1 /*
   2  * Copyright (c) 2003, 2013, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 #include "precompiled.hpp"
  26 #include "interp_masm_x86.hpp"
  27 #include "interpreter/interpreter.hpp"
  28 #include "interpreter/interpreterRuntime.hpp"
  29 #include "oops/arrayOop.hpp"
  30 #include "oops/markOop.hpp"
  31 #include "oops/methodData.hpp"
  32 #include "oops/method.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 #include "runtime/thread.inline.hpp"
  40 
  41 
  42 // Implementation of InterpreterMacroAssembler
  43 
  44 #ifdef CC_INTERP
  45 void InterpreterMacroAssembler::get_method(Register reg) {
  46   movptr(reg, Address(rbp, -((int)sizeof(BytecodeInterpreter) + 2 * wordSize)));
  47   movptr(reg, Address(reg, byte_offset_of(BytecodeInterpreter, _method)));
  48 }
  49 #endif // CC_INTERP
  50 
  51 #ifndef CC_INTERP
  52 
  53 void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point,
  54                                                   int number_of_arguments) {
  55   // interpreter specific
  56   //
  57   // Note: No need to save/restore bcp & locals (r13 & r14) pointer
  58   //       since these are callee saved registers and no blocking/
  59   //       GC can happen in leaf calls.
  60   // Further Note: DO NOT save/restore bcp/locals. If a caller has
  61   // already saved them so that it can use esi/edi as temporaries
  62   // then a save/restore here will DESTROY the copy the caller
  63   // saved! There used to be a save_bcp() that only happened in
  64   // the ASSERT path (no restore_bcp). Which caused bizarre failures
  65   // when jvm built with ASSERTs.
  66 #ifdef ASSERT
  67   {
  68     Label L;
  69     cmpptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
  70     jcc(Assembler::equal, L);
  71     stop("InterpreterMacroAssembler::call_VM_leaf_base:"
  72          " last_sp != NULL");
  73     bind(L);
  74   }
  75 #endif
  76   // super call
  77   MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments);
  78   // interpreter specific
  79   // Used to ASSERT that r13/r14 were equal to frame's bcp/locals
  80   // but since they may not have been saved (and we don't want to
  81   // save thme here (see note above) the assert is invalid.
  82 }
  83 
  84 void InterpreterMacroAssembler::call_VM_base(Register oop_result,
  85                                              Register java_thread,
  86                                              Register last_java_sp,
  87                                              address  entry_point,
  88                                              int      number_of_arguments,
  89                                              bool     check_exceptions) {
  90   // interpreter specific
  91   //
  92   // Note: Could avoid restoring locals ptr (callee saved) - however doesn't
  93   //       really make a difference for these runtime calls, since they are
  94   //       slow anyway. Btw., bcp must be saved/restored since it may change
  95   //       due to GC.
  96   // assert(java_thread == noreg , "not expecting a precomputed java thread");
  97   save_bcp();
  98 #ifdef ASSERT
  99   {
 100     Label L;
 101     cmpptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
 102     jcc(Assembler::equal, L);
 103     stop("InterpreterMacroAssembler::call_VM_leaf_base:"
 104          " last_sp != NULL");
 105     bind(L);
 106   }
 107 #endif /* ASSERT */
 108   // super call
 109   MacroAssembler::call_VM_base(oop_result, noreg, last_java_sp,
 110                                entry_point, number_of_arguments,
 111                                check_exceptions);
 112   // interpreter specific
 113   restore_bcp();
 114   restore_locals();
 115 }
 116 
 117 
 118 void InterpreterMacroAssembler::check_and_handle_popframe(Register java_thread) {
 119   if (JvmtiExport::can_pop_frame()) {
 120     Label L;
 121     // Initiate popframe handling only if it is not already being
 122     // processed.  If the flag has the popframe_processing bit set, it
 123     // means that this code is called *during* popframe handling - we
 124     // don't want to reenter.
 125     // This method is only called just after the call into the vm in
 126     // call_VM_base, so the arg registers are available.
 127     movl(c_rarg0, Address(r15_thread, JavaThread::popframe_condition_offset()));
 128     testl(c_rarg0, JavaThread::popframe_pending_bit);
 129     jcc(Assembler::zero, L);
 130     testl(c_rarg0, JavaThread::popframe_processing_bit);
 131     jcc(Assembler::notZero, L);
 132     // Call Interpreter::remove_activation_preserving_args_entry() to get the
 133     // address of the same-named entrypoint in the generated interpreter code.
 134     call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry));
 135     jmp(rax);
 136     bind(L);
 137   }
 138 }
 139 
 140 
 141 void InterpreterMacroAssembler::load_earlyret_value(TosState state) {
 142   movptr(rcx, Address(r15_thread, JavaThread::jvmti_thread_state_offset()));
 143   const Address tos_addr(rcx, JvmtiThreadState::earlyret_tos_offset());
 144   const Address oop_addr(rcx, JvmtiThreadState::earlyret_oop_offset());
 145   const Address val_addr(rcx, JvmtiThreadState::earlyret_value_offset());
 146   switch (state) {
 147     case atos: movptr(rax, oop_addr);
 148                movptr(oop_addr, (int32_t)NULL_WORD);
 149                verify_oop(rax, state);              break;
 150     case ltos: movptr(rax, val_addr);                 break;
 151     case btos:                                   // fall through
 152     case ctos:                                   // fall through
 153     case stos:                                   // fall through
 154     case itos: movl(rax, val_addr);                 break;
 155     case ftos: movflt(xmm0, val_addr);              break;
 156     case dtos: movdbl(xmm0, val_addr);              break;
 157     case vtos: /* nothing to do */                  break;
 158     default  : ShouldNotReachHere();
 159   }
 160   // Clean up tos value in the thread object
 161   movl(tos_addr,  (int) ilgl);
 162   movl(val_addr,  (int32_t) NULL_WORD);
 163 }
 164 
 165 
 166 void InterpreterMacroAssembler::check_and_handle_earlyret(Register java_thread) {
 167   if (JvmtiExport::can_force_early_return()) {
 168     Label L;
 169     movptr(c_rarg0, Address(r15_thread, JavaThread::jvmti_thread_state_offset()));
 170     testptr(c_rarg0, c_rarg0);
 171     jcc(Assembler::zero, L); // if (thread->jvmti_thread_state() == NULL) exit;
 172 
 173     // Initiate earlyret handling only if it is not already being processed.
 174     // If the flag has the earlyret_processing bit set, it means that this code
 175     // is called *during* earlyret handling - we don't want to reenter.
 176     movl(c_rarg0, Address(c_rarg0, JvmtiThreadState::earlyret_state_offset()));
 177     cmpl(c_rarg0, JvmtiThreadState::earlyret_pending);
 178     jcc(Assembler::notEqual, L);
 179 
 180     // Call Interpreter::remove_activation_early_entry() to get the address of the
 181     // same-named entrypoint in the generated interpreter code.
 182     movptr(c_rarg0, Address(r15_thread, JavaThread::jvmti_thread_state_offset()));
 183     movl(c_rarg0, Address(c_rarg0, JvmtiThreadState::earlyret_tos_offset()));
 184     call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), c_rarg0);
 185     jmp(rax);
 186     bind(L);
 187   }
 188 }
 189 
 190 
 191 void InterpreterMacroAssembler::get_unsigned_2_byte_index_at_bcp(
 192   Register reg,
 193   int bcp_offset) {
 194   assert(bcp_offset >= 0, "bcp is still pointing to start of bytecode");
 195   load_unsigned_short(reg, Address(r13, bcp_offset));
 196   bswapl(reg);
 197   shrl(reg, 16);
 198 }
 199 
 200 
 201 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register index,
 202                                                        int bcp_offset,
 203                                                        size_t index_size) {
 204   assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
 205   if (index_size == sizeof(u2)) {
 206     load_unsigned_short(index, Address(r13, bcp_offset));
 207   } else if (index_size == sizeof(u4)) {
 208     movl(index, Address(r13, bcp_offset));
 209     // Check if the secondary index definition is still ~x, otherwise
 210     // we have to change the following assembler code to calculate the
 211     // plain index.
 212     assert(ConstantPool::decode_invokedynamic_index(~123) == 123, "else change next line");
 213     notl(index);  // convert to plain index
 214   } else if (index_size == sizeof(u1)) {
 215     load_unsigned_byte(index, Address(r13, bcp_offset));
 216   } else {
 217     ShouldNotReachHere();
 218   }
 219 }
 220 
 221 
 222 void InterpreterMacroAssembler::get_cache_and_index_at_bcp(Register cache,
 223                                                            Register index,
 224                                                            int bcp_offset,
 225                                                            size_t index_size) {
 226   assert_different_registers(cache, index);
 227   get_cache_index_at_bcp(index, bcp_offset, index_size);
 228   movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
 229   assert(sizeof(ConstantPoolCacheEntry) == 4 * wordSize, "adjust code below");
 230   // convert from field index to ConstantPoolCacheEntry index
 231   assert(exact_log2(in_words(ConstantPoolCacheEntry::size())) == 2, "else change next line");
 232   shll(index, 2);
 233 }
 234 
 235 
 236 void InterpreterMacroAssembler::get_cache_and_index_and_bytecode_at_bcp(Register cache,
 237                                                                         Register index,
 238                                                                         Register bytecode,
 239                                                                         int byte_no,
 240                                                                         int bcp_offset,
 241                                                                         size_t index_size) {
 242   get_cache_and_index_at_bcp(cache, index, bcp_offset, index_size);
 243   // We use a 32-bit load here since the layout of 64-bit words on
 244   // little-endian machines allow us that.
 245   movl(bytecode, Address(cache, index, Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset()));
 246   const int shift_count = (1 + byte_no) * BitsPerByte;
 247   assert((byte_no == TemplateTable::f1_byte && shift_count == ConstantPoolCacheEntry::bytecode_1_shift) ||
 248          (byte_no == TemplateTable::f2_byte && shift_count == ConstantPoolCacheEntry::bytecode_2_shift),
 249          "correct shift count");
 250   shrl(bytecode, shift_count);
 251   assert(ConstantPoolCacheEntry::bytecode_1_mask == ConstantPoolCacheEntry::bytecode_2_mask, "common mask");
 252   andl(bytecode, ConstantPoolCacheEntry::bytecode_1_mask);
 253 }
 254 
 255 
 256 void InterpreterMacroAssembler::get_cache_entry_pointer_at_bcp(Register cache,
 257                                                                Register tmp,
 258                                                                int bcp_offset,
 259                                                                size_t index_size) {
 260   assert(cache != tmp, "must use different register");
 261   get_cache_index_at_bcp(tmp, bcp_offset, index_size);
 262   assert(sizeof(ConstantPoolCacheEntry) == 4 * wordSize, "adjust code below");
 263   // convert from field index to ConstantPoolCacheEntry index
 264   // and from word offset to byte offset
 265   assert(exact_log2(in_bytes(ConstantPoolCacheEntry::size_in_bytes())) == 2 + LogBytesPerWord, "else change next line");
 266   shll(tmp, 2 + LogBytesPerWord);
 267   movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
 268   // skip past the header
 269   addptr(cache, in_bytes(ConstantPoolCache::base_offset()));
 270   addptr(cache, tmp);  // construct pointer to cache entry
 271 }
 272 
 273 // Load object from cpool->resolved_references(index)
 274 void InterpreterMacroAssembler::load_resolved_reference_at_index(
 275                                            Register result, Register index) {
 276   assert_different_registers(result, index);
 277   // convert from field index to resolved_references() index and from
 278   // word index to byte offset. Since this is a java object, it can be compressed
 279   Register tmp = index;  // reuse
 280   shll(tmp, LogBytesPerHeapOop);
 281 
 282   get_constant_pool(result);
 283   // load pointer for resolved_references[] objArray
 284   movptr(result, Address(result, ConstantPool::resolved_references_offset_in_bytes()));
 285   // JNIHandles::resolve(obj);
 286   movptr(result, Address(result, 0));
 287   // Add in the index
 288   addptr(result, tmp);
 289   load_heap_oop(result, Address(result, arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
 290 }
 291 
 292 // Generate a subtype check: branch to ok_is_subtype if sub_klass is a
 293 // subtype of super_klass.
 294 //
 295 // Args:
 296 //      rax: superklass
 297 //      Rsub_klass: subklass
 298 //
 299 // Kills:
 300 //      rcx, rdi
 301 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass,
 302                                                   Label& ok_is_subtype) {
 303   assert(Rsub_klass != rax, "rax holds superklass");
 304   assert(Rsub_klass != r14, "r14 holds locals");
 305   assert(Rsub_klass != r13, "r13 holds bcp");
 306   assert(Rsub_klass != rcx, "rcx holds 2ndary super array length");
 307   assert(Rsub_klass != rdi, "rdi holds 2ndary super array scan ptr");
 308 
 309   // Profile the not-null value's klass.
 310   profile_typecheck(rcx, Rsub_klass, rdi); // blows rcx, reloads rdi
 311 
 312   // Do the check.
 313   check_klass_subtype(Rsub_klass, rax, rcx, ok_is_subtype); // blows rcx
 314 
 315   // Profile the failure of the check.
 316   profile_typecheck_failed(rcx); // blows rcx
 317 }
 318 
 319 
 320 
 321 // Java Expression Stack
 322 
 323 void InterpreterMacroAssembler::pop_ptr(Register r) {
 324   pop(r);
 325 }
 326 
 327 void InterpreterMacroAssembler::pop_i(Register r) {
 328   // XXX can't use pop currently, upper half non clean
 329   movl(r, Address(rsp, 0));
 330   addptr(rsp, wordSize);
 331 }
 332 
 333 void InterpreterMacroAssembler::pop_l(Register r) {
 334   movq(r, Address(rsp, 0));
 335   addptr(rsp, 2 * Interpreter::stackElementSize);
 336 }
 337 
 338 void InterpreterMacroAssembler::pop_f(XMMRegister r) {
 339   movflt(r, Address(rsp, 0));
 340   addptr(rsp, wordSize);
 341 }
 342 
 343 void InterpreterMacroAssembler::pop_d(XMMRegister r) {
 344   movdbl(r, Address(rsp, 0));
 345   addptr(rsp, 2 * Interpreter::stackElementSize);
 346 }
 347 
 348 void InterpreterMacroAssembler::push_ptr(Register r) {
 349   push(r);
 350 }
 351 
 352 void InterpreterMacroAssembler::push_i(Register r) {
 353   push(r);
 354 }
 355 
 356 void InterpreterMacroAssembler::push_l(Register r) {
 357   subptr(rsp, 2 * wordSize);
 358   movq(Address(rsp, 0), r);
 359 }
 360 
 361 void InterpreterMacroAssembler::push_f(XMMRegister r) {
 362   subptr(rsp, wordSize);
 363   movflt(Address(rsp, 0), r);
 364 }
 365 
 366 void InterpreterMacroAssembler::push_d(XMMRegister r) {
 367   subptr(rsp, 2 * wordSize);
 368   movdbl(Address(rsp, 0), r);
 369 }
 370 
 371 void InterpreterMacroAssembler::pop(TosState state) {
 372   switch (state) {
 373   case atos: pop_ptr();                 break;
 374   case btos:
 375   case ctos:
 376   case stos:
 377   case itos: pop_i();                   break;
 378   case ltos: pop_l();                   break;
 379   case ftos: pop_f();                   break;
 380   case dtos: pop_d();                   break;
 381   case vtos: /* nothing to do */        break;
 382   default:   ShouldNotReachHere();
 383   }
 384   verify_oop(rax, state);
 385 }
 386 
 387 void InterpreterMacroAssembler::push(TosState state) {
 388   verify_oop(rax, state);
 389   switch (state) {
 390   case atos: push_ptr();                break;
 391   case btos:
 392   case ctos:
 393   case stos:
 394   case itos: push_i();                  break;
 395   case ltos: push_l();                  break;
 396   case ftos: push_f();                  break;
 397   case dtos: push_d();                  break;
 398   case vtos: /* nothing to do */        break;
 399   default  : ShouldNotReachHere();
 400   }
 401 }
 402 
 403 
 404 // Helpers for swap and dup
 405 void InterpreterMacroAssembler::load_ptr(int n, Register val) {
 406   movptr(val, Address(rsp, Interpreter::expr_offset_in_bytes(n)));
 407 }
 408 
 409 void InterpreterMacroAssembler::store_ptr(int n, Register val) {
 410   movptr(Address(rsp, Interpreter::expr_offset_in_bytes(n)), val);
 411 }
 412 
 413 
 414 void InterpreterMacroAssembler::prepare_to_jump_from_interpreted() {
 415   // set sender sp
 416   lea(r13, Address(rsp, wordSize));
 417   // record last_sp
 418   movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), r13);
 419 }
 420 
 421 
 422 // Jump to from_interpreted entry of a call unless single stepping is possible
 423 // in this thread in which case we must call the i2i entry
 424 void InterpreterMacroAssembler::jump_from_interpreted(Register method, Register temp) {
 425   prepare_to_jump_from_interpreted();
 426 
 427   if (JvmtiExport::can_post_interpreter_events()) {
 428     Label run_compiled_code;
 429     // JVMTI events, such as single-stepping, are implemented partly by avoiding running
 430     // compiled code in threads for which the event is enabled.  Check here for
 431     // interp_only_mode if these events CAN be enabled.
 432     // interp_only is an int, on little endian it is sufficient to test the byte only
 433     // Is a cmpl faster?
 434     cmpb(Address(r15_thread, JavaThread::interp_only_mode_offset()), 0);
 435     jccb(Assembler::zero, run_compiled_code);
 436     jmp(Address(method, Method::interpreter_entry_offset()));
 437     bind(run_compiled_code);
 438   }
 439 
 440   jmp(Address(method, Method::from_interpreted_offset()));
 441 
 442 }
 443 
 444 
 445 // The following two routines provide a hook so that an implementation
 446 // can schedule the dispatch in two parts.  amd64 does not do this.
 447 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int step) {
 448   // Nothing amd64 specific to be done here
 449 }
 450 
 451 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int step) {
 452   dispatch_next(state, step);
 453 }
 454 
 455 void InterpreterMacroAssembler::dispatch_base(TosState state,
 456                                               address* table,
 457                                               bool verifyoop) {
 458   verify_FPU(1, state);
 459   if (VerifyActivationFrameSize) {
 460     Label L;
 461     mov(rcx, rbp);
 462     subptr(rcx, rsp);
 463     int32_t min_frame_size =
 464       (frame::link_offset - frame::interpreter_frame_initial_sp_offset) *
 465       wordSize;
 466     cmpptr(rcx, (int32_t)min_frame_size);
 467     jcc(Assembler::greaterEqual, L);
 468     stop("broken stack frame");
 469     bind(L);
 470   }
 471   if (verifyoop) {
 472     verify_oop(rax, state);
 473   }
 474   lea(rscratch1, ExternalAddress((address)table));
 475   jmp(Address(rscratch1, rbx, Address::times_8));
 476 }
 477 
 478 void InterpreterMacroAssembler::dispatch_only(TosState state) {
 479   dispatch_base(state, Interpreter::dispatch_table(state));
 480 }
 481 
 482 void InterpreterMacroAssembler::dispatch_only_normal(TosState state) {
 483   dispatch_base(state, Interpreter::normal_table(state));
 484 }
 485 
 486 void InterpreterMacroAssembler::dispatch_only_noverify(TosState state) {
 487   dispatch_base(state, Interpreter::normal_table(state), false);
 488 }
 489 
 490 
 491 void InterpreterMacroAssembler::dispatch_next(TosState state, int step) {
 492   // load next bytecode (load before advancing r13 to prevent AGI)
 493   load_unsigned_byte(rbx, Address(r13, step));
 494   // advance r13
 495   increment(r13, step);
 496   dispatch_base(state, Interpreter::dispatch_table(state));
 497 }
 498 
 499 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) {
 500   // load current bytecode
 501   load_unsigned_byte(rbx, Address(r13, 0));
 502   dispatch_base(state, table);
 503 }
 504 
 505 // remove activation
 506 //
 507 // Unlock the receiver if this is a synchronized method.
 508 // Unlock any Java monitors from syncronized blocks.
 509 // Remove the activation from the stack.
 510 //
 511 // If there are locked Java monitors
 512 //    If throw_monitor_exception
 513 //       throws IllegalMonitorStateException
 514 //    Else if install_monitor_exception
 515 //       installs IllegalMonitorStateException
 516 //    Else
 517 //       no error processing
 518 void InterpreterMacroAssembler::remove_activation(
 519         TosState state,
 520         Register ret_addr,
 521         bool throw_monitor_exception,
 522         bool install_monitor_exception,
 523         bool notify_jvmdi) {
 524   // Note: Registers rdx xmm0 may be in use for the
 525   // result check if synchronized method
 526   Label unlocked, unlock, no_unlock;
 527 
 528   // get the value of _do_not_unlock_if_synchronized into rdx
 529   const Address do_not_unlock_if_synchronized(r15_thread,
 530     in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
 531   movbool(rdx, do_not_unlock_if_synchronized);
 532   movbool(do_not_unlock_if_synchronized, false); // reset the flag
 533 
 534  // get method access flags
 535   movptr(rbx, Address(rbp, frame::interpreter_frame_method_offset * wordSize));
 536   movl(rcx, Address(rbx, Method::access_flags_offset()));
 537   testl(rcx, JVM_ACC_SYNCHRONIZED);
 538   jcc(Assembler::zero, unlocked);
 539 
 540   // Don't unlock anything if the _do_not_unlock_if_synchronized flag
 541   // is set.
 542   testbool(rdx);
 543   jcc(Assembler::notZero, no_unlock);
 544 
 545   // unlock monitor
 546   push(state); // save result
 547 
 548   // BasicObjectLock will be first in list, since this is a
 549   // synchronized method. However, need to check that the object has
 550   // not been unlocked by an explicit monitorexit bytecode.
 551   const Address monitor(rbp, frame::interpreter_frame_initial_sp_offset *
 552                         wordSize - (int) sizeof(BasicObjectLock));
 553   // We use c_rarg1 so that if we go slow path it will be the correct
 554   // register for unlock_object to pass to VM directly
 555   lea(c_rarg1, monitor); // address of first monitor
 556 
 557   movptr(rax, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
 558   testptr(rax, rax);
 559   jcc(Assembler::notZero, unlock);
 560 
 561   pop(state);
 562   if (throw_monitor_exception) {
 563     // Entry already unlocked, need to throw exception
 564     call_VM(noreg, CAST_FROM_FN_PTR(address,
 565                    InterpreterRuntime::throw_illegal_monitor_state_exception));
 566     should_not_reach_here();
 567   } else {
 568     // Monitor already unlocked during a stack unroll. If requested,
 569     // install an illegal_monitor_state_exception.  Continue with
 570     // stack unrolling.
 571     if (install_monitor_exception) {
 572       call_VM(noreg, CAST_FROM_FN_PTR(address,
 573                      InterpreterRuntime::new_illegal_monitor_state_exception));
 574     }
 575     jmp(unlocked);
 576   }
 577 
 578   bind(unlock);
 579   unlock_object(c_rarg1);
 580   pop(state);
 581 
 582   // Check that for block-structured locking (i.e., that all locked
 583   // objects has been unlocked)
 584   bind(unlocked);
 585 
 586   // rax: Might contain return value
 587 
 588   // Check that all monitors are unlocked
 589   {
 590     Label loop, exception, entry, restart;
 591     const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
 592     const Address monitor_block_top(
 593         rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
 594     const Address monitor_block_bot(
 595         rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
 596 
 597     bind(restart);
 598     // We use c_rarg1 so that if we go slow path it will be the correct
 599     // register for unlock_object to pass to VM directly
 600     movptr(c_rarg1, monitor_block_top); // points to current entry, starting
 601                                   // with top-most entry
 602     lea(rbx, monitor_block_bot);  // points to word before bottom of
 603                                   // monitor block
 604     jmp(entry);
 605 
 606     // Entry already locked, need to throw exception
 607     bind(exception);
 608 
 609     if (throw_monitor_exception) {
 610       // Throw exception
 611       MacroAssembler::call_VM(noreg,
 612                               CAST_FROM_FN_PTR(address, InterpreterRuntime::
 613                                    throw_illegal_monitor_state_exception));
 614       should_not_reach_here();
 615     } else {
 616       // Stack unrolling. Unlock object and install illegal_monitor_exception.
 617       // Unlock does not block, so don't have to worry about the frame.
 618       // We don't have to preserve c_rarg1 since we are going to throw an exception.
 619 
 620       push(state);
 621       unlock_object(c_rarg1);
 622       pop(state);
 623 
 624       if (install_monitor_exception) {
 625         call_VM(noreg, CAST_FROM_FN_PTR(address,
 626                                         InterpreterRuntime::
 627                                         new_illegal_monitor_state_exception));
 628       }
 629 
 630       jmp(restart);
 631     }
 632 
 633     bind(loop);
 634     // check if current entry is used
 635     cmpptr(Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()), (int32_t) NULL);
 636     jcc(Assembler::notEqual, exception);
 637 
 638     addptr(c_rarg1, entry_size); // otherwise advance to next entry
 639     bind(entry);
 640     cmpptr(c_rarg1, rbx); // check if bottom reached
 641     jcc(Assembler::notEqual, loop); // if not at bottom then check this entry
 642   }
 643 
 644   bind(no_unlock);
 645 
 646   // jvmti support
 647   if (notify_jvmdi) {
 648     notify_method_exit(state, NotifyJVMTI);    // preserve TOSCA
 649   } else {
 650     notify_method_exit(state, SkipNotifyJVMTI); // preserve TOSCA
 651   }
 652 
 653   // remove activation
 654   // get sender sp
 655   movptr(rbx,
 656          Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize));
 657   leave();                           // remove frame anchor
 658   pop(ret_addr);                     // get return address
 659   mov(rsp, rbx);                     // set sp to sender sp
 660 }
 661 
 662 #endif // C_INTERP
 663 
 664 void InterpreterMacroAssembler::get_method_counters(Register method,
 665                                                     Register mcs, Label& skip) {
 666   Label has_counters;
 667   movptr(mcs, Address(method, Method::method_counters_offset()));
 668   testptr(mcs, mcs);
 669   jcc(Assembler::notZero, has_counters);
 670   call_VM(noreg, CAST_FROM_FN_PTR(address,
 671           InterpreterRuntime::build_method_counters), method);
 672   movptr(mcs, Address(method,Method::method_counters_offset()));
 673   testptr(mcs, mcs);
 674   jcc(Assembler::zero, skip); // No MethodCounters allocated, OutOfMemory
 675   bind(has_counters);
 676 }
 677 
 678 
 679 // Lock object
 680 //
 681 // Args:
 682 //      c_rarg1: BasicObjectLock to be used for locking
 683 //
 684 // Kills:
 685 //      rax
 686 //      c_rarg0, c_rarg1, c_rarg2, c_rarg3, .. (param regs)
 687 //      rscratch1, rscratch2 (scratch regs)
 688 void InterpreterMacroAssembler::lock_object(Register lock_reg) {
 689   assert(lock_reg == c_rarg1, "The argument is only for looks. It must be c_rarg1");
 690 
 691   if (UseHeavyMonitors) {
 692     call_VM(noreg,
 693             CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
 694             lock_reg);
 695   } else {
 696     Label done;
 697 
 698     const Register swap_reg = rax; // Must use rax for cmpxchg instruction
 699     const Register obj_reg = c_rarg3; // Will contain the oop
 700 
 701     const int obj_offset = BasicObjectLock::obj_offset_in_bytes();
 702     const int lock_offset = BasicObjectLock::lock_offset_in_bytes ();
 703     const int mark_offset = lock_offset +
 704                             BasicLock::displaced_header_offset_in_bytes();
 705 
 706     Label slow_case;
 707 
 708     // Load object pointer into obj_reg %c_rarg3
 709     movptr(obj_reg, Address(lock_reg, obj_offset));
 710 
 711     if (UseBiasedLocking) {
 712       biased_locking_enter(lock_reg, obj_reg, swap_reg, rscratch1, false, done, &slow_case);
 713     }
 714 
 715     // Load immediate 1 into swap_reg %rax
 716     movl(swap_reg, 1);
 717 
 718     // Load (object->mark() | 1) into swap_reg %rax
 719     orptr(swap_reg, Address(obj_reg, 0));
 720 
 721     // Save (object->mark() | 1) into BasicLock's displaced header
 722     movptr(Address(lock_reg, mark_offset), swap_reg);
 723 
 724     assert(lock_offset == 0,
 725            "displached header must be first word in BasicObjectLock");
 726 
 727     if (os::is_MP()) lock();
 728     cmpxchgptr(lock_reg, Address(obj_reg, 0));
 729     if (PrintBiasedLockingStatistics) {
 730       cond_inc32(Assembler::zero,
 731                  ExternalAddress((address) BiasedLocking::fast_path_entry_count_addr()));
 732     }
 733     jcc(Assembler::zero, done);
 734 
 735     // Test if the oopMark is an obvious stack pointer, i.e.,
 736     //  1) (mark & 7) == 0, and
 737     //  2) rsp <= mark < mark + os::pagesize()
 738     //
 739     // These 3 tests can be done by evaluating the following
 740     // expression: ((mark - rsp) & (7 - os::vm_page_size())),
 741     // assuming both stack pointer and pagesize have their
 742     // least significant 3 bits clear.
 743     // NOTE: the oopMark is in swap_reg %rax as the result of cmpxchg
 744     subptr(swap_reg, rsp);
 745     andptr(swap_reg, 7 - os::vm_page_size());
 746 
 747     // Save the test result, for recursive case, the result is zero
 748     movptr(Address(lock_reg, mark_offset), swap_reg);
 749 
 750     if (PrintBiasedLockingStatistics) {
 751       cond_inc32(Assembler::zero,
 752                  ExternalAddress((address) BiasedLocking::fast_path_entry_count_addr()));
 753     }
 754     jcc(Assembler::zero, done);
 755 
 756     bind(slow_case);
 757 
 758     // Call the runtime routine for slow case
 759     call_VM(noreg,
 760             CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
 761             lock_reg);
 762 
 763     bind(done);
 764   }
 765 }
 766 
 767 
 768 // Unlocks an object. Used in monitorexit bytecode and
 769 // remove_activation.  Throws an IllegalMonitorException if object is
 770 // not locked by current thread.
 771 //
 772 // Args:
 773 //      c_rarg1: BasicObjectLock for lock
 774 //
 775 // Kills:
 776 //      rax
 777 //      c_rarg0, c_rarg1, c_rarg2, c_rarg3, ... (param regs)
 778 //      rscratch1, rscratch2 (scratch regs)
 779 void InterpreterMacroAssembler::unlock_object(Register lock_reg) {
 780   assert(lock_reg == c_rarg1, "The argument is only for looks. It must be rarg1");
 781 
 782   if (UseHeavyMonitors) {
 783     call_VM(noreg,
 784             CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit),
 785             lock_reg);
 786   } else {
 787     Label done;
 788 
 789     const Register swap_reg   = rax;  // Must use rax for cmpxchg instruction
 790     const Register header_reg = c_rarg2;  // Will contain the old oopMark
 791     const Register obj_reg    = c_rarg3;  // Will contain the oop
 792 
 793     save_bcp(); // Save in case of exception
 794 
 795     // Convert from BasicObjectLock structure to object and BasicLock
 796     // structure Store the BasicLock address into %rax
 797     lea(swap_reg, Address(lock_reg, BasicObjectLock::lock_offset_in_bytes()));
 798 
 799     // Load oop into obj_reg(%c_rarg3)
 800     movptr(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()));
 801 
 802     // Free entry
 803     movptr(Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()), (int32_t)NULL_WORD);
 804 
 805     if (UseBiasedLocking) {
 806       biased_locking_exit(obj_reg, header_reg, done);
 807     }
 808 
 809     // Load the old header from BasicLock structure
 810     movptr(header_reg, Address(swap_reg,
 811                                BasicLock::displaced_header_offset_in_bytes()));
 812 
 813     // Test for recursion
 814     testptr(header_reg, header_reg);
 815 
 816     // zero for recursive case
 817     jcc(Assembler::zero, done);
 818 
 819     // Atomic swap back the old header
 820     if (os::is_MP()) lock();
 821     cmpxchgptr(header_reg, Address(obj_reg, 0));
 822 
 823     // zero for recursive case
 824     jcc(Assembler::zero, done);
 825 
 826     // Call the runtime routine for slow case.
 827     movptr(Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()),
 828          obj_reg); // restore obj
 829     call_VM(noreg,
 830             CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit),
 831             lock_reg);
 832 
 833     bind(done);
 834 
 835     restore_bcp();
 836   }
 837 }
 838 
 839 #ifndef CC_INTERP
 840 
 841 void InterpreterMacroAssembler::test_method_data_pointer(Register mdp,
 842                                                          Label& zero_continue) {
 843   assert(ProfileInterpreter, "must be profiling interpreter");
 844   movptr(mdp, Address(rbp, frame::interpreter_frame_mdx_offset * wordSize));
 845   testptr(mdp, mdp);
 846   jcc(Assembler::zero, zero_continue);
 847 }
 848 
 849 
 850 // Set the method data pointer for the current bcp.
 851 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
 852   assert(ProfileInterpreter, "must be profiling interpreter");
 853   Label set_mdp;
 854   push(rax);
 855   push(rbx);
 856 
 857   get_method(rbx);
 858   // Test MDO to avoid the call if it is NULL.
 859   movptr(rax, Address(rbx, in_bytes(Method::method_data_offset())));
 860   testptr(rax, rax);
 861   jcc(Assembler::zero, set_mdp);
 862   // rbx: method
 863   // r13: bcp
 864   call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), rbx, r13);
 865   // rax: mdi
 866   // mdo is guaranteed to be non-zero here, we checked for it before the call.
 867   movptr(rbx, Address(rbx, in_bytes(Method::method_data_offset())));
 868   addptr(rbx, in_bytes(MethodData::data_offset()));
 869   addptr(rax, rbx);
 870   bind(set_mdp);
 871   movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), rax);
 872   pop(rbx);
 873   pop(rax);
 874 }
 875 
 876 void InterpreterMacroAssembler::verify_method_data_pointer() {
 877   assert(ProfileInterpreter, "must be profiling interpreter");
 878 #ifdef ASSERT
 879   Label verify_continue;
 880   push(rax);
 881   push(rbx);
 882   push(c_rarg3);
 883   push(c_rarg2);
 884   test_method_data_pointer(c_rarg3, verify_continue); // If mdp is zero, continue
 885   get_method(rbx);
 886 
 887   // If the mdp is valid, it will point to a DataLayout header which is
 888   // consistent with the bcp.  The converse is highly probable also.
 889   load_unsigned_short(c_rarg2,
 890                       Address(c_rarg3, in_bytes(DataLayout::bci_offset())));
 891   addptr(c_rarg2, Address(rbx, Method::const_offset()));
 892   lea(c_rarg2, Address(c_rarg2, ConstMethod::codes_offset()));
 893   cmpptr(c_rarg2, r13);
 894   jcc(Assembler::equal, verify_continue);
 895   // rbx: method
 896   // r13: bcp
 897   // c_rarg3: mdp
 898   call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp),
 899                rbx, r13, c_rarg3);
 900   bind(verify_continue);
 901   pop(c_rarg2);
 902   pop(c_rarg3);
 903   pop(rbx);
 904   pop(rax);
 905 #endif // ASSERT
 906 }
 907 
 908 
 909 void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in,
 910                                                 int constant,
 911                                                 Register value) {
 912   assert(ProfileInterpreter, "must be profiling interpreter");
 913   Address data(mdp_in, constant);
 914   movptr(data, value);
 915 }
 916 
 917 
 918 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
 919                                                       int constant,
 920                                                       bool decrement) {
 921   // Counter address
 922   Address data(mdp_in, constant);
 923 
 924   increment_mdp_data_at(data, decrement);
 925 }
 926 
 927 void InterpreterMacroAssembler::increment_mdp_data_at(Address data,
 928                                                       bool decrement) {
 929   assert(ProfileInterpreter, "must be profiling interpreter");
 930   // %%% this does 64bit counters at best it is wasting space
 931   // at worst it is a rare bug when counters overflow
 932 
 933   if (decrement) {
 934     // Decrement the register.  Set condition codes.
 935     addptr(data, (int32_t) -DataLayout::counter_increment);
 936     // If the decrement causes the counter to overflow, stay negative
 937     Label L;
 938     jcc(Assembler::negative, L);
 939     addptr(data, (int32_t) DataLayout::counter_increment);
 940     bind(L);
 941   } else {
 942     assert(DataLayout::counter_increment == 1,
 943            "flow-free idiom only works with 1");
 944     // Increment the register.  Set carry flag.
 945     addptr(data, DataLayout::counter_increment);
 946     // If the increment causes the counter to overflow, pull back by 1.
 947     sbbptr(data, (int32_t)0);
 948   }
 949 }
 950 
 951 
 952 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
 953                                                       Register reg,
 954                                                       int constant,
 955                                                       bool decrement) {
 956   Address data(mdp_in, reg, Address::times_1, constant);
 957 
 958   increment_mdp_data_at(data, decrement);
 959 }
 960 
 961 void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in,
 962                                                 int flag_byte_constant) {
 963   assert(ProfileInterpreter, "must be profiling interpreter");
 964   int header_offset = in_bytes(DataLayout::header_offset());
 965   int header_bits = DataLayout::flag_mask_to_header_mask(flag_byte_constant);
 966   // Set the flag
 967   orl(Address(mdp_in, header_offset), header_bits);
 968 }
 969 
 970 
 971 
 972 void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in,
 973                                                  int offset,
 974                                                  Register value,
 975                                                  Register test_value_out,
 976                                                  Label& not_equal_continue) {
 977   assert(ProfileInterpreter, "must be profiling interpreter");
 978   if (test_value_out == noreg) {
 979     cmpptr(value, Address(mdp_in, offset));
 980   } else {
 981     // Put the test value into a register, so caller can use it:
 982     movptr(test_value_out, Address(mdp_in, offset));
 983     cmpptr(test_value_out, value);
 984   }
 985   jcc(Assembler::notEqual, not_equal_continue);
 986 }
 987 
 988 
 989 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
 990                                                      int offset_of_disp) {
 991   assert(ProfileInterpreter, "must be profiling interpreter");
 992   Address disp_address(mdp_in, offset_of_disp);
 993   addptr(mdp_in, disp_address);
 994   movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), mdp_in);
 995 }
 996 
 997 
 998 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
 999                                                      Register reg,
1000                                                      int offset_of_disp) {
1001   assert(ProfileInterpreter, "must be profiling interpreter");
1002   Address disp_address(mdp_in, reg, Address::times_1, 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_constant(Register mdp_in,
1009                                                        int constant) {
1010   assert(ProfileInterpreter, "must be profiling interpreter");
1011   addptr(mdp_in, constant);
1012   movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), mdp_in);
1013 }
1014 
1015 
1016 void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) {
1017   assert(ProfileInterpreter, "must be profiling interpreter");
1018   push(return_bci); // save/restore across call_VM
1019   call_VM(noreg,
1020           CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret),
1021           return_bci);
1022   pop(return_bci);
1023 }
1024 
1025 
1026 void InterpreterMacroAssembler::profile_taken_branch(Register mdp,
1027                                                      Register bumped_count) {
1028   if (ProfileInterpreter) {
1029     Label profile_continue;
1030 
1031     // If no method data exists, go to profile_continue.
1032     // Otherwise, assign to mdp
1033     test_method_data_pointer(mdp, profile_continue);
1034 
1035     // We are taking a branch.  Increment the taken count.
1036     // We inline increment_mdp_data_at to return bumped_count in a register
1037     //increment_mdp_data_at(mdp, in_bytes(JumpData::taken_offset()));
1038     Address data(mdp, in_bytes(JumpData::taken_offset()));
1039     movptr(bumped_count, data);
1040     assert(DataLayout::counter_increment == 1,
1041             "flow-free idiom only works with 1");
1042     addptr(bumped_count, DataLayout::counter_increment);
1043     sbbptr(bumped_count, 0);
1044     movptr(data, bumped_count); // Store back out
1045 
1046     // The method data pointer needs to be updated to reflect the new target.
1047     update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset()));
1048     bind(profile_continue);
1049   }
1050 }
1051 
1052 
1053 void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp) {
1054   if (ProfileInterpreter) {
1055     Label profile_continue;
1056 
1057     // If no method data exists, go to profile_continue.
1058     test_method_data_pointer(mdp, profile_continue);
1059 
1060     // We are taking a branch.  Increment the not taken count.
1061     increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset()));
1062 
1063     // The method data pointer needs to be updated to correspond to
1064     // the next bytecode
1065     update_mdp_by_constant(mdp, in_bytes(BranchData::branch_data_size()));
1066     bind(profile_continue);
1067   }
1068 }
1069 
1070 void InterpreterMacroAssembler::profile_call(Register mdp) {
1071   if (ProfileInterpreter) {
1072     Label profile_continue;
1073 
1074     // If no method data exists, go to profile_continue.
1075     test_method_data_pointer(mdp, profile_continue);
1076 
1077     // We are making a call.  Increment the count.
1078     increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1079 
1080     // The method data pointer needs to be updated to reflect the new target.
1081     update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size()));
1082     bind(profile_continue);
1083   }
1084 }
1085 
1086 
1087 void InterpreterMacroAssembler::profile_final_call(Register mdp) {
1088   if (ProfileInterpreter) {
1089     Label profile_continue;
1090 
1091     // If no method data exists, go to profile_continue.
1092     test_method_data_pointer(mdp, profile_continue);
1093 
1094     // We are making a call.  Increment the count.
1095     increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1096 
1097     // The method data pointer needs to be updated to reflect the new target.
1098     update_mdp_by_constant(mdp,
1099                            in_bytes(VirtualCallData::
1100                                     virtual_call_data_size()));
1101     bind(profile_continue);
1102   }
1103 }
1104 
1105 
1106 void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
1107                                                      Register mdp,
1108                                                      Register reg2,
1109                                                      bool receiver_can_be_null) {
1110   if (ProfileInterpreter) {
1111     Label profile_continue;
1112 
1113     // If no method data exists, go to profile_continue.
1114     test_method_data_pointer(mdp, profile_continue);
1115 
1116     Label skip_receiver_profile;
1117     if (receiver_can_be_null) {
1118       Label not_null;
1119       testptr(receiver, receiver);
1120       jccb(Assembler::notZero, not_null);
1121       // We are making a call.  Increment the count for null receiver.
1122       increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1123       jmp(skip_receiver_profile);
1124       bind(not_null);
1125     }
1126 
1127     // Record the receiver type.
1128     record_klass_in_profile(receiver, mdp, reg2, true);
1129     bind(skip_receiver_profile);
1130 
1131     // The method data pointer needs to be updated to reflect the new target.
1132     update_mdp_by_constant(mdp,
1133                            in_bytes(VirtualCallData::
1134                                     virtual_call_data_size()));
1135     bind(profile_continue);
1136   }
1137 }
1138 
1139 // This routine creates a state machine for updating the multi-row
1140 // type profile at a virtual call site (or other type-sensitive bytecode).
1141 // The machine visits each row (of receiver/count) until the receiver type
1142 // is found, or until it runs out of rows.  At the same time, it remembers
1143 // the location of the first empty row.  (An empty row records null for its
1144 // receiver, and can be allocated for a newly-observed receiver type.)
1145 // Because there are two degrees of freedom in the state, a simple linear
1146 // search will not work; it must be a decision tree.  Hence this helper
1147 // function is recursive, to generate the required tree structured code.
1148 // It's the interpreter, so we are trading off code space for speed.
1149 // See below for example code.
1150 void InterpreterMacroAssembler::record_klass_in_profile_helper(
1151                                         Register receiver, Register mdp,
1152                                         Register reg2, int start_row,
1153                                         Label& done, bool is_virtual_call) {
1154   if (TypeProfileWidth == 0) {
1155     if (is_virtual_call) {
1156       increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1157     }
1158     return;
1159   }
1160 
1161   int last_row = VirtualCallData::row_limit() - 1;
1162   assert(start_row <= last_row, "must be work left to do");
1163   // Test this row for both the receiver and for null.
1164   // Take any of three different outcomes:
1165   //   1. found receiver => increment count and goto done
1166   //   2. found null => keep looking for case 1, maybe allocate this cell
1167   //   3. found something else => keep looking for cases 1 and 2
1168   // Case 3 is handled by a recursive call.
1169   for (int row = start_row; row <= last_row; row++) {
1170     Label next_test;
1171     bool test_for_null_also = (row == start_row);
1172 
1173     // See if the receiver is receiver[n].
1174     int recvr_offset = in_bytes(VirtualCallData::receiver_offset(row));
1175     test_mdp_data_at(mdp, recvr_offset, receiver,
1176                      (test_for_null_also ? reg2 : noreg),
1177                      next_test);
1178     // (Reg2 now contains the receiver from the CallData.)
1179 
1180     // The receiver is receiver[n].  Increment count[n].
1181     int count_offset = in_bytes(VirtualCallData::receiver_count_offset(row));
1182     increment_mdp_data_at(mdp, count_offset);
1183     jmp(done);
1184     bind(next_test);
1185 
1186     if (test_for_null_also) {
1187       Label found_null;
1188       // Failed the equality check on receiver[n]...  Test for null.
1189       testptr(reg2, reg2);
1190       if (start_row == last_row) {
1191         // The only thing left to do is handle the null case.
1192         if (is_virtual_call) {
1193           jccb(Assembler::zero, found_null);
1194           // Receiver did not match any saved receiver and there is no empty row for it.
1195           // Increment total counter to indicate polymorphic case.
1196           increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1197           jmp(done);
1198           bind(found_null);
1199         } else {
1200           jcc(Assembler::notZero, done);
1201         }
1202         break;
1203       }
1204       // Since null is rare, make it be the branch-taken case.
1205       jcc(Assembler::zero, found_null);
1206 
1207       // Put all the "Case 3" tests here.
1208       record_klass_in_profile_helper(receiver, mdp, reg2, start_row + 1, done, is_virtual_call);
1209 
1210       // Found a null.  Keep searching for a matching receiver,
1211       // but remember that this is an empty (unused) slot.
1212       bind(found_null);
1213     }
1214   }
1215 
1216   // In the fall-through case, we found no matching receiver, but we
1217   // observed the receiver[start_row] is NULL.
1218 
1219   // Fill in the receiver field and increment the count.
1220   int recvr_offset = in_bytes(VirtualCallData::receiver_offset(start_row));
1221   set_mdp_data_at(mdp, recvr_offset, receiver);
1222   int count_offset = in_bytes(VirtualCallData::receiver_count_offset(start_row));
1223   movl(reg2, DataLayout::counter_increment);
1224   set_mdp_data_at(mdp, count_offset, reg2);
1225   if (start_row > 0) {
1226     jmp(done);
1227   }
1228 }
1229 
1230 // Example state machine code for three profile rows:
1231 //   // main copy of decision tree, rooted at row[1]
1232 //   if (row[0].rec == rec) { row[0].incr(); goto done; }
1233 //   if (row[0].rec != NULL) {
1234 //     // inner copy of decision tree, rooted at row[1]
1235 //     if (row[1].rec == rec) { row[1].incr(); goto done; }
1236 //     if (row[1].rec != NULL) {
1237 //       // degenerate decision tree, rooted at row[2]
1238 //       if (row[2].rec == rec) { row[2].incr(); goto done; }
1239 //       if (row[2].rec != NULL) { count.incr(); goto done; } // overflow
1240 //       row[2].init(rec); goto done;
1241 //     } else {
1242 //       // remember row[1] is empty
1243 //       if (row[2].rec == rec) { row[2].incr(); goto done; }
1244 //       row[1].init(rec); goto done;
1245 //     }
1246 //   } else {
1247 //     // remember row[0] is empty
1248 //     if (row[1].rec == rec) { row[1].incr(); goto done; }
1249 //     if (row[2].rec == rec) { row[2].incr(); goto done; }
1250 //     row[0].init(rec); goto done;
1251 //   }
1252 //   done:
1253 
1254 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver,
1255                                                         Register mdp, Register reg2,
1256                                                         bool is_virtual_call) {
1257   assert(ProfileInterpreter, "must be profiling");
1258   Label done;
1259 
1260   record_klass_in_profile_helper(receiver, mdp, reg2, 0, done, is_virtual_call);
1261 
1262   bind (done);
1263 }
1264 
1265 void InterpreterMacroAssembler::profile_ret(Register return_bci,
1266                                             Register mdp) {
1267   if (ProfileInterpreter) {
1268     Label profile_continue;
1269     uint row;
1270 
1271     // If no method data exists, go to profile_continue.
1272     test_method_data_pointer(mdp, profile_continue);
1273 
1274     // Update the total ret count.
1275     increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1276 
1277     for (row = 0; row < RetData::row_limit(); row++) {
1278       Label next_test;
1279 
1280       // See if return_bci is equal to bci[n]:
1281       test_mdp_data_at(mdp,
1282                        in_bytes(RetData::bci_offset(row)),
1283                        return_bci, noreg,
1284                        next_test);
1285 
1286       // return_bci is equal to bci[n].  Increment the count.
1287       increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row)));
1288 
1289       // The method data pointer needs to be updated to reflect the new target.
1290       update_mdp_by_offset(mdp,
1291                            in_bytes(RetData::bci_displacement_offset(row)));
1292       jmp(profile_continue);
1293       bind(next_test);
1294     }
1295 
1296     update_mdp_for_ret(return_bci);
1297 
1298     bind(profile_continue);
1299   }
1300 }
1301 
1302 
1303 void InterpreterMacroAssembler::profile_null_seen(Register mdp) {
1304   if (ProfileInterpreter) {
1305     Label profile_continue;
1306 
1307     // If no method data exists, go to profile_continue.
1308     test_method_data_pointer(mdp, profile_continue);
1309 
1310     set_mdp_flag_at(mdp, BitData::null_seen_byte_constant());
1311 
1312     // The method data pointer needs to be updated.
1313     int mdp_delta = in_bytes(BitData::bit_data_size());
1314     if (TypeProfileCasts) {
1315       mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1316     }
1317     update_mdp_by_constant(mdp, mdp_delta);
1318 
1319     bind(profile_continue);
1320   }
1321 }
1322 
1323 
1324 void InterpreterMacroAssembler::profile_typecheck_failed(Register mdp) {
1325   if (ProfileInterpreter && TypeProfileCasts) {
1326     Label profile_continue;
1327 
1328     // If no method data exists, go to profile_continue.
1329     test_method_data_pointer(mdp, profile_continue);
1330 
1331     int count_offset = in_bytes(CounterData::count_offset());
1332     // Back up the address, since we have already bumped the mdp.
1333     count_offset -= in_bytes(VirtualCallData::virtual_call_data_size());
1334 
1335     // *Decrement* the counter.  We expect to see zero or small negatives.
1336     increment_mdp_data_at(mdp, count_offset, true);
1337 
1338     bind (profile_continue);
1339   }
1340 }
1341 
1342 
1343 void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) {
1344   if (ProfileInterpreter) {
1345     Label profile_continue;
1346 
1347     // If no method data exists, go to profile_continue.
1348     test_method_data_pointer(mdp, profile_continue);
1349 
1350     // The method data pointer needs to be updated.
1351     int mdp_delta = in_bytes(BitData::bit_data_size());
1352     if (TypeProfileCasts) {
1353       mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1354 
1355       // Record the object type.
1356       record_klass_in_profile(klass, mdp, reg2, false);
1357     }
1358     update_mdp_by_constant(mdp, mdp_delta);
1359 
1360     bind(profile_continue);
1361   }
1362 }
1363 
1364 
1365 void InterpreterMacroAssembler::profile_switch_default(Register mdp) {
1366   if (ProfileInterpreter) {
1367     Label profile_continue;
1368 
1369     // If no method data exists, go to profile_continue.
1370     test_method_data_pointer(mdp, profile_continue);
1371 
1372     // Update the default case count
1373     increment_mdp_data_at(mdp,
1374                           in_bytes(MultiBranchData::default_count_offset()));
1375 
1376     // The method data pointer needs to be updated.
1377     update_mdp_by_offset(mdp,
1378                          in_bytes(MultiBranchData::
1379                                   default_displacement_offset()));
1380 
1381     bind(profile_continue);
1382   }
1383 }
1384 
1385 
1386 void InterpreterMacroAssembler::profile_switch_case(Register index,
1387                                                     Register mdp,
1388                                                     Register reg2) {
1389   if (ProfileInterpreter) {
1390     Label profile_continue;
1391 
1392     // If no method data exists, go to profile_continue.
1393     test_method_data_pointer(mdp, profile_continue);
1394 
1395     // Build the base (index * per_case_size_in_bytes()) +
1396     // case_array_offset_in_bytes()
1397     movl(reg2, in_bytes(MultiBranchData::per_case_size()));
1398     imulptr(index, reg2); // XXX l ?
1399     addptr(index, in_bytes(MultiBranchData::case_array_offset())); // XXX l ?
1400 
1401     // Update the case count
1402     increment_mdp_data_at(mdp,
1403                           index,
1404                           in_bytes(MultiBranchData::relative_count_offset()));
1405 
1406     // The method data pointer needs to be updated.
1407     update_mdp_by_offset(mdp,
1408                          index,
1409                          in_bytes(MultiBranchData::
1410                                   relative_displacement_offset()));
1411 
1412     bind(profile_continue);
1413   }
1414 }
1415 
1416 
1417 
1418 void InterpreterMacroAssembler::verify_oop(Register reg, TosState state) {
1419   if (state == atos) {
1420     MacroAssembler::verify_oop(reg);
1421   }
1422 }
1423 
1424 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) {
1425 }
1426 
1427 // Jump if ((*counter_addr += increment) & mask) satisfies the condition.
1428 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr,
1429                                                         int increment, int mask,
1430                                                         Register scratch, bool preloaded,
1431                                                         Condition cond, Label* where) {
1432   if (!preloaded) {
1433     movl(scratch, counter_addr);
1434   }
1435   incrementl(scratch, increment);
1436   movl(counter_addr, scratch);
1437   andl(scratch, mask);
1438   jcc(cond, *where);
1439 }
1440 #endif // !CC_INTERP
1441 
1442 
1443 void InterpreterMacroAssembler::notify_method_entry() {
1444   // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1445   // track stack depth.  If it is possible to enter interp_only_mode we add
1446   // the code to check if the event should be sent.
1447   if (JvmtiExport::can_post_interpreter_events()) {
1448     Label L;
1449     movl(rdx, Address(r15_thread, JavaThread::interp_only_mode_offset()));
1450     testl(rdx, rdx);
1451     jcc(Assembler::zero, L);
1452     call_VM(noreg, CAST_FROM_FN_PTR(address,
1453                                     InterpreterRuntime::post_method_entry));
1454     bind(L);
1455   }
1456 
1457   {
1458     SkipIfEqual skip(this, &DTraceMethodProbes, false);
1459     get_method(c_rarg1);
1460     call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
1461                  r15_thread, c_rarg1);
1462   }
1463 
1464   // RedefineClasses() tracing support for obsolete method entry
1465   if (RC_TRACE_IN_RANGE(0x00001000, 0x00002000)) {
1466     get_method(c_rarg1);
1467     call_VM_leaf(
1468       CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
1469       r15_thread, c_rarg1);
1470   }
1471 }
1472 
1473 
1474 void InterpreterMacroAssembler::notify_method_exit(
1475     TosState state, NotifyMethodExitMode mode) {
1476   // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1477   // track stack depth.  If it is possible to enter interp_only_mode we add
1478   // the code to check if the event should be sent.
1479   if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
1480     Label L;
1481     // Note: frame::interpreter_frame_result has a dependency on how the
1482     // method result is saved across the call to post_method_exit. If this
1483     // is changed then the interpreter_frame_result implementation will
1484     // need to be updated too.
1485 
1486     // For c++ interpreter the result is always stored at a known location in the frame
1487     // template interpreter will leave it on the top of the stack.
1488     NOT_CC_INTERP(push(state);)
1489     movl(rdx, Address(r15_thread, JavaThread::interp_only_mode_offset()));
1490     testl(rdx, rdx);
1491     jcc(Assembler::zero, L);
1492     call_VM(noreg,
1493             CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
1494     bind(L);
1495     NOT_CC_INTERP(pop(state));
1496   }
1497 
1498   {
1499     SkipIfEqual skip(this, &DTraceMethodProbes, false);
1500     NOT_CC_INTERP(push(state));
1501     get_method(c_rarg1);
1502     call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
1503                  r15_thread, c_rarg1);
1504     NOT_CC_INTERP(pop(state));
1505   }
1506 }
1507