1 /*
   2  * Copyright (c) 1997, 2024, 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 "compiler/compiler_globals.hpp"
  27 #include "interp_masm_x86.hpp"
  28 #include "interpreter/interpreter.hpp"
  29 #include "interpreter/interpreterRuntime.hpp"
  30 #include "logging/log.hpp"
  31 #include "oops/arrayOop.hpp"
  32 #include "oops/markWord.hpp"
  33 #include "oops/methodData.hpp"
  34 #include "oops/method.hpp"
  35 #include "oops/resolvedFieldEntry.hpp"
  36 #include "oops/resolvedIndyEntry.hpp"
  37 #include "oops/resolvedMethodEntry.hpp"
  38 #include "prims/jvmtiExport.hpp"
  39 #include "prims/jvmtiThreadState.hpp"
  40 #include "runtime/basicLock.hpp"
  41 #include "runtime/frame.inline.hpp"
  42 #include "runtime/javaThread.hpp"
  43 #include "runtime/safepointMechanism.hpp"
  44 #include "runtime/sharedRuntime.hpp"
  45 #include "utilities/powerOfTwo.hpp"
  46 
  47 // Implementation of InterpreterMacroAssembler
  48 
  49 void InterpreterMacroAssembler::jump_to_entry(address entry) {
  50   assert(entry, "Entry must have been generated by now");
  51   jump(RuntimeAddress(entry));
  52 }
  53 
  54 void InterpreterMacroAssembler::profile_obj_type(Register obj, const Address& mdo_addr) {
  55   Label update, next, none;
  56 
  57 #ifdef _LP64
  58   assert_different_registers(obj, rscratch1, mdo_addr.base(), mdo_addr.index());
  59 #else
  60   assert_different_registers(obj, mdo_addr.base(), mdo_addr.index());
  61 #endif
  62 
  63   interp_verify_oop(obj, atos);
  64 
  65   testptr(obj, obj);
  66   jccb(Assembler::notZero, update);
  67   testptr(mdo_addr, TypeEntries::null_seen);
  68   jccb(Assembler::notZero, next); // null already seen. Nothing to do anymore.
  69   // atomic update to prevent overwriting Klass* with 0
  70   lock();
  71   orptr(mdo_addr, TypeEntries::null_seen);
  72   jmpb(next);
  73 
  74   bind(update);
  75   load_klass(obj, obj, rscratch1);
  76 #ifdef _LP64
  77   mov(rscratch1, obj);
  78 #endif
  79 
  80   xorptr(obj, mdo_addr);
  81   testptr(obj, TypeEntries::type_klass_mask);
  82   jccb(Assembler::zero, next); // klass seen before, nothing to
  83                                // do. The unknown bit may have been
  84                                // set already but no need to check.
  85 
  86   testptr(obj, TypeEntries::type_unknown);
  87   jccb(Assembler::notZero, next); // already unknown. Nothing to do anymore.
  88 
  89   cmpptr(mdo_addr, 0);
  90   jccb(Assembler::equal, none);
  91   cmpptr(mdo_addr, TypeEntries::null_seen);
  92   jccb(Assembler::equal, none);
  93 #ifdef _LP64
  94   // There is a chance that the checks above (re-reading profiling
  95   // data from memory) fail if another thread has just set the
  96   // profiling to this obj's klass
  97   mov(obj, rscratch1);
  98   xorptr(obj, mdo_addr);
  99   testptr(obj, TypeEntries::type_klass_mask);
 100   jccb(Assembler::zero, next);
 101 #endif
 102 
 103   // different than before. Cannot keep accurate profile.
 104   orptr(mdo_addr, TypeEntries::type_unknown);
 105   jmpb(next);
 106 
 107   bind(none);
 108   // first time here. Set profile type.
 109   movptr(mdo_addr, obj);
 110 #ifdef ASSERT
 111   andptr(obj, TypeEntries::type_klass_mask);
 112   verify_klass_ptr(obj);
 113 #endif
 114 
 115   bind(next);
 116 }
 117 
 118 void InterpreterMacroAssembler::profile_arguments_type(Register mdp, Register callee, Register tmp, bool is_virtual) {
 119   if (!ProfileInterpreter) {
 120     return;
 121   }
 122 
 123   if (MethodData::profile_arguments() || MethodData::profile_return()) {
 124     Label profile_continue;
 125 
 126     test_method_data_pointer(mdp, profile_continue);
 127 
 128     int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size());
 129 
 130     cmpb(Address(mdp, in_bytes(DataLayout::tag_offset()) - off_to_start), is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag);
 131     jcc(Assembler::notEqual, profile_continue);
 132 
 133     if (MethodData::profile_arguments()) {
 134       Label done;
 135       int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset());
 136       addptr(mdp, off_to_args);
 137 
 138       for (int i = 0; i < TypeProfileArgsLimit; i++) {
 139         if (i > 0 || MethodData::profile_return()) {
 140           // If return value type is profiled we may have no argument to profile
 141           movptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args));
 142           subl(tmp, i*TypeStackSlotEntries::per_arg_count());
 143           cmpl(tmp, TypeStackSlotEntries::per_arg_count());
 144           jcc(Assembler::less, done);
 145         }
 146         movptr(tmp, Address(callee, Method::const_offset()));
 147         load_unsigned_short(tmp, Address(tmp, ConstMethod::size_of_parameters_offset()));
 148         // stack offset o (zero based) from the start of the argument
 149         // list, for n arguments translates into offset n - o - 1 from
 150         // the end of the argument list
 151         subptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))-off_to_args));
 152         subl(tmp, 1);
 153         Address arg_addr = argument_address(tmp);
 154         movptr(tmp, arg_addr);
 155 
 156         Address mdo_arg_addr(mdp, in_bytes(TypeEntriesAtCall::argument_type_offset(i))-off_to_args);
 157         profile_obj_type(tmp, mdo_arg_addr);
 158 
 159         int to_add = in_bytes(TypeStackSlotEntries::per_arg_size());
 160         addptr(mdp, to_add);
 161         off_to_args += to_add;
 162       }
 163 
 164       if (MethodData::profile_return()) {
 165         movptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args));
 166         subl(tmp, TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count());
 167       }
 168 
 169       bind(done);
 170 
 171       if (MethodData::profile_return()) {
 172         // We're right after the type profile for the last
 173         // argument. tmp is the number of cells left in the
 174         // CallTypeData/VirtualCallTypeData to reach its end. Non null
 175         // if there's a return to profile.
 176         assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type");
 177         shll(tmp, log2i_exact((int)DataLayout::cell_size));
 178         addptr(mdp, tmp);
 179       }
 180       movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp);
 181     } else {
 182       assert(MethodData::profile_return(), "either profile call args or call ret");
 183       update_mdp_by_constant(mdp, in_bytes(TypeEntriesAtCall::return_only_size()));
 184     }
 185 
 186     // mdp points right after the end of the
 187     // CallTypeData/VirtualCallTypeData, right after the cells for the
 188     // return value type if there's one
 189 
 190     bind(profile_continue);
 191   }
 192 }
 193 
 194 void InterpreterMacroAssembler::profile_return_type(Register mdp, Register ret, Register tmp) {
 195   assert_different_registers(mdp, ret, tmp, _bcp_register);
 196   if (ProfileInterpreter && MethodData::profile_return()) {
 197     Label profile_continue;
 198 
 199     test_method_data_pointer(mdp, profile_continue);
 200 
 201     if (MethodData::profile_return_jsr292_only()) {
 202       assert(Method::intrinsic_id_size_in_bytes() == 2, "assuming Method::_intrinsic_id is u2");
 203 
 204       // If we don't profile all invoke bytecodes we must make sure
 205       // it's a bytecode we indeed profile. We can't go back to the
 206       // beginning of the ProfileData we intend to update to check its
 207       // type because we're right after it and we don't known its
 208       // length
 209       Label do_profile;
 210       cmpb(Address(_bcp_register, 0), Bytecodes::_invokedynamic);
 211       jcc(Assembler::equal, do_profile);
 212       cmpb(Address(_bcp_register, 0), Bytecodes::_invokehandle);
 213       jcc(Assembler::equal, do_profile);
 214       get_method(tmp);
 215       cmpw(Address(tmp, Method::intrinsic_id_offset()), static_cast<int>(vmIntrinsics::_compiledLambdaForm));
 216       jcc(Assembler::notEqual, profile_continue);
 217 
 218       bind(do_profile);
 219     }
 220 
 221     Address mdo_ret_addr(mdp, -in_bytes(ReturnTypeEntry::size()));
 222     mov(tmp, ret);
 223     profile_obj_type(tmp, mdo_ret_addr);
 224 
 225     bind(profile_continue);
 226   }
 227 }
 228 
 229 void InterpreterMacroAssembler::profile_parameters_type(Register mdp, Register tmp1, Register tmp2) {
 230   if (ProfileInterpreter && MethodData::profile_parameters()) {
 231     Label profile_continue;
 232 
 233     test_method_data_pointer(mdp, profile_continue);
 234 
 235     // Load the offset of the area within the MDO used for
 236     // parameters. If it's negative we're not profiling any parameters
 237     movl(tmp1, Address(mdp, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset())));
 238     testl(tmp1, tmp1);
 239     jcc(Assembler::negative, profile_continue);
 240 
 241     // Compute a pointer to the area for parameters from the offset
 242     // and move the pointer to the slot for the last
 243     // parameters. Collect profiling from last parameter down.
 244     // mdo start + parameters offset + array length - 1
 245     addptr(mdp, tmp1);
 246     movptr(tmp1, Address(mdp, ArrayData::array_len_offset()));
 247     decrement(tmp1, TypeStackSlotEntries::per_arg_count());
 248 
 249     Label loop;
 250     bind(loop);
 251 
 252     int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0));
 253     int type_base = in_bytes(ParametersTypeData::type_offset(0));
 254     Address::ScaleFactor per_arg_scale = Address::times(DataLayout::cell_size);
 255     Address arg_off(mdp, tmp1, per_arg_scale, off_base);
 256     Address arg_type(mdp, tmp1, per_arg_scale, type_base);
 257 
 258     // load offset on the stack from the slot for this parameter
 259     movptr(tmp2, arg_off);
 260     negptr(tmp2);
 261     // read the parameter from the local area
 262     movptr(tmp2, Address(_locals_register, tmp2, Interpreter::stackElementScale()));
 263 
 264     // profile the parameter
 265     profile_obj_type(tmp2, arg_type);
 266 
 267     // go to next parameter
 268     decrement(tmp1, TypeStackSlotEntries::per_arg_count());
 269     jcc(Assembler::positive, loop);
 270 
 271     bind(profile_continue);
 272   }
 273 }
 274 
 275 void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point,
 276                                                   int number_of_arguments) {
 277   // interpreter specific
 278   //
 279   // Note: No need to save/restore bcp & locals registers
 280   //       since these are callee saved registers and no blocking/
 281   //       GC can happen in leaf calls.
 282   // Further Note: DO NOT save/restore bcp/locals. If a caller has
 283   // already saved them so that it can use rsi/rdi as temporaries
 284   // then a save/restore here will DESTROY the copy the caller
 285   // saved! There used to be a save_bcp() that only happened in
 286   // the ASSERT path (no restore_bcp). Which caused bizarre failures
 287   // when jvm built with ASSERTs.
 288 #ifdef ASSERT
 289   {
 290     Label L;
 291     cmpptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
 292     jcc(Assembler::equal, L);
 293     stop("InterpreterMacroAssembler::call_VM_leaf_base:"
 294          " last_sp != null");
 295     bind(L);
 296   }
 297 #endif
 298   // super call
 299   MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments);
 300   // interpreter specific
 301   // LP64: Used to ASSERT that r13/r14 were equal to frame's bcp/locals
 302   // but since they may not have been saved (and we don't want to
 303   // save them here (see note above) the assert is invalid.
 304 }
 305 
 306 void InterpreterMacroAssembler::call_VM_base(Register oop_result,
 307                                              Register java_thread,
 308                                              Register last_java_sp,
 309                                              address  entry_point,
 310                                              int      number_of_arguments,
 311                                              bool     check_exceptions) {
 312   // interpreter specific
 313   //
 314   // Note: Could avoid restoring locals ptr (callee saved) - however doesn't
 315   //       really make a difference for these runtime calls, since they are
 316   //       slow anyway. Btw., bcp must be saved/restored since it may change
 317   //       due to GC.
 318   NOT_LP64(assert(java_thread == noreg , "not expecting a precomputed java thread");)
 319   save_bcp();
 320 #ifdef ASSERT
 321   {
 322     Label L;
 323     cmpptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
 324     jcc(Assembler::equal, L);
 325     stop("InterpreterMacroAssembler::call_VM_base:"
 326          " last_sp isn't null");
 327     bind(L);
 328   }
 329 #endif /* ASSERT */
 330   // super call
 331   MacroAssembler::call_VM_base(oop_result, noreg, last_java_sp,
 332                                entry_point, number_of_arguments,
 333                                check_exceptions);
 334   // interpreter specific
 335   restore_bcp();
 336   restore_locals();
 337 }
 338 
 339 void InterpreterMacroAssembler::check_and_handle_popframe(Register java_thread) {
 340   if (JvmtiExport::can_pop_frame()) {
 341     Label L;
 342     // Initiate popframe handling only if it is not already being
 343     // processed.  If the flag has the popframe_processing bit set, it
 344     // means that this code is called *during* popframe handling - we
 345     // don't want to reenter.
 346     // This method is only called just after the call into the vm in
 347     // call_VM_base, so the arg registers are available.
 348     Register pop_cond = NOT_LP64(java_thread) // Not clear if any other register is available on 32 bit
 349                         LP64_ONLY(c_rarg0);
 350     movl(pop_cond, Address(java_thread, JavaThread::popframe_condition_offset()));
 351     testl(pop_cond, JavaThread::popframe_pending_bit);
 352     jcc(Assembler::zero, L);
 353     testl(pop_cond, JavaThread::popframe_processing_bit);
 354     jcc(Assembler::notZero, L);
 355     // Call Interpreter::remove_activation_preserving_args_entry() to get the
 356     // address of the same-named entrypoint in the generated interpreter code.
 357     call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry));
 358     jmp(rax);
 359     bind(L);
 360     NOT_LP64(get_thread(java_thread);)
 361   }
 362 }
 363 
 364 void InterpreterMacroAssembler::load_earlyret_value(TosState state) {
 365   Register thread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
 366   NOT_LP64(get_thread(thread);)
 367   movptr(rcx, Address(thread, JavaThread::jvmti_thread_state_offset()));
 368   const Address tos_addr(rcx, JvmtiThreadState::earlyret_tos_offset());
 369   const Address oop_addr(rcx, JvmtiThreadState::earlyret_oop_offset());
 370   const Address val_addr(rcx, JvmtiThreadState::earlyret_value_offset());
 371 #ifdef _LP64
 372   switch (state) {
 373     case atos: movptr(rax, oop_addr);
 374                movptr(oop_addr, NULL_WORD);
 375                interp_verify_oop(rax, state);         break;
 376     case ltos: movptr(rax, val_addr);                 break;
 377     case btos:                                   // fall through
 378     case ztos:                                   // fall through
 379     case ctos:                                   // fall through
 380     case stos:                                   // fall through
 381     case itos: movl(rax, val_addr);                 break;
 382     case ftos: load_float(val_addr);                break;
 383     case dtos: load_double(val_addr);               break;
 384     case vtos: /* nothing to do */                  break;
 385     default  : ShouldNotReachHere();
 386   }
 387   // Clean up tos value in the thread object
 388   movl(tos_addr, ilgl);
 389   movl(val_addr, NULL_WORD);
 390 #else
 391   const Address val_addr1(rcx, JvmtiThreadState::earlyret_value_offset()
 392                              + in_ByteSize(wordSize));
 393   switch (state) {
 394     case atos: movptr(rax, oop_addr);
 395                movptr(oop_addr, NULL_WORD);
 396                interp_verify_oop(rax, state);         break;
 397     case ltos:
 398                movl(rdx, val_addr1);               // fall through
 399     case btos:                                     // fall through
 400     case ztos:                                     // fall through
 401     case ctos:                                     // fall through
 402     case stos:                                     // fall through
 403     case itos: movl(rax, val_addr);                   break;
 404     case ftos: load_float(val_addr);                  break;
 405     case dtos: load_double(val_addr);                 break;
 406     case vtos: /* nothing to do */                    break;
 407     default  : ShouldNotReachHere();
 408   }
 409 #endif // _LP64
 410   // Clean up tos value in the thread object
 411   movl(tos_addr, ilgl);
 412   movptr(val_addr, NULL_WORD);
 413   NOT_LP64(movptr(val_addr1, NULL_WORD);)
 414 }
 415 
 416 
 417 void InterpreterMacroAssembler::check_and_handle_earlyret(Register java_thread) {
 418   if (JvmtiExport::can_force_early_return()) {
 419     Label L;
 420     Register tmp = LP64_ONLY(c_rarg0) NOT_LP64(java_thread);
 421     Register rthread = LP64_ONLY(r15_thread) NOT_LP64(java_thread);
 422 
 423     movptr(tmp, Address(rthread, JavaThread::jvmti_thread_state_offset()));
 424     testptr(tmp, tmp);
 425     jcc(Assembler::zero, L); // if (thread->jvmti_thread_state() == nullptr) exit;
 426 
 427     // Initiate earlyret handling only if it is not already being processed.
 428     // If the flag has the earlyret_processing bit set, it means that this code
 429     // is called *during* earlyret handling - we don't want to reenter.
 430     movl(tmp, Address(tmp, JvmtiThreadState::earlyret_state_offset()));
 431     cmpl(tmp, JvmtiThreadState::earlyret_pending);
 432     jcc(Assembler::notEqual, L);
 433 
 434     // Call Interpreter::remove_activation_early_entry() to get the address of the
 435     // same-named entrypoint in the generated interpreter code.
 436     NOT_LP64(get_thread(java_thread);)
 437     movptr(tmp, Address(rthread, JavaThread::jvmti_thread_state_offset()));
 438 #ifdef _LP64
 439     movl(tmp, Address(tmp, JvmtiThreadState::earlyret_tos_offset()));
 440     call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), tmp);
 441 #else
 442     pushl(Address(tmp, JvmtiThreadState::earlyret_tos_offset()));
 443     call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), 1);
 444 #endif // _LP64
 445     jmp(rax);
 446     bind(L);
 447     NOT_LP64(get_thread(java_thread);)
 448   }
 449 }
 450 
 451 void InterpreterMacroAssembler::get_unsigned_2_byte_index_at_bcp(Register reg, int bcp_offset) {
 452   assert(bcp_offset >= 0, "bcp is still pointing to start of bytecode");
 453   load_unsigned_short(reg, Address(_bcp_register, bcp_offset));
 454   bswapl(reg);
 455   shrl(reg, 16);
 456 }
 457 
 458 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register index,
 459                                                        int bcp_offset,
 460                                                        size_t index_size) {
 461   assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
 462   if (index_size == sizeof(u2)) {
 463     load_unsigned_short(index, Address(_bcp_register, bcp_offset));
 464   } else if (index_size == sizeof(u4)) {
 465     movl(index, Address(_bcp_register, bcp_offset));
 466   } else if (index_size == sizeof(u1)) {
 467     load_unsigned_byte(index, Address(_bcp_register, bcp_offset));
 468   } else {
 469     ShouldNotReachHere();
 470   }
 471 }
 472 
 473 // Load object from cpool->resolved_references(index)
 474 void InterpreterMacroAssembler::load_resolved_reference_at_index(Register result,
 475                                                                  Register index,
 476                                                                  Register tmp) {
 477   assert_different_registers(result, index);
 478 
 479   get_constant_pool(result);
 480   // load pointer for resolved_references[] objArray
 481   movptr(result, Address(result, ConstantPool::cache_offset()));
 482   movptr(result, Address(result, ConstantPoolCache::resolved_references_offset()));
 483   resolve_oop_handle(result, tmp);
 484   load_heap_oop(result, Address(result, index,
 485                                 UseCompressedOops ? Address::times_4 : Address::times_ptr,
 486                                 arrayOopDesc::base_offset_in_bytes(T_OBJECT)), tmp);
 487 }
 488 
 489 // load cpool->resolved_klass_at(index)
 490 void InterpreterMacroAssembler::load_resolved_klass_at_index(Register klass,
 491                                                              Register cpool,
 492                                                              Register index) {
 493   assert_different_registers(cpool, index);
 494 
 495   movw(index, Address(cpool, index, Address::times_ptr, sizeof(ConstantPool)));
 496   Register resolved_klasses = cpool;
 497   movptr(resolved_klasses, Address(cpool, ConstantPool::resolved_klasses_offset()));
 498   movptr(klass, Address(resolved_klasses, index, Address::times_ptr, Array<Klass*>::base_offset_in_bytes()));
 499 }
 500 
 501 // Generate a subtype check: branch to ok_is_subtype if sub_klass is a
 502 // subtype of super_klass.
 503 //
 504 // Args:
 505 //      rax: superklass
 506 //      Rsub_klass: subklass
 507 //
 508 // Kills:
 509 //      rcx, rdi
 510 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass,
 511                                                   Label& ok_is_subtype) {
 512   assert(Rsub_klass != rax, "rax holds superklass");
 513   LP64_ONLY(assert(Rsub_klass != r14, "r14 holds locals");)
 514   LP64_ONLY(assert(Rsub_klass != r13, "r13 holds bcp");)
 515   assert(Rsub_klass != rcx, "rcx holds 2ndary super array length");
 516   assert(Rsub_klass != rdi, "rdi holds 2ndary super array scan ptr");
 517 
 518   // Profile the not-null value's klass.
 519   profile_typecheck(rcx, Rsub_klass, rdi); // blows rcx, reloads rdi
 520 
 521   // Do the check.
 522   check_klass_subtype(Rsub_klass, rax, rcx, ok_is_subtype); // blows rcx
 523 }
 524 
 525 
 526 #ifndef _LP64
 527 void InterpreterMacroAssembler::f2ieee() {
 528   if (IEEEPrecision) {
 529     fstp_s(Address(rsp, 0));
 530     fld_s(Address(rsp, 0));
 531   }
 532 }
 533 
 534 
 535 void InterpreterMacroAssembler::d2ieee() {
 536   if (IEEEPrecision) {
 537     fstp_d(Address(rsp, 0));
 538     fld_d(Address(rsp, 0));
 539   }
 540 }
 541 #endif // _LP64
 542 
 543 // Java Expression Stack
 544 
 545 void InterpreterMacroAssembler::pop_ptr(Register r) {
 546   pop(r);
 547 }
 548 
 549 void InterpreterMacroAssembler::push_ptr(Register r) {
 550   push(r);
 551 }
 552 
 553 void InterpreterMacroAssembler::push_i(Register r) {
 554   push(r);
 555 }
 556 
 557 void InterpreterMacroAssembler::push_i_or_ptr(Register r) {
 558   push(r);
 559 }
 560 
 561 void InterpreterMacroAssembler::push_f(XMMRegister r) {
 562   subptr(rsp, wordSize);
 563   movflt(Address(rsp, 0), r);
 564 }
 565 
 566 void InterpreterMacroAssembler::pop_f(XMMRegister r) {
 567   movflt(r, Address(rsp, 0));
 568   addptr(rsp, wordSize);
 569 }
 570 
 571 void InterpreterMacroAssembler::push_d(XMMRegister r) {
 572   subptr(rsp, 2 * wordSize);
 573   movdbl(Address(rsp, 0), r);
 574 }
 575 
 576 void InterpreterMacroAssembler::pop_d(XMMRegister r) {
 577   movdbl(r, Address(rsp, 0));
 578   addptr(rsp, 2 * Interpreter::stackElementSize);
 579 }
 580 
 581 #ifdef _LP64
 582 void InterpreterMacroAssembler::pop_i(Register r) {
 583   // XXX can't use pop currently, upper half non clean
 584   movl(r, Address(rsp, 0));
 585   addptr(rsp, wordSize);
 586 }
 587 
 588 void InterpreterMacroAssembler::pop_l(Register r) {
 589   movq(r, Address(rsp, 0));
 590   addptr(rsp, 2 * Interpreter::stackElementSize);
 591 }
 592 
 593 void InterpreterMacroAssembler::push_l(Register r) {
 594   subptr(rsp, 2 * wordSize);
 595   movptr(Address(rsp, Interpreter::expr_offset_in_bytes(0)), r         );
 596   movptr(Address(rsp, Interpreter::expr_offset_in_bytes(1)), NULL_WORD );
 597 }
 598 
 599 void InterpreterMacroAssembler::pop(TosState state) {
 600   switch (state) {
 601   case atos: pop_ptr();                 break;
 602   case btos:
 603   case ztos:
 604   case ctos:
 605   case stos:
 606   case itos: pop_i();                   break;
 607   case ltos: pop_l();                   break;
 608   case ftos: pop_f(xmm0);               break;
 609   case dtos: pop_d(xmm0);               break;
 610   case vtos: /* nothing to do */        break;
 611   default:   ShouldNotReachHere();
 612   }
 613   interp_verify_oop(rax, state);
 614 }
 615 
 616 void InterpreterMacroAssembler::push(TosState state) {
 617   interp_verify_oop(rax, state);
 618   switch (state) {
 619   case atos: push_ptr();                break;
 620   case btos:
 621   case ztos:
 622   case ctos:
 623   case stos:
 624   case itos: push_i();                  break;
 625   case ltos: push_l();                  break;
 626   case ftos: push_f(xmm0);              break;
 627   case dtos: push_d(xmm0);              break;
 628   case vtos: /* nothing to do */        break;
 629   default  : ShouldNotReachHere();
 630   }
 631 }
 632 #else
 633 void InterpreterMacroAssembler::pop_i(Register r) {
 634   pop(r);
 635 }
 636 
 637 void InterpreterMacroAssembler::pop_l(Register lo, Register hi) {
 638   pop(lo);
 639   pop(hi);
 640 }
 641 
 642 void InterpreterMacroAssembler::pop_f() {
 643   fld_s(Address(rsp, 0));
 644   addptr(rsp, 1 * wordSize);
 645 }
 646 
 647 void InterpreterMacroAssembler::pop_d() {
 648   fld_d(Address(rsp, 0));
 649   addptr(rsp, 2 * wordSize);
 650 }
 651 
 652 
 653 void InterpreterMacroAssembler::pop(TosState state) {
 654   switch (state) {
 655     case atos: pop_ptr(rax);                                 break;
 656     case btos:                                               // fall through
 657     case ztos:                                               // fall through
 658     case ctos:                                               // fall through
 659     case stos:                                               // fall through
 660     case itos: pop_i(rax);                                   break;
 661     case ltos: pop_l(rax, rdx);                              break;
 662     case ftos:
 663       if (UseSSE >= 1) {
 664         pop_f(xmm0);
 665       } else {
 666         pop_f();
 667       }
 668       break;
 669     case dtos:
 670       if (UseSSE >= 2) {
 671         pop_d(xmm0);
 672       } else {
 673         pop_d();
 674       }
 675       break;
 676     case vtos: /* nothing to do */                           break;
 677     default  : ShouldNotReachHere();
 678   }
 679   interp_verify_oop(rax, state);
 680 }
 681 
 682 
 683 void InterpreterMacroAssembler::push_l(Register lo, Register hi) {
 684   push(hi);
 685   push(lo);
 686 }
 687 
 688 void InterpreterMacroAssembler::push_f() {
 689   // Do not schedule for no AGI! Never write beyond rsp!
 690   subptr(rsp, 1 * wordSize);
 691   fstp_s(Address(rsp, 0));
 692 }
 693 
 694 void InterpreterMacroAssembler::push_d() {
 695   // Do not schedule for no AGI! Never write beyond rsp!
 696   subptr(rsp, 2 * wordSize);
 697   fstp_d(Address(rsp, 0));
 698 }
 699 
 700 
 701 void InterpreterMacroAssembler::push(TosState state) {
 702   interp_verify_oop(rax, state);
 703   switch (state) {
 704     case atos: push_ptr(rax); break;
 705     case btos:                                               // fall through
 706     case ztos:                                               // fall through
 707     case ctos:                                               // fall through
 708     case stos:                                               // fall through
 709     case itos: push_i(rax);                                    break;
 710     case ltos: push_l(rax, rdx);                               break;
 711     case ftos:
 712       if (UseSSE >= 1) {
 713         push_f(xmm0);
 714       } else {
 715         push_f();
 716       }
 717       break;
 718     case dtos:
 719       if (UseSSE >= 2) {
 720         push_d(xmm0);
 721       } else {
 722         push_d();
 723       }
 724       break;
 725     case vtos: /* nothing to do */                             break;
 726     default  : ShouldNotReachHere();
 727   }
 728 }
 729 #endif // _LP64
 730 
 731 
 732 // Helpers for swap and dup
 733 void InterpreterMacroAssembler::load_ptr(int n, Register val) {
 734   movptr(val, Address(rsp, Interpreter::expr_offset_in_bytes(n)));
 735 }
 736 
 737 void InterpreterMacroAssembler::store_ptr(int n, Register val) {
 738   movptr(Address(rsp, Interpreter::expr_offset_in_bytes(n)), val);
 739 }
 740 
 741 
 742 void InterpreterMacroAssembler::prepare_to_jump_from_interpreted() {
 743   // set sender sp
 744   lea(_bcp_register, Address(rsp, wordSize));
 745   // record last_sp
 746   mov(rcx, _bcp_register);
 747   subptr(rcx, rbp);
 748   sarptr(rcx, LogBytesPerWord);
 749   movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), rcx);
 750 }
 751 
 752 
 753 // Jump to from_interpreted entry of a call unless single stepping is possible
 754 // in this thread in which case we must call the i2i entry
 755 void InterpreterMacroAssembler::jump_from_interpreted(Register method, Register temp) {
 756   prepare_to_jump_from_interpreted();
 757 
 758   if (JvmtiExport::can_post_interpreter_events()) {
 759     Label run_compiled_code;
 760     // JVMTI events, such as single-stepping, are implemented partly by avoiding running
 761     // compiled code in threads for which the event is enabled.  Check here for
 762     // interp_only_mode if these events CAN be enabled.
 763     // interp_only is an int, on little endian it is sufficient to test the byte only
 764     // Is a cmpl faster?
 765     LP64_ONLY(temp = r15_thread;)
 766     NOT_LP64(get_thread(temp);)
 767     cmpb(Address(temp, JavaThread::interp_only_mode_offset()), 0);
 768     jccb(Assembler::zero, run_compiled_code);
 769     jmp(Address(method, Method::interpreter_entry_offset()));
 770     bind(run_compiled_code);
 771   }
 772 
 773   jmp(Address(method, Method::from_interpreted_offset()));
 774 }
 775 
 776 // The following two routines provide a hook so that an implementation
 777 // can schedule the dispatch in two parts.  x86 does not do this.
 778 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int step) {
 779   // Nothing x86 specific to be done here
 780 }
 781 
 782 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int step) {
 783   dispatch_next(state, step);
 784 }
 785 
 786 void InterpreterMacroAssembler::dispatch_base(TosState state,
 787                                               address* table,
 788                                               bool verifyoop,
 789                                               bool generate_poll) {
 790   verify_FPU(1, state);
 791   if (VerifyActivationFrameSize) {
 792     Label L;
 793     mov(rcx, rbp);
 794     subptr(rcx, rsp);
 795     int32_t min_frame_size =
 796       (frame::link_offset - frame::interpreter_frame_initial_sp_offset) *
 797       wordSize;
 798     cmpptr(rcx, min_frame_size);
 799     jcc(Assembler::greaterEqual, L);
 800     stop("broken stack frame");
 801     bind(L);
 802   }
 803   if (verifyoop) {
 804     interp_verify_oop(rax, state);
 805   }
 806 
 807   address* const safepoint_table = Interpreter::safept_table(state);
 808 #ifdef _LP64
 809   Label no_safepoint, dispatch;
 810   if (table != safepoint_table && generate_poll) {
 811     NOT_PRODUCT(block_comment("Thread-local Safepoint poll"));
 812     testb(Address(r15_thread, JavaThread::polling_word_offset()), SafepointMechanism::poll_bit());
 813 
 814     jccb(Assembler::zero, no_safepoint);
 815     lea(rscratch1, ExternalAddress((address)safepoint_table));
 816     jmpb(dispatch);
 817   }
 818 
 819   bind(no_safepoint);
 820   lea(rscratch1, ExternalAddress((address)table));
 821   bind(dispatch);
 822   jmp(Address(rscratch1, rbx, Address::times_8));
 823 
 824 #else
 825   Address index(noreg, rbx, Address::times_ptr);
 826   if (table != safepoint_table && generate_poll) {
 827     NOT_PRODUCT(block_comment("Thread-local Safepoint poll"));
 828     Label no_safepoint;
 829     const Register thread = rcx;
 830     get_thread(thread);
 831     testb(Address(thread, JavaThread::polling_word_offset()), SafepointMechanism::poll_bit());
 832 
 833     jccb(Assembler::zero, no_safepoint);
 834     ArrayAddress dispatch_addr(ExternalAddress((address)safepoint_table), index);
 835     jump(dispatch_addr, noreg);
 836     bind(no_safepoint);
 837   }
 838 
 839   {
 840     ArrayAddress dispatch_addr(ExternalAddress((address)table), index);
 841     jump(dispatch_addr, noreg);
 842   }
 843 #endif // _LP64
 844 }
 845 
 846 void InterpreterMacroAssembler::dispatch_only(TosState state, bool generate_poll) {
 847   dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll);
 848 }
 849 
 850 void InterpreterMacroAssembler::dispatch_only_normal(TosState state) {
 851   dispatch_base(state, Interpreter::normal_table(state));
 852 }
 853 
 854 void InterpreterMacroAssembler::dispatch_only_noverify(TosState state) {
 855   dispatch_base(state, Interpreter::normal_table(state), false);
 856 }
 857 
 858 
 859 void InterpreterMacroAssembler::dispatch_next(TosState state, int step, bool generate_poll) {
 860   // load next bytecode (load before advancing _bcp_register to prevent AGI)
 861   load_unsigned_byte(rbx, Address(_bcp_register, step));
 862   // advance _bcp_register
 863   increment(_bcp_register, step);
 864   dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll);
 865 }
 866 
 867 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) {
 868   // load current bytecode
 869   load_unsigned_byte(rbx, Address(_bcp_register, 0));
 870   dispatch_base(state, table);
 871 }
 872 
 873 void InterpreterMacroAssembler::narrow(Register result) {
 874 
 875   // Get method->_constMethod->_result_type
 876   movptr(rcx, Address(rbp, frame::interpreter_frame_method_offset * wordSize));
 877   movptr(rcx, Address(rcx, Method::const_offset()));
 878   load_unsigned_byte(rcx, Address(rcx, ConstMethod::result_type_offset()));
 879 
 880   Label done, notBool, notByte, notChar;
 881 
 882   // common case first
 883   cmpl(rcx, T_INT);
 884   jcc(Assembler::equal, done);
 885 
 886   // mask integer result to narrower return type.
 887   cmpl(rcx, T_BOOLEAN);
 888   jcc(Assembler::notEqual, notBool);
 889   andl(result, 0x1);
 890   jmp(done);
 891 
 892   bind(notBool);
 893   cmpl(rcx, T_BYTE);
 894   jcc(Assembler::notEqual, notByte);
 895   LP64_ONLY(movsbl(result, result);)
 896   NOT_LP64(shll(result, 24);)      // truncate upper 24 bits
 897   NOT_LP64(sarl(result, 24);)      // and sign-extend byte
 898   jmp(done);
 899 
 900   bind(notByte);
 901   cmpl(rcx, T_CHAR);
 902   jcc(Assembler::notEqual, notChar);
 903   LP64_ONLY(movzwl(result, result);)
 904   NOT_LP64(andl(result, 0xFFFF);)  // truncate upper 16 bits
 905   jmp(done);
 906 
 907   bind(notChar);
 908   // cmpl(rcx, T_SHORT);  // all that's left
 909   // jcc(Assembler::notEqual, done);
 910   LP64_ONLY(movswl(result, result);)
 911   NOT_LP64(shll(result, 16);)      // truncate upper 16 bits
 912   NOT_LP64(sarl(result, 16);)      // and sign-extend short
 913 
 914   // Nothing to do for T_INT
 915   bind(done);
 916 }
 917 
 918 // remove activation
 919 //
 920 // Apply stack watermark barrier.
 921 // Unlock the receiver if this is a synchronized method.
 922 // Unlock any Java monitors from synchronized blocks.
 923 // Remove the activation from the stack.
 924 //
 925 // If there are locked Java monitors
 926 //    If throw_monitor_exception
 927 //       throws IllegalMonitorStateException
 928 //    Else if install_monitor_exception
 929 //       installs IllegalMonitorStateException
 930 //    Else
 931 //       no error processing
 932 void InterpreterMacroAssembler::remove_activation(
 933         TosState state,
 934         Register ret_addr,
 935         bool throw_monitor_exception,
 936         bool install_monitor_exception,
 937         bool notify_jvmdi) {
 938   // Note: Registers rdx xmm0 may be in use for the
 939   // result check if synchronized method
 940   Label unlocked, unlock, no_unlock;
 941 
 942   const Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
 943   const Register robj    = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
 944   const Register rmon    = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
 945                               // monitor pointers need different register
 946                               // because rdx may have the result in it
 947   NOT_LP64(get_thread(rthread);)
 948 
 949   // The below poll is for the stack watermark barrier. It allows fixing up frames lazily,
 950   // that would normally not be safe to use. Such bad returns into unsafe territory of
 951   // the stack, will call InterpreterRuntime::at_unwind.
 952   Label slow_path;
 953   Label fast_path;
 954   safepoint_poll(slow_path, rthread, true /* at_return */, false /* in_nmethod */);
 955   jmp(fast_path);
 956   bind(slow_path);
 957   push(state);
 958   set_last_Java_frame(rthread, noreg, rbp, (address)pc(), rscratch1);
 959   super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::at_unwind), rthread);
 960   NOT_LP64(get_thread(rthread);) // call_VM clobbered it, restore
 961   reset_last_Java_frame(rthread, true);
 962   pop(state);
 963   bind(fast_path);
 964 
 965   // get the value of _do_not_unlock_if_synchronized into rdx
 966   const Address do_not_unlock_if_synchronized(rthread,
 967     in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
 968   movbool(rbx, do_not_unlock_if_synchronized);
 969   movbool(do_not_unlock_if_synchronized, false); // reset the flag
 970 
 971  // get method access flags
 972   movptr(rcx, Address(rbp, frame::interpreter_frame_method_offset * wordSize));
 973   movl(rcx, Address(rcx, Method::access_flags_offset()));
 974   testl(rcx, JVM_ACC_SYNCHRONIZED);
 975   jcc(Assembler::zero, unlocked);
 976 
 977   // Don't unlock anything if the _do_not_unlock_if_synchronized flag
 978   // is set.
 979   testbool(rbx);
 980   jcc(Assembler::notZero, no_unlock);
 981 
 982   // unlock monitor
 983   push(state); // save result
 984 
 985   // BasicObjectLock will be first in list, since this is a
 986   // synchronized method. However, need to check that the object has
 987   // not been unlocked by an explicit monitorexit bytecode.
 988   const Address monitor(rbp, frame::interpreter_frame_initial_sp_offset *
 989                         wordSize - (int) sizeof(BasicObjectLock));
 990   // We use c_rarg1/rdx so that if we go slow path it will be the correct
 991   // register for unlock_object to pass to VM directly
 992   lea(robj, monitor); // address of first monitor
 993 
 994   movptr(rax, Address(robj, BasicObjectLock::obj_offset()));
 995   testptr(rax, rax);
 996   jcc(Assembler::notZero, unlock);
 997 
 998   pop(state);
 999   if (throw_monitor_exception) {
1000     // Entry already unlocked, need to throw exception
1001     NOT_LP64(empty_FPU_stack();)  // remove possible return value from FPU-stack, otherwise stack could overflow
1002     call_VM(noreg, CAST_FROM_FN_PTR(address,
1003                    InterpreterRuntime::throw_illegal_monitor_state_exception));
1004     should_not_reach_here();
1005   } else {
1006     // Monitor already unlocked during a stack unroll. If requested,
1007     // install an illegal_monitor_state_exception.  Continue with
1008     // stack unrolling.
1009     if (install_monitor_exception) {
1010       NOT_LP64(empty_FPU_stack();)
1011       call_VM(noreg, CAST_FROM_FN_PTR(address,
1012                      InterpreterRuntime::new_illegal_monitor_state_exception));
1013     }
1014     jmp(unlocked);
1015   }
1016 
1017   bind(unlock);
1018   unlock_object(robj);
1019   pop(state);
1020 
1021   // Check that for block-structured locking (i.e., that all locked
1022   // objects has been unlocked)
1023   bind(unlocked);
1024 
1025   // rax, rdx: Might contain return value
1026 
1027   // Check that all monitors are unlocked
1028   {
1029     Label loop, exception, entry, restart;
1030     const int entry_size = frame::interpreter_frame_monitor_size_in_bytes();
1031     const Address monitor_block_top(
1032         rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
1033     const Address monitor_block_bot(
1034         rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
1035 
1036     bind(restart);
1037     // We use c_rarg1 so that if we go slow path it will be the correct
1038     // register for unlock_object to pass to VM directly
1039     movptr(rmon, monitor_block_top); // derelativize pointer
1040     lea(rmon, Address(rbp, rmon, Address::times_ptr));
1041     // c_rarg1 points to current entry, starting with top-most entry
1042 
1043     lea(rbx, monitor_block_bot);  // points to word before bottom of
1044                                   // monitor block
1045     jmp(entry);
1046 
1047     // Entry already locked, need to throw exception
1048     bind(exception);
1049 
1050     if (throw_monitor_exception) {
1051       // Throw exception
1052       NOT_LP64(empty_FPU_stack();)
1053       MacroAssembler::call_VM(noreg,
1054                               CAST_FROM_FN_PTR(address, InterpreterRuntime::
1055                                    throw_illegal_monitor_state_exception));
1056       should_not_reach_here();
1057     } else {
1058       // Stack unrolling. Unlock object and install illegal_monitor_exception.
1059       // Unlock does not block, so don't have to worry about the frame.
1060       // We don't have to preserve c_rarg1 since we are going to throw an exception.
1061 
1062       push(state);
1063       mov(robj, rmon);   // nop if robj and rmon are the same
1064       unlock_object(robj);
1065       pop(state);
1066 
1067       if (install_monitor_exception) {
1068         NOT_LP64(empty_FPU_stack();)
1069         call_VM(noreg, CAST_FROM_FN_PTR(address,
1070                                         InterpreterRuntime::
1071                                         new_illegal_monitor_state_exception));
1072       }
1073 
1074       jmp(restart);
1075     }
1076 
1077     bind(loop);
1078     // check if current entry is used
1079     cmpptr(Address(rmon, BasicObjectLock::obj_offset()), NULL_WORD);
1080     jcc(Assembler::notEqual, exception);
1081 
1082     addptr(rmon, entry_size); // otherwise advance to next entry
1083     bind(entry);
1084     cmpptr(rmon, rbx); // check if bottom reached
1085     jcc(Assembler::notEqual, loop); // if not at bottom then check this entry
1086   }
1087 
1088   bind(no_unlock);
1089 
1090   // jvmti support
1091   if (notify_jvmdi) {
1092     notify_method_exit(state, NotifyJVMTI);    // preserve TOSCA
1093   } else {
1094     notify_method_exit(state, SkipNotifyJVMTI); // preserve TOSCA
1095   }
1096 
1097   // remove activation
1098   // get sender sp
1099   movptr(rbx,
1100          Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize));
1101   if (StackReservedPages > 0) {
1102     // testing if reserved zone needs to be re-enabled
1103     Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
1104     Label no_reserved_zone_enabling;
1105 
1106     NOT_LP64(get_thread(rthread);)
1107 
1108     // check if already enabled - if so no re-enabling needed
1109     assert(sizeof(StackOverflow::StackGuardState) == 4, "unexpected size");
1110     cmpl(Address(rthread, JavaThread::stack_guard_state_offset()), StackOverflow::stack_guard_enabled);
1111     jcc(Assembler::equal, no_reserved_zone_enabling);
1112 
1113     cmpptr(rbx, Address(rthread, JavaThread::reserved_stack_activation_offset()));
1114     jcc(Assembler::lessEqual, no_reserved_zone_enabling);
1115 
1116     call_VM_leaf(
1117       CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), rthread);
1118     call_VM(noreg, CAST_FROM_FN_PTR(address,
1119                    InterpreterRuntime::throw_delayed_StackOverflowError));
1120     should_not_reach_here();
1121 
1122     bind(no_reserved_zone_enabling);
1123   }
1124   leave();                           // remove frame anchor
1125   pop(ret_addr);                     // get return address
1126   mov(rsp, rbx);                     // set sp to sender sp
1127   pop_cont_fastpath();
1128 }
1129 
1130 void InterpreterMacroAssembler::get_method_counters(Register method,
1131                                                     Register mcs, Label& skip) {
1132   Label has_counters;
1133   movptr(mcs, Address(method, Method::method_counters_offset()));
1134   testptr(mcs, mcs);
1135   jcc(Assembler::notZero, has_counters);
1136   call_VM(noreg, CAST_FROM_FN_PTR(address,
1137           InterpreterRuntime::build_method_counters), method);
1138   movptr(mcs, Address(method,Method::method_counters_offset()));
1139   testptr(mcs, mcs);
1140   jcc(Assembler::zero, skip); // No MethodCounters allocated, OutOfMemory
1141   bind(has_counters);
1142 }
1143 
1144 
1145 // Lock object
1146 //
1147 // Args:
1148 //      rdx, c_rarg1: BasicObjectLock to be used for locking
1149 //
1150 // Kills:
1151 //      rax, rbx
1152 void InterpreterMacroAssembler::lock_object(Register lock_reg) {
1153   assert(lock_reg == LP64_ONLY(c_rarg1) NOT_LP64(rdx),
1154          "The argument is only for looks. It must be c_rarg1");
1155 
1156   if (LockingMode == LM_MONITOR) {
1157     call_VM(noreg,
1158             CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
1159             lock_reg);
1160   } else {
1161     Label count_locking, done, slow_case;
1162 
1163     const Register swap_reg = rax; // Must use rax for cmpxchg instruction
1164     const Register tmp_reg = rbx;
1165     const Register obj_reg = LP64_ONLY(c_rarg3) NOT_LP64(rcx); // Will contain the oop
1166     const Register rklass_decode_tmp = rscratch1;
1167 
1168     const int obj_offset = in_bytes(BasicObjectLock::obj_offset());
1169     const int lock_offset = in_bytes(BasicObjectLock::lock_offset());
1170     const int mark_offset = lock_offset +
1171                             BasicLock::displaced_header_offset_in_bytes();
1172 
1173     // Load object pointer into obj_reg
1174     movptr(obj_reg, Address(lock_reg, obj_offset));
1175 
1176     if (DiagnoseSyncOnValueBasedClasses != 0) {
1177       load_klass(tmp_reg, obj_reg, rklass_decode_tmp);
1178       testb(Address(tmp_reg, Klass::misc_flags_offset()), KlassFlags::_misc_is_value_based_class);
1179       jcc(Assembler::notZero, slow_case);
1180     }
1181 
1182     if (LockingMode == LM_LIGHTWEIGHT) {
1183 #ifdef _LP64
1184       const Register thread = r15_thread;
1185       lightweight_lock(lock_reg, obj_reg, swap_reg, thread, tmp_reg, slow_case);
1186 #else
1187       // Lacking registers and thread on x86_32. Always take slow path.
1188       jmp(slow_case);
1189 #endif
1190     } else if (LockingMode == LM_LEGACY) {
1191       // Load immediate 1 into swap_reg %rax
1192       movl(swap_reg, 1);
1193 
1194       // Load (object->mark() | 1) into swap_reg %rax
1195       orptr(swap_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
1196 
1197       // Save (object->mark() | 1) into BasicLock's displaced header
1198       movptr(Address(lock_reg, mark_offset), swap_reg);
1199 
1200       assert(lock_offset == 0,
1201              "displaced header must be first word in BasicObjectLock");
1202 
1203       lock();
1204       cmpxchgptr(lock_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
1205       jcc(Assembler::zero, count_locking);
1206 
1207       const int zero_bits = LP64_ONLY(7) NOT_LP64(3);
1208 
1209       // Fast check for recursive lock.
1210       //
1211       // Can apply the optimization only if this is a stack lock
1212       // allocated in this thread. For efficiency, we can focus on
1213       // recently allocated stack locks (instead of reading the stack
1214       // base and checking whether 'mark' points inside the current
1215       // thread stack):
1216       //  1) (mark & zero_bits) == 0, and
1217       //  2) rsp <= mark < mark + os::pagesize()
1218       //
1219       // Warning: rsp + os::pagesize can overflow the stack base. We must
1220       // neither apply the optimization for an inflated lock allocated
1221       // just above the thread stack (this is why condition 1 matters)
1222       // nor apply the optimization if the stack lock is inside the stack
1223       // of another thread. The latter is avoided even in case of overflow
1224       // because we have guard pages at the end of all stacks. Hence, if
1225       // we go over the stack base and hit the stack of another thread,
1226       // this should not be in a writeable area that could contain a
1227       // stack lock allocated by that thread. As a consequence, a stack
1228       // lock less than page size away from rsp is guaranteed to be
1229       // owned by the current thread.
1230       //
1231       // These 3 tests can be done by evaluating the following
1232       // expression: ((mark - rsp) & (zero_bits - os::vm_page_size())),
1233       // assuming both stack pointer and pagesize have their
1234       // least significant bits clear.
1235       // NOTE: the mark is in swap_reg %rax as the result of cmpxchg
1236       subptr(swap_reg, rsp);
1237       andptr(swap_reg, zero_bits - (int)os::vm_page_size());
1238 
1239       // Save the test result, for recursive case, the result is zero
1240       movptr(Address(lock_reg, mark_offset), swap_reg);
1241       jcc(Assembler::notZero, slow_case);
1242 
1243       bind(count_locking);
1244     }
1245     inc_held_monitor_count();
1246     jmp(done);
1247 
1248     bind(slow_case);
1249 
1250     // Call the runtime routine for slow case
1251     call_VM(noreg,
1252             CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
1253             lock_reg);
1254     bind(done);
1255   }
1256 }
1257 
1258 
1259 // Unlocks an object. Used in monitorexit bytecode and
1260 // remove_activation.  Throws an IllegalMonitorException if object is
1261 // not locked by current thread.
1262 //
1263 // Args:
1264 //      rdx, c_rarg1: BasicObjectLock for lock
1265 //
1266 // Kills:
1267 //      rax
1268 //      c_rarg0, c_rarg1, c_rarg2, c_rarg3, ... (param regs)
1269 //      rscratch1 (scratch reg)
1270 // rax, rbx, rcx, rdx
1271 void InterpreterMacroAssembler::unlock_object(Register lock_reg) {
1272   assert(lock_reg == LP64_ONLY(c_rarg1) NOT_LP64(rdx),
1273          "The argument is only for looks. It must be c_rarg1");
1274 
1275   if (LockingMode == LM_MONITOR) {
1276     call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
1277   } else {
1278     Label count_locking, done, slow_case;
1279 
1280     const Register swap_reg   = rax;  // Must use rax for cmpxchg instruction
1281     const Register header_reg = LP64_ONLY(c_rarg2) NOT_LP64(rbx);  // Will contain the old oopMark
1282     const Register obj_reg    = LP64_ONLY(c_rarg3) NOT_LP64(rcx);  // Will contain the oop
1283 
1284     save_bcp(); // Save in case of exception
1285 
1286     if (LockingMode != LM_LIGHTWEIGHT) {
1287       // Convert from BasicObjectLock structure to object and BasicLock
1288       // structure Store the BasicLock address into %rax
1289       lea(swap_reg, Address(lock_reg, BasicObjectLock::lock_offset()));
1290     }
1291 
1292     // Load oop into obj_reg(%c_rarg3)
1293     movptr(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset()));
1294 
1295     // Free entry
1296     movptr(Address(lock_reg, BasicObjectLock::obj_offset()), NULL_WORD);
1297 
1298     if (LockingMode == LM_LIGHTWEIGHT) {
1299 #ifdef _LP64
1300       lightweight_unlock(obj_reg, swap_reg, r15_thread, header_reg, slow_case);
1301 #else
1302       // Lacking registers and thread on x86_32. Always take slow path.
1303       jmp(slow_case);
1304 #endif
1305     } else if (LockingMode == LM_LEGACY) {
1306       // Load the old header from BasicLock structure
1307       movptr(header_reg, Address(swap_reg,
1308                                  BasicLock::displaced_header_offset_in_bytes()));
1309 
1310       // Test for recursion
1311       testptr(header_reg, header_reg);
1312 
1313       // zero for recursive case
1314       jcc(Assembler::zero, count_locking);
1315 
1316       // Atomic swap back the old header
1317       lock();
1318       cmpxchgptr(header_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
1319 
1320       // zero for simple unlock of a stack-lock case
1321       jcc(Assembler::notZero, slow_case);
1322 
1323       bind(count_locking);
1324     }
1325     dec_held_monitor_count();
1326     jmp(done);
1327 
1328     bind(slow_case);
1329     // Call the runtime routine for slow case.
1330     movptr(Address(lock_reg, BasicObjectLock::obj_offset()), obj_reg); // restore obj
1331     call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
1332 
1333     bind(done);
1334 
1335     restore_bcp();
1336   }
1337 }
1338 
1339 void InterpreterMacroAssembler::test_method_data_pointer(Register mdp,
1340                                                          Label& zero_continue) {
1341   assert(ProfileInterpreter, "must be profiling interpreter");
1342   movptr(mdp, Address(rbp, frame::interpreter_frame_mdp_offset * wordSize));
1343   testptr(mdp, mdp);
1344   jcc(Assembler::zero, zero_continue);
1345 }
1346 
1347 
1348 // Set the method data pointer for the current bcp.
1349 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
1350   assert(ProfileInterpreter, "must be profiling interpreter");
1351   Label set_mdp;
1352   push(rax);
1353   push(rbx);
1354 
1355   get_method(rbx);
1356   // Test MDO to avoid the call if it is null.
1357   movptr(rax, Address(rbx, in_bytes(Method::method_data_offset())));
1358   testptr(rax, rax);
1359   jcc(Assembler::zero, set_mdp);
1360   // rbx: method
1361   // _bcp_register: bcp
1362   call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), rbx, _bcp_register);
1363   // rax: mdi
1364   // mdo is guaranteed to be non-zero here, we checked for it before the call.
1365   movptr(rbx, Address(rbx, in_bytes(Method::method_data_offset())));
1366   addptr(rbx, in_bytes(MethodData::data_offset()));
1367   addptr(rax, rbx);
1368   bind(set_mdp);
1369   movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), rax);
1370   pop(rbx);
1371   pop(rax);
1372 }
1373 
1374 void InterpreterMacroAssembler::verify_method_data_pointer() {
1375   assert(ProfileInterpreter, "must be profiling interpreter");
1376 #ifdef ASSERT
1377   Label verify_continue;
1378   push(rax);
1379   push(rbx);
1380   Register arg3_reg = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
1381   Register arg2_reg = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
1382   push(arg3_reg);
1383   push(arg2_reg);
1384   test_method_data_pointer(arg3_reg, verify_continue); // If mdp is zero, continue
1385   get_method(rbx);
1386 
1387   // If the mdp is valid, it will point to a DataLayout header which is
1388   // consistent with the bcp.  The converse is highly probable also.
1389   load_unsigned_short(arg2_reg,
1390                       Address(arg3_reg, in_bytes(DataLayout::bci_offset())));
1391   addptr(arg2_reg, Address(rbx, Method::const_offset()));
1392   lea(arg2_reg, Address(arg2_reg, ConstMethod::codes_offset()));
1393   cmpptr(arg2_reg, _bcp_register);
1394   jcc(Assembler::equal, verify_continue);
1395   // rbx: method
1396   // _bcp_register: bcp
1397   // c_rarg3: mdp
1398   call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp),
1399                rbx, _bcp_register, arg3_reg);
1400   bind(verify_continue);
1401   pop(arg2_reg);
1402   pop(arg3_reg);
1403   pop(rbx);
1404   pop(rax);
1405 #endif // ASSERT
1406 }
1407 
1408 
1409 void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in,
1410                                                 int constant,
1411                                                 Register value) {
1412   assert(ProfileInterpreter, "must be profiling interpreter");
1413   Address data(mdp_in, constant);
1414   movptr(data, value);
1415 }
1416 
1417 
1418 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
1419                                                       int constant,
1420                                                       bool decrement) {
1421   // Counter address
1422   Address data(mdp_in, constant);
1423 
1424   increment_mdp_data_at(data, decrement);
1425 }
1426 
1427 void InterpreterMacroAssembler::increment_mdp_data_at(Address data,
1428                                                       bool decrement) {
1429   assert(ProfileInterpreter, "must be profiling interpreter");
1430   // %%% this does 64bit counters at best it is wasting space
1431   // at worst it is a rare bug when counters overflow
1432 
1433   if (decrement) {
1434     // Decrement the register.  Set condition codes.
1435     addptr(data, -DataLayout::counter_increment);
1436     // If the decrement causes the counter to overflow, stay negative
1437     Label L;
1438     jcc(Assembler::negative, L);
1439     addptr(data, DataLayout::counter_increment);
1440     bind(L);
1441   } else {
1442     assert(DataLayout::counter_increment == 1,
1443            "flow-free idiom only works with 1");
1444     // Increment the register.  Set carry flag.
1445     addptr(data, DataLayout::counter_increment);
1446     // If the increment causes the counter to overflow, pull back by 1.
1447     sbbptr(data, 0);
1448   }
1449 }
1450 
1451 
1452 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
1453                                                       Register reg,
1454                                                       int constant,
1455                                                       bool decrement) {
1456   Address data(mdp_in, reg, Address::times_1, constant);
1457 
1458   increment_mdp_data_at(data, decrement);
1459 }
1460 
1461 void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in,
1462                                                 int flag_byte_constant) {
1463   assert(ProfileInterpreter, "must be profiling interpreter");
1464   int header_offset = in_bytes(DataLayout::flags_offset());
1465   int header_bits = flag_byte_constant;
1466   // Set the flag
1467   orb(Address(mdp_in, header_offset), header_bits);
1468 }
1469 
1470 
1471 
1472 void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in,
1473                                                  int offset,
1474                                                  Register value,
1475                                                  Register test_value_out,
1476                                                  Label& not_equal_continue) {
1477   assert(ProfileInterpreter, "must be profiling interpreter");
1478   if (test_value_out == noreg) {
1479     cmpptr(value, Address(mdp_in, offset));
1480   } else {
1481     // Put the test value into a register, so caller can use it:
1482     movptr(test_value_out, Address(mdp_in, offset));
1483     cmpptr(test_value_out, value);
1484   }
1485   jcc(Assembler::notEqual, not_equal_continue);
1486 }
1487 
1488 
1489 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1490                                                      int offset_of_disp) {
1491   assert(ProfileInterpreter, "must be profiling interpreter");
1492   Address disp_address(mdp_in, offset_of_disp);
1493   addptr(mdp_in, disp_address);
1494   movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in);
1495 }
1496 
1497 
1498 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1499                                                      Register reg,
1500                                                      int offset_of_disp) {
1501   assert(ProfileInterpreter, "must be profiling interpreter");
1502   Address disp_address(mdp_in, reg, Address::times_1, offset_of_disp);
1503   addptr(mdp_in, disp_address);
1504   movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in);
1505 }
1506 
1507 
1508 void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in,
1509                                                        int constant) {
1510   assert(ProfileInterpreter, "must be profiling interpreter");
1511   addptr(mdp_in, constant);
1512   movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in);
1513 }
1514 
1515 
1516 void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) {
1517   assert(ProfileInterpreter, "must be profiling interpreter");
1518   push(return_bci); // save/restore across call_VM
1519   call_VM(noreg,
1520           CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret),
1521           return_bci);
1522   pop(return_bci);
1523 }
1524 
1525 
1526 void InterpreterMacroAssembler::profile_taken_branch(Register mdp,
1527                                                      Register bumped_count) {
1528   if (ProfileInterpreter) {
1529     Label profile_continue;
1530 
1531     // If no method data exists, go to profile_continue.
1532     // Otherwise, assign to mdp
1533     test_method_data_pointer(mdp, profile_continue);
1534 
1535     // We are taking a branch.  Increment the taken count.
1536     // We inline increment_mdp_data_at to return bumped_count in a register
1537     //increment_mdp_data_at(mdp, in_bytes(JumpData::taken_offset()));
1538     Address data(mdp, in_bytes(JumpData::taken_offset()));
1539     movptr(bumped_count, data);
1540     assert(DataLayout::counter_increment == 1,
1541             "flow-free idiom only works with 1");
1542     addptr(bumped_count, DataLayout::counter_increment);
1543     sbbptr(bumped_count, 0);
1544     movptr(data, bumped_count); // Store back out
1545 
1546     // The method data pointer needs to be updated to reflect the new target.
1547     update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset()));
1548     bind(profile_continue);
1549   }
1550 }
1551 
1552 
1553 void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp) {
1554   if (ProfileInterpreter) {
1555     Label profile_continue;
1556 
1557     // If no method data exists, go to profile_continue.
1558     test_method_data_pointer(mdp, profile_continue);
1559 
1560     // We are taking a branch.  Increment the not taken count.
1561     increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset()));
1562 
1563     // The method data pointer needs to be updated to correspond to
1564     // the next bytecode
1565     update_mdp_by_constant(mdp, in_bytes(BranchData::branch_data_size()));
1566     bind(profile_continue);
1567   }
1568 }
1569 
1570 void InterpreterMacroAssembler::profile_call(Register mdp) {
1571   if (ProfileInterpreter) {
1572     Label profile_continue;
1573 
1574     // If no method data exists, go to profile_continue.
1575     test_method_data_pointer(mdp, profile_continue);
1576 
1577     // We are making a call.  Increment the count.
1578     increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1579 
1580     // The method data pointer needs to be updated to reflect the new target.
1581     update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size()));
1582     bind(profile_continue);
1583   }
1584 }
1585 
1586 
1587 void InterpreterMacroAssembler::profile_final_call(Register mdp) {
1588   if (ProfileInterpreter) {
1589     Label profile_continue;
1590 
1591     // If no method data exists, go to profile_continue.
1592     test_method_data_pointer(mdp, profile_continue);
1593 
1594     // We are making a call.  Increment the count.
1595     increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1596 
1597     // The method data pointer needs to be updated to reflect the new target.
1598     update_mdp_by_constant(mdp,
1599                            in_bytes(VirtualCallData::
1600                                     virtual_call_data_size()));
1601     bind(profile_continue);
1602   }
1603 }
1604 
1605 
1606 void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
1607                                                      Register mdp,
1608                                                      Register reg2,
1609                                                      bool receiver_can_be_null) {
1610   if (ProfileInterpreter) {
1611     Label profile_continue;
1612 
1613     // If no method data exists, go to profile_continue.
1614     test_method_data_pointer(mdp, profile_continue);
1615 
1616     Label skip_receiver_profile;
1617     if (receiver_can_be_null) {
1618       Label not_null;
1619       testptr(receiver, receiver);
1620       jccb(Assembler::notZero, not_null);
1621       // We are making a call.  Increment the count for null receiver.
1622       increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1623       jmp(skip_receiver_profile);
1624       bind(not_null);
1625     }
1626 
1627     // Record the receiver type.
1628     record_klass_in_profile(receiver, mdp, reg2, true);
1629     bind(skip_receiver_profile);
1630 
1631     // The method data pointer needs to be updated to reflect the new target.
1632     update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size()));
1633     bind(profile_continue);
1634   }
1635 }
1636 
1637 // This routine creates a state machine for updating the multi-row
1638 // type profile at a virtual call site (or other type-sensitive bytecode).
1639 // The machine visits each row (of receiver/count) until the receiver type
1640 // is found, or until it runs out of rows.  At the same time, it remembers
1641 // the location of the first empty row.  (An empty row records null for its
1642 // receiver, and can be allocated for a newly-observed receiver type.)
1643 // Because there are two degrees of freedom in the state, a simple linear
1644 // search will not work; it must be a decision tree.  Hence this helper
1645 // function is recursive, to generate the required tree structured code.
1646 // It's the interpreter, so we are trading off code space for speed.
1647 // See below for example code.
1648 void InterpreterMacroAssembler::record_klass_in_profile_helper(
1649                                         Register receiver, Register mdp,
1650                                         Register reg2, int start_row,
1651                                         Label& done, bool is_virtual_call) {
1652   if (TypeProfileWidth == 0) {
1653     increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1654   } else {
1655     record_item_in_profile_helper(receiver, mdp, reg2, 0, done, TypeProfileWidth,
1656                                   &VirtualCallData::receiver_offset, &VirtualCallData::receiver_count_offset);
1657   }
1658 }
1659 
1660 void InterpreterMacroAssembler::record_item_in_profile_helper(Register item, Register mdp, Register reg2, int start_row,
1661                                                               Label& done, int total_rows,
1662                                                               OffsetFunction item_offset_fn,
1663                                                               OffsetFunction item_count_offset_fn) {
1664   int last_row = total_rows - 1;
1665   assert(start_row <= last_row, "must be work left to do");
1666   // Test this row for both the item and for null.
1667   // Take any of three different outcomes:
1668   //   1. found item => increment count and goto done
1669   //   2. found null => keep looking for case 1, maybe allocate this cell
1670   //   3. found something else => keep looking for cases 1 and 2
1671   // Case 3 is handled by a recursive call.
1672   for (int row = start_row; row <= last_row; row++) {
1673     Label next_test;
1674     bool test_for_null_also = (row == start_row);
1675 
1676     // See if the item is item[n].
1677     int item_offset = in_bytes(item_offset_fn(row));
1678     test_mdp_data_at(mdp, item_offset, item,
1679                      (test_for_null_also ? reg2 : noreg),
1680                      next_test);
1681     // (Reg2 now contains the item from the CallData.)
1682 
1683     // The item is item[n].  Increment count[n].
1684     int count_offset = in_bytes(item_count_offset_fn(row));
1685     increment_mdp_data_at(mdp, count_offset);
1686     jmp(done);
1687     bind(next_test);
1688 
1689     if (test_for_null_also) {
1690       // Failed the equality check on item[n]...  Test for null.
1691       testptr(reg2, reg2);
1692       if (start_row == last_row) {
1693         // The only thing left to do is handle the null case.
1694         Label found_null;
1695         jccb(Assembler::zero, found_null);
1696         // Item did not match any saved item and there is no empty row for it.
1697         // Increment total counter to indicate polymorphic case.
1698         increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1699         jmp(done);
1700         bind(found_null);
1701         break;
1702       }
1703       Label found_null;
1704       // Since null is rare, make it be the branch-taken case.
1705       jcc(Assembler::zero, found_null);
1706 
1707       // Put all the "Case 3" tests here.
1708       record_item_in_profile_helper(item, mdp, reg2, start_row + 1, done, total_rows,
1709                                     item_offset_fn, item_count_offset_fn);
1710 
1711       // Found a null.  Keep searching for a matching item,
1712       // but remember that this is an empty (unused) slot.
1713       bind(found_null);
1714     }
1715   }
1716 
1717   // In the fall-through case, we found no matching item, but we
1718   // observed the item[start_row] is null.
1719 
1720   // Fill in the item field and increment the count.
1721   int item_offset = in_bytes(item_offset_fn(start_row));
1722   set_mdp_data_at(mdp, item_offset, item);
1723   int count_offset = in_bytes(item_count_offset_fn(start_row));
1724   movl(reg2, DataLayout::counter_increment);
1725   set_mdp_data_at(mdp, count_offset, reg2);
1726   if (start_row > 0) {
1727     jmp(done);
1728   }
1729 }
1730 
1731 // Example state machine code for three profile rows:
1732 //   // main copy of decision tree, rooted at row[1]
1733 //   if (row[0].rec == rec) { row[0].incr(); goto done; }
1734 //   if (row[0].rec != nullptr) {
1735 //     // inner copy of decision tree, rooted at row[1]
1736 //     if (row[1].rec == rec) { row[1].incr(); goto done; }
1737 //     if (row[1].rec != nullptr) {
1738 //       // degenerate decision tree, rooted at row[2]
1739 //       if (row[2].rec == rec) { row[2].incr(); goto done; }
1740 //       if (row[2].rec != nullptr) { count.incr(); goto done; } // overflow
1741 //       row[2].init(rec); goto done;
1742 //     } else {
1743 //       // remember row[1] is empty
1744 //       if (row[2].rec == rec) { row[2].incr(); goto done; }
1745 //       row[1].init(rec); goto done;
1746 //     }
1747 //   } else {
1748 //     // remember row[0] is empty
1749 //     if (row[1].rec == rec) { row[1].incr(); goto done; }
1750 //     if (row[2].rec == rec) { row[2].incr(); goto done; }
1751 //     row[0].init(rec); goto done;
1752 //   }
1753 //   done:
1754 
1755 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver,
1756                                                         Register mdp, Register reg2,
1757                                                         bool is_virtual_call) {
1758   assert(ProfileInterpreter, "must be profiling");
1759   Label done;
1760 
1761   record_klass_in_profile_helper(receiver, mdp, reg2, 0, done, is_virtual_call);
1762 
1763   bind (done);
1764 }
1765 
1766 void InterpreterMacroAssembler::profile_ret(Register return_bci,
1767                                             Register mdp) {
1768   if (ProfileInterpreter) {
1769     Label profile_continue;
1770     uint row;
1771 
1772     // If no method data exists, go to profile_continue.
1773     test_method_data_pointer(mdp, profile_continue);
1774 
1775     // Update the total ret count.
1776     increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1777 
1778     for (row = 0; row < RetData::row_limit(); row++) {
1779       Label next_test;
1780 
1781       // See if return_bci is equal to bci[n]:
1782       test_mdp_data_at(mdp,
1783                        in_bytes(RetData::bci_offset(row)),
1784                        return_bci, noreg,
1785                        next_test);
1786 
1787       // return_bci is equal to bci[n].  Increment the count.
1788       increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row)));
1789 
1790       // The method data pointer needs to be updated to reflect the new target.
1791       update_mdp_by_offset(mdp,
1792                            in_bytes(RetData::bci_displacement_offset(row)));
1793       jmp(profile_continue);
1794       bind(next_test);
1795     }
1796 
1797     update_mdp_for_ret(return_bci);
1798 
1799     bind(profile_continue);
1800   }
1801 }
1802 
1803 
1804 void InterpreterMacroAssembler::profile_null_seen(Register mdp) {
1805   if (ProfileInterpreter) {
1806     Label profile_continue;
1807 
1808     // If no method data exists, go to profile_continue.
1809     test_method_data_pointer(mdp, profile_continue);
1810 
1811     set_mdp_flag_at(mdp, BitData::null_seen_byte_constant());
1812 
1813     // The method data pointer needs to be updated.
1814     int mdp_delta = in_bytes(BitData::bit_data_size());
1815     if (TypeProfileCasts) {
1816       mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1817     }
1818     update_mdp_by_constant(mdp, mdp_delta);
1819 
1820     bind(profile_continue);
1821   }
1822 }
1823 
1824 
1825 void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) {
1826   if (ProfileInterpreter) {
1827     Label profile_continue;
1828 
1829     // If no method data exists, go to profile_continue.
1830     test_method_data_pointer(mdp, profile_continue);
1831 
1832     // The method data pointer needs to be updated.
1833     int mdp_delta = in_bytes(BitData::bit_data_size());
1834     if (TypeProfileCasts) {
1835       mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1836 
1837       // Record the object type.
1838       record_klass_in_profile(klass, mdp, reg2, false);
1839       NOT_LP64(assert(reg2 == rdi, "we know how to fix this blown reg");)
1840       NOT_LP64(restore_locals();)         // Restore EDI
1841     }
1842     update_mdp_by_constant(mdp, mdp_delta);
1843 
1844     bind(profile_continue);
1845   }
1846 }
1847 
1848 
1849 void InterpreterMacroAssembler::profile_switch_default(Register mdp) {
1850   if (ProfileInterpreter) {
1851     Label profile_continue;
1852 
1853     // If no method data exists, go to profile_continue.
1854     test_method_data_pointer(mdp, profile_continue);
1855 
1856     // Update the default case count
1857     increment_mdp_data_at(mdp,
1858                           in_bytes(MultiBranchData::default_count_offset()));
1859 
1860     // The method data pointer needs to be updated.
1861     update_mdp_by_offset(mdp,
1862                          in_bytes(MultiBranchData::
1863                                   default_displacement_offset()));
1864 
1865     bind(profile_continue);
1866   }
1867 }
1868 
1869 
1870 void InterpreterMacroAssembler::profile_switch_case(Register index,
1871                                                     Register mdp,
1872                                                     Register reg2) {
1873   if (ProfileInterpreter) {
1874     Label profile_continue;
1875 
1876     // If no method data exists, go to profile_continue.
1877     test_method_data_pointer(mdp, profile_continue);
1878 
1879     // Build the base (index * per_case_size_in_bytes()) +
1880     // case_array_offset_in_bytes()
1881     movl(reg2, in_bytes(MultiBranchData::per_case_size()));
1882     imulptr(index, reg2); // XXX l ?
1883     addptr(index, in_bytes(MultiBranchData::case_array_offset())); // XXX l ?
1884 
1885     // Update the case count
1886     increment_mdp_data_at(mdp,
1887                           index,
1888                           in_bytes(MultiBranchData::relative_count_offset()));
1889 
1890     // The method data pointer needs to be updated.
1891     update_mdp_by_offset(mdp,
1892                          index,
1893                          in_bytes(MultiBranchData::
1894                                   relative_displacement_offset()));
1895 
1896     bind(profile_continue);
1897   }
1898 }
1899 
1900 
1901 
1902 void InterpreterMacroAssembler::_interp_verify_oop(Register reg, TosState state, const char* file, int line) {
1903   if (state == atos) {
1904     MacroAssembler::_verify_oop_checked(reg, "broken oop", file, line);
1905   }
1906 }
1907 
1908 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) {
1909 #ifndef _LP64
1910   if ((state == ftos && UseSSE < 1) ||
1911       (state == dtos && UseSSE < 2)) {
1912     MacroAssembler::verify_FPU(stack_depth);
1913   }
1914 #endif
1915 }
1916 
1917 // Jump if ((*counter_addr += increment) & mask) == 0
1918 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr, Address mask,
1919                                                         Register scratch, Label* where) {
1920   // This update is actually not atomic and can lose a number of updates
1921   // under heavy contention, but the alternative of using the (contended)
1922   // atomic update here penalizes profiling paths too much.
1923   movl(scratch, counter_addr);
1924   incrementl(scratch, InvocationCounter::count_increment);
1925   movl(counter_addr, scratch);
1926   andl(scratch, mask);
1927   if (where != nullptr) {
1928     jcc(Assembler::zero, *where);
1929   }
1930 }
1931 
1932 void InterpreterMacroAssembler::notify_method_entry() {
1933   // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1934   // track stack depth.  If it is possible to enter interp_only_mode we add
1935   // the code to check if the event should be sent.
1936   Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
1937   Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rbx);
1938   if (JvmtiExport::can_post_interpreter_events()) {
1939     Label L;
1940     NOT_LP64(get_thread(rthread);)
1941     movl(rdx, Address(rthread, JavaThread::interp_only_mode_offset()));
1942     testl(rdx, rdx);
1943     jcc(Assembler::zero, L);
1944     call_VM(noreg, CAST_FROM_FN_PTR(address,
1945                                     InterpreterRuntime::post_method_entry));
1946     bind(L);
1947   }
1948 
1949   if (DTraceMethodProbes) {
1950     NOT_LP64(get_thread(rthread);)
1951     get_method(rarg);
1952     call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
1953                  rthread, rarg);
1954   }
1955 
1956   // RedefineClasses() tracing support for obsolete method entry
1957   if (log_is_enabled(Trace, redefine, class, obsolete)) {
1958     NOT_LP64(get_thread(rthread);)
1959     get_method(rarg);
1960     call_VM_leaf(
1961       CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
1962       rthread, rarg);
1963   }
1964 }
1965 
1966 
1967 void InterpreterMacroAssembler::notify_method_exit(
1968     TosState state, NotifyMethodExitMode mode) {
1969   // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1970   // track stack depth.  If it is possible to enter interp_only_mode we add
1971   // the code to check if the event should be sent.
1972   Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
1973   Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rbx);
1974   if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
1975     Label L;
1976     // Note: frame::interpreter_frame_result has a dependency on how the
1977     // method result is saved across the call to post_method_exit. If this
1978     // is changed then the interpreter_frame_result implementation will
1979     // need to be updated too.
1980 
1981     // template interpreter will leave the result on the top of the stack.
1982     push(state);
1983     NOT_LP64(get_thread(rthread);)
1984     movl(rdx, Address(rthread, JavaThread::interp_only_mode_offset()));
1985     testl(rdx, rdx);
1986     jcc(Assembler::zero, L);
1987     call_VM(noreg,
1988             CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
1989     bind(L);
1990     pop(state);
1991   }
1992 
1993   if (DTraceMethodProbes) {
1994     push(state);
1995     NOT_LP64(get_thread(rthread);)
1996     get_method(rarg);
1997     call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
1998                  rthread, rarg);
1999     pop(state);
2000   }
2001 }
2002 
2003 void InterpreterMacroAssembler::load_resolved_indy_entry(Register cache, Register index) {
2004   // Get index out of bytecode pointer
2005   get_cache_index_at_bcp(index, 1, sizeof(u4));
2006   // Get address of invokedynamic array
2007   movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
2008   movptr(cache, Address(cache, in_bytes(ConstantPoolCache::invokedynamic_entries_offset())));
2009   if (is_power_of_2(sizeof(ResolvedIndyEntry))) {
2010     shll(index, log2i_exact(sizeof(ResolvedIndyEntry))); // Scale index by power of 2
2011   } else {
2012     imull(index, index, sizeof(ResolvedIndyEntry)); // Scale the index to be the entry index * sizeof(ResolvedIndyEntry)
2013   }
2014   lea(cache, Address(cache, index, Address::times_1, Array<ResolvedIndyEntry>::base_offset_in_bytes()));
2015 }
2016 
2017 void InterpreterMacroAssembler::load_field_entry(Register cache, Register index, int bcp_offset) {
2018   // Get index out of bytecode pointer
2019   movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
2020   get_cache_index_at_bcp(index, bcp_offset, sizeof(u2));
2021 
2022   movptr(cache, Address(cache, ConstantPoolCache::field_entries_offset()));
2023   // Take shortcut if the size is a power of 2
2024   if (is_power_of_2(sizeof(ResolvedFieldEntry))) {
2025     shll(index, log2i_exact(sizeof(ResolvedFieldEntry))); // Scale index by power of 2
2026   } else {
2027     imull(index, index, sizeof(ResolvedFieldEntry)); // Scale the index to be the entry index * sizeof(ResolvedFieldEntry)
2028   }
2029   lea(cache, Address(cache, index, Address::times_1, Array<ResolvedFieldEntry>::base_offset_in_bytes()));
2030 }
2031 
2032 void InterpreterMacroAssembler::load_method_entry(Register cache, Register index, int bcp_offset) {
2033   // Get index out of bytecode pointer
2034   movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
2035   get_cache_index_at_bcp(index, bcp_offset, sizeof(u2));
2036 
2037   movptr(cache, Address(cache, ConstantPoolCache::method_entries_offset()));
2038   imull(index, index, sizeof(ResolvedMethodEntry)); // Scale the index to be the entry index * sizeof(ResolvedMethodEntry)
2039   lea(cache, Address(cache, index, Address::times_1, Array<ResolvedMethodEntry>::base_offset_in_bytes()));
2040 }