1 /*
   2  * Copyright (c) 1997, 2025, 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 "compiler/compiler_globals.hpp"
  26 #include "interp_masm_x86.hpp"
  27 #include "interpreter/interpreter.hpp"
  28 #include "interpreter/interpreterRuntime.hpp"
  29 #include "logging/log.hpp"
  30 #include "oops/arrayOop.hpp"
  31 #include "oops/markWord.hpp"
  32 #include "oops/methodData.hpp"
  33 #include "oops/method.hpp"
  34 #include "oops/resolvedFieldEntry.hpp"
  35 #include "oops/resolvedIndyEntry.hpp"
  36 #include "oops/resolvedMethodEntry.hpp"
  37 #include "prims/jvmtiExport.hpp"
  38 #include "prims/jvmtiThreadState.hpp"
  39 #include "runtime/basicLock.hpp"
  40 #include "runtime/frame.inline.hpp"
  41 #include "runtime/javaThread.hpp"
  42 #include "runtime/safepointMechanism.hpp"
  43 #include "runtime/sharedRuntime.hpp"
  44 #include "utilities/powerOfTwo.hpp"
  45 
  46 // Implementation of InterpreterMacroAssembler
  47 
  48 void InterpreterMacroAssembler::jump_to_entry(address entry) {
  49   assert(entry, "Entry must have been generated by now");
  50   jump(RuntimeAddress(entry));
  51 }
  52 
  53 void InterpreterMacroAssembler::profile_obj_type(Register obj, const Address& mdo_addr) {
  54   Label update, next, none;
  55 
  56   assert_different_registers(obj, rscratch1, mdo_addr.base(), mdo_addr.index());
  57 
  58   interp_verify_oop(obj, atos);
  59 
  60   testptr(obj, obj);
  61   jccb(Assembler::notZero, update);
  62   testptr(mdo_addr, TypeEntries::null_seen);
  63   jccb(Assembler::notZero, next); // null already seen. Nothing to do anymore.
  64   // atomic update to prevent overwriting Klass* with 0
  65   lock();
  66   orptr(mdo_addr, TypeEntries::null_seen);
  67   jmpb(next);
  68 
  69   bind(update);
  70   load_klass(obj, obj, rscratch1);
  71   mov(rscratch1, obj);
  72 
  73   xorptr(obj, mdo_addr);
  74   testptr(obj, TypeEntries::type_klass_mask);
  75   jccb(Assembler::zero, next); // klass seen before, nothing to
  76                                // do. The unknown bit may have been
  77                                // set already but no need to check.
  78 
  79   testptr(obj, TypeEntries::type_unknown);
  80   jccb(Assembler::notZero, next); // already unknown. Nothing to do anymore.
  81 
  82   cmpptr(mdo_addr, 0);
  83   jccb(Assembler::equal, none);
  84   cmpptr(mdo_addr, TypeEntries::null_seen);
  85   jccb(Assembler::equal, none);
  86 
  87   // There is a chance that the checks above (re-reading profiling
  88   // data from memory) fail if another thread has just set the
  89   // profiling to this obj's klass
  90   mov(obj, rscratch1);
  91   xorptr(obj, mdo_addr);
  92   testptr(obj, TypeEntries::type_klass_mask);
  93   jccb(Assembler::zero, next);
  94 
  95   // different than before. Cannot keep accurate profile.
  96   orptr(mdo_addr, TypeEntries::type_unknown);
  97   jmpb(next);
  98 
  99   bind(none);
 100   // first time here. Set profile type.
 101   movptr(mdo_addr, obj);
 102 #ifdef ASSERT
 103   andptr(obj, TypeEntries::type_klass_mask);
 104   verify_klass_ptr(obj);
 105 #endif
 106 
 107   bind(next);
 108 }
 109 
 110 void InterpreterMacroAssembler::profile_arguments_type(Register mdp, Register callee, Register tmp, bool is_virtual) {
 111   if (!ProfileInterpreter) {
 112     return;
 113   }
 114 
 115   if (MethodData::profile_arguments() || MethodData::profile_return()) {
 116     Label profile_continue;
 117 
 118     test_method_data_pointer(mdp, profile_continue);
 119 
 120     int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size());
 121 
 122     cmpb(Address(mdp, in_bytes(DataLayout::tag_offset()) - off_to_start), is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag);
 123     jcc(Assembler::notEqual, profile_continue);
 124 
 125     if (MethodData::profile_arguments()) {
 126       Label done;
 127       int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset());
 128       addptr(mdp, off_to_args);
 129 
 130       for (int i = 0; i < TypeProfileArgsLimit; i++) {
 131         if (i > 0 || MethodData::profile_return()) {
 132           // If return value type is profiled we may have no argument to profile
 133           movptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args));
 134           subl(tmp, i*TypeStackSlotEntries::per_arg_count());
 135           cmpl(tmp, TypeStackSlotEntries::per_arg_count());
 136           jcc(Assembler::less, done);
 137         }
 138         movptr(tmp, Address(callee, Method::const_offset()));
 139         load_unsigned_short(tmp, Address(tmp, ConstMethod::size_of_parameters_offset()));
 140         // stack offset o (zero based) from the start of the argument
 141         // list, for n arguments translates into offset n - o - 1 from
 142         // the end of the argument list
 143         subptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))-off_to_args));
 144         subl(tmp, 1);
 145         Address arg_addr = argument_address(tmp);
 146         movptr(tmp, arg_addr);
 147 
 148         Address mdo_arg_addr(mdp, in_bytes(TypeEntriesAtCall::argument_type_offset(i))-off_to_args);
 149         profile_obj_type(tmp, mdo_arg_addr);
 150 
 151         int to_add = in_bytes(TypeStackSlotEntries::per_arg_size());
 152         addptr(mdp, to_add);
 153         off_to_args += to_add;
 154       }
 155 
 156       if (MethodData::profile_return()) {
 157         movptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args));
 158         subl(tmp, TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count());
 159       }
 160 
 161       bind(done);
 162 
 163       if (MethodData::profile_return()) {
 164         // We're right after the type profile for the last
 165         // argument. tmp is the number of cells left in the
 166         // CallTypeData/VirtualCallTypeData to reach its end. Non null
 167         // if there's a return to profile.
 168         assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type");
 169         shll(tmp, log2i_exact((int)DataLayout::cell_size));
 170         addptr(mdp, tmp);
 171       }
 172       movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp);
 173     } else {
 174       assert(MethodData::profile_return(), "either profile call args or call ret");
 175       update_mdp_by_constant(mdp, in_bytes(TypeEntriesAtCall::return_only_size()));
 176     }
 177 
 178     // mdp points right after the end of the
 179     // CallTypeData/VirtualCallTypeData, right after the cells for the
 180     // return value type if there's one
 181 
 182     bind(profile_continue);
 183   }
 184 }
 185 
 186 void InterpreterMacroAssembler::profile_return_type(Register mdp, Register ret, Register tmp) {
 187   assert_different_registers(mdp, ret, tmp, _bcp_register);
 188   if (ProfileInterpreter && MethodData::profile_return()) {
 189     Label profile_continue;
 190 
 191     test_method_data_pointer(mdp, profile_continue);
 192 
 193     if (MethodData::profile_return_jsr292_only()) {
 194       assert(Method::intrinsic_id_size_in_bytes() == 2, "assuming Method::_intrinsic_id is u2");
 195 
 196       // If we don't profile all invoke bytecodes we must make sure
 197       // it's a bytecode we indeed profile. We can't go back to the
 198       // beginning of the ProfileData we intend to update to check its
 199       // type because we're right after it and we don't known its
 200       // length
 201       Label do_profile;
 202       cmpb(Address(_bcp_register, 0), Bytecodes::_invokedynamic);
 203       jcc(Assembler::equal, do_profile);
 204       cmpb(Address(_bcp_register, 0), Bytecodes::_invokehandle);
 205       jcc(Assembler::equal, do_profile);
 206       get_method(tmp);
 207       cmpw(Address(tmp, Method::intrinsic_id_offset()), static_cast<int>(vmIntrinsics::_compiledLambdaForm));
 208       jcc(Assembler::notEqual, profile_continue);
 209 
 210       bind(do_profile);
 211     }
 212 
 213     Address mdo_ret_addr(mdp, -in_bytes(ReturnTypeEntry::size()));
 214     mov(tmp, ret);
 215     profile_obj_type(tmp, mdo_ret_addr);
 216 
 217     bind(profile_continue);
 218   }
 219 }
 220 
 221 void InterpreterMacroAssembler::profile_parameters_type(Register mdp, Register tmp1, Register tmp2) {
 222   if (ProfileInterpreter && MethodData::profile_parameters()) {
 223     Label profile_continue;
 224 
 225     test_method_data_pointer(mdp, profile_continue);
 226 
 227     // Load the offset of the area within the MDO used for
 228     // parameters. If it's negative we're not profiling any parameters
 229     movl(tmp1, Address(mdp, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset())));
 230     testl(tmp1, tmp1);
 231     jcc(Assembler::negative, profile_continue);
 232 
 233     // Compute a pointer to the area for parameters from the offset
 234     // and move the pointer to the slot for the last
 235     // parameters. Collect profiling from last parameter down.
 236     // mdo start + parameters offset + array length - 1
 237     addptr(mdp, tmp1);
 238     movptr(tmp1, Address(mdp, ArrayData::array_len_offset()));
 239     decrement(tmp1, TypeStackSlotEntries::per_arg_count());
 240 
 241     Label loop;
 242     bind(loop);
 243 
 244     int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0));
 245     int type_base = in_bytes(ParametersTypeData::type_offset(0));
 246     Address::ScaleFactor per_arg_scale = Address::times(DataLayout::cell_size);
 247     Address arg_off(mdp, tmp1, per_arg_scale, off_base);
 248     Address arg_type(mdp, tmp1, per_arg_scale, type_base);
 249 
 250     // load offset on the stack from the slot for this parameter
 251     movptr(tmp2, arg_off);
 252     negptr(tmp2);
 253     // read the parameter from the local area
 254     movptr(tmp2, Address(_locals_register, tmp2, Interpreter::stackElementScale()));
 255 
 256     // profile the parameter
 257     profile_obj_type(tmp2, arg_type);
 258 
 259     // go to next parameter
 260     decrement(tmp1, TypeStackSlotEntries::per_arg_count());
 261     jcc(Assembler::positive, loop);
 262 
 263     bind(profile_continue);
 264   }
 265 }
 266 
 267 void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point,
 268                                                   int number_of_arguments) {
 269   // interpreter specific
 270   //
 271   // Note: No need to save/restore bcp & locals registers
 272   //       since these are callee saved registers and no blocking/
 273   //       GC can happen in leaf calls.
 274   // Further Note: DO NOT save/restore bcp/locals. If a caller has
 275   // already saved them so that it can use rsi/rdi as temporaries
 276   // then a save/restore here will DESTROY the copy the caller
 277   // saved! There used to be a save_bcp() that only happened in
 278   // the ASSERT path (no restore_bcp). Which caused bizarre failures
 279   // when jvm built with ASSERTs.
 280 #ifdef ASSERT
 281   {
 282     Label L;
 283     cmpptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
 284     jcc(Assembler::equal, L);
 285     stop("InterpreterMacroAssembler::call_VM_leaf_base:"
 286          " last_sp != null");
 287     bind(L);
 288   }
 289 #endif
 290   // super call
 291   MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments);
 292   // interpreter specific
 293   // LP64: Used to ASSERT that r13/r14 were equal to frame's bcp/locals
 294   // but since they may not have been saved (and we don't want to
 295   // save them here (see note above) the assert is invalid.
 296 }
 297 
 298 void InterpreterMacroAssembler::call_VM_base(Register oop_result,
 299                                              Register last_java_sp,
 300                                              address  entry_point,
 301                                              int      number_of_arguments,
 302                                              bool     check_exceptions) {
 303   // interpreter specific
 304   //
 305   // Note: Could avoid restoring locals ptr (callee saved) - however doesn't
 306   //       really make a difference for these runtime calls, since they are
 307   //       slow anyway. Btw., bcp must be saved/restored since it may change
 308   //       due to GC.
 309   save_bcp();
 310 #ifdef ASSERT
 311   {
 312     Label L;
 313     cmpptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
 314     jcc(Assembler::equal, L);
 315     stop("InterpreterMacroAssembler::call_VM_base:"
 316          " last_sp isn't null");
 317     bind(L);
 318   }
 319 #endif /* ASSERT */
 320   // super call
 321   MacroAssembler::call_VM_base(oop_result, last_java_sp,
 322                                entry_point, number_of_arguments,
 323                                check_exceptions);
 324   // interpreter specific
 325   restore_bcp();
 326   restore_locals();
 327 }
 328 
 329 void InterpreterMacroAssembler::call_VM_preemptable(Register oop_result,
 330                                                     address entry_point,
 331                                                     Register arg_1) {
 332   assert(arg_1 == c_rarg1, "");
 333   Label resume_pc, not_preempted;
 334 
 335 #ifdef ASSERT
 336   {
 337     Label L;
 338     cmpptr(Address(r15_thread, JavaThread::preempt_alternate_return_offset()), NULL_WORD);
 339     jcc(Assembler::equal, L);
 340     stop("Should not have alternate return address set");
 341     bind(L);
 342   }
 343 #endif /* ASSERT */
 344 
 345   // Force freeze slow path.
 346   push_cont_fastpath();
 347 
 348   // Make VM call. In case of preemption set last_pc to the one we want to resume to.
 349   // Note: call_VM_helper requires last_Java_pc for anchor to be at the top of the stack.
 350   lea(rscratch1, resume_pc);
 351   push(rscratch1);
 352   MacroAssembler::call_VM_helper(oop_result, entry_point, 1, false /*check_exceptions*/);
 353   pop(rscratch1);
 354 
 355   pop_cont_fastpath();
 356 
 357   // Check if preempted.
 358   movptr(rscratch1, Address(r15_thread, JavaThread::preempt_alternate_return_offset()));
 359   cmpptr(rscratch1, NULL_WORD);
 360   jccb(Assembler::zero, not_preempted);
 361   movptr(Address(r15_thread, JavaThread::preempt_alternate_return_offset()), NULL_WORD);
 362   jmp(rscratch1);
 363 
 364   // In case of preemption, this is where we will resume once we finally acquire the monitor.
 365   bind(resume_pc);
 366   restore_after_resume(false /* is_native */);
 367 
 368   bind(not_preempted);
 369 }
 370 
 371 void InterpreterMacroAssembler::restore_after_resume(bool is_native) {
 372   lea(rscratch1, ExternalAddress(Interpreter::cont_resume_interpreter_adapter()));
 373   call(rscratch1);
 374   if (is_native) {
 375     // On resume we need to set up stack as expected.
 376     push(dtos);
 377     push(ltos);
 378   }
 379 }
 380 
 381 void InterpreterMacroAssembler::check_and_handle_popframe() {
 382   if (JvmtiExport::can_pop_frame()) {
 383     Label L;
 384     // Initiate popframe handling only if it is not already being
 385     // processed.  If the flag has the popframe_processing bit set, it
 386     // means that this code is called *during* popframe handling - we
 387     // don't want to reenter.
 388     // This method is only called just after the call into the vm in
 389     // call_VM_base, so the arg registers are available.
 390     Register pop_cond = c_rarg0;
 391     movl(pop_cond, Address(r15_thread, JavaThread::popframe_condition_offset()));
 392     testl(pop_cond, JavaThread::popframe_pending_bit);
 393     jcc(Assembler::zero, L);
 394     testl(pop_cond, JavaThread::popframe_processing_bit);
 395     jcc(Assembler::notZero, L);
 396     // Call Interpreter::remove_activation_preserving_args_entry() to get the
 397     // address of the same-named entrypoint in the generated interpreter code.
 398     call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry));
 399     jmp(rax);
 400     bind(L);
 401   }
 402 }
 403 
 404 void InterpreterMacroAssembler::load_earlyret_value(TosState state) {
 405   movptr(rcx, Address(r15_thread, JavaThread::jvmti_thread_state_offset()));
 406   const Address tos_addr(rcx, JvmtiThreadState::earlyret_tos_offset());
 407   const Address oop_addr(rcx, JvmtiThreadState::earlyret_oop_offset());
 408   const Address val_addr(rcx, JvmtiThreadState::earlyret_value_offset());
 409 
 410   switch (state) {
 411     case atos: movptr(rax, oop_addr);
 412                movptr(oop_addr, NULL_WORD);
 413                interp_verify_oop(rax, state);         break;
 414     case ltos: movptr(rax, val_addr);                 break;
 415     case btos:                                   // fall through
 416     case ztos:                                   // fall through
 417     case ctos:                                   // fall through
 418     case stos:                                   // fall through
 419     case itos: movl(rax, val_addr);                 break;
 420     case ftos: load_float(val_addr);                break;
 421     case dtos: load_double(val_addr);               break;
 422     case vtos: /* nothing to do */                  break;
 423     default  : ShouldNotReachHere();
 424   }
 425 
 426   // Clean up tos value in the thread object
 427   movl(tos_addr, ilgl);
 428   movptr(val_addr, NULL_WORD);
 429 }
 430 
 431 
 432 void InterpreterMacroAssembler::check_and_handle_earlyret() {
 433   if (JvmtiExport::can_force_early_return()) {
 434     Label L;
 435     Register tmp = c_rarg0;
 436     Register rthread = r15_thread;
 437 
 438     movptr(tmp, Address(rthread, JavaThread::jvmti_thread_state_offset()));
 439     testptr(tmp, tmp);
 440     jcc(Assembler::zero, L); // if (thread->jvmti_thread_state() == nullptr) exit;
 441 
 442     // Initiate earlyret handling only if it is not already being processed.
 443     // If the flag has the earlyret_processing bit set, it means that this code
 444     // is called *during* earlyret handling - we don't want to reenter.
 445     movl(tmp, Address(tmp, JvmtiThreadState::earlyret_state_offset()));
 446     cmpl(tmp, JvmtiThreadState::earlyret_pending);
 447     jcc(Assembler::notEqual, L);
 448 
 449     // Call Interpreter::remove_activation_early_entry() to get the address of the
 450     // same-named entrypoint in the generated interpreter code.
 451     movptr(tmp, Address(rthread, JavaThread::jvmti_thread_state_offset()));
 452     movl(tmp, Address(tmp, JvmtiThreadState::earlyret_tos_offset()));
 453     call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), tmp);
 454     jmp(rax);
 455     bind(L);
 456   }
 457 }
 458 
 459 void InterpreterMacroAssembler::get_unsigned_2_byte_index_at_bcp(Register reg, int bcp_offset) {
 460   assert(bcp_offset >= 0, "bcp is still pointing to start of bytecode");
 461   load_unsigned_short(reg, Address(_bcp_register, bcp_offset));
 462   bswapl(reg);
 463   shrl(reg, 16);
 464 }
 465 
 466 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register index,
 467                                                        int bcp_offset,
 468                                                        size_t index_size) {
 469   assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
 470   if (index_size == sizeof(u2)) {
 471     load_unsigned_short(index, Address(_bcp_register, bcp_offset));
 472   } else if (index_size == sizeof(u4)) {
 473     movl(index, Address(_bcp_register, bcp_offset));
 474   } else if (index_size == sizeof(u1)) {
 475     load_unsigned_byte(index, Address(_bcp_register, bcp_offset));
 476   } else {
 477     ShouldNotReachHere();
 478   }
 479 }
 480 
 481 // Load object from cpool->resolved_references(index)
 482 void InterpreterMacroAssembler::load_resolved_reference_at_index(Register result,
 483                                                                  Register index,
 484                                                                  Register tmp) {
 485   assert_different_registers(result, index);
 486 
 487   get_constant_pool(result);
 488   // load pointer for resolved_references[] objArray
 489   movptr(result, Address(result, ConstantPool::cache_offset()));
 490   movptr(result, Address(result, ConstantPoolCache::resolved_references_offset()));
 491   resolve_oop_handle(result, tmp);
 492   load_heap_oop(result, Address(result, index,
 493                                 UseCompressedOops ? Address::times_4 : Address::times_ptr,
 494                                 arrayOopDesc::base_offset_in_bytes(T_OBJECT)), tmp);
 495 }
 496 
 497 // load cpool->resolved_klass_at(index)
 498 void InterpreterMacroAssembler::load_resolved_klass_at_index(Register klass,
 499                                                              Register cpool,
 500                                                              Register index) {
 501   assert_different_registers(cpool, index);
 502 
 503   movw(index, Address(cpool, index, Address::times_ptr, sizeof(ConstantPool)));
 504   Register resolved_klasses = cpool;
 505   movptr(resolved_klasses, Address(cpool, ConstantPool::resolved_klasses_offset()));
 506   movptr(klass, Address(resolved_klasses, index, Address::times_ptr, Array<Klass*>::base_offset_in_bytes()));
 507 }
 508 
 509 // Generate a subtype check: branch to ok_is_subtype if sub_klass is a
 510 // subtype of super_klass.
 511 //
 512 // Args:
 513 //      rax: superklass
 514 //      Rsub_klass: subklass
 515 //
 516 // Kills:
 517 //      rcx, rdi
 518 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass,
 519                                                   Label& ok_is_subtype) {
 520   assert(Rsub_klass != rax, "rax holds superklass");
 521   LP64_ONLY(assert(Rsub_klass != r14, "r14 holds locals");)
 522   LP64_ONLY(assert(Rsub_klass != r13, "r13 holds bcp");)
 523   assert(Rsub_klass != rcx, "rcx holds 2ndary super array length");
 524   assert(Rsub_klass != rdi, "rdi holds 2ndary super array scan ptr");
 525 
 526   // Profile the not-null value's klass.
 527   profile_typecheck(rcx, Rsub_klass, rdi); // blows rcx, reloads rdi
 528 
 529   // Do the check.
 530   check_klass_subtype(Rsub_klass, rax, rcx, ok_is_subtype); // blows rcx
 531 }
 532 
 533 
 534 // Java Expression Stack
 535 
 536 void InterpreterMacroAssembler::pop_ptr(Register r) {
 537   pop(r);
 538 }
 539 
 540 void InterpreterMacroAssembler::push_ptr(Register r) {
 541   push(r);
 542 }
 543 
 544 void InterpreterMacroAssembler::push_i(Register r) {
 545   push(r);
 546 }
 547 
 548 void InterpreterMacroAssembler::push_i_or_ptr(Register r) {
 549   push(r);
 550 }
 551 
 552 void InterpreterMacroAssembler::push_f(XMMRegister r) {
 553   subptr(rsp, wordSize);
 554   movflt(Address(rsp, 0), r);
 555 }
 556 
 557 void InterpreterMacroAssembler::pop_f(XMMRegister r) {
 558   movflt(r, Address(rsp, 0));
 559   addptr(rsp, wordSize);
 560 }
 561 
 562 void InterpreterMacroAssembler::push_d(XMMRegister r) {
 563   subptr(rsp, 2 * wordSize);
 564   movdbl(Address(rsp, 0), r);
 565 }
 566 
 567 void InterpreterMacroAssembler::pop_d(XMMRegister r) {
 568   movdbl(r, Address(rsp, 0));
 569   addptr(rsp, 2 * Interpreter::stackElementSize);
 570 }
 571 
 572 void InterpreterMacroAssembler::pop_i(Register r) {
 573   // XXX can't use pop currently, upper half non clean
 574   movl(r, Address(rsp, 0));
 575   addptr(rsp, wordSize);
 576 }
 577 
 578 void InterpreterMacroAssembler::pop_l(Register r) {
 579   movq(r, Address(rsp, 0));
 580   addptr(rsp, 2 * Interpreter::stackElementSize);
 581 }
 582 
 583 void InterpreterMacroAssembler::push_l(Register r) {
 584   subptr(rsp, 2 * wordSize);
 585   movptr(Address(rsp, Interpreter::expr_offset_in_bytes(0)), r         );
 586   movptr(Address(rsp, Interpreter::expr_offset_in_bytes(1)), NULL_WORD );
 587 }
 588 
 589 void InterpreterMacroAssembler::pop(TosState state) {
 590   switch (state) {
 591   case atos: pop_ptr();                 break;
 592   case btos:
 593   case ztos:
 594   case ctos:
 595   case stos:
 596   case itos: pop_i();                   break;
 597   case ltos: pop_l();                   break;
 598   case ftos: pop_f(xmm0);               break;
 599   case dtos: pop_d(xmm0);               break;
 600   case vtos: /* nothing to do */        break;
 601   default:   ShouldNotReachHere();
 602   }
 603   interp_verify_oop(rax, state);
 604 }
 605 
 606 void InterpreterMacroAssembler::push(TosState state) {
 607   interp_verify_oop(rax, state);
 608   switch (state) {
 609   case atos: push_ptr();                break;
 610   case btos:
 611   case ztos:
 612   case ctos:
 613   case stos:
 614   case itos: push_i();                  break;
 615   case ltos: push_l();                  break;
 616   case ftos: push_f(xmm0);              break;
 617   case dtos: push_d(xmm0);              break;
 618   case vtos: /* nothing to do */        break;
 619   default  : ShouldNotReachHere();
 620   }
 621 }
 622 
 623 // Helpers for swap and dup
 624 void InterpreterMacroAssembler::load_ptr(int n, Register val) {
 625   movptr(val, Address(rsp, Interpreter::expr_offset_in_bytes(n)));
 626 }
 627 
 628 void InterpreterMacroAssembler::store_ptr(int n, Register val) {
 629   movptr(Address(rsp, Interpreter::expr_offset_in_bytes(n)), val);
 630 }
 631 
 632 
 633 void InterpreterMacroAssembler::prepare_to_jump_from_interpreted() {
 634   // set sender sp
 635   lea(_bcp_register, Address(rsp, wordSize));
 636   // record last_sp
 637   mov(rcx, _bcp_register);
 638   subptr(rcx, rbp);
 639   sarptr(rcx, LogBytesPerWord);
 640   movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), rcx);
 641 }
 642 
 643 
 644 // Jump to from_interpreted entry of a call unless single stepping is possible
 645 // in this thread in which case we must call the i2i entry
 646 void InterpreterMacroAssembler::jump_from_interpreted(Register method, Register temp) {
 647   prepare_to_jump_from_interpreted();
 648 
 649   if (JvmtiExport::can_post_interpreter_events()) {
 650     Label run_compiled_code;
 651     // JVMTI events, such as single-stepping, are implemented partly by avoiding running
 652     // compiled code in threads for which the event is enabled.  Check here for
 653     // interp_only_mode if these events CAN be enabled.
 654     // interp_only is an int, on little endian it is sufficient to test the byte only
 655     // Is a cmpl faster?
 656     cmpb(Address(r15_thread, JavaThread::interp_only_mode_offset()), 0);
 657     jccb(Assembler::zero, run_compiled_code);
 658     jmp(Address(method, Method::interpreter_entry_offset()));
 659     bind(run_compiled_code);
 660   }
 661 
 662   jmp(Address(method, Method::from_interpreted_offset()));
 663 }
 664 
 665 // The following two routines provide a hook so that an implementation
 666 // can schedule the dispatch in two parts.  x86 does not do this.
 667 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int step) {
 668   // Nothing x86 specific to be done here
 669 }
 670 
 671 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int step) {
 672   dispatch_next(state, step);
 673 }
 674 
 675 void InterpreterMacroAssembler::dispatch_base(TosState state,
 676                                               address* table,
 677                                               bool verifyoop,
 678                                               bool generate_poll) {
 679   if (VerifyActivationFrameSize) {
 680     Label L;
 681     mov(rcx, rbp);
 682     subptr(rcx, rsp);
 683     int32_t min_frame_size =
 684       (frame::link_offset - frame::interpreter_frame_initial_sp_offset) *
 685       wordSize;
 686     cmpptr(rcx, min_frame_size);
 687     jcc(Assembler::greaterEqual, L);
 688     stop("broken stack frame");
 689     bind(L);
 690   }
 691   if (verifyoop) {
 692     interp_verify_oop(rax, state);
 693   }
 694 
 695   address* const safepoint_table = Interpreter::safept_table(state);
 696   Label no_safepoint, dispatch;
 697   if (table != safepoint_table && generate_poll) {
 698     NOT_PRODUCT(block_comment("Thread-local Safepoint poll"));
 699     testb(Address(r15_thread, JavaThread::polling_word_offset()), SafepointMechanism::poll_bit());
 700 
 701     jccb(Assembler::zero, no_safepoint);
 702     lea(rscratch1, ExternalAddress((address)safepoint_table));
 703     jmpb(dispatch);
 704   }
 705 
 706   bind(no_safepoint);
 707   lea(rscratch1, ExternalAddress((address)table));
 708   bind(dispatch);
 709   jmp(Address(rscratch1, rbx, Address::times_8));
 710 }
 711 
 712 void InterpreterMacroAssembler::dispatch_only(TosState state, bool generate_poll) {
 713   dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll);
 714 }
 715 
 716 void InterpreterMacroAssembler::dispatch_only_normal(TosState state) {
 717   dispatch_base(state, Interpreter::normal_table(state));
 718 }
 719 
 720 void InterpreterMacroAssembler::dispatch_only_noverify(TosState state) {
 721   dispatch_base(state, Interpreter::normal_table(state), false);
 722 }
 723 
 724 
 725 void InterpreterMacroAssembler::dispatch_next(TosState state, int step, bool generate_poll) {
 726   // load next bytecode (load before advancing _bcp_register to prevent AGI)
 727   load_unsigned_byte(rbx, Address(_bcp_register, step));
 728   // advance _bcp_register
 729   increment(_bcp_register, step);
 730   dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll);
 731 }
 732 
 733 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) {
 734   // load current bytecode
 735   load_unsigned_byte(rbx, Address(_bcp_register, 0));
 736   dispatch_base(state, table);
 737 }
 738 
 739 void InterpreterMacroAssembler::narrow(Register result) {
 740 
 741   // Get method->_constMethod->_result_type
 742   movptr(rcx, Address(rbp, frame::interpreter_frame_method_offset * wordSize));
 743   movptr(rcx, Address(rcx, Method::const_offset()));
 744   load_unsigned_byte(rcx, Address(rcx, ConstMethod::result_type_offset()));
 745 
 746   Label done, notBool, notByte, notChar;
 747 
 748   // common case first
 749   cmpl(rcx, T_INT);
 750   jcc(Assembler::equal, done);
 751 
 752   // mask integer result to narrower return type.
 753   cmpl(rcx, T_BOOLEAN);
 754   jcc(Assembler::notEqual, notBool);
 755   andl(result, 0x1);
 756   jmp(done);
 757 
 758   bind(notBool);
 759   cmpl(rcx, T_BYTE);
 760   jcc(Assembler::notEqual, notByte);
 761   movsbl(result, result);
 762   jmp(done);
 763 
 764   bind(notByte);
 765   cmpl(rcx, T_CHAR);
 766   jcc(Assembler::notEqual, notChar);
 767   movzwl(result, result);
 768   jmp(done);
 769 
 770   bind(notChar);
 771   // cmpl(rcx, T_SHORT);  // all that's left
 772   // jcc(Assembler::notEqual, done);
 773   movswl(result, result);
 774 
 775   // Nothing to do for T_INT
 776   bind(done);
 777 }
 778 
 779 // remove activation
 780 //
 781 // Apply stack watermark barrier.
 782 // Unlock the receiver if this is a synchronized method.
 783 // Unlock any Java monitors from synchronized blocks.
 784 // Remove the activation from the stack.
 785 //
 786 // If there are locked Java monitors
 787 //    If throw_monitor_exception
 788 //       throws IllegalMonitorStateException
 789 //    Else if install_monitor_exception
 790 //       installs IllegalMonitorStateException
 791 //    Else
 792 //       no error processing
 793 void InterpreterMacroAssembler::remove_activation(
 794         TosState state,
 795         Register ret_addr,
 796         bool throw_monitor_exception,
 797         bool install_monitor_exception,
 798         bool notify_jvmdi) {
 799   // Note: Registers rdx xmm0 may be in use for the
 800   // result check if synchronized method
 801   Label unlocked, unlock, no_unlock;
 802 
 803   const Register rthread = r15_thread;
 804   const Register robj    = c_rarg1;
 805   const Register rmon    = c_rarg1;
 806 
 807   // The below poll is for the stack watermark barrier. It allows fixing up frames lazily,
 808   // that would normally not be safe to use. Such bad returns into unsafe territory of
 809   // the stack, will call InterpreterRuntime::at_unwind.
 810   Label slow_path;
 811   Label fast_path;
 812   safepoint_poll(slow_path, true /* at_return */, false /* in_nmethod */);
 813   jmp(fast_path);
 814   bind(slow_path);
 815   push(state);
 816   set_last_Java_frame(noreg, rbp, (address)pc(), rscratch1);
 817   super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::at_unwind), rthread);
 818   reset_last_Java_frame(true);
 819   pop(state);
 820   bind(fast_path);
 821 
 822   // get the value of _do_not_unlock_if_synchronized into rdx
 823   const Address do_not_unlock_if_synchronized(rthread,
 824     in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
 825   movbool(rbx, do_not_unlock_if_synchronized);
 826   movbool(do_not_unlock_if_synchronized, false); // reset the flag
 827 
 828  // get method access flags
 829   movptr(rcx, Address(rbp, frame::interpreter_frame_method_offset * wordSize));
 830   load_unsigned_short(rcx, Address(rcx, Method::access_flags_offset()));
 831   testl(rcx, JVM_ACC_SYNCHRONIZED);
 832   jcc(Assembler::zero, unlocked);
 833 
 834   // Don't unlock anything if the _do_not_unlock_if_synchronized flag
 835   // is set.
 836   testbool(rbx);
 837   jcc(Assembler::notZero, no_unlock);
 838 
 839   // unlock monitor
 840   push(state); // save result
 841 
 842   // BasicObjectLock will be first in list, since this is a
 843   // synchronized method. However, need to check that the object has
 844   // not been unlocked by an explicit monitorexit bytecode.
 845   const Address monitor(rbp, frame::interpreter_frame_initial_sp_offset *
 846                         wordSize - (int) sizeof(BasicObjectLock));
 847   // We use c_rarg1/rdx so that if we go slow path it will be the correct
 848   // register for unlock_object to pass to VM directly
 849   lea(robj, monitor); // address of first monitor
 850 
 851   movptr(rax, Address(robj, BasicObjectLock::obj_offset()));
 852   testptr(rax, rax);
 853   jcc(Assembler::notZero, unlock);
 854 
 855   pop(state);
 856   if (throw_monitor_exception) {
 857     // Entry already unlocked, need to throw exception
 858     call_VM(noreg, CAST_FROM_FN_PTR(address,
 859                    InterpreterRuntime::throw_illegal_monitor_state_exception));
 860     should_not_reach_here();
 861   } else {
 862     // Monitor already unlocked during a stack unroll. If requested,
 863     // install an illegal_monitor_state_exception.  Continue with
 864     // stack unrolling.
 865     if (install_monitor_exception) {
 866       call_VM(noreg, CAST_FROM_FN_PTR(address,
 867                      InterpreterRuntime::new_illegal_monitor_state_exception));
 868     }
 869     jmp(unlocked);
 870   }
 871 
 872   bind(unlock);
 873   unlock_object(robj);
 874   pop(state);
 875 
 876   // Check that for block-structured locking (i.e., that all locked
 877   // objects has been unlocked)
 878   bind(unlocked);
 879 
 880   // rax, rdx: Might contain return value
 881 
 882   // Check that all monitors are unlocked
 883   {
 884     Label loop, exception, entry, restart;
 885     const int entry_size = frame::interpreter_frame_monitor_size_in_bytes();
 886     const Address monitor_block_top(
 887         rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
 888     const Address monitor_block_bot(
 889         rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
 890 
 891     bind(restart);
 892     // We use c_rarg1 so that if we go slow path it will be the correct
 893     // register for unlock_object to pass to VM directly
 894     movptr(rmon, monitor_block_top); // derelativize pointer
 895     lea(rmon, Address(rbp, rmon, Address::times_ptr));
 896     // c_rarg1 points to current entry, starting with top-most entry
 897 
 898     lea(rbx, monitor_block_bot);  // points to word before bottom of
 899                                   // monitor block
 900     jmp(entry);
 901 
 902     // Entry already locked, need to throw exception
 903     bind(exception);
 904 
 905     if (throw_monitor_exception) {
 906       // Throw exception
 907       MacroAssembler::call_VM(noreg,
 908                               CAST_FROM_FN_PTR(address, InterpreterRuntime::
 909                                    throw_illegal_monitor_state_exception));
 910       should_not_reach_here();
 911     } else {
 912       // Stack unrolling. Unlock object and install illegal_monitor_exception.
 913       // Unlock does not block, so don't have to worry about the frame.
 914       // We don't have to preserve c_rarg1 since we are going to throw an exception.
 915 
 916       push(state);
 917       mov(robj, rmon);   // nop if robj and rmon are the same
 918       unlock_object(robj);
 919       pop(state);
 920 
 921       if (install_monitor_exception) {
 922         call_VM(noreg, CAST_FROM_FN_PTR(address,
 923                                         InterpreterRuntime::
 924                                         new_illegal_monitor_state_exception));
 925       }
 926 
 927       jmp(restart);
 928     }
 929 
 930     bind(loop);
 931     // check if current entry is used
 932     cmpptr(Address(rmon, BasicObjectLock::obj_offset()), NULL_WORD);
 933     jcc(Assembler::notEqual, exception);
 934 
 935     addptr(rmon, entry_size); // otherwise advance to next entry
 936     bind(entry);
 937     cmpptr(rmon, rbx); // check if bottom reached
 938     jcc(Assembler::notEqual, loop); // if not at bottom then check this entry
 939   }
 940 
 941   bind(no_unlock);
 942 
 943   // jvmti support
 944   if (notify_jvmdi) {
 945     notify_method_exit(state, NotifyJVMTI);    // preserve TOSCA
 946   } else {
 947     notify_method_exit(state, SkipNotifyJVMTI); // preserve TOSCA
 948   }
 949 
 950   // remove activation
 951   // get sender sp
 952   movptr(rbx,
 953          Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize));
 954   if (StackReservedPages > 0) {
 955     // testing if reserved zone needs to be re-enabled
 956     Register rthread = r15_thread;
 957     Label no_reserved_zone_enabling;
 958 
 959     // check if already enabled - if so no re-enabling needed
 960     assert(sizeof(StackOverflow::StackGuardState) == 4, "unexpected size");
 961     cmpl(Address(rthread, JavaThread::stack_guard_state_offset()), StackOverflow::stack_guard_enabled);
 962     jcc(Assembler::equal, no_reserved_zone_enabling);
 963 
 964     cmpptr(rbx, Address(rthread, JavaThread::reserved_stack_activation_offset()));
 965     jcc(Assembler::lessEqual, no_reserved_zone_enabling);
 966 
 967     call_VM_leaf(
 968       CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), rthread);
 969     call_VM(noreg, CAST_FROM_FN_PTR(address,
 970                    InterpreterRuntime::throw_delayed_StackOverflowError));
 971     should_not_reach_here();
 972 
 973     bind(no_reserved_zone_enabling);
 974   }
 975   leave();                           // remove frame anchor
 976   pop(ret_addr);                     // get return address
 977   mov(rsp, rbx);                     // set sp to sender sp
 978   pop_cont_fastpath();
 979 }
 980 
 981 void InterpreterMacroAssembler::get_method_counters(Register method,
 982                                                     Register mcs, Label& skip) {
 983   Label has_counters;
 984   movptr(mcs, Address(method, Method::method_counters_offset()));
 985   testptr(mcs, mcs);
 986   jcc(Assembler::notZero, has_counters);
 987   call_VM(noreg, CAST_FROM_FN_PTR(address,
 988           InterpreterRuntime::build_method_counters), method);
 989   movptr(mcs, Address(method,Method::method_counters_offset()));
 990   testptr(mcs, mcs);
 991   jcc(Assembler::zero, skip); // No MethodCounters allocated, OutOfMemory
 992   bind(has_counters);
 993 }
 994 
 995 
 996 // Lock object
 997 //
 998 // Args:
 999 //      rdx, c_rarg1: BasicObjectLock to be used for locking
1000 //
1001 // Kills:
1002 //      rax, rbx
1003 void InterpreterMacroAssembler::lock_object(Register lock_reg) {
1004   assert(lock_reg == c_rarg1, "The argument is only for looks. It must be c_rarg1");
1005 
1006   if (LockingMode == LM_MONITOR) {
1007     call_VM_preemptable(noreg,
1008             CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
1009             lock_reg);
1010   } else {
1011     Label count_locking, done, slow_case;
1012 
1013     const Register swap_reg = rax; // Must use rax for cmpxchg instruction
1014     const Register tmp_reg = rbx;
1015     const Register obj_reg = c_rarg3; // Will contain the oop
1016     const Register rklass_decode_tmp = rscratch1;
1017 
1018     const int obj_offset = in_bytes(BasicObjectLock::obj_offset());
1019     const int lock_offset = in_bytes(BasicObjectLock::lock_offset());
1020     const int mark_offset = lock_offset +
1021                             BasicLock::displaced_header_offset_in_bytes();
1022 
1023     // Load object pointer into obj_reg
1024     movptr(obj_reg, Address(lock_reg, obj_offset));
1025 
1026     if (DiagnoseSyncOnValueBasedClasses != 0) {
1027       load_klass(tmp_reg, obj_reg, rklass_decode_tmp);
1028       testb(Address(tmp_reg, Klass::misc_flags_offset()), KlassFlags::_misc_is_value_based_class);
1029       jcc(Assembler::notZero, slow_case);
1030     }
1031 
1032     if (LockingMode == LM_LIGHTWEIGHT) {
1033       lightweight_lock(lock_reg, obj_reg, swap_reg, tmp_reg, slow_case);
1034     } else if (LockingMode == LM_LEGACY) {
1035       // Load immediate 1 into swap_reg %rax
1036       movl(swap_reg, 1);
1037 
1038       // Load (object->mark() | 1) into swap_reg %rax
1039       orptr(swap_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
1040 
1041       // Save (object->mark() | 1) into BasicLock's displaced header
1042       movptr(Address(lock_reg, mark_offset), swap_reg);
1043 
1044       assert(lock_offset == 0,
1045              "displaced header must be first word in BasicObjectLock");
1046 
1047       lock();
1048       cmpxchgptr(lock_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
1049       jcc(Assembler::zero, count_locking);
1050 
1051       const int zero_bits = 7;
1052 
1053       // Fast check for recursive lock.
1054       //
1055       // Can apply the optimization only if this is a stack lock
1056       // allocated in this thread. For efficiency, we can focus on
1057       // recently allocated stack locks (instead of reading the stack
1058       // base and checking whether 'mark' points inside the current
1059       // thread stack):
1060       //  1) (mark & zero_bits) == 0, and
1061       //  2) rsp <= mark < mark + os::pagesize()
1062       //
1063       // Warning: rsp + os::pagesize can overflow the stack base. We must
1064       // neither apply the optimization for an inflated lock allocated
1065       // just above the thread stack (this is why condition 1 matters)
1066       // nor apply the optimization if the stack lock is inside the stack
1067       // of another thread. The latter is avoided even in case of overflow
1068       // because we have guard pages at the end of all stacks. Hence, if
1069       // we go over the stack base and hit the stack of another thread,
1070       // this should not be in a writeable area that could contain a
1071       // stack lock allocated by that thread. As a consequence, a stack
1072       // lock less than page size away from rsp is guaranteed to be
1073       // owned by the current thread.
1074       //
1075       // These 3 tests can be done by evaluating the following
1076       // expression: ((mark - rsp) & (zero_bits - os::vm_page_size())),
1077       // assuming both stack pointer and pagesize have their
1078       // least significant bits clear.
1079       // NOTE: the mark is in swap_reg %rax as the result of cmpxchg
1080       subptr(swap_reg, rsp);
1081       andptr(swap_reg, zero_bits - (int)os::vm_page_size());
1082 
1083       // Save the test result, for recursive case, the result is zero
1084       movptr(Address(lock_reg, mark_offset), swap_reg);
1085       jcc(Assembler::notZero, slow_case);
1086 
1087       bind(count_locking);
1088       inc_held_monitor_count();
1089     }
1090     jmp(done);
1091 
1092     bind(slow_case);
1093 
1094     // Call the runtime routine for slow case
1095     call_VM_preemptable(noreg,
1096             CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
1097             lock_reg);
1098     bind(done);
1099   }
1100 }
1101 
1102 
1103 // Unlocks an object. Used in monitorexit bytecode and
1104 // remove_activation.  Throws an IllegalMonitorException if object is
1105 // not locked by current thread.
1106 //
1107 // Args:
1108 //      rdx, c_rarg1: BasicObjectLock for lock
1109 //
1110 // Kills:
1111 //      rax
1112 //      c_rarg0, c_rarg1, c_rarg2, c_rarg3, ... (param regs)
1113 //      rscratch1 (scratch reg)
1114 // rax, rbx, rcx, rdx
1115 void InterpreterMacroAssembler::unlock_object(Register lock_reg) {
1116   assert(lock_reg == c_rarg1, "The argument is only for looks. It must be c_rarg1");
1117 
1118   if (LockingMode == LM_MONITOR) {
1119     call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
1120   } else {
1121     Label count_locking, done, slow_case;
1122 
1123     const Register swap_reg   = rax;  // Must use rax for cmpxchg instruction
1124     const Register header_reg = c_rarg2;  // Will contain the old oopMark
1125     const Register obj_reg    = c_rarg3;  // Will contain the oop
1126 
1127     save_bcp(); // Save in case of exception
1128 
1129     if (LockingMode != LM_LIGHTWEIGHT) {
1130       // Convert from BasicObjectLock structure to object and BasicLock
1131       // structure Store the BasicLock address into %rax
1132       lea(swap_reg, Address(lock_reg, BasicObjectLock::lock_offset()));
1133     }
1134 
1135     // Load oop into obj_reg(%c_rarg3)
1136     movptr(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset()));
1137 
1138     // Free entry
1139     movptr(Address(lock_reg, BasicObjectLock::obj_offset()), NULL_WORD);
1140 
1141     if (LockingMode == LM_LIGHTWEIGHT) {
1142       lightweight_unlock(obj_reg, swap_reg, header_reg, slow_case);
1143     } else if (LockingMode == LM_LEGACY) {
1144       // Load the old header from BasicLock structure
1145       movptr(header_reg, Address(swap_reg,
1146                                  BasicLock::displaced_header_offset_in_bytes()));
1147 
1148       // Test for recursion
1149       testptr(header_reg, header_reg);
1150 
1151       // zero for recursive case
1152       jcc(Assembler::zero, count_locking);
1153 
1154       // Atomic swap back the old header
1155       lock();
1156       cmpxchgptr(header_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
1157 
1158       // zero for simple unlock of a stack-lock case
1159       jcc(Assembler::notZero, slow_case);
1160 
1161       bind(count_locking);
1162       dec_held_monitor_count();
1163     }
1164     jmp(done);
1165 
1166     bind(slow_case);
1167     // Call the runtime routine for slow case.
1168     movptr(Address(lock_reg, BasicObjectLock::obj_offset()), obj_reg); // restore obj
1169     call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
1170 
1171     bind(done);
1172 
1173     restore_bcp();
1174   }
1175 }
1176 
1177 void InterpreterMacroAssembler::test_method_data_pointer(Register mdp,
1178                                                          Label& zero_continue) {
1179   assert(ProfileInterpreter, "must be profiling interpreter");
1180   movptr(mdp, Address(rbp, frame::interpreter_frame_mdp_offset * wordSize));
1181   testptr(mdp, mdp);
1182   jcc(Assembler::zero, zero_continue);
1183 }
1184 
1185 
1186 // Set the method data pointer for the current bcp.
1187 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
1188   assert(ProfileInterpreter, "must be profiling interpreter");
1189   Label set_mdp;
1190   push(rax);
1191   push(rbx);
1192 
1193   get_method(rbx);
1194   // Test MDO to avoid the call if it is null.
1195   movptr(rax, Address(rbx, in_bytes(Method::method_data_offset())));
1196   testptr(rax, rax);
1197   jcc(Assembler::zero, set_mdp);
1198   // rbx: method
1199   // _bcp_register: bcp
1200   call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), rbx, _bcp_register);
1201   // rax: mdi
1202   // mdo is guaranteed to be non-zero here, we checked for it before the call.
1203   movptr(rbx, Address(rbx, in_bytes(Method::method_data_offset())));
1204   addptr(rbx, in_bytes(MethodData::data_offset()));
1205   addptr(rax, rbx);
1206   bind(set_mdp);
1207   movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), rax);
1208   pop(rbx);
1209   pop(rax);
1210 }
1211 
1212 void InterpreterMacroAssembler::verify_method_data_pointer() {
1213   assert(ProfileInterpreter, "must be profiling interpreter");
1214 #ifdef ASSERT
1215   Label verify_continue;
1216   push(rax);
1217   push(rbx);
1218   Register arg3_reg = c_rarg3;
1219   Register arg2_reg = c_rarg2;
1220   push(arg3_reg);
1221   push(arg2_reg);
1222   test_method_data_pointer(arg3_reg, verify_continue); // If mdp is zero, continue
1223   get_method(rbx);
1224 
1225   // If the mdp is valid, it will point to a DataLayout header which is
1226   // consistent with the bcp.  The converse is highly probable also.
1227   load_unsigned_short(arg2_reg,
1228                       Address(arg3_reg, in_bytes(DataLayout::bci_offset())));
1229   addptr(arg2_reg, Address(rbx, Method::const_offset()));
1230   lea(arg2_reg, Address(arg2_reg, ConstMethod::codes_offset()));
1231   cmpptr(arg2_reg, _bcp_register);
1232   jcc(Assembler::equal, verify_continue);
1233   // rbx: method
1234   // _bcp_register: bcp
1235   // c_rarg3: mdp
1236   call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp),
1237                rbx, _bcp_register, arg3_reg);
1238   bind(verify_continue);
1239   pop(arg2_reg);
1240   pop(arg3_reg);
1241   pop(rbx);
1242   pop(rax);
1243 #endif // ASSERT
1244 }
1245 
1246 
1247 void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in,
1248                                                 int constant,
1249                                                 Register value) {
1250   assert(ProfileInterpreter, "must be profiling interpreter");
1251   Address data(mdp_in, constant);
1252   movptr(data, value);
1253 }
1254 
1255 
1256 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
1257                                                       int constant,
1258                                                       bool decrement) {
1259   // Counter address
1260   Address data(mdp_in, constant);
1261 
1262   increment_mdp_data_at(data, decrement);
1263 }
1264 
1265 void InterpreterMacroAssembler::increment_mdp_data_at(Address data,
1266                                                       bool decrement) {
1267   assert(ProfileInterpreter, "must be profiling interpreter");
1268   // %%% this does 64bit counters at best it is wasting space
1269   // at worst it is a rare bug when counters overflow
1270 
1271   if (decrement) {
1272     // Decrement the register.  Set condition codes.
1273     addptr(data, -DataLayout::counter_increment);
1274     // If the decrement causes the counter to overflow, stay negative
1275     Label L;
1276     jcc(Assembler::negative, L);
1277     addptr(data, DataLayout::counter_increment);
1278     bind(L);
1279   } else {
1280     assert(DataLayout::counter_increment == 1,
1281            "flow-free idiom only works with 1");
1282     // Increment the register.  Set carry flag.
1283     addptr(data, DataLayout::counter_increment);
1284     // If the increment causes the counter to overflow, pull back by 1.
1285     sbbptr(data, 0);
1286   }
1287 }
1288 
1289 
1290 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
1291                                                       Register reg,
1292                                                       int constant,
1293                                                       bool decrement) {
1294   Address data(mdp_in, reg, Address::times_1, constant);
1295 
1296   increment_mdp_data_at(data, decrement);
1297 }
1298 
1299 void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in,
1300                                                 int flag_byte_constant) {
1301   assert(ProfileInterpreter, "must be profiling interpreter");
1302   int header_offset = in_bytes(DataLayout::flags_offset());
1303   int header_bits = flag_byte_constant;
1304   // Set the flag
1305   orb(Address(mdp_in, header_offset), header_bits);
1306 }
1307 
1308 
1309 
1310 void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in,
1311                                                  int offset,
1312                                                  Register value,
1313                                                  Register test_value_out,
1314                                                  Label& not_equal_continue) {
1315   assert(ProfileInterpreter, "must be profiling interpreter");
1316   if (test_value_out == noreg) {
1317     cmpptr(value, Address(mdp_in, offset));
1318   } else {
1319     // Put the test value into a register, so caller can use it:
1320     movptr(test_value_out, Address(mdp_in, offset));
1321     cmpptr(test_value_out, value);
1322   }
1323   jcc(Assembler::notEqual, not_equal_continue);
1324 }
1325 
1326 
1327 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1328                                                      int offset_of_disp) {
1329   assert(ProfileInterpreter, "must be profiling interpreter");
1330   Address disp_address(mdp_in, offset_of_disp);
1331   addptr(mdp_in, disp_address);
1332   movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in);
1333 }
1334 
1335 
1336 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1337                                                      Register reg,
1338                                                      int offset_of_disp) {
1339   assert(ProfileInterpreter, "must be profiling interpreter");
1340   Address disp_address(mdp_in, reg, Address::times_1, offset_of_disp);
1341   addptr(mdp_in, disp_address);
1342   movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in);
1343 }
1344 
1345 
1346 void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in,
1347                                                        int constant) {
1348   assert(ProfileInterpreter, "must be profiling interpreter");
1349   addptr(mdp_in, constant);
1350   movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in);
1351 }
1352 
1353 
1354 void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) {
1355   assert(ProfileInterpreter, "must be profiling interpreter");
1356   push(return_bci); // save/restore across call_VM
1357   call_VM(noreg,
1358           CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret),
1359           return_bci);
1360   pop(return_bci);
1361 }
1362 
1363 
1364 void InterpreterMacroAssembler::profile_taken_branch(Register mdp,
1365                                                      Register bumped_count) {
1366   if (ProfileInterpreter) {
1367     Label profile_continue;
1368 
1369     // If no method data exists, go to profile_continue.
1370     // Otherwise, assign to mdp
1371     test_method_data_pointer(mdp, profile_continue);
1372 
1373     // We are taking a branch.  Increment the taken count.
1374     // We inline increment_mdp_data_at to return bumped_count in a register
1375     //increment_mdp_data_at(mdp, in_bytes(JumpData::taken_offset()));
1376     Address data(mdp, in_bytes(JumpData::taken_offset()));
1377     movptr(bumped_count, data);
1378     assert(DataLayout::counter_increment == 1,
1379             "flow-free idiom only works with 1");
1380     addptr(bumped_count, DataLayout::counter_increment);
1381     sbbptr(bumped_count, 0);
1382     movptr(data, bumped_count); // Store back out
1383 
1384     // The method data pointer needs to be updated to reflect the new target.
1385     update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset()));
1386     bind(profile_continue);
1387   }
1388 }
1389 
1390 
1391 void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp) {
1392   if (ProfileInterpreter) {
1393     Label profile_continue;
1394 
1395     // If no method data exists, go to profile_continue.
1396     test_method_data_pointer(mdp, profile_continue);
1397 
1398     // We are taking a branch.  Increment the not taken count.
1399     increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset()));
1400 
1401     // The method data pointer needs to be updated to correspond to
1402     // the next bytecode
1403     update_mdp_by_constant(mdp, in_bytes(BranchData::branch_data_size()));
1404     bind(profile_continue);
1405   }
1406 }
1407 
1408 void InterpreterMacroAssembler::profile_call(Register mdp) {
1409   if (ProfileInterpreter) {
1410     Label profile_continue;
1411 
1412     // If no method data exists, go to profile_continue.
1413     test_method_data_pointer(mdp, profile_continue);
1414 
1415     // We are making a call.  Increment the count.
1416     increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1417 
1418     // The method data pointer needs to be updated to reflect the new target.
1419     update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size()));
1420     bind(profile_continue);
1421   }
1422 }
1423 
1424 
1425 void InterpreterMacroAssembler::profile_final_call(Register mdp) {
1426   if (ProfileInterpreter) {
1427     Label profile_continue;
1428 
1429     // If no method data exists, go to profile_continue.
1430     test_method_data_pointer(mdp, profile_continue);
1431 
1432     // We are making a call.  Increment the count.
1433     increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1434 
1435     // The method data pointer needs to be updated to reflect the new target.
1436     update_mdp_by_constant(mdp,
1437                            in_bytes(VirtualCallData::
1438                                     virtual_call_data_size()));
1439     bind(profile_continue);
1440   }
1441 }
1442 
1443 
1444 void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
1445                                                      Register mdp,
1446                                                      Register reg2,
1447                                                      bool receiver_can_be_null) {
1448   if (ProfileInterpreter) {
1449     Label profile_continue;
1450 
1451     // If no method data exists, go to profile_continue.
1452     test_method_data_pointer(mdp, profile_continue);
1453 
1454     Label skip_receiver_profile;
1455     if (receiver_can_be_null) {
1456       Label not_null;
1457       testptr(receiver, receiver);
1458       jccb(Assembler::notZero, not_null);
1459       // We are making a call.  Increment the count for null receiver.
1460       increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1461       jmp(skip_receiver_profile);
1462       bind(not_null);
1463     }
1464 
1465     // Record the receiver type.
1466     record_klass_in_profile(receiver, mdp, reg2, true);
1467     bind(skip_receiver_profile);
1468 
1469     // The method data pointer needs to be updated to reflect the new target.
1470     update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size()));
1471     bind(profile_continue);
1472   }
1473 }
1474 
1475 // This routine creates a state machine for updating the multi-row
1476 // type profile at a virtual call site (or other type-sensitive bytecode).
1477 // The machine visits each row (of receiver/count) until the receiver type
1478 // is found, or until it runs out of rows.  At the same time, it remembers
1479 // the location of the first empty row.  (An empty row records null for its
1480 // receiver, and can be allocated for a newly-observed receiver type.)
1481 // Because there are two degrees of freedom in the state, a simple linear
1482 // search will not work; it must be a decision tree.  Hence this helper
1483 // function is recursive, to generate the required tree structured code.
1484 // It's the interpreter, so we are trading off code space for speed.
1485 // See below for example code.
1486 void InterpreterMacroAssembler::record_klass_in_profile_helper(
1487                                         Register receiver, Register mdp,
1488                                         Register reg2, int start_row,
1489                                         Label& done, bool is_virtual_call) {
1490   if (TypeProfileWidth == 0) {
1491     increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1492   } else {
1493     record_item_in_profile_helper(receiver, mdp, reg2, 0, done, TypeProfileWidth,
1494                                   &VirtualCallData::receiver_offset, &VirtualCallData::receiver_count_offset);
1495   }
1496 }
1497 
1498 void InterpreterMacroAssembler::record_item_in_profile_helper(Register item, Register mdp, Register reg2, int start_row,
1499                                                               Label& done, int total_rows,
1500                                                               OffsetFunction item_offset_fn,
1501                                                               OffsetFunction item_count_offset_fn) {
1502   int last_row = total_rows - 1;
1503   assert(start_row <= last_row, "must be work left to do");
1504   // Test this row for both the item and for null.
1505   // Take any of three different outcomes:
1506   //   1. found item => increment count and goto done
1507   //   2. found null => keep looking for case 1, maybe allocate this cell
1508   //   3. found something else => keep looking for cases 1 and 2
1509   // Case 3 is handled by a recursive call.
1510   for (int row = start_row; row <= last_row; row++) {
1511     Label next_test;
1512     bool test_for_null_also = (row == start_row);
1513 
1514     // See if the item is item[n].
1515     int item_offset = in_bytes(item_offset_fn(row));
1516     test_mdp_data_at(mdp, item_offset, item,
1517                      (test_for_null_also ? reg2 : noreg),
1518                      next_test);
1519     // (Reg2 now contains the item from the CallData.)
1520 
1521     // The item is item[n].  Increment count[n].
1522     int count_offset = in_bytes(item_count_offset_fn(row));
1523     increment_mdp_data_at(mdp, count_offset);
1524     jmp(done);
1525     bind(next_test);
1526 
1527     if (test_for_null_also) {
1528       // Failed the equality check on item[n]...  Test for null.
1529       testptr(reg2, reg2);
1530       if (start_row == last_row) {
1531         // The only thing left to do is handle the null case.
1532         Label found_null;
1533         jccb(Assembler::zero, found_null);
1534         // Item did not match any saved item and there is no empty row for it.
1535         // Increment total counter to indicate polymorphic case.
1536         increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1537         jmp(done);
1538         bind(found_null);
1539         break;
1540       }
1541       Label found_null;
1542       // Since null is rare, make it be the branch-taken case.
1543       jcc(Assembler::zero, found_null);
1544 
1545       // Put all the "Case 3" tests here.
1546       record_item_in_profile_helper(item, mdp, reg2, start_row + 1, done, total_rows,
1547                                     item_offset_fn, item_count_offset_fn);
1548 
1549       // Found a null.  Keep searching for a matching item,
1550       // but remember that this is an empty (unused) slot.
1551       bind(found_null);
1552     }
1553   }
1554 
1555   // In the fall-through case, we found no matching item, but we
1556   // observed the item[start_row] is null.
1557 
1558   // Fill in the item field and increment the count.
1559   int item_offset = in_bytes(item_offset_fn(start_row));
1560   set_mdp_data_at(mdp, item_offset, item);
1561   int count_offset = in_bytes(item_count_offset_fn(start_row));
1562   movl(reg2, DataLayout::counter_increment);
1563   set_mdp_data_at(mdp, count_offset, reg2);
1564   if (start_row > 0) {
1565     jmp(done);
1566   }
1567 }
1568 
1569 // Example state machine code for three profile rows:
1570 //   // main copy of decision tree, rooted at row[1]
1571 //   if (row[0].rec == rec) { row[0].incr(); goto done; }
1572 //   if (row[0].rec != nullptr) {
1573 //     // inner copy of decision tree, rooted at row[1]
1574 //     if (row[1].rec == rec) { row[1].incr(); goto done; }
1575 //     if (row[1].rec != nullptr) {
1576 //       // degenerate decision tree, rooted at row[2]
1577 //       if (row[2].rec == rec) { row[2].incr(); goto done; }
1578 //       if (row[2].rec != nullptr) { count.incr(); goto done; } // overflow
1579 //       row[2].init(rec); goto done;
1580 //     } else {
1581 //       // remember row[1] is empty
1582 //       if (row[2].rec == rec) { row[2].incr(); goto done; }
1583 //       row[1].init(rec); goto done;
1584 //     }
1585 //   } else {
1586 //     // remember row[0] is empty
1587 //     if (row[1].rec == rec) { row[1].incr(); goto done; }
1588 //     if (row[2].rec == rec) { row[2].incr(); goto done; }
1589 //     row[0].init(rec); goto done;
1590 //   }
1591 //   done:
1592 
1593 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver,
1594                                                         Register mdp, Register reg2,
1595                                                         bool is_virtual_call) {
1596   assert(ProfileInterpreter, "must be profiling");
1597   Label done;
1598 
1599   record_klass_in_profile_helper(receiver, mdp, reg2, 0, done, is_virtual_call);
1600 
1601   bind (done);
1602 }
1603 
1604 void InterpreterMacroAssembler::profile_ret(Register return_bci,
1605                                             Register mdp) {
1606   if (ProfileInterpreter) {
1607     Label profile_continue;
1608     uint row;
1609 
1610     // If no method data exists, go to profile_continue.
1611     test_method_data_pointer(mdp, profile_continue);
1612 
1613     // Update the total ret count.
1614     increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1615 
1616     for (row = 0; row < RetData::row_limit(); row++) {
1617       Label next_test;
1618 
1619       // See if return_bci is equal to bci[n]:
1620       test_mdp_data_at(mdp,
1621                        in_bytes(RetData::bci_offset(row)),
1622                        return_bci, noreg,
1623                        next_test);
1624 
1625       // return_bci is equal to bci[n].  Increment the count.
1626       increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row)));
1627 
1628       // The method data pointer needs to be updated to reflect the new target.
1629       update_mdp_by_offset(mdp,
1630                            in_bytes(RetData::bci_displacement_offset(row)));
1631       jmp(profile_continue);
1632       bind(next_test);
1633     }
1634 
1635     update_mdp_for_ret(return_bci);
1636 
1637     bind(profile_continue);
1638   }
1639 }
1640 
1641 
1642 void InterpreterMacroAssembler::profile_null_seen(Register mdp) {
1643   if (ProfileInterpreter) {
1644     Label profile_continue;
1645 
1646     // If no method data exists, go to profile_continue.
1647     test_method_data_pointer(mdp, profile_continue);
1648 
1649     set_mdp_flag_at(mdp, BitData::null_seen_byte_constant());
1650 
1651     // The method data pointer needs to be updated.
1652     int mdp_delta = in_bytes(BitData::bit_data_size());
1653     if (TypeProfileCasts) {
1654       mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1655     }
1656     update_mdp_by_constant(mdp, mdp_delta);
1657 
1658     bind(profile_continue);
1659   }
1660 }
1661 
1662 
1663 void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) {
1664   if (ProfileInterpreter) {
1665     Label profile_continue;
1666 
1667     // If no method data exists, go to profile_continue.
1668     test_method_data_pointer(mdp, profile_continue);
1669 
1670     // The method data pointer needs to be updated.
1671     int mdp_delta = in_bytes(BitData::bit_data_size());
1672     if (TypeProfileCasts) {
1673       mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1674 
1675       // Record the object type.
1676       record_klass_in_profile(klass, mdp, reg2, false);
1677     }
1678     update_mdp_by_constant(mdp, mdp_delta);
1679 
1680     bind(profile_continue);
1681   }
1682 }
1683 
1684 
1685 void InterpreterMacroAssembler::profile_switch_default(Register mdp) {
1686   if (ProfileInterpreter) {
1687     Label profile_continue;
1688 
1689     // If no method data exists, go to profile_continue.
1690     test_method_data_pointer(mdp, profile_continue);
1691 
1692     // Update the default case count
1693     increment_mdp_data_at(mdp,
1694                           in_bytes(MultiBranchData::default_count_offset()));
1695 
1696     // The method data pointer needs to be updated.
1697     update_mdp_by_offset(mdp,
1698                          in_bytes(MultiBranchData::
1699                                   default_displacement_offset()));
1700 
1701     bind(profile_continue);
1702   }
1703 }
1704 
1705 
1706 void InterpreterMacroAssembler::profile_switch_case(Register index,
1707                                                     Register mdp,
1708                                                     Register reg2) {
1709   if (ProfileInterpreter) {
1710     Label profile_continue;
1711 
1712     // If no method data exists, go to profile_continue.
1713     test_method_data_pointer(mdp, profile_continue);
1714 
1715     // Build the base (index * per_case_size_in_bytes()) +
1716     // case_array_offset_in_bytes()
1717     movl(reg2, in_bytes(MultiBranchData::per_case_size()));
1718     imulptr(index, reg2); // XXX l ?
1719     addptr(index, in_bytes(MultiBranchData::case_array_offset())); // XXX l ?
1720 
1721     // Update the case count
1722     increment_mdp_data_at(mdp,
1723                           index,
1724                           in_bytes(MultiBranchData::relative_count_offset()));
1725 
1726     // The method data pointer needs to be updated.
1727     update_mdp_by_offset(mdp,
1728                          index,
1729                          in_bytes(MultiBranchData::
1730                                   relative_displacement_offset()));
1731 
1732     bind(profile_continue);
1733   }
1734 }
1735 
1736 
1737 
1738 void InterpreterMacroAssembler::_interp_verify_oop(Register reg, TosState state, const char* file, int line) {
1739   if (state == atos) {
1740     MacroAssembler::_verify_oop_checked(reg, "broken oop", file, line);
1741   }
1742 }
1743 
1744 
1745 // Jump if ((*counter_addr += increment) & mask) == 0
1746 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr, Address mask,
1747                                                         Register scratch, Label* where) {
1748   // This update is actually not atomic and can lose a number of updates
1749   // under heavy contention, but the alternative of using the (contended)
1750   // atomic update here penalizes profiling paths too much.
1751   movl(scratch, counter_addr);
1752   incrementl(scratch, InvocationCounter::count_increment);
1753   movl(counter_addr, scratch);
1754   andl(scratch, mask);
1755   if (where != nullptr) {
1756     jcc(Assembler::zero, *where);
1757   }
1758 }
1759 
1760 void InterpreterMacroAssembler::notify_method_entry() {
1761   // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1762   // track stack depth.  If it is possible to enter interp_only_mode we add
1763   // the code to check if the event should be sent.
1764   Register rthread = r15_thread;
1765   Register rarg = c_rarg1;
1766   if (JvmtiExport::can_post_interpreter_events()) {
1767     Label L;
1768     movl(rdx, Address(rthread, JavaThread::interp_only_mode_offset()));
1769     testl(rdx, rdx);
1770     jcc(Assembler::zero, L);
1771     call_VM(noreg, CAST_FROM_FN_PTR(address,
1772                                     InterpreterRuntime::post_method_entry));
1773     bind(L);
1774   }
1775 
1776   if (DTraceMethodProbes) {
1777     get_method(rarg);
1778     call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
1779                  rthread, rarg);
1780   }
1781 
1782   // RedefineClasses() tracing support for obsolete method entry
1783   if (log_is_enabled(Trace, redefine, class, obsolete)) {
1784     get_method(rarg);
1785     call_VM_leaf(
1786       CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
1787       rthread, rarg);
1788   }
1789 }
1790 
1791 
1792 void InterpreterMacroAssembler::notify_method_exit(
1793     TosState state, NotifyMethodExitMode mode) {
1794   // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1795   // track stack depth.  If it is possible to enter interp_only_mode we add
1796   // the code to check if the event should be sent.
1797   Register rthread = r15_thread;
1798   Register rarg = c_rarg1;
1799   if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
1800     Label L;
1801     // Note: frame::interpreter_frame_result has a dependency on how the
1802     // method result is saved across the call to post_method_exit. If this
1803     // is changed then the interpreter_frame_result implementation will
1804     // need to be updated too.
1805 
1806     // template interpreter will leave the result on the top of the stack.
1807     push(state);
1808     movl(rdx, Address(rthread, JavaThread::interp_only_mode_offset()));
1809     testl(rdx, rdx);
1810     jcc(Assembler::zero, L);
1811     call_VM(noreg,
1812             CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
1813     bind(L);
1814     pop(state);
1815   }
1816 
1817   if (DTraceMethodProbes) {
1818     push(state);
1819     get_method(rarg);
1820     call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
1821                  rthread, rarg);
1822     pop(state);
1823   }
1824 }
1825 
1826 void InterpreterMacroAssembler::load_resolved_indy_entry(Register cache, Register index) {
1827   // Get index out of bytecode pointer
1828   get_cache_index_at_bcp(index, 1, sizeof(u4));
1829   // Get address of invokedynamic array
1830   movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
1831   movptr(cache, Address(cache, in_bytes(ConstantPoolCache::invokedynamic_entries_offset())));
1832   if (is_power_of_2(sizeof(ResolvedIndyEntry))) {
1833     shll(index, log2i_exact(sizeof(ResolvedIndyEntry))); // Scale index by power of 2
1834   } else {
1835     imull(index, index, sizeof(ResolvedIndyEntry)); // Scale the index to be the entry index * sizeof(ResolvedIndyEntry)
1836   }
1837   lea(cache, Address(cache, index, Address::times_1, Array<ResolvedIndyEntry>::base_offset_in_bytes()));
1838 }
1839 
1840 void InterpreterMacroAssembler::load_field_entry(Register cache, Register index, int bcp_offset) {
1841   // Get index out of bytecode pointer
1842   movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
1843   get_cache_index_at_bcp(index, bcp_offset, sizeof(u2));
1844 
1845   movptr(cache, Address(cache, ConstantPoolCache::field_entries_offset()));
1846   // Take shortcut if the size is a power of 2
1847   if (is_power_of_2(sizeof(ResolvedFieldEntry))) {
1848     shll(index, log2i_exact(sizeof(ResolvedFieldEntry))); // Scale index by power of 2
1849   } else {
1850     imull(index, index, sizeof(ResolvedFieldEntry)); // Scale the index to be the entry index * sizeof(ResolvedFieldEntry)
1851   }
1852   lea(cache, Address(cache, index, Address::times_1, Array<ResolvedFieldEntry>::base_offset_in_bytes()));
1853 }
1854 
1855 void InterpreterMacroAssembler::load_method_entry(Register cache, Register index, int bcp_offset) {
1856   // Get index out of bytecode pointer
1857   movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
1858   get_cache_index_at_bcp(index, bcp_offset, sizeof(u2));
1859 
1860   movptr(cache, Address(cache, ConstantPoolCache::method_entries_offset()));
1861   imull(index, index, sizeof(ResolvedMethodEntry)); // Scale the index to be the entry index * sizeof(ResolvedMethodEntry)
1862   lea(cache, Address(cache, index, Address::times_1, Array<ResolvedMethodEntry>::base_offset_in_bytes()));
1863 }