671     movl(swap_reg, 1);
 672 
 673     // Load (object->mark() | 1) into swap_reg %rax
 674     orptr(swap_reg, Address(obj_reg, 0));
 675 
 676     // Save (object->mark() | 1) into BasicLock's displaced header
 677     movptr(Address(lock_reg, mark_offset), swap_reg);
 678 
 679     assert(lock_offset == 0,
 680            "displached header must be first word in BasicObjectLock");
 681 
 682     if (os::is_MP()) lock();
 683     cmpxchgptr(lock_reg, Address(obj_reg, 0));
 684     if (PrintBiasedLockingStatistics) {
 685       cond_inc32(Assembler::zero,
 686                  ExternalAddress((address) BiasedLocking::fast_path_entry_count_addr()));
 687     }
 688     jcc(Assembler::zero, done);
 689 
 690     // Test if the oopMark is an obvious stack pointer, i.e.,
 691     //  1) (mark & 7) == 0, and
 692     //  2) rsp <= mark < mark + os::pagesize()
 693     //
 694     // These 3 tests can be done by evaluating the following
 695     // expression: ((mark - rsp) & (7 - os::vm_page_size())),
 696     // assuming both stack pointer and pagesize have their
 697     // least significant 3 bits clear.
 698     // NOTE: the oopMark is in swap_reg %rax as the result of cmpxchg
 699     subptr(swap_reg, rsp);
 700     andptr(swap_reg, 7 - os::vm_page_size());
 701 
 702     // Save the test result, for recursive case, the result is zero
 703     movptr(Address(lock_reg, mark_offset), swap_reg);
 704 
 705     if (PrintBiasedLockingStatistics) {
 706       cond_inc32(Assembler::zero,
 707                  ExternalAddress((address) BiasedLocking::fast_path_entry_count_addr()));
 708     }
 709     jcc(Assembler::zero, done);
 710 
 711     bind(slow_case);
 712 
 713     // Call the runtime routine for slow case
 714     call_VM(noreg,
 715             CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
 716             lock_reg);
 717 
 718     bind(done);
 719   }
 720 }
 721 
 722 
 723 // Unlocks an object. Used in monitorexit bytecode and
 724 // remove_activation.  Throws an IllegalMonitorException if object is
 725 // not locked by current thread.
 726 //
 727 // Args:
 728 //      c_rarg1: BasicObjectLock for lock
 729 //
 730 // Kills:
 731 //      rax
 732 //      c_rarg0, c_rarg1, c_rarg2, c_rarg3, ... (param regs)
 733 //      rscratch1, rscratch2 (scratch regs)
 734 void InterpreterMacroAssembler::unlock_object(Register lock_reg) {
 735   assert(lock_reg == c_rarg1, "The argument is only for looks. It must be rarg1");
 736 
 737   if (UseHeavyMonitors) {
 738     call_VM(noreg,
 739             CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit),
 740             lock_reg);
 741   } else {
 742     Label done;
 743 
 744     const Register swap_reg   = rax;  // Must use rax for cmpxchg instruction
 745     const Register header_reg = c_rarg2;  // Will contain the old oopMark
 746     const Register obj_reg    = c_rarg3;  // Will contain the oop
 747 
 748     save_bcp(); // Save in case of exception
 749 
 750     // Convert from BasicObjectLock structure to object and BasicLock
 751     // structure Store the BasicLock address into %rax
 752     lea(swap_reg, Address(lock_reg, BasicObjectLock::lock_offset_in_bytes()));
 753 
 754     // Load oop into obj_reg(%c_rarg3)
 755     movptr(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()));
 756 
 757     // Free entry
 758     movptr(Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()), (int32_t)NULL_WORD);
 759 
 760     if (UseBiasedLocking) {
 761       biased_locking_exit(obj_reg, header_reg, done);
 762     }
 763 
 764     // Load the old header from BasicLock structure
 765     movptr(header_reg, Address(swap_reg,
 766                                BasicLock::displaced_header_offset_in_bytes()));
 767 
 768     // Test for recursion
 769     testptr(header_reg, header_reg);
 770 
 771     // zero for recursive case
 772     jcc(Assembler::zero, done);
 773 
 774     // Atomic swap back the old header
 775     if (os::is_MP()) lock();
 776     cmpxchgptr(header_reg, Address(obj_reg, 0));
 777 
 778     // zero for recursive case
 779     jcc(Assembler::zero, done);
 780 
 781     // Call the runtime routine for slow case.
 782     movptr(Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()),
 783          obj_reg); // restore obj
 784     call_VM(noreg,
 785             CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit),
 786             lock_reg);
 787 
 788     bind(done);
 789 
 790     restore_bcp();
 791   }
 792 }
 793 
 794 #ifndef CC_INTERP
 795 
 796 void InterpreterMacroAssembler::test_method_data_pointer(Register mdp,
 797                                                          Label& zero_continue) {
 798   assert(ProfileInterpreter, "must be profiling interpreter");
 799   movptr(mdp, Address(rbp, frame::interpreter_frame_mdx_offset * wordSize));
 800   testptr(mdp, mdp);
 801   jcc(Assembler::zero, zero_continue);
 802 }
 803 
 804 
 805 // Set the method data pointer for the current bcp.
 806 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
 807   assert(ProfileInterpreter, "must be profiling interpreter");
 808   Label set_mdp;
 809   push(rax);
 810   push(rbx);
 811 
 812   get_method(rbx);
 813   // Test MDO to avoid the call if it is NULL.
 814   movptr(rax, Address(rbx, in_bytes(methodOopDesc::method_data_offset())));
 815   testptr(rax, rax);
 816   jcc(Assembler::zero, set_mdp);
 817   // rbx: method
 818   // r13: bcp
 819   call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), rbx, r13);
 820   // rax: mdi
 821   // mdo is guaranteed to be non-zero here, we checked for it before the call.
 822   movptr(rbx, Address(rbx, in_bytes(methodOopDesc::method_data_offset())));
 823   addptr(rbx, in_bytes(methodDataOopDesc::data_offset()));
 824   addptr(rax, rbx);
 825   bind(set_mdp);
 826   movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), rax);
 827   pop(rbx);
 828   pop(rax);
 829 }
 830 
 831 void InterpreterMacroAssembler::verify_method_data_pointer() {
 832   assert(ProfileInterpreter, "must be profiling interpreter");
 833 #ifdef ASSERT
 834   Label verify_continue;
 835   push(rax);
 836   push(rbx);
 837   push(c_rarg3);
 838   push(c_rarg2);
 839   test_method_data_pointer(c_rarg3, verify_continue); // If mdp is zero, continue
 840   get_method(rbx);
 841 
 842   // If the mdp is valid, it will point to a DataLayout header which is
 843   // consistent with the bcp.  The converse is highly probable also.
 844   load_unsigned_short(c_rarg2,
 845                       Address(c_rarg3, in_bytes(DataLayout::bci_offset())));
 846   addptr(c_rarg2, Address(rbx, methodOopDesc::const_offset()));
 847   lea(c_rarg2, Address(c_rarg2, constMethodOopDesc::codes_offset()));
 848   cmpptr(c_rarg2, r13);
 849   jcc(Assembler::equal, verify_continue);
 850   // rbx: method
 851   // r13: bcp
 852   // c_rarg3: mdp
 853   call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp),
 854                rbx, r13, c_rarg3);
 855   bind(verify_continue);
 856   pop(c_rarg2);
 857   pop(c_rarg3);
 858   pop(rbx);
 859   pop(rax);
 860 #endif // ASSERT
 861 }
 862 
 863 
 864 void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in,
 865                                                 int constant,
 866                                                 Register value) {
 867   assert(ProfileInterpreter, "must be profiling interpreter");
 868   Address data(mdp_in, constant);
 869   movptr(data, value);
 870 }
 871 
 872 
 873 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
 874                                                       int constant,
 875                                                       bool decrement) {
 876   // Counter address
 877   Address data(mdp_in, constant);
 878 
 879   increment_mdp_data_at(data, decrement);
 880 }
 881 
 882 void InterpreterMacroAssembler::increment_mdp_data_at(Address data,
 883                                                       bool decrement) {
 884   assert(ProfileInterpreter, "must be profiling interpreter");
 885   // %%% this does 64bit counters at best it is wasting space
 886   // at worst it is a rare bug when counters overflow
 887 
 888   if (decrement) {
 889     // Decrement the register.  Set condition codes.
 890     addptr(data, (int32_t) -DataLayout::counter_increment);
 891     // If the decrement causes the counter to overflow, stay negative
 892     Label L;
 893     jcc(Assembler::negative, L);
 894     addptr(data, (int32_t) DataLayout::counter_increment);
 895     bind(L);
 896   } else {
 897     assert(DataLayout::counter_increment == 1,
 898            "flow-free idiom only works with 1");
 899     // Increment the register.  Set carry flag.
 900     addptr(data, DataLayout::counter_increment);
 901     // If the increment causes the counter to overflow, pull back by 1.
 902     sbbptr(data, (int32_t)0);
 903   }
 904 }
 905 
 906 
 907 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
 908                                                       Register reg,
 909                                                       int constant,
 910                                                       bool decrement) {
 911   Address data(mdp_in, reg, Address::times_1, constant);
 912 
 913   increment_mdp_data_at(data, decrement);
 914 }
 915 
 916 void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in,
 917                                                 int flag_byte_constant) {
 918   assert(ProfileInterpreter, "must be profiling interpreter");
 919   int header_offset = in_bytes(DataLayout::header_offset());
 920   int header_bits = DataLayout::flag_mask_to_header_mask(flag_byte_constant);
 921   // Set the flag
 922   orl(Address(mdp_in, header_offset), header_bits);
 923 }
 924 
 925 
 926 
 927 void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in,
 928                                                  int offset,
 929                                                  Register value,
 930                                                  Register test_value_out,
 931                                                  Label& not_equal_continue) {
 932   assert(ProfileInterpreter, "must be profiling interpreter");
 933   if (test_value_out == noreg) {
 934     cmpptr(value, Address(mdp_in, offset));
 935   } else {
 936     // Put the test value into a register, so caller can use it:
 937     movptr(test_value_out, Address(mdp_in, offset));
 938     cmpptr(test_value_out, value);
 939   }
 940   jcc(Assembler::notEqual, not_equal_continue);
 941 }
 942 
 943 
 944 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
 945                                                      int offset_of_disp) {
 946   assert(ProfileInterpreter, "must be profiling interpreter");
 947   Address disp_address(mdp_in, offset_of_disp);
 948   addptr(mdp_in, disp_address);
 949   movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), mdp_in);
 950 }
 951 
 952 
 953 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
 954                                                      Register reg,
 955                                                      int offset_of_disp) {
 956   assert(ProfileInterpreter, "must be profiling interpreter");
 957   Address disp_address(mdp_in, reg, Address::times_1, offset_of_disp);
 958   addptr(mdp_in, disp_address);
 959   movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), mdp_in);
 960 }
 961 
 962 
 963 void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in,
 964                                                        int constant) {
 965   assert(ProfileInterpreter, "must be profiling interpreter");
 966   addptr(mdp_in, constant);
 967   movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), mdp_in);
 968 }
 969 
 970 
 971 void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) {
 972   assert(ProfileInterpreter, "must be profiling interpreter");
 973   push(return_bci); // save/restore across call_VM
 974   call_VM(noreg,
 975           CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret),
 976           return_bci);
 977   pop(return_bci);
 978 }
 979 
 980 
 981 void InterpreterMacroAssembler::profile_taken_branch(Register mdp,
 982                                                      Register bumped_count) {
 983   if (ProfileInterpreter) {
 984     Label profile_continue;
 985 
 986     // If no method data exists, go to profile_continue.
 987     // Otherwise, assign to mdp
 988     test_method_data_pointer(mdp, profile_continue);
 989 
 990     // We are taking a branch.  Increment the taken count.
 991     // We inline increment_mdp_data_at to return bumped_count in a register
 992     //increment_mdp_data_at(mdp, in_bytes(JumpData::taken_offset()));
 993     Address data(mdp, in_bytes(JumpData::taken_offset()));
 994     movptr(bumped_count, data);
 995     assert(DataLayout::counter_increment == 1,
 996             "flow-free idiom only works with 1");
 997     addptr(bumped_count, DataLayout::counter_increment);
 998     sbbptr(bumped_count, 0);
 999     movptr(data, bumped_count); // Store back out
1000 
1001     // The method data pointer needs to be updated to reflect the new target.
1002     update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset()));
1003     bind(profile_continue);
1004   }
1005 }
1006 
1007 
1008 void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp) {
1009   if (ProfileInterpreter) {
1010     Label profile_continue;
1011 
1012     // If no method data exists, go to profile_continue.
1013     test_method_data_pointer(mdp, profile_continue);
1014 
1015     // We are taking a branch.  Increment the not taken count.
1016     increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset()));
1017 
1018     // The method data pointer needs to be updated to correspond to
1019     // the next bytecode
1020     update_mdp_by_constant(mdp, in_bytes(BranchData::branch_data_size()));
1021     bind(profile_continue);
1022   }
1023 }
1024 
1025 
1026 void InterpreterMacroAssembler::profile_call(Register mdp) {
1027   if (ProfileInterpreter) {
1028     Label profile_continue;
1029 
1030     // If no method data exists, go to profile_continue.
1031     test_method_data_pointer(mdp, profile_continue);
1032 
1033     // We are making a call.  Increment the count.
1034     increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1035 
1036     // The method data pointer needs to be updated to reflect the new target.
1037     update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size()));
1038     bind(profile_continue);
1039   }
1040 }
1041 
1042 
1043 void InterpreterMacroAssembler::profile_final_call(Register mdp) {
1044   if (ProfileInterpreter) {
1045     Label profile_continue;
1046 
1047     // If no method data exists, go to profile_continue.
1048     test_method_data_pointer(mdp, profile_continue);
1049 
1050     // We are making a call.  Increment the count.
1051     increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1052 
1053     // The method data pointer needs to be updated to reflect the new target.
1054     update_mdp_by_constant(mdp,
1055                            in_bytes(VirtualCallData::
1056                                     virtual_call_data_size()));
1057     bind(profile_continue);
1058   }
1059 }
1060 
1061 
1062 void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
1063                                                      Register mdp,
1064                                                      Register reg2,
1065                                                      bool receiver_can_be_null) {
1066   if (ProfileInterpreter) {
1067     Label profile_continue;
1068 
1069     // If no method data exists, go to profile_continue.
1070     test_method_data_pointer(mdp, profile_continue);
1071 
1072     Label skip_receiver_profile;
1073     if (receiver_can_be_null) {
1074       Label not_null;
1075       testptr(receiver, receiver);
1076       jccb(Assembler::notZero, not_null);
1077       // We are making a call.  Increment the count for null receiver.
1078       increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1079       jmp(skip_receiver_profile);
1080       bind(not_null);
1081     }
1082 
1083     // Record the receiver type.
1084     record_klass_in_profile(receiver, mdp, reg2, true);
1085     bind(skip_receiver_profile);
1086 
1087     // The method data pointer needs to be updated to reflect the new target.
1088     update_mdp_by_constant(mdp,
1089                            in_bytes(VirtualCallData::
1090                                     virtual_call_data_size()));
1091     bind(profile_continue);
1092   }
1093 }
1094 
1095 // This routine creates a state machine for updating the multi-row
1096 // type profile at a virtual call site (or other type-sensitive bytecode).
1097 // The machine visits each row (of receiver/count) until the receiver type
1098 // is found, or until it runs out of rows.  At the same time, it remembers
1099 // the location of the first empty row.  (An empty row records null for its
1100 // receiver, and can be allocated for a newly-observed receiver type.)
1101 // Because there are two degrees of freedom in the state, a simple linear
1102 // search will not work; it must be a decision tree.  Hence this helper
1103 // function is recursive, to generate the required tree structured code.
1104 // It's the interpreter, so we are trading off code space for speed.
1105 // See below for example code.
1106 void InterpreterMacroAssembler::record_klass_in_profile_helper(
1107                                         Register receiver, Register mdp,
1108                                         Register reg2, int start_row,
1109                                         Label& done, bool is_virtual_call) {
1110   if (TypeProfileWidth == 0) {
1111     if (is_virtual_call) {
1112       increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1113     }
1114     return;
1115   }
1116 
1117   int last_row = VirtualCallData::row_limit() - 1;
1118   assert(start_row <= last_row, "must be work left to do");
1119   // Test this row for both the receiver and for null.
1120   // Take any of three different outcomes:
1121   //   1. found receiver => increment count and goto done
1122   //   2. found null => keep looking for case 1, maybe allocate this cell
1123   //   3. found something else => keep looking for cases 1 and 2
1124   // Case 3 is handled by a recursive call.
1125   for (int row = start_row; row <= last_row; row++) {
1126     Label next_test;
1127     bool test_for_null_also = (row == start_row);
1128 
1129     // See if the receiver is receiver[n].
1130     int recvr_offset = in_bytes(VirtualCallData::receiver_offset(row));
1131     test_mdp_data_at(mdp, recvr_offset, receiver,
1132                      (test_for_null_also ? reg2 : noreg),
1133                      next_test);
1134     // (Reg2 now contains the receiver from the CallData.)
1135 
1136     // The receiver is receiver[n].  Increment count[n].
1137     int count_offset = in_bytes(VirtualCallData::receiver_count_offset(row));
1138     increment_mdp_data_at(mdp, count_offset);
1139     jmp(done);
1140     bind(next_test);
1141 
1142     if (test_for_null_also) {
1143       Label found_null;
1144       // Failed the equality check on receiver[n]...  Test for null.
1145       testptr(reg2, reg2);
1146       if (start_row == last_row) {
1147         // The only thing left to do is handle the null case.
1148         if (is_virtual_call) {
1149           jccb(Assembler::zero, found_null);
1150           // Receiver did not match any saved receiver and there is no empty row for it.
1151           // Increment total counter to indicate polymorphic case.
1152           increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1153           jmp(done);
1154           bind(found_null);
1155         } else {
1156           jcc(Assembler::notZero, done);
1157         }
1158         break;
1159       }
1160       // Since null is rare, make it be the branch-taken case.
1161       jcc(Assembler::zero, found_null);
1162 
1163       // Put all the "Case 3" tests here.
1164       record_klass_in_profile_helper(receiver, mdp, reg2, start_row + 1, done, is_virtual_call);
1165 
1166       // Found a null.  Keep searching for a matching receiver,
1167       // but remember that this is an empty (unused) slot.
1168       bind(found_null);
1169     }
1170   }
1171 
1172   // In the fall-through case, we found no matching receiver, but we
1173   // observed the receiver[start_row] is NULL.
1174 
1175   // Fill in the receiver field and increment the count.
1176   int recvr_offset = in_bytes(VirtualCallData::receiver_offset(start_row));
1177   set_mdp_data_at(mdp, recvr_offset, receiver);
1178   int count_offset = in_bytes(VirtualCallData::receiver_count_offset(start_row));
1179   movl(reg2, DataLayout::counter_increment);
1180   set_mdp_data_at(mdp, count_offset, reg2);
1181   if (start_row > 0) {
1182     jmp(done);
1183   }
1184 }
1185 
1186 // Example state machine code for three profile rows:
1187 //   // main copy of decision tree, rooted at row[1]
1188 //   if (row[0].rec == rec) { row[0].incr(); goto done; }
1189 //   if (row[0].rec != NULL) {
1190 //     // inner copy of decision tree, rooted at row[1]
1191 //     if (row[1].rec == rec) { row[1].incr(); goto done; }
1192 //     if (row[1].rec != NULL) {
1193 //       // degenerate decision tree, rooted at row[2]
1194 //       if (row[2].rec == rec) { row[2].incr(); goto done; }
1195 //       if (row[2].rec != NULL) { count.incr(); goto done; } // overflow
1196 //       row[2].init(rec); goto done;
1197 //     } else {
1198 //       // remember row[1] is empty
1199 //       if (row[2].rec == rec) { row[2].incr(); goto done; }
1200 //       row[1].init(rec); goto done;
1201 //     }
1202 //   } else {
1203 //     // remember row[0] is empty
1204 //     if (row[1].rec == rec) { row[1].incr(); goto done; }
1205 //     if (row[2].rec == rec) { row[2].incr(); goto done; }
1206 //     row[0].init(rec); goto done;
1207 //   }
1208 //   done:
1209 
1210 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver,
1211                                                         Register mdp, Register reg2,
1212                                                         bool is_virtual_call) {
1213   assert(ProfileInterpreter, "must be profiling");
1214   Label done;
1215 
1216   record_klass_in_profile_helper(receiver, mdp, reg2, 0, done, is_virtual_call);
1217 
1218   bind (done);
1219 }
1220 
1221 void InterpreterMacroAssembler::profile_ret(Register return_bci,
1222                                             Register mdp) {
1223   if (ProfileInterpreter) {
1224     Label profile_continue;
1225     uint row;
1226 
1227     // If no method data exists, go to profile_continue.
1228     test_method_data_pointer(mdp, profile_continue);
1229 
1230     // Update the total ret count.
1231     increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1232 
1233     for (row = 0; row < RetData::row_limit(); row++) {
1234       Label next_test;
1235 
1236       // See if return_bci is equal to bci[n]:
1237       test_mdp_data_at(mdp,
1238                        in_bytes(RetData::bci_offset(row)),
1239                        return_bci, noreg,
1240                        next_test);
1241 
1242       // return_bci is equal to bci[n].  Increment the count.
1243       increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row)));
1244 
1245       // The method data pointer needs to be updated to reflect the new target.
1246       update_mdp_by_offset(mdp,
1247                            in_bytes(RetData::bci_displacement_offset(row)));
1248       jmp(profile_continue);
1249       bind(next_test);
1250     }
1251 
1252     update_mdp_for_ret(return_bci);
1253 
1254     bind(profile_continue);
1255   }
1256 }
1257 
1258 
1259 void InterpreterMacroAssembler::profile_null_seen(Register mdp) {
1260   if (ProfileInterpreter) {
1261     Label profile_continue;
1262 
1263     // If no method data exists, go to profile_continue.
1264     test_method_data_pointer(mdp, profile_continue);
1265 
1266     set_mdp_flag_at(mdp, BitData::null_seen_byte_constant());
1267 
1268     // The method data pointer needs to be updated.
1269     int mdp_delta = in_bytes(BitData::bit_data_size());
1270     if (TypeProfileCasts) {
1271       mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1272     }
1273     update_mdp_by_constant(mdp, mdp_delta);
1274 
1275     bind(profile_continue);
1276   }
1277 }
1278 
1279 
1280 void InterpreterMacroAssembler::profile_typecheck_failed(Register mdp) {
1281   if (ProfileInterpreter && TypeProfileCasts) {
1282     Label profile_continue;
1283 
1284     // If no method data exists, go to profile_continue.
1285     test_method_data_pointer(mdp, profile_continue);
1286 
1287     int count_offset = in_bytes(CounterData::count_offset());
1288     // Back up the address, since we have already bumped the mdp.
1289     count_offset -= in_bytes(VirtualCallData::virtual_call_data_size());
1290 
1291     // *Decrement* the counter.  We expect to see zero or small negatives.
1292     increment_mdp_data_at(mdp, count_offset, true);
1293 
1294     bind (profile_continue);
1295   }
1296 }
1297 
1298 
1299 void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) {
1300   if (ProfileInterpreter) {
1301     Label profile_continue;
1302 
1303     // If no method data exists, go to profile_continue.
1304     test_method_data_pointer(mdp, profile_continue);
1305 
1306     // The method data pointer needs to be updated.
1307     int mdp_delta = in_bytes(BitData::bit_data_size());
1308     if (TypeProfileCasts) {
1309       mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1310 
1311       // Record the object type.
1312       record_klass_in_profile(klass, mdp, reg2, false);
1313     }
1314     update_mdp_by_constant(mdp, mdp_delta);
1315 
1316     bind(profile_continue);
1317   }
1318 }
1319 
1320 
1321 void InterpreterMacroAssembler::profile_switch_default(Register mdp) {
1322   if (ProfileInterpreter) {
1323     Label profile_continue;
1324 
1325     // If no method data exists, go to profile_continue.
1326     test_method_data_pointer(mdp, profile_continue);
1327 
1328     // Update the default case count
1329     increment_mdp_data_at(mdp,
1330                           in_bytes(MultiBranchData::default_count_offset()));
1331 
1332     // The method data pointer needs to be updated.
1333     update_mdp_by_offset(mdp,
1334                          in_bytes(MultiBranchData::
1335                                   default_displacement_offset()));
1336 
1337     bind(profile_continue);
1338   }
1339 }
1340 
1341 
1342 void InterpreterMacroAssembler::profile_switch_case(Register index,
1343                                                     Register mdp,
1344                                                     Register reg2) {
1345   if (ProfileInterpreter) {
1346     Label profile_continue;
1347 
1348     // If no method data exists, go to profile_continue.
1349     test_method_data_pointer(mdp, profile_continue);
1350 
1351     // Build the base (index * per_case_size_in_bytes()) +
1352     // case_array_offset_in_bytes()
1353     movl(reg2, in_bytes(MultiBranchData::per_case_size()));
1354     imulptr(index, reg2); // XXX l ?
1355     addptr(index, in_bytes(MultiBranchData::case_array_offset())); // XXX l ?
1356 
1357     // Update the case count
1358     increment_mdp_data_at(mdp,
1359                           index,
1360                           in_bytes(MultiBranchData::relative_count_offset()));
1361 
1362     // The method data pointer needs to be updated.
1363     update_mdp_by_offset(mdp,
1364                          index,
1365                          in_bytes(MultiBranchData::
1366                                   relative_displacement_offset()));
1367 
1368     bind(profile_continue);
1369   }
1370 }
1371 
1372 
1373 
1374 void InterpreterMacroAssembler::verify_oop(Register reg, TosState state) {
1375   if (state == atos) {
1376     MacroAssembler::verify_oop(reg);
1377   }
1378 }
1379 
1380 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) {
1381 }
1382 #endif // !CC_INTERP
1383 
1384 
1385 void InterpreterMacroAssembler::notify_method_entry() {
1386   // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1387   // track stack depth.  If it is possible to enter interp_only_mode we add
1388   // the code to check if the event should be sent.
1389   if (JvmtiExport::can_post_interpreter_events()) {
1390     Label L;
1391     movl(rdx, Address(r15_thread, JavaThread::interp_only_mode_offset()));
1392     testl(rdx, rdx);
1393     jcc(Assembler::zero, L);
1394     call_VM(noreg, CAST_FROM_FN_PTR(address,
1395                                     InterpreterRuntime::post_method_entry));
1396     bind(L);
1397   }
1398 
1399   {
1400     SkipIfEqual skip(this, &DTraceMethodProbes, false);
1401     get_method(c_rarg1);
1402     call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
1403                  r15_thread, c_rarg1);
1404   }
1405 
1406   // RedefineClasses() tracing support for obsolete method entry
1407   if (RC_TRACE_IN_RANGE(0x00001000, 0x00002000)) {
1408     get_method(c_rarg1);
1409     call_VM_leaf(
1410       CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
1411       r15_thread, c_rarg1);
1412   }
1413 }
1414 
1415 
1416 void InterpreterMacroAssembler::notify_method_exit(
1417     TosState state, NotifyMethodExitMode mode) {
1418   // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1419   // track stack depth.  If it is possible to enter interp_only_mode we add
1420   // the code to check if the event should be sent.
1421   if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
1422     Label L;
1423     // Note: frame::interpreter_frame_result has a dependency on how the
1424     // method result is saved across the call to post_method_exit. If this
1425     // is changed then the interpreter_frame_result implementation will
1426     // need to be updated too.
1427 
1428     // For c++ interpreter the result is always stored at a known location in the frame
1429     // template interpreter will leave it on the top of the stack.
1430     NOT_CC_INTERP(push(state);)
1431     movl(rdx, Address(r15_thread, JavaThread::interp_only_mode_offset()));
1432     testl(rdx, rdx);
1433     jcc(Assembler::zero, L);
1434     call_VM(noreg,
1435             CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
1436     bind(L);
1437     NOT_CC_INTERP(pop(state));
1438   }
1439 
1440   {
1441     SkipIfEqual skip(this, &DTraceMethodProbes, false);
1442     NOT_CC_INTERP(push(state));
1443     get_method(c_rarg1);
1444     call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
1445                  r15_thread, c_rarg1);
1446     NOT_CC_INTERP(pop(state));
1447   }
1448 }
1449 
1450 // Jump if ((*counter_addr += increment) & mask) satisfies the condition.
1451 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr,
1452                                                         int increment, int mask,
1453                                                         Register scratch, bool preloaded,
1454                                                         Condition cond, Label* where) {
1455   if (!preloaded) {
1456     movl(scratch, counter_addr);
1457   }
1458   incrementl(scratch, increment);
1459   movl(counter_addr, scratch);
1460   andl(scratch, mask);
1461   jcc(cond, *where);
1462 }