1 /*
   2  * Copyright (c) 2003, 2010, 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/biasedLocking.hpp"
  37 #include "runtime/sharedRuntime.hpp"
  38 #include "runtime/synchronizer.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 EnableInvokeDynamic");
 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(EnableMethodHandles, "tiny index used only for EnableMethodHandles");
 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::super_call_VM_leaf(address entry_point) {
 385   MacroAssembler::call_VM_leaf_base(entry_point, 0);
 386 }
 387 
 388 
 389 void InterpreterMacroAssembler::super_call_VM_leaf(address entry_point,
 390                                                    Register arg_1) {
 391   if (c_rarg0 != arg_1) {
 392     mov(c_rarg0, arg_1);
 393   }
 394   MacroAssembler::call_VM_leaf_base(entry_point, 1);
 395 }
 396 
 397 
 398 void InterpreterMacroAssembler::super_call_VM_leaf(address entry_point,
 399                                                    Register arg_1,
 400                                                    Register arg_2) {
 401   assert(c_rarg0 != arg_2, "smashed argument");
 402   assert(c_rarg1 != arg_1, "smashed argument");
 403   if (c_rarg0 != arg_1) {
 404     mov(c_rarg0, arg_1);
 405   }
 406   if (c_rarg1 != arg_2) {
 407     mov(c_rarg1, arg_2);
 408   }
 409   MacroAssembler::call_VM_leaf_base(entry_point, 2);
 410 }
 411 
 412 void InterpreterMacroAssembler::super_call_VM_leaf(address entry_point,
 413                                                    Register arg_1,
 414                                                    Register arg_2,
 415                                                    Register arg_3) {
 416   assert(c_rarg0 != arg_2, "smashed argument");
 417   assert(c_rarg0 != arg_3, "smashed argument");
 418   assert(c_rarg1 != arg_1, "smashed argument");
 419   assert(c_rarg1 != arg_3, "smashed argument");
 420   assert(c_rarg2 != arg_1, "smashed argument");
 421   assert(c_rarg2 != arg_2, "smashed argument");
 422   if (c_rarg0 != arg_1) {
 423     mov(c_rarg0, arg_1);
 424   }
 425   if (c_rarg1 != arg_2) {
 426     mov(c_rarg1, arg_2);
 427   }
 428   if (c_rarg2 != arg_3) {
 429     mov(c_rarg2, arg_3);
 430   }
 431   MacroAssembler::call_VM_leaf_base(entry_point, 3);
 432 }
 433 
 434 void InterpreterMacroAssembler::prepare_to_jump_from_interpreted() {
 435   // set sender sp
 436   lea(r13, Address(rsp, wordSize));
 437   // record last_sp
 438   movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), r13);
 439 }
 440 
 441 
 442 // Jump to from_interpreted entry of a call unless single stepping is possible
 443 // in this thread in which case we must call the i2i entry
 444 void InterpreterMacroAssembler::jump_from_interpreted(Register method, Register temp) {
 445   prepare_to_jump_from_interpreted();
 446 
 447   if (JvmtiExport::can_post_interpreter_events()) {
 448     Label run_compiled_code;
 449     // JVMTI events, such as single-stepping, are implemented partly by avoiding running
 450     // compiled code in threads for which the event is enabled.  Check here for
 451     // interp_only_mode if these events CAN be enabled.
 452     get_thread(temp);
 453     // interp_only is an int, on little endian it is sufficient to test the byte only
 454     // Is a cmpl faster (ce
 455     cmpb(Address(temp, JavaThread::interp_only_mode_offset()), 0);
 456     jcc(Assembler::zero, run_compiled_code);
 457     jmp(Address(method, methodOopDesc::interpreter_entry_offset()));
 458     bind(run_compiled_code);
 459   }
 460 
 461   jmp(Address(method, methodOopDesc::from_interpreted_offset()));
 462 
 463 }
 464 
 465 
 466 // The following two routines provide a hook so that an implementation
 467 // can schedule the dispatch in two parts.  amd64 does not do this.
 468 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int step) {
 469   // Nothing amd64 specific to be done here
 470 }
 471 
 472 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int step) {
 473   dispatch_next(state, step);
 474 }
 475 
 476 void InterpreterMacroAssembler::dispatch_base(TosState state,
 477                                               address* table,
 478                                               bool verifyoop) {
 479   verify_FPU(1, state);
 480   if (VerifyActivationFrameSize) {
 481     Label L;
 482     mov(rcx, rbp);
 483     subptr(rcx, rsp);
 484     int32_t min_frame_size =
 485       (frame::link_offset - frame::interpreter_frame_initial_sp_offset) *
 486       wordSize;
 487     cmpptr(rcx, (int32_t)min_frame_size);
 488     jcc(Assembler::greaterEqual, L);
 489     stop("broken stack frame");
 490     bind(L);
 491   }
 492   if (verifyoop) {
 493     verify_oop(rax, state);
 494   }
 495   lea(rscratch1, ExternalAddress((address)table));
 496   jmp(Address(rscratch1, rbx, Address::times_8));
 497 }
 498 
 499 void InterpreterMacroAssembler::dispatch_only(TosState state) {
 500   dispatch_base(state, Interpreter::dispatch_table(state));
 501 }
 502 
 503 void InterpreterMacroAssembler::dispatch_only_normal(TosState state) {
 504   dispatch_base(state, Interpreter::normal_table(state));
 505 }
 506 
 507 void InterpreterMacroAssembler::dispatch_only_noverify(TosState state) {
 508   dispatch_base(state, Interpreter::normal_table(state), false);
 509 }
 510 
 511 
 512 void InterpreterMacroAssembler::dispatch_next(TosState state, int step) {
 513   // load next bytecode (load before advancing r13 to prevent AGI)
 514   load_unsigned_byte(rbx, Address(r13, step));
 515   // advance r13
 516   increment(r13, step);
 517   dispatch_base(state, Interpreter::dispatch_table(state));
 518 }
 519 
 520 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) {
 521   // load current bytecode
 522   load_unsigned_byte(rbx, Address(r13, 0));
 523   dispatch_base(state, table);
 524 }
 525 
 526 // remove activation
 527 //
 528 // Unlock the receiver if this is a synchronized method.
 529 // Unlock any Java monitors from syncronized blocks.
 530 // Remove the activation from the stack.
 531 //
 532 // If there are locked Java monitors
 533 //    If throw_monitor_exception
 534 //       throws IllegalMonitorStateException
 535 //    Else if install_monitor_exception
 536 //       installs IllegalMonitorStateException
 537 //    Else
 538 //       no error processing
 539 void InterpreterMacroAssembler::remove_activation(
 540         TosState state,
 541         Register ret_addr,
 542         bool throw_monitor_exception,
 543         bool install_monitor_exception,
 544         bool notify_jvmdi) {
 545   // Note: Registers rdx xmm0 may be in use for the
 546   // result check if synchronized method
 547   Label unlocked, unlock, no_unlock;
 548 
 549   // get the value of _do_not_unlock_if_synchronized into rdx
 550   const Address do_not_unlock_if_synchronized(r15_thread,
 551     in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
 552   movbool(rdx, do_not_unlock_if_synchronized);
 553   movbool(do_not_unlock_if_synchronized, false); // reset the flag
 554 
 555  // get method access flags
 556   movptr(rbx, Address(rbp, frame::interpreter_frame_method_offset * wordSize));
 557   movl(rcx, Address(rbx, methodOopDesc::access_flags_offset()));
 558   testl(rcx, JVM_ACC_SYNCHRONIZED);
 559   jcc(Assembler::zero, unlocked);
 560 
 561   // Don't unlock anything if the _do_not_unlock_if_synchronized flag
 562   // is set.
 563   testbool(rdx);
 564   jcc(Assembler::notZero, no_unlock);
 565 
 566   // unlock monitor
 567   push(state); // save result
 568 
 569   // BasicObjectLock will be first in list, since this is a
 570   // synchronized method. However, need to check that the object has
 571   // not been unlocked by an explicit monitorexit bytecode.
 572   const Address monitor(rbp, frame::interpreter_frame_initial_sp_offset *
 573                         wordSize - (int) sizeof(BasicObjectLock));
 574   // We use c_rarg1 so that if we go slow path it will be the correct
 575   // register for unlock_object to pass to VM directly
 576   lea(c_rarg1, monitor); // address of first monitor
 577 
 578   movptr(rax, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
 579   testptr(rax, rax);
 580   jcc(Assembler::notZero, unlock);
 581 
 582   pop(state);
 583   if (throw_monitor_exception) {
 584     // Entry already unlocked, need to throw exception
 585     call_VM(noreg, CAST_FROM_FN_PTR(address,
 586                    InterpreterRuntime::throw_illegal_monitor_state_exception));
 587     should_not_reach_here();
 588   } else {
 589     // Monitor already unlocked during a stack unroll. If requested,
 590     // install an illegal_monitor_state_exception.  Continue with
 591     // stack unrolling.
 592     if (install_monitor_exception) {
 593       call_VM(noreg, CAST_FROM_FN_PTR(address,
 594                      InterpreterRuntime::new_illegal_monitor_state_exception));
 595     }
 596     jmp(unlocked);
 597   }
 598 
 599   bind(unlock);
 600   unlock_object(c_rarg1);
 601   pop(state);
 602 
 603   // Check that for block-structured locking (i.e., that all locked
 604   // objects has been unlocked)
 605   bind(unlocked);
 606 
 607   // rax: Might contain return value
 608 
 609   // Check that all monitors are unlocked
 610   {
 611     Label loop, exception, entry, restart;
 612     const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
 613     const Address monitor_block_top(
 614         rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
 615     const Address monitor_block_bot(
 616         rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
 617 
 618     bind(restart);
 619     // We use c_rarg1 so that if we go slow path it will be the correct
 620     // register for unlock_object to pass to VM directly
 621     movptr(c_rarg1, monitor_block_top); // points to current entry, starting
 622                                   // with top-most entry
 623     lea(rbx, monitor_block_bot);  // points to word before bottom of
 624                                   // monitor block
 625     jmp(entry);
 626 
 627     // Entry already locked, need to throw exception
 628     bind(exception);
 629 
 630     if (throw_monitor_exception) {
 631       // Throw exception
 632       MacroAssembler::call_VM(noreg,
 633                               CAST_FROM_FN_PTR(address, InterpreterRuntime::
 634                                    throw_illegal_monitor_state_exception));
 635       should_not_reach_here();
 636     } else {
 637       // Stack unrolling. Unlock object and install illegal_monitor_exception.
 638       // Unlock does not block, so don't have to worry about the frame.
 639       // We don't have to preserve c_rarg1 since we are going to throw an exception.
 640 
 641       push(state);
 642       unlock_object(c_rarg1);
 643       pop(state);
 644 
 645       if (install_monitor_exception) {
 646         call_VM(noreg, CAST_FROM_FN_PTR(address,
 647                                         InterpreterRuntime::
 648                                         new_illegal_monitor_state_exception));
 649       }
 650 
 651       jmp(restart);
 652     }
 653 
 654     bind(loop);
 655     // check if current entry is used
 656     cmpptr(Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()), (int32_t) NULL);
 657     jcc(Assembler::notEqual, exception);
 658 
 659     addptr(c_rarg1, entry_size); // otherwise advance to next entry
 660     bind(entry);
 661     cmpptr(c_rarg1, rbx); // check if bottom reached
 662     jcc(Assembler::notEqual, loop); // if not at bottom then check this entry
 663   }
 664 
 665   bind(no_unlock);
 666 
 667   // jvmti support
 668   if (notify_jvmdi) {
 669     notify_method_exit(state, NotifyJVMTI);    // preserve TOSCA
 670   } else {
 671     notify_method_exit(state, SkipNotifyJVMTI); // preserve TOSCA
 672   }
 673 
 674   // remove activation
 675   // get sender sp
 676   movptr(rbx,
 677          Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize));
 678   leave();                           // remove frame anchor
 679   pop(ret_addr);                     // get return address
 680   mov(rsp, rbx);                     // set sp to sender sp
 681 }
 682 
 683 #endif // C_INTERP
 684 
 685 // Lock object
 686 //
 687 // Args:
 688 //      c_rarg1: BasicObjectLock to be used for locking
 689 //
 690 // Kills:
 691 //      rax
 692 //      c_rarg0, c_rarg1, c_rarg2, c_rarg3, .. (param regs)
 693 //      rscratch1, rscratch2 (scratch regs)
 694 void InterpreterMacroAssembler::lock_object(Register lock_reg) {
 695   assert(lock_reg == c_rarg1, "The argument is only for looks. It must be c_rarg1");
 696 
 697   if (UseHeavyMonitors) {
 698     call_VM(noreg,
 699             CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
 700             lock_reg);
 701   } else {
 702     Label done;
 703 
 704     const Register swap_reg = rax; // Must use rax for cmpxchg instruction
 705     const Register obj_reg = c_rarg3; // Will contain the oop
 706 
 707     const int obj_offset = BasicObjectLock::obj_offset_in_bytes();
 708     const int lock_offset = BasicObjectLock::lock_offset_in_bytes ();
 709     const int mark_offset = lock_offset +
 710                             BasicLock::displaced_header_offset_in_bytes();
 711 
 712     Label slow_case;
 713 
 714     // Load object pointer into obj_reg %c_rarg3
 715     movptr(obj_reg, Address(lock_reg, obj_offset));
 716 
 717     if (UseBiasedLocking) {
 718       biased_locking_enter(lock_reg, obj_reg, swap_reg, rscratch1, false, done, &slow_case);
 719     }
 720 
 721     // Load immediate 1 into swap_reg %rax
 722     movl(swap_reg, 1);
 723 
 724     // Load (object->mark() | 1) into swap_reg %rax
 725     orptr(swap_reg, Address(obj_reg, 0));
 726 
 727     // Save (object->mark() | 1) into BasicLock's displaced header
 728     movptr(Address(lock_reg, mark_offset), swap_reg);
 729 
 730     assert(lock_offset == 0,
 731            "displached header must be first word in BasicObjectLock");
 732 
 733     if (os::is_MP()) lock();
 734     cmpxchgptr(lock_reg, Address(obj_reg, 0));
 735     if (PrintBiasedLockingStatistics) {
 736       cond_inc32(Assembler::zero,
 737                  ExternalAddress((address) BiasedLocking::fast_path_entry_count_addr()));
 738     }
 739     jcc(Assembler::zero, done);
 740 
 741     // Test if the oopMark is an obvious stack pointer, i.e.,
 742     //  1) (mark & 7) == 0, and
 743     //  2) rsp <= mark < mark + os::pagesize()
 744     //
 745     // These 3 tests can be done by evaluating the following
 746     // expression: ((mark - rsp) & (7 - os::vm_page_size())),
 747     // assuming both stack pointer and pagesize have their
 748     // least significant 3 bits clear.
 749     // NOTE: the oopMark is in swap_reg %rax as the result of cmpxchg
 750     subptr(swap_reg, rsp);
 751     andptr(swap_reg, 7 - os::vm_page_size());
 752 
 753     // Save the test result, for recursive case, the result is zero
 754     movptr(Address(lock_reg, mark_offset), swap_reg);
 755 
 756     if (PrintBiasedLockingStatistics) {
 757       cond_inc32(Assembler::zero,
 758                  ExternalAddress((address) BiasedLocking::fast_path_entry_count_addr()));
 759     }
 760     jcc(Assembler::zero, done);
 761 
 762     bind(slow_case);
 763 
 764     // Call the runtime routine for slow case
 765     call_VM(noreg,
 766             CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
 767             lock_reg);
 768 
 769     bind(done);
 770   }
 771 }
 772 
 773 
 774 // Unlocks an object. Used in monitorexit bytecode and
 775 // remove_activation.  Throws an IllegalMonitorException if object is
 776 // not locked by current thread.
 777 //
 778 // Args:
 779 //      c_rarg1: BasicObjectLock for lock
 780 //
 781 // Kills:
 782 //      rax
 783 //      c_rarg0, c_rarg1, c_rarg2, c_rarg3, ... (param regs)
 784 //      rscratch1, rscratch2 (scratch regs)
 785 void InterpreterMacroAssembler::unlock_object(Register lock_reg) {
 786   assert(lock_reg == c_rarg1, "The argument is only for looks. It must be rarg1");
 787 
 788   if (UseHeavyMonitors) {
 789     call_VM(noreg,
 790             CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit),
 791             lock_reg);
 792   } else {
 793     Label done;
 794 
 795     const Register swap_reg   = rax;  // Must use rax for cmpxchg instruction
 796     const Register header_reg = c_rarg2;  // Will contain the old oopMark
 797     const Register obj_reg    = c_rarg3;  // Will contain the oop
 798 
 799     save_bcp(); // Save in case of exception
 800 
 801     // Convert from BasicObjectLock structure to object and BasicLock
 802     // structure Store the BasicLock address into %rax
 803     lea(swap_reg, Address(lock_reg, BasicObjectLock::lock_offset_in_bytes()));
 804 
 805     // Load oop into obj_reg(%c_rarg3)
 806     movptr(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()));
 807 
 808     // Free entry
 809     movptr(Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()), (int32_t)NULL_WORD);
 810 
 811     if (UseBiasedLocking) {
 812       biased_locking_exit(obj_reg, header_reg, done);
 813     }
 814 
 815     // Load the old header from BasicLock structure
 816     movptr(header_reg, Address(swap_reg,
 817                                BasicLock::displaced_header_offset_in_bytes()));
 818 
 819     // Test for recursion
 820     testptr(header_reg, header_reg);
 821 
 822     // zero for recursive case
 823     jcc(Assembler::zero, done);
 824 
 825     // Atomic swap back the old header
 826     if (os::is_MP()) lock();
 827     cmpxchgptr(header_reg, Address(obj_reg, 0));
 828 
 829     // zero for recursive case
 830     jcc(Assembler::zero, done);
 831 
 832     // Call the runtime routine for slow case.
 833     movptr(Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()),
 834          obj_reg); // restore obj
 835     call_VM(noreg,
 836             CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit),
 837             lock_reg);
 838 
 839     bind(done);
 840 
 841     restore_bcp();
 842   }
 843 }
 844 
 845 #ifndef CC_INTERP
 846 
 847 void InterpreterMacroAssembler::test_method_data_pointer(Register mdp,
 848                                                          Label& zero_continue) {
 849   assert(ProfileInterpreter, "must be profiling interpreter");
 850   movptr(mdp, Address(rbp, frame::interpreter_frame_mdx_offset * wordSize));
 851   testptr(mdp, mdp);
 852   jcc(Assembler::zero, zero_continue);
 853 }
 854 
 855 
 856 // Set the method data pointer for the current bcp.
 857 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
 858   assert(ProfileInterpreter, "must be profiling interpreter");
 859   Label zero_continue;
 860   push(rax);
 861   push(rbx);
 862 
 863   get_method(rbx);
 864   // Test MDO to avoid the call if it is NULL.
 865   movptr(rax, Address(rbx, in_bytes(methodOopDesc::method_data_offset())));
 866   testptr(rax, rax);
 867   jcc(Assembler::zero, zero_continue);
 868 
 869   // rbx: method
 870   // r13: bcp
 871   call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), rbx, r13);
 872   // rax: mdi
 873 
 874   movptr(rbx, Address(rbx, in_bytes(methodOopDesc::method_data_offset())));
 875   testptr(rbx, rbx);
 876   jcc(Assembler::zero, zero_continue);
 877   addptr(rbx, in_bytes(methodDataOopDesc::data_offset()));
 878   addptr(rbx, rax);
 879   movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), rbx);
 880 
 881   bind(zero_continue);
 882   pop(rbx);
 883   pop(rax);
 884 }
 885 
 886 void InterpreterMacroAssembler::verify_method_data_pointer() {
 887   assert(ProfileInterpreter, "must be profiling interpreter");
 888 #ifdef ASSERT
 889   Label verify_continue;
 890   push(rax);
 891   push(rbx);
 892   push(c_rarg3);
 893   push(c_rarg2);
 894   test_method_data_pointer(c_rarg3, verify_continue); // If mdp is zero, continue
 895   get_method(rbx);
 896 
 897   // If the mdp is valid, it will point to a DataLayout header which is
 898   // consistent with the bcp.  The converse is highly probable also.
 899   load_unsigned_short(c_rarg2,
 900                       Address(c_rarg3, in_bytes(DataLayout::bci_offset())));
 901   addptr(c_rarg2, Address(rbx, methodOopDesc::const_offset()));
 902   lea(c_rarg2, Address(c_rarg2, constMethodOopDesc::codes_offset()));
 903   cmpptr(c_rarg2, r13);
 904   jcc(Assembler::equal, verify_continue);
 905   // rbx: method
 906   // r13: bcp
 907   // c_rarg3: mdp
 908   call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp),
 909                rbx, r13, c_rarg3);
 910   bind(verify_continue);
 911   pop(c_rarg2);
 912   pop(c_rarg3);
 913   pop(rbx);
 914   pop(rax);
 915 #endif // ASSERT
 916 }
 917 
 918 
 919 void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in,
 920                                                 int constant,
 921                                                 Register value) {
 922   assert(ProfileInterpreter, "must be profiling interpreter");
 923   Address data(mdp_in, constant);
 924   movptr(data, value);
 925 }
 926 
 927 
 928 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
 929                                                       int constant,
 930                                                       bool decrement) {
 931   // Counter address
 932   Address data(mdp_in, constant);
 933 
 934   increment_mdp_data_at(data, decrement);
 935 }
 936 
 937 void InterpreterMacroAssembler::increment_mdp_data_at(Address data,
 938                                                       bool decrement) {
 939   assert(ProfileInterpreter, "must be profiling interpreter");
 940   // %%% this does 64bit counters at best it is wasting space
 941   // at worst it is a rare bug when counters overflow
 942 
 943   if (decrement) {
 944     // Decrement the register.  Set condition codes.
 945     addptr(data, (int32_t) -DataLayout::counter_increment);
 946     // If the decrement causes the counter to overflow, stay negative
 947     Label L;
 948     jcc(Assembler::negative, L);
 949     addptr(data, (int32_t) DataLayout::counter_increment);
 950     bind(L);
 951   } else {
 952     assert(DataLayout::counter_increment == 1,
 953            "flow-free idiom only works with 1");
 954     // Increment the register.  Set carry flag.
 955     addptr(data, DataLayout::counter_increment);
 956     // If the increment causes the counter to overflow, pull back by 1.
 957     sbbptr(data, (int32_t)0);
 958   }
 959 }
 960 
 961 
 962 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
 963                                                       Register reg,
 964                                                       int constant,
 965                                                       bool decrement) {
 966   Address data(mdp_in, reg, Address::times_1, constant);
 967 
 968   increment_mdp_data_at(data, decrement);
 969 }
 970 
 971 void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in,
 972                                                 int flag_byte_constant) {
 973   assert(ProfileInterpreter, "must be profiling interpreter");
 974   int header_offset = in_bytes(DataLayout::header_offset());
 975   int header_bits = DataLayout::flag_mask_to_header_mask(flag_byte_constant);
 976   // Set the flag
 977   orl(Address(mdp_in, header_offset), header_bits);
 978 }
 979 
 980 
 981 
 982 void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in,
 983                                                  int offset,
 984                                                  Register value,
 985                                                  Register test_value_out,
 986                                                  Label& not_equal_continue) {
 987   assert(ProfileInterpreter, "must be profiling interpreter");
 988   if (test_value_out == noreg) {
 989     cmpptr(value, Address(mdp_in, offset));
 990   } else {
 991     // Put the test value into a register, so caller can use it:
 992     movptr(test_value_out, Address(mdp_in, offset));
 993     cmpptr(test_value_out, value);
 994   }
 995   jcc(Assembler::notEqual, not_equal_continue);
 996 }
 997 
 998 
 999 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1000                                                      int offset_of_disp) {
1001   assert(ProfileInterpreter, "must be profiling interpreter");
1002   Address disp_address(mdp_in, offset_of_disp);
1003   addptr(mdp_in, disp_address);
1004   movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), mdp_in);
1005 }
1006 
1007 
1008 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1009                                                      Register reg,
1010                                                      int offset_of_disp) {
1011   assert(ProfileInterpreter, "must be profiling interpreter");
1012   Address disp_address(mdp_in, reg, Address::times_1, offset_of_disp);
1013   addptr(mdp_in, disp_address);
1014   movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), mdp_in);
1015 }
1016 
1017 
1018 void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in,
1019                                                        int constant) {
1020   assert(ProfileInterpreter, "must be profiling interpreter");
1021   addptr(mdp_in, constant);
1022   movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), mdp_in);
1023 }
1024 
1025 
1026 void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) {
1027   assert(ProfileInterpreter, "must be profiling interpreter");
1028   push(return_bci); // save/restore across call_VM
1029   call_VM(noreg,
1030           CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret),
1031           return_bci);
1032   pop(return_bci);
1033 }
1034 
1035 
1036 void InterpreterMacroAssembler::profile_taken_branch(Register mdp,
1037                                                      Register bumped_count) {
1038   if (ProfileInterpreter) {
1039     Label profile_continue;
1040 
1041     // If no method data exists, go to profile_continue.
1042     // Otherwise, assign to mdp
1043     test_method_data_pointer(mdp, profile_continue);
1044 
1045     // We are taking a branch.  Increment the taken count.
1046     // We inline increment_mdp_data_at to return bumped_count in a register
1047     //increment_mdp_data_at(mdp, in_bytes(JumpData::taken_offset()));
1048     Address data(mdp, in_bytes(JumpData::taken_offset()));
1049     movptr(bumped_count, data);
1050     assert(DataLayout::counter_increment == 1,
1051             "flow-free idiom only works with 1");
1052     addptr(bumped_count, DataLayout::counter_increment);
1053     sbbptr(bumped_count, 0);
1054     movptr(data, bumped_count); // Store back out
1055 
1056     // The method data pointer needs to be updated to reflect the new target.
1057     update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset()));
1058     bind(profile_continue);
1059   }
1060 }
1061 
1062 
1063 void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp) {
1064   if (ProfileInterpreter) {
1065     Label profile_continue;
1066 
1067     // If no method data exists, go to profile_continue.
1068     test_method_data_pointer(mdp, profile_continue);
1069 
1070     // We are taking a branch.  Increment the not taken count.
1071     increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset()));
1072 
1073     // The method data pointer needs to be updated to correspond to
1074     // the next bytecode
1075     update_mdp_by_constant(mdp, in_bytes(BranchData::branch_data_size()));
1076     bind(profile_continue);
1077   }
1078 }
1079 
1080 
1081 void InterpreterMacroAssembler::profile_call(Register mdp) {
1082   if (ProfileInterpreter) {
1083     Label profile_continue;
1084 
1085     // If no method data exists, go to profile_continue.
1086     test_method_data_pointer(mdp, profile_continue);
1087 
1088     // We are making a call.  Increment the count.
1089     increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1090 
1091     // The method data pointer needs to be updated to reflect the new target.
1092     update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size()));
1093     bind(profile_continue);
1094   }
1095 }
1096 
1097 
1098 void InterpreterMacroAssembler::profile_final_call(Register mdp) {
1099   if (ProfileInterpreter) {
1100     Label profile_continue;
1101 
1102     // If no method data exists, go to profile_continue.
1103     test_method_data_pointer(mdp, profile_continue);
1104 
1105     // We are making a call.  Increment the count.
1106     increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1107 
1108     // The method data pointer needs to be updated to reflect the new target.
1109     update_mdp_by_constant(mdp,
1110                            in_bytes(VirtualCallData::
1111                                     virtual_call_data_size()));
1112     bind(profile_continue);
1113   }
1114 }
1115 
1116 
1117 void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
1118                                                      Register mdp,
1119                                                      Register reg2,
1120                                                      bool receiver_can_be_null) {
1121   if (ProfileInterpreter) {
1122     Label profile_continue;
1123 
1124     // If no method data exists, go to profile_continue.
1125     test_method_data_pointer(mdp, profile_continue);
1126 
1127     Label skip_receiver_profile;
1128     if (receiver_can_be_null) {
1129       Label not_null;
1130       testptr(receiver, receiver);
1131       jccb(Assembler::notZero, not_null);
1132       // We are making a call.  Increment the count for null receiver.
1133       increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1134       jmp(skip_receiver_profile);
1135       bind(not_null);
1136     }
1137 
1138     // Record the receiver type.
1139     record_klass_in_profile(receiver, mdp, reg2, true);
1140     bind(skip_receiver_profile);
1141 
1142     // The method data pointer needs to be updated to reflect the new target.
1143     update_mdp_by_constant(mdp,
1144                            in_bytes(VirtualCallData::
1145                                     virtual_call_data_size()));
1146     bind(profile_continue);
1147   }
1148 }
1149 
1150 // This routine creates a state machine for updating the multi-row
1151 // type profile at a virtual call site (or other type-sensitive bytecode).
1152 // The machine visits each row (of receiver/count) until the receiver type
1153 // is found, or until it runs out of rows.  At the same time, it remembers
1154 // the location of the first empty row.  (An empty row records null for its
1155 // receiver, and can be allocated for a newly-observed receiver type.)
1156 // Because there are two degrees of freedom in the state, a simple linear
1157 // search will not work; it must be a decision tree.  Hence this helper
1158 // function is recursive, to generate the required tree structured code.
1159 // It's the interpreter, so we are trading off code space for speed.
1160 // See below for example code.
1161 void InterpreterMacroAssembler::record_klass_in_profile_helper(
1162                                         Register receiver, Register mdp,
1163                                         Register reg2, int start_row,
1164                                         Label& done, bool is_virtual_call) {
1165   if (TypeProfileWidth == 0) {
1166     if (is_virtual_call) {
1167       increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1168     }
1169     return;
1170   }
1171 
1172   int last_row = VirtualCallData::row_limit() - 1;
1173   assert(start_row <= last_row, "must be work left to do");
1174   // Test this row for both the receiver and for null.
1175   // Take any of three different outcomes:
1176   //   1. found receiver => increment count and goto done
1177   //   2. found null => keep looking for case 1, maybe allocate this cell
1178   //   3. found something else => keep looking for cases 1 and 2
1179   // Case 3 is handled by a recursive call.
1180   for (int row = start_row; row <= last_row; row++) {
1181     Label next_test;
1182     bool test_for_null_also = (row == start_row);
1183 
1184     // See if the receiver is receiver[n].
1185     int recvr_offset = in_bytes(VirtualCallData::receiver_offset(row));
1186     test_mdp_data_at(mdp, recvr_offset, receiver,
1187                      (test_for_null_also ? reg2 : noreg),
1188                      next_test);
1189     // (Reg2 now contains the receiver from the CallData.)
1190 
1191     // The receiver is receiver[n].  Increment count[n].
1192     int count_offset = in_bytes(VirtualCallData::receiver_count_offset(row));
1193     increment_mdp_data_at(mdp, count_offset);
1194     jmp(done);
1195     bind(next_test);
1196 
1197     if (test_for_null_also) {
1198       Label found_null;
1199       // Failed the equality check on receiver[n]...  Test for null.
1200       testptr(reg2, reg2);
1201       if (start_row == last_row) {
1202         // The only thing left to do is handle the null case.
1203         if (is_virtual_call) {
1204           jccb(Assembler::zero, found_null);
1205           // Receiver did not match any saved receiver and there is no empty row for it.
1206           // Increment total counter to indicate polymorphic case.
1207           increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1208           jmp(done);
1209           bind(found_null);
1210         } else {
1211           jcc(Assembler::notZero, done);
1212         }
1213         break;
1214       }
1215       // Since null is rare, make it be the branch-taken case.
1216       jcc(Assembler::zero, found_null);
1217 
1218       // Put all the "Case 3" tests here.
1219       record_klass_in_profile_helper(receiver, mdp, reg2, start_row + 1, done, is_virtual_call);
1220 
1221       // Found a null.  Keep searching for a matching receiver,
1222       // but remember that this is an empty (unused) slot.
1223       bind(found_null);
1224     }
1225   }
1226 
1227   // In the fall-through case, we found no matching receiver, but we
1228   // observed the receiver[start_row] is NULL.
1229 
1230   // Fill in the receiver field and increment the count.
1231   int recvr_offset = in_bytes(VirtualCallData::receiver_offset(start_row));
1232   set_mdp_data_at(mdp, recvr_offset, receiver);
1233   int count_offset = in_bytes(VirtualCallData::receiver_count_offset(start_row));
1234   movl(reg2, DataLayout::counter_increment);
1235   set_mdp_data_at(mdp, count_offset, reg2);
1236   if (start_row > 0) {
1237     jmp(done);
1238   }
1239 }
1240 
1241 // Example state machine code for three profile rows:
1242 //   // main copy of decision tree, rooted at row[1]
1243 //   if (row[0].rec == rec) { row[0].incr(); goto done; }
1244 //   if (row[0].rec != NULL) {
1245 //     // inner copy of decision tree, rooted at row[1]
1246 //     if (row[1].rec == rec) { row[1].incr(); goto done; }
1247 //     if (row[1].rec != NULL) {
1248 //       // degenerate decision tree, rooted at row[2]
1249 //       if (row[2].rec == rec) { row[2].incr(); goto done; }
1250 //       if (row[2].rec != NULL) { count.incr(); goto done; } // overflow
1251 //       row[2].init(rec); goto done;
1252 //     } else {
1253 //       // remember row[1] is empty
1254 //       if (row[2].rec == rec) { row[2].incr(); goto done; }
1255 //       row[1].init(rec); goto done;
1256 //     }
1257 //   } else {
1258 //     // remember row[0] is empty
1259 //     if (row[1].rec == rec) { row[1].incr(); goto done; }
1260 //     if (row[2].rec == rec) { row[2].incr(); goto done; }
1261 //     row[0].init(rec); goto done;
1262 //   }
1263 //   done:
1264 
1265 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver,
1266                                                         Register mdp, Register reg2,
1267                                                         bool is_virtual_call) {
1268   assert(ProfileInterpreter, "must be profiling");
1269   Label done;
1270 
1271   record_klass_in_profile_helper(receiver, mdp, reg2, 0, done, is_virtual_call);
1272 
1273   bind (done);
1274 }
1275 
1276 void InterpreterMacroAssembler::profile_ret(Register return_bci,
1277                                             Register mdp) {
1278   if (ProfileInterpreter) {
1279     Label profile_continue;
1280     uint row;
1281 
1282     // If no method data exists, go to profile_continue.
1283     test_method_data_pointer(mdp, profile_continue);
1284 
1285     // Update the total ret count.
1286     increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1287 
1288     for (row = 0; row < RetData::row_limit(); row++) {
1289       Label next_test;
1290 
1291       // See if return_bci is equal to bci[n]:
1292       test_mdp_data_at(mdp,
1293                        in_bytes(RetData::bci_offset(row)),
1294                        return_bci, noreg,
1295                        next_test);
1296 
1297       // return_bci is equal to bci[n].  Increment the count.
1298       increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row)));
1299 
1300       // The method data pointer needs to be updated to reflect the new target.
1301       update_mdp_by_offset(mdp,
1302                            in_bytes(RetData::bci_displacement_offset(row)));
1303       jmp(profile_continue);
1304       bind(next_test);
1305     }
1306 
1307     update_mdp_for_ret(return_bci);
1308 
1309     bind(profile_continue);
1310   }
1311 }
1312 
1313 
1314 void InterpreterMacroAssembler::profile_null_seen(Register mdp) {
1315   if (ProfileInterpreter) {
1316     Label profile_continue;
1317 
1318     // If no method data exists, go to profile_continue.
1319     test_method_data_pointer(mdp, profile_continue);
1320 
1321     set_mdp_flag_at(mdp, BitData::null_seen_byte_constant());
1322 
1323     // The method data pointer needs to be updated.
1324     int mdp_delta = in_bytes(BitData::bit_data_size());
1325     if (TypeProfileCasts) {
1326       mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1327     }
1328     update_mdp_by_constant(mdp, mdp_delta);
1329 
1330     bind(profile_continue);
1331   }
1332 }
1333 
1334 
1335 void InterpreterMacroAssembler::profile_typecheck_failed(Register mdp) {
1336   if (ProfileInterpreter && TypeProfileCasts) {
1337     Label profile_continue;
1338 
1339     // If no method data exists, go to profile_continue.
1340     test_method_data_pointer(mdp, profile_continue);
1341 
1342     int count_offset = in_bytes(CounterData::count_offset());
1343     // Back up the address, since we have already bumped the mdp.
1344     count_offset -= in_bytes(VirtualCallData::virtual_call_data_size());
1345 
1346     // *Decrement* the counter.  We expect to see zero or small negatives.
1347     increment_mdp_data_at(mdp, count_offset, true);
1348 
1349     bind (profile_continue);
1350   }
1351 }
1352 
1353 
1354 void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) {
1355   if (ProfileInterpreter) {
1356     Label profile_continue;
1357 
1358     // If no method data exists, go to profile_continue.
1359     test_method_data_pointer(mdp, profile_continue);
1360 
1361     // The method data pointer needs to be updated.
1362     int mdp_delta = in_bytes(BitData::bit_data_size());
1363     if (TypeProfileCasts) {
1364       mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1365 
1366       // Record the object type.
1367       record_klass_in_profile(klass, mdp, reg2, false);
1368     }
1369     update_mdp_by_constant(mdp, mdp_delta);
1370 
1371     bind(profile_continue);
1372   }
1373 }
1374 
1375 
1376 void InterpreterMacroAssembler::profile_switch_default(Register mdp) {
1377   if (ProfileInterpreter) {
1378     Label profile_continue;
1379 
1380     // If no method data exists, go to profile_continue.
1381     test_method_data_pointer(mdp, profile_continue);
1382 
1383     // Update the default case count
1384     increment_mdp_data_at(mdp,
1385                           in_bytes(MultiBranchData::default_count_offset()));
1386 
1387     // The method data pointer needs to be updated.
1388     update_mdp_by_offset(mdp,
1389                          in_bytes(MultiBranchData::
1390                                   default_displacement_offset()));
1391 
1392     bind(profile_continue);
1393   }
1394 }
1395 
1396 
1397 void InterpreterMacroAssembler::profile_switch_case(Register index,
1398                                                     Register mdp,
1399                                                     Register reg2) {
1400   if (ProfileInterpreter) {
1401     Label profile_continue;
1402 
1403     // If no method data exists, go to profile_continue.
1404     test_method_data_pointer(mdp, profile_continue);
1405 
1406     // Build the base (index * per_case_size_in_bytes()) +
1407     // case_array_offset_in_bytes()
1408     movl(reg2, in_bytes(MultiBranchData::per_case_size()));
1409     imulptr(index, reg2); // XXX l ?
1410     addptr(index, in_bytes(MultiBranchData::case_array_offset())); // XXX l ?
1411 
1412     // Update the case count
1413     increment_mdp_data_at(mdp,
1414                           index,
1415                           in_bytes(MultiBranchData::relative_count_offset()));
1416 
1417     // The method data pointer needs to be updated.
1418     update_mdp_by_offset(mdp,
1419                          index,
1420                          in_bytes(MultiBranchData::
1421                                   relative_displacement_offset()));
1422 
1423     bind(profile_continue);
1424   }
1425 }
1426 
1427 
1428 
1429 void InterpreterMacroAssembler::verify_oop(Register reg, TosState state) {
1430   if (state == atos) {
1431     MacroAssembler::verify_oop(reg);
1432   }
1433 }
1434 
1435 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) {
1436 }
1437 #endif // !CC_INTERP
1438 
1439 
1440 void InterpreterMacroAssembler::notify_method_entry() {
1441   // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1442   // track stack depth.  If it is possible to enter interp_only_mode we add
1443   // the code to check if the event should be sent.
1444   if (JvmtiExport::can_post_interpreter_events()) {
1445     Label L;
1446     movl(rdx, Address(r15_thread, JavaThread::interp_only_mode_offset()));
1447     testl(rdx, rdx);
1448     jcc(Assembler::zero, L);
1449     call_VM(noreg, CAST_FROM_FN_PTR(address,
1450                                     InterpreterRuntime::post_method_entry));
1451     bind(L);
1452   }
1453 
1454   {
1455     SkipIfEqual skip(this, &DTraceMethodProbes, false);
1456     get_method(c_rarg1);
1457     call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
1458                  r15_thread, c_rarg1);
1459   }
1460 
1461   // RedefineClasses() tracing support for obsolete method entry
1462   if (RC_TRACE_IN_RANGE(0x00001000, 0x00002000)) {
1463     get_method(c_rarg1);
1464     call_VM_leaf(
1465       CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
1466       r15_thread, c_rarg1);
1467   }
1468 }
1469 
1470 
1471 void InterpreterMacroAssembler::notify_method_exit(
1472     TosState state, NotifyMethodExitMode mode) {
1473   // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1474   // track stack depth.  If it is possible to enter interp_only_mode we add
1475   // the code to check if the event should be sent.
1476   if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
1477     Label L;
1478     // Note: frame::interpreter_frame_result has a dependency on how the
1479     // method result is saved across the call to post_method_exit. If this
1480     // is changed then the interpreter_frame_result implementation will
1481     // need to be updated too.
1482 
1483     // For c++ interpreter the result is always stored at a known location in the frame
1484     // template interpreter will leave it on the top of the stack.
1485     NOT_CC_INTERP(push(state);)
1486     movl(rdx, Address(r15_thread, JavaThread::interp_only_mode_offset()));
1487     testl(rdx, rdx);
1488     jcc(Assembler::zero, L);
1489     call_VM(noreg,
1490             CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
1491     bind(L);
1492     NOT_CC_INTERP(pop(state));
1493   }
1494 
1495   {
1496     SkipIfEqual skip(this, &DTraceMethodProbes, false);
1497     NOT_CC_INTERP(push(state));
1498     get_method(c_rarg1);
1499     call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
1500                  r15_thread, c_rarg1);
1501     NOT_CC_INTERP(pop(state));
1502   }
1503 }
1504 
1505 // Jump if ((*counter_addr += increment) & mask) satisfies the condition.
1506 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr,
1507                                                         int increment, int mask,
1508                                                         Register scratch, bool preloaded,
1509                                                         Condition cond, Label* where) {
1510   if (!preloaded) {
1511     movl(scratch, counter_addr);
1512   }
1513   incrementl(scratch, increment);
1514   movl(counter_addr, scratch);
1515   andl(scratch, mask);
1516   jcc(cond, *where);
1517 }