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/constMethodFlags.hpp" 32 #include "oops/markWord.hpp" 33 #include "oops/methodData.hpp" 34 #include "oops/method.hpp" 35 #include "oops/inlineKlass.hpp" 36 #include "oops/resolvedFieldEntry.hpp" 37 #include "oops/resolvedIndyEntry.hpp" 38 #include "oops/resolvedMethodEntry.hpp" 39 #include "prims/jvmtiExport.hpp" 40 #include "prims/jvmtiThreadState.hpp" 41 #include "runtime/basicLock.hpp" 42 #include "runtime/frame.inline.hpp" 43 #include "runtime/javaThread.hpp" 44 #include "runtime/safepointMechanism.hpp" 45 #include "runtime/sharedRuntime.hpp" 46 #include "utilities/powerOfTwo.hpp" 47 48 // Implementation of InterpreterMacroAssembler 49 50 void InterpreterMacroAssembler::jump_to_entry(address entry) { 51 assert(entry, "Entry must have been generated by now"); 52 jump(RuntimeAddress(entry)); 53 } 54 55 void InterpreterMacroAssembler::profile_obj_type(Register obj, const Address& mdo_addr) { 56 Label update, next, none; 57 58 assert_different_registers(obj, rscratch1, mdo_addr.base(), mdo_addr.index()); 59 60 interp_verify_oop(obj, atos); 61 62 testptr(obj, obj); 63 jccb(Assembler::notZero, update); 64 testptr(mdo_addr, TypeEntries::null_seen); 65 jccb(Assembler::notZero, next); // null already seen. Nothing to do anymore. 66 // atomic update to prevent overwriting Klass* with 0 67 lock(); 68 orptr(mdo_addr, TypeEntries::null_seen); 69 jmpb(next); 70 71 bind(update); 72 load_klass(obj, obj, rscratch1); 73 mov(rscratch1, obj); 74 75 xorptr(obj, mdo_addr); 76 testptr(obj, TypeEntries::type_klass_mask); 77 jccb(Assembler::zero, next); // klass seen before, nothing to 78 // do. The unknown bit may have been 79 // set already but no need to check. 80 81 testptr(obj, TypeEntries::type_unknown); 82 jccb(Assembler::notZero, next); // already unknown. Nothing to do anymore. 83 84 cmpptr(mdo_addr, 0); 85 jccb(Assembler::equal, none); 86 cmpptr(mdo_addr, TypeEntries::null_seen); 87 jccb(Assembler::equal, none); 88 89 // There is a chance that the checks above (re-reading profiling 90 // data from memory) fail if another thread has just set the 91 // profiling to this obj's klass 92 mov(obj, rscratch1); 93 xorptr(obj, mdo_addr); 94 testptr(obj, TypeEntries::type_klass_mask); 95 jccb(Assembler::zero, next); 96 97 // different than before. Cannot keep accurate profile. 98 orptr(mdo_addr, TypeEntries::type_unknown); 99 jmpb(next); 100 101 bind(none); 102 // first time here. Set profile type. 103 movptr(mdo_addr, obj); 104 #ifdef ASSERT 105 andptr(obj, TypeEntries::type_klass_mask); 106 verify_klass_ptr(obj); 107 #endif 108 109 bind(next); 110 } 111 112 void InterpreterMacroAssembler::profile_arguments_type(Register mdp, Register callee, Register tmp, bool is_virtual) { 113 if (!ProfileInterpreter) { 114 return; 115 } 116 117 if (MethodData::profile_arguments() || MethodData::profile_return()) { 118 Label profile_continue; 119 120 test_method_data_pointer(mdp, profile_continue); 121 122 int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size()); 123 124 cmpb(Address(mdp, in_bytes(DataLayout::tag_offset()) - off_to_start), is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag); 125 jcc(Assembler::notEqual, profile_continue); 126 127 if (MethodData::profile_arguments()) { 128 Label done; 129 int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset()); 130 addptr(mdp, off_to_args); 131 132 for (int i = 0; i < TypeProfileArgsLimit; i++) { 133 if (i > 0 || MethodData::profile_return()) { 134 // If return value type is profiled we may have no argument to profile 135 movptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args)); 136 subl(tmp, i*TypeStackSlotEntries::per_arg_count()); 137 cmpl(tmp, TypeStackSlotEntries::per_arg_count()); 138 jcc(Assembler::less, done); 139 } 140 movptr(tmp, Address(callee, Method::const_offset())); 141 load_unsigned_short(tmp, Address(tmp, ConstMethod::size_of_parameters_offset())); 142 // stack offset o (zero based) from the start of the argument 143 // list, for n arguments translates into offset n - o - 1 from 144 // the end of the argument list 145 subptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))-off_to_args)); 146 subl(tmp, 1); 147 Address arg_addr = argument_address(tmp); 148 movptr(tmp, arg_addr); 149 150 Address mdo_arg_addr(mdp, in_bytes(TypeEntriesAtCall::argument_type_offset(i))-off_to_args); 151 profile_obj_type(tmp, mdo_arg_addr); 152 153 int to_add = in_bytes(TypeStackSlotEntries::per_arg_size()); 154 addptr(mdp, to_add); 155 off_to_args += to_add; 156 } 157 158 if (MethodData::profile_return()) { 159 movptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args)); 160 subl(tmp, TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count()); 161 } 162 163 bind(done); 164 165 if (MethodData::profile_return()) { 166 // We're right after the type profile for the last 167 // argument. tmp is the number of cells left in the 168 // CallTypeData/VirtualCallTypeData to reach its end. Non null 169 // if there's a return to profile. 170 assert(SingleTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type"); 171 shll(tmp, log2i_exact((int)DataLayout::cell_size)); 172 addptr(mdp, tmp); 173 } 174 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp); 175 } else { 176 assert(MethodData::profile_return(), "either profile call args or call ret"); 177 update_mdp_by_constant(mdp, in_bytes(TypeEntriesAtCall::return_only_size())); 178 } 179 180 // mdp points right after the end of the 181 // CallTypeData/VirtualCallTypeData, right after the cells for the 182 // return value type if there's one 183 184 bind(profile_continue); 185 } 186 } 187 188 void InterpreterMacroAssembler::profile_return_type(Register mdp, Register ret, Register tmp) { 189 assert_different_registers(mdp, ret, tmp, _bcp_register); 190 if (ProfileInterpreter && MethodData::profile_return()) { 191 Label profile_continue; 192 193 test_method_data_pointer(mdp, profile_continue); 194 195 if (MethodData::profile_return_jsr292_only()) { 196 assert(Method::intrinsic_id_size_in_bytes() == 2, "assuming Method::_intrinsic_id is u2"); 197 198 // If we don't profile all invoke bytecodes we must make sure 199 // it's a bytecode we indeed profile. We can't go back to the 200 // beginning of the ProfileData we intend to update to check its 201 // type because we're right after it and we don't known its 202 // length 203 Label do_profile; 204 cmpb(Address(_bcp_register, 0), Bytecodes::_invokedynamic); 205 jcc(Assembler::equal, do_profile); 206 cmpb(Address(_bcp_register, 0), Bytecodes::_invokehandle); 207 jcc(Assembler::equal, do_profile); 208 get_method(tmp); 209 cmpw(Address(tmp, Method::intrinsic_id_offset()), static_cast<int>(vmIntrinsics::_compiledLambdaForm)); 210 jcc(Assembler::notEqual, profile_continue); 211 212 bind(do_profile); 213 } 214 215 Address mdo_ret_addr(mdp, -in_bytes(SingleTypeEntry::size())); 216 mov(tmp, ret); 217 profile_obj_type(tmp, mdo_ret_addr); 218 219 bind(profile_continue); 220 } 221 } 222 223 void InterpreterMacroAssembler::profile_parameters_type(Register mdp, Register tmp1, Register tmp2) { 224 if (ProfileInterpreter && MethodData::profile_parameters()) { 225 Label profile_continue; 226 227 test_method_data_pointer(mdp, profile_continue); 228 229 // Load the offset of the area within the MDO used for 230 // parameters. If it's negative we're not profiling any parameters 231 movl(tmp1, Address(mdp, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset()))); 232 testl(tmp1, tmp1); 233 jcc(Assembler::negative, profile_continue); 234 235 // Compute a pointer to the area for parameters from the offset 236 // and move the pointer to the slot for the last 237 // parameters. Collect profiling from last parameter down. 238 // mdo start + parameters offset + array length - 1 239 addptr(mdp, tmp1); 240 movptr(tmp1, Address(mdp, ArrayData::array_len_offset())); 241 decrement(tmp1, TypeStackSlotEntries::per_arg_count()); 242 243 Label loop; 244 bind(loop); 245 246 int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0)); 247 int type_base = in_bytes(ParametersTypeData::type_offset(0)); 248 Address::ScaleFactor per_arg_scale = Address::times(DataLayout::cell_size); 249 Address arg_off(mdp, tmp1, per_arg_scale, off_base); 250 Address arg_type(mdp, tmp1, per_arg_scale, type_base); 251 252 // load offset on the stack from the slot for this parameter 253 movptr(tmp2, arg_off); 254 negptr(tmp2); 255 // read the parameter from the local area 256 movptr(tmp2, Address(_locals_register, tmp2, Interpreter::stackElementScale())); 257 258 // profile the parameter 259 profile_obj_type(tmp2, arg_type); 260 261 // go to next parameter 262 decrement(tmp1, TypeStackSlotEntries::per_arg_count()); 263 jcc(Assembler::positive, loop); 264 265 bind(profile_continue); 266 } 267 } 268 269 void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point, 270 int number_of_arguments) { 271 // interpreter specific 272 // 273 // Note: No need to save/restore bcp & locals registers 274 // since these are callee saved registers and no blocking/ 275 // GC can happen in leaf calls. 276 // Further Note: DO NOT save/restore bcp/locals. If a caller has 277 // already saved them so that it can use rsi/rdi as temporaries 278 // then a save/restore here will DESTROY the copy the caller 279 // saved! There used to be a save_bcp() that only happened in 280 // the ASSERT path (no restore_bcp). Which caused bizarre failures 281 // when jvm built with ASSERTs. 282 #ifdef ASSERT 283 { 284 Label L; 285 cmpptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD); 286 jcc(Assembler::equal, L); 287 stop("InterpreterMacroAssembler::call_VM_leaf_base:" 288 " last_sp != null"); 289 bind(L); 290 } 291 #endif 292 // super call 293 MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments); 294 // interpreter specific 295 // LP64: Used to ASSERT that r13/r14 were equal to frame's bcp/locals 296 // but since they may not have been saved (and we don't want to 297 // save them here (see note above) the assert is invalid. 298 } 299 300 void InterpreterMacroAssembler::call_VM_base(Register oop_result, 301 Register last_java_sp, 302 address entry_point, 303 int number_of_arguments, 304 bool check_exceptions) { 305 // interpreter specific 306 // 307 // Note: Could avoid restoring locals ptr (callee saved) - however doesn't 308 // really make a difference for these runtime calls, since they are 309 // slow anyway. Btw., bcp must be saved/restored since it may change 310 // due to GC. 311 save_bcp(); 312 #ifdef ASSERT 313 { 314 Label L; 315 cmpptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD); 316 jcc(Assembler::equal, L); 317 stop("InterpreterMacroAssembler::call_VM_base:" 318 " last_sp isn't null"); 319 bind(L); 320 } 321 #endif /* ASSERT */ 322 // super call 323 MacroAssembler::call_VM_base(oop_result, last_java_sp, 324 entry_point, number_of_arguments, 325 check_exceptions); 326 // interpreter specific 327 restore_bcp(); 328 restore_locals(); 329 } 330 331 void InterpreterMacroAssembler::call_VM_preemptable(Register oop_result, 332 address entry_point, 333 Register arg_1) { 334 assert(arg_1 == c_rarg1, ""); 335 Label resume_pc, not_preempted; 336 337 #ifdef ASSERT 338 { 339 Label L; 340 cmpptr(Address(r15_thread, JavaThread::preempt_alternate_return_offset()), NULL_WORD); 341 jcc(Assembler::equal, L); 342 stop("Should not have alternate return address set"); 343 bind(L); 344 } 345 #endif /* ASSERT */ 346 347 // Force freeze slow path. 348 push_cont_fastpath(); 349 350 // Make VM call. In case of preemption set last_pc to the one we want to resume to. 351 // Note: call_VM_helper requires last_Java_pc for anchor to be at the top of the stack. 352 lea(rscratch1, resume_pc); 353 push(rscratch1); 354 MacroAssembler::call_VM_helper(oop_result, entry_point, 1, false /*check_exceptions*/); 355 pop(rscratch1); 356 357 pop_cont_fastpath(); 358 359 // Check if preempted. 360 movptr(rscratch1, Address(r15_thread, JavaThread::preempt_alternate_return_offset())); 361 cmpptr(rscratch1, NULL_WORD); 362 jccb(Assembler::zero, not_preempted); 363 movptr(Address(r15_thread, JavaThread::preempt_alternate_return_offset()), NULL_WORD); 364 jmp(rscratch1); 365 366 // In case of preemption, this is where we will resume once we finally acquire the monitor. 367 bind(resume_pc); 368 restore_after_resume(false /* is_native */); 369 370 bind(not_preempted); 371 } 372 373 void InterpreterMacroAssembler::restore_after_resume(bool is_native) { 374 lea(rscratch1, ExternalAddress(Interpreter::cont_resume_interpreter_adapter())); 375 call(rscratch1); 376 if (is_native) { 377 // On resume we need to set up stack as expected. 378 push(dtos); 379 push(ltos); 380 } 381 } 382 383 void InterpreterMacroAssembler::check_and_handle_popframe() { 384 if (JvmtiExport::can_pop_frame()) { 385 Label L; 386 // Initiate popframe handling only if it is not already being 387 // processed. If the flag has the popframe_processing bit set, it 388 // means that this code is called *during* popframe handling - we 389 // don't want to reenter. 390 // This method is only called just after the call into the vm in 391 // call_VM_base, so the arg registers are available. 392 Register pop_cond = c_rarg0; 393 movl(pop_cond, Address(r15_thread, JavaThread::popframe_condition_offset())); 394 testl(pop_cond, JavaThread::popframe_pending_bit); 395 jcc(Assembler::zero, L); 396 testl(pop_cond, JavaThread::popframe_processing_bit); 397 jcc(Assembler::notZero, L); 398 // Call Interpreter::remove_activation_preserving_args_entry() to get the 399 // address of the same-named entrypoint in the generated interpreter code. 400 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry)); 401 jmp(rax); 402 bind(L); 403 } 404 } 405 406 void InterpreterMacroAssembler::load_earlyret_value(TosState state) { 407 movptr(rcx, Address(r15_thread, JavaThread::jvmti_thread_state_offset())); 408 const Address tos_addr(rcx, JvmtiThreadState::earlyret_tos_offset()); 409 const Address oop_addr(rcx, JvmtiThreadState::earlyret_oop_offset()); 410 const Address val_addr(rcx, JvmtiThreadState::earlyret_value_offset()); 411 412 switch (state) { 413 case atos: movptr(rax, oop_addr); 414 movptr(oop_addr, NULL_WORD); 415 interp_verify_oop(rax, state); break; 416 case ltos: movptr(rax, val_addr); break; 417 case btos: // fall through 418 case ztos: // fall through 419 case ctos: // fall through 420 case stos: // fall through 421 case itos: movl(rax, val_addr); break; 422 case ftos: movflt(xmm0, val_addr); break; 423 case dtos: movdbl(xmm0, val_addr); break; 424 case vtos: /* nothing to do */ break; 425 default : ShouldNotReachHere(); 426 } 427 428 // Clean up tos value in the thread object 429 movl(tos_addr, ilgl); 430 movptr(val_addr, NULL_WORD); 431 } 432 433 434 void InterpreterMacroAssembler::check_and_handle_earlyret() { 435 if (JvmtiExport::can_force_early_return()) { 436 Label L; 437 Register tmp = c_rarg0; 438 Register rthread = r15_thread; 439 440 movptr(tmp, Address(rthread, JavaThread::jvmti_thread_state_offset())); 441 testptr(tmp, tmp); 442 jcc(Assembler::zero, L); // if (thread->jvmti_thread_state() == nullptr) exit; 443 444 // Initiate earlyret handling only if it is not already being processed. 445 // If the flag has the earlyret_processing bit set, it means that this code 446 // is called *during* earlyret handling - we don't want to reenter. 447 movl(tmp, Address(tmp, JvmtiThreadState::earlyret_state_offset())); 448 cmpl(tmp, JvmtiThreadState::earlyret_pending); 449 jcc(Assembler::notEqual, L); 450 451 // Call Interpreter::remove_activation_early_entry() to get the address of the 452 // same-named entrypoint in the generated interpreter code. 453 movptr(tmp, Address(rthread, JavaThread::jvmti_thread_state_offset())); 454 movl(tmp, Address(tmp, JvmtiThreadState::earlyret_tos_offset())); 455 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), tmp); 456 jmp(rax); 457 bind(L); 458 } 459 } 460 461 void InterpreterMacroAssembler::get_unsigned_2_byte_index_at_bcp(Register reg, int bcp_offset) { 462 assert(bcp_offset >= 0, "bcp is still pointing to start of bytecode"); 463 load_unsigned_short(reg, Address(_bcp_register, bcp_offset)); 464 bswapl(reg); 465 shrl(reg, 16); 466 } 467 468 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register index, 469 int bcp_offset, 470 size_t index_size) { 471 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode"); 472 if (index_size == sizeof(u2)) { 473 load_unsigned_short(index, Address(_bcp_register, bcp_offset)); 474 } else if (index_size == sizeof(u4)) { 475 movl(index, Address(_bcp_register, bcp_offset)); 476 } else if (index_size == sizeof(u1)) { 477 load_unsigned_byte(index, Address(_bcp_register, bcp_offset)); 478 } else { 479 ShouldNotReachHere(); 480 } 481 } 482 483 // Load object from cpool->resolved_references(index) 484 void InterpreterMacroAssembler::load_resolved_reference_at_index(Register result, 485 Register index, 486 Register tmp) { 487 assert_different_registers(result, index); 488 489 get_constant_pool(result); 490 // load pointer for resolved_references[] objArray 491 movptr(result, Address(result, ConstantPool::cache_offset())); 492 movptr(result, Address(result, ConstantPoolCache::resolved_references_offset())); 493 resolve_oop_handle(result, tmp); 494 load_heap_oop(result, Address(result, index, 495 UseCompressedOops ? Address::times_4 : Address::times_ptr, 496 arrayOopDesc::base_offset_in_bytes(T_OBJECT)), tmp); 497 } 498 499 // load cpool->resolved_klass_at(index) 500 void InterpreterMacroAssembler::load_resolved_klass_at_index(Register klass, 501 Register cpool, 502 Register index) { 503 assert_different_registers(cpool, index); 504 505 movw(index, Address(cpool, index, Address::times_ptr, sizeof(ConstantPool))); 506 Register resolved_klasses = cpool; 507 movptr(resolved_klasses, Address(cpool, ConstantPool::resolved_klasses_offset())); 508 movptr(klass, Address(resolved_klasses, index, Address::times_ptr, Array<Klass*>::base_offset_in_bytes())); 509 } 510 511 // Generate a subtype check: branch to ok_is_subtype if sub_klass is a 512 // subtype of super_klass. 513 // 514 // Args: 515 // rax: superklass 516 // Rsub_klass: subklass 517 // 518 // Kills: 519 // rcx, rdi 520 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass, 521 Label& ok_is_subtype, 522 bool profile) { 523 assert(Rsub_klass != rax, "rax holds superklass"); 524 LP64_ONLY(assert(Rsub_klass != r14, "r14 holds locals");) 525 LP64_ONLY(assert(Rsub_klass != r13, "r13 holds bcp");) 526 assert(Rsub_klass != rcx, "rcx holds 2ndary super array length"); 527 assert(Rsub_klass != rdi, "rdi holds 2ndary super array scan ptr"); 528 529 // Profile the not-null value's klass. 530 if (profile) { 531 profile_typecheck(rcx, Rsub_klass, rdi); // blows rcx, reloads rdi 532 } 533 534 // Do the check. 535 check_klass_subtype(Rsub_klass, rax, rcx, ok_is_subtype); // blows rcx 536 } 537 538 539 // Java Expression Stack 540 541 void InterpreterMacroAssembler::pop_ptr(Register r) { 542 pop(r); 543 } 544 545 void InterpreterMacroAssembler::push_ptr(Register r) { 546 push(r); 547 } 548 549 void InterpreterMacroAssembler::push_i(Register r) { 550 push(r); 551 } 552 553 void InterpreterMacroAssembler::push_i_or_ptr(Register r) { 554 push(r); 555 } 556 557 void InterpreterMacroAssembler::push_f(XMMRegister r) { 558 subptr(rsp, wordSize); 559 movflt(Address(rsp, 0), r); 560 } 561 562 void InterpreterMacroAssembler::pop_f(XMMRegister r) { 563 movflt(r, Address(rsp, 0)); 564 addptr(rsp, wordSize); 565 } 566 567 void InterpreterMacroAssembler::push_d(XMMRegister r) { 568 subptr(rsp, 2 * wordSize); 569 movdbl(Address(rsp, 0), r); 570 } 571 572 void InterpreterMacroAssembler::pop_d(XMMRegister r) { 573 movdbl(r, Address(rsp, 0)); 574 addptr(rsp, 2 * Interpreter::stackElementSize); 575 } 576 577 void InterpreterMacroAssembler::pop_i(Register r) { 578 // XXX can't use pop currently, upper half non clean 579 movl(r, Address(rsp, 0)); 580 addptr(rsp, wordSize); 581 } 582 583 void InterpreterMacroAssembler::pop_l(Register r) { 584 movq(r, Address(rsp, 0)); 585 addptr(rsp, 2 * Interpreter::stackElementSize); 586 } 587 588 void InterpreterMacroAssembler::push_l(Register r) { 589 subptr(rsp, 2 * wordSize); 590 movptr(Address(rsp, Interpreter::expr_offset_in_bytes(0)), r ); 591 movptr(Address(rsp, Interpreter::expr_offset_in_bytes(1)), NULL_WORD ); 592 } 593 594 void InterpreterMacroAssembler::pop(TosState state) { 595 switch (state) { 596 case atos: pop_ptr(); break; 597 case btos: 598 case ztos: 599 case ctos: 600 case stos: 601 case itos: pop_i(); break; 602 case ltos: pop_l(); break; 603 case ftos: pop_f(xmm0); break; 604 case dtos: pop_d(xmm0); break; 605 case vtos: /* nothing to do */ break; 606 default: ShouldNotReachHere(); 607 } 608 interp_verify_oop(rax, state); 609 } 610 611 void InterpreterMacroAssembler::push(TosState state) { 612 interp_verify_oop(rax, state); 613 switch (state) { 614 case atos: push_ptr(); break; 615 case btos: 616 case ztos: 617 case ctos: 618 case stos: 619 case itos: push_i(); break; 620 case ltos: push_l(); break; 621 case ftos: push_f(xmm0); break; 622 case dtos: push_d(xmm0); break; 623 case vtos: /* nothing to do */ break; 624 default : ShouldNotReachHere(); 625 } 626 } 627 628 // Helpers for swap and dup 629 void InterpreterMacroAssembler::load_ptr(int n, Register val) { 630 movptr(val, Address(rsp, Interpreter::expr_offset_in_bytes(n))); 631 } 632 633 void InterpreterMacroAssembler::store_ptr(int n, Register val) { 634 movptr(Address(rsp, Interpreter::expr_offset_in_bytes(n)), val); 635 } 636 637 638 void InterpreterMacroAssembler::prepare_to_jump_from_interpreted() { 639 // set sender sp 640 lea(_bcp_register, Address(rsp, wordSize)); 641 // record last_sp 642 mov(rcx, _bcp_register); 643 subptr(rcx, rbp); 644 sarptr(rcx, LogBytesPerWord); 645 movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), rcx); 646 } 647 648 649 // Jump to from_interpreted entry of a call unless single stepping is possible 650 // in this thread in which case we must call the i2i entry 651 void InterpreterMacroAssembler::jump_from_interpreted(Register method, Register temp) { 652 prepare_to_jump_from_interpreted(); 653 654 if (JvmtiExport::can_post_interpreter_events()) { 655 Label run_compiled_code; 656 // JVMTI events, such as single-stepping, are implemented partly by avoiding running 657 // compiled code in threads for which the event is enabled. Check here for 658 // interp_only_mode if these events CAN be enabled. 659 // interp_only is an int, on little endian it is sufficient to test the byte only 660 // Is a cmpl faster? 661 cmpb(Address(r15_thread, JavaThread::interp_only_mode_offset()), 0); 662 jccb(Assembler::zero, run_compiled_code); 663 jmp(Address(method, Method::interpreter_entry_offset())); 664 bind(run_compiled_code); 665 } 666 667 jmp(Address(method, Method::from_interpreted_offset())); 668 } 669 670 // The following two routines provide a hook so that an implementation 671 // can schedule the dispatch in two parts. x86 does not do this. 672 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int step) { 673 // Nothing x86 specific to be done here 674 } 675 676 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int step) { 677 dispatch_next(state, step); 678 } 679 680 void InterpreterMacroAssembler::dispatch_base(TosState state, 681 address* table, 682 bool verifyoop, 683 bool generate_poll) { 684 if (VerifyActivationFrameSize) { 685 Label L; 686 mov(rcx, rbp); 687 subptr(rcx, rsp); 688 int32_t min_frame_size = 689 (frame::link_offset - frame::interpreter_frame_initial_sp_offset) * 690 wordSize; 691 cmpptr(rcx, min_frame_size); 692 jcc(Assembler::greaterEqual, L); 693 stop("broken stack frame"); 694 bind(L); 695 } 696 if (verifyoop) { 697 interp_verify_oop(rax, state); 698 } 699 700 address* const safepoint_table = Interpreter::safept_table(state); 701 Label no_safepoint, dispatch; 702 if (table != safepoint_table && generate_poll) { 703 NOT_PRODUCT(block_comment("Thread-local Safepoint poll")); 704 testb(Address(r15_thread, JavaThread::polling_word_offset()), SafepointMechanism::poll_bit()); 705 706 jccb(Assembler::zero, no_safepoint); 707 lea(rscratch1, ExternalAddress((address)safepoint_table)); 708 jmpb(dispatch); 709 } 710 711 bind(no_safepoint); 712 lea(rscratch1, ExternalAddress((address)table)); 713 bind(dispatch); 714 jmp(Address(rscratch1, rbx, Address::times_8)); 715 } 716 717 void InterpreterMacroAssembler::dispatch_only(TosState state, bool generate_poll) { 718 dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll); 719 } 720 721 void InterpreterMacroAssembler::dispatch_only_normal(TosState state) { 722 dispatch_base(state, Interpreter::normal_table(state)); 723 } 724 725 void InterpreterMacroAssembler::dispatch_only_noverify(TosState state) { 726 dispatch_base(state, Interpreter::normal_table(state), false); 727 } 728 729 730 void InterpreterMacroAssembler::dispatch_next(TosState state, int step, bool generate_poll) { 731 // load next bytecode (load before advancing _bcp_register to prevent AGI) 732 load_unsigned_byte(rbx, Address(_bcp_register, step)); 733 // advance _bcp_register 734 increment(_bcp_register, step); 735 dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll); 736 } 737 738 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) { 739 // load current bytecode 740 load_unsigned_byte(rbx, Address(_bcp_register, 0)); 741 dispatch_base(state, table); 742 } 743 744 void InterpreterMacroAssembler::narrow(Register result) { 745 746 // Get method->_constMethod->_result_type 747 movptr(rcx, Address(rbp, frame::interpreter_frame_method_offset * wordSize)); 748 movptr(rcx, Address(rcx, Method::const_offset())); 749 load_unsigned_byte(rcx, Address(rcx, ConstMethod::result_type_offset())); 750 751 Label done, notBool, notByte, notChar; 752 753 // common case first 754 cmpl(rcx, T_INT); 755 jcc(Assembler::equal, done); 756 757 // mask integer result to narrower return type. 758 cmpl(rcx, T_BOOLEAN); 759 jcc(Assembler::notEqual, notBool); 760 andl(result, 0x1); 761 jmp(done); 762 763 bind(notBool); 764 cmpl(rcx, T_BYTE); 765 jcc(Assembler::notEqual, notByte); 766 movsbl(result, result); 767 jmp(done); 768 769 bind(notByte); 770 cmpl(rcx, T_CHAR); 771 jcc(Assembler::notEqual, notChar); 772 movzwl(result, result); 773 jmp(done); 774 775 bind(notChar); 776 // cmpl(rcx, T_SHORT); // all that's left 777 // jcc(Assembler::notEqual, done); 778 movswl(result, result); 779 780 // Nothing to do for T_INT 781 bind(done); 782 } 783 784 // remove activation 785 // 786 // Apply stack watermark barrier. 787 // Unlock the receiver if this is a synchronized method. 788 // Unlock any Java monitors from synchronized blocks. 789 // Remove the activation from the stack. 790 // 791 // If there are locked Java monitors 792 // If throw_monitor_exception 793 // throws IllegalMonitorStateException 794 // Else if install_monitor_exception 795 // installs IllegalMonitorStateException 796 // Else 797 // no error processing 798 void InterpreterMacroAssembler::remove_activation( 799 TosState state, 800 Register ret_addr, 801 bool throw_monitor_exception, 802 bool install_monitor_exception, 803 bool notify_jvmdi) { 804 // Note: Registers rdx xmm0 may be in use for the 805 // result check if synchronized method 806 Label unlocked, unlock, no_unlock; 807 808 const Register rthread = r15_thread; 809 const Register robj = c_rarg1; 810 const Register rmon = c_rarg1; 811 812 // The below poll is for the stack watermark barrier. It allows fixing up frames lazily, 813 // that would normally not be safe to use. Such bad returns into unsafe territory of 814 // the stack, will call InterpreterRuntime::at_unwind. 815 Label slow_path; 816 Label fast_path; 817 safepoint_poll(slow_path, true /* at_return */, false /* in_nmethod */); 818 jmp(fast_path); 819 bind(slow_path); 820 push(state); 821 set_last_Java_frame(noreg, rbp, (address)pc(), rscratch1); 822 super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::at_unwind), rthread); 823 reset_last_Java_frame(true); 824 pop(state); 825 bind(fast_path); 826 827 // get the value of _do_not_unlock_if_synchronized into rdx 828 const Address do_not_unlock_if_synchronized(rthread, 829 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset())); 830 movbool(rbx, do_not_unlock_if_synchronized); 831 movbool(do_not_unlock_if_synchronized, false); // reset the flag 832 833 // get method access flags 834 movptr(rcx, Address(rbp, frame::interpreter_frame_method_offset * wordSize)); 835 load_unsigned_short(rcx, Address(rcx, Method::access_flags_offset())); 836 testl(rcx, JVM_ACC_SYNCHRONIZED); 837 jcc(Assembler::zero, unlocked); 838 839 // Don't unlock anything if the _do_not_unlock_if_synchronized flag 840 // is set. 841 testbool(rbx); 842 jcc(Assembler::notZero, no_unlock); 843 844 // unlock monitor 845 push(state); // save result 846 847 // BasicObjectLock will be first in list, since this is a 848 // synchronized method. However, need to check that the object has 849 // not been unlocked by an explicit monitorexit bytecode. 850 const Address monitor(rbp, frame::interpreter_frame_initial_sp_offset * 851 wordSize - (int) sizeof(BasicObjectLock)); 852 // We use c_rarg1/rdx so that if we go slow path it will be the correct 853 // register for unlock_object to pass to VM directly 854 lea(robj, monitor); // address of first monitor 855 856 movptr(rax, Address(robj, BasicObjectLock::obj_offset())); 857 testptr(rax, rax); 858 jcc(Assembler::notZero, unlock); 859 860 pop(state); 861 if (throw_monitor_exception) { 862 // Entry already unlocked, need to throw exception 863 call_VM(noreg, CAST_FROM_FN_PTR(address, 864 InterpreterRuntime::throw_illegal_monitor_state_exception)); 865 should_not_reach_here(); 866 } else { 867 // Monitor already unlocked during a stack unroll. If requested, 868 // install an illegal_monitor_state_exception. Continue with 869 // stack unrolling. 870 if (install_monitor_exception) { 871 call_VM(noreg, CAST_FROM_FN_PTR(address, 872 InterpreterRuntime::new_illegal_monitor_state_exception)); 873 } 874 jmp(unlocked); 875 } 876 877 bind(unlock); 878 unlock_object(robj); 879 pop(state); 880 881 // Check that for block-structured locking (i.e., that all locked 882 // objects has been unlocked) 883 bind(unlocked); 884 885 // rax, rdx: Might contain return value 886 887 // Check that all monitors are unlocked 888 { 889 Label loop, exception, entry, restart; 890 const int entry_size = frame::interpreter_frame_monitor_size_in_bytes(); 891 const Address monitor_block_top( 892 rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize); 893 const Address monitor_block_bot( 894 rbp, frame::interpreter_frame_initial_sp_offset * wordSize); 895 896 bind(restart); 897 // We use c_rarg1 so that if we go slow path it will be the correct 898 // register for unlock_object to pass to VM directly 899 movptr(rmon, monitor_block_top); // derelativize pointer 900 lea(rmon, Address(rbp, rmon, Address::times_ptr)); 901 // c_rarg1 points to current entry, starting with top-most entry 902 903 lea(rbx, monitor_block_bot); // points to word before bottom of 904 // monitor block 905 jmp(entry); 906 907 // Entry already locked, need to throw exception 908 bind(exception); 909 910 if (throw_monitor_exception) { 911 // Throw exception 912 MacroAssembler::call_VM(noreg, 913 CAST_FROM_FN_PTR(address, InterpreterRuntime:: 914 throw_illegal_monitor_state_exception)); 915 should_not_reach_here(); 916 } else { 917 // Stack unrolling. Unlock object and install illegal_monitor_exception. 918 // Unlock does not block, so don't have to worry about the frame. 919 // We don't have to preserve c_rarg1 since we are going to throw an exception. 920 921 push(state); 922 mov(robj, rmon); // nop if robj and rmon are the same 923 unlock_object(robj); 924 pop(state); 925 926 if (install_monitor_exception) { 927 call_VM(noreg, CAST_FROM_FN_PTR(address, 928 InterpreterRuntime:: 929 new_illegal_monitor_state_exception)); 930 } 931 932 jmp(restart); 933 } 934 935 bind(loop); 936 // check if current entry is used 937 cmpptr(Address(rmon, BasicObjectLock::obj_offset()), NULL_WORD); 938 jcc(Assembler::notEqual, exception); 939 940 addptr(rmon, entry_size); // otherwise advance to next entry 941 bind(entry); 942 cmpptr(rmon, rbx); // check if bottom reached 943 jcc(Assembler::notEqual, loop); // if not at bottom then check this entry 944 } 945 946 bind(no_unlock); 947 948 // jvmti support 949 if (notify_jvmdi) { 950 notify_method_exit(state, NotifyJVMTI); // preserve TOSCA 951 } else { 952 notify_method_exit(state, SkipNotifyJVMTI); // preserve TOSCA 953 } 954 955 if (StackReservedPages > 0) { 956 movptr(rbx, 957 Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); 958 // testing if reserved zone needs to be re-enabled 959 Register rthread = r15_thread; 960 Label no_reserved_zone_enabling; 961 962 // check if already enabled - if so no re-enabling needed 963 assert(sizeof(StackOverflow::StackGuardState) == 4, "unexpected size"); 964 cmpl(Address(rthread, JavaThread::stack_guard_state_offset()), StackOverflow::stack_guard_enabled); 965 jcc(Assembler::equal, no_reserved_zone_enabling); 966 967 cmpptr(rbx, Address(rthread, JavaThread::reserved_stack_activation_offset())); 968 jcc(Assembler::lessEqual, no_reserved_zone_enabling); 969 970 call_VM_leaf( 971 CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), rthread); 972 call_VM(noreg, CAST_FROM_FN_PTR(address, 973 InterpreterRuntime::throw_delayed_StackOverflowError)); 974 should_not_reach_here(); 975 976 bind(no_reserved_zone_enabling); 977 } 978 979 // remove activation 980 // get sender sp 981 movptr(rbx, 982 Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); 983 984 if (state == atos && InlineTypeReturnedAsFields) { 985 // Check if we are returning an non-null inline type and load its fields into registers 986 Label skip; 987 test_oop_is_not_inline_type(rax, rscratch1, skip); 988 989 #ifndef _LP64 990 super_call_VM_leaf(StubRoutines::load_inline_type_fields_in_regs()); 991 #else 992 // Load fields from a buffered value with an inline class specific handler 993 load_klass(rdi, rax, rscratch1); 994 movptr(rdi, Address(rdi, InstanceKlass::adr_inlineklass_fixed_block_offset())); 995 movptr(rdi, Address(rdi, InlineKlass::unpack_handler_offset())); 996 // Unpack handler can be null if inline type is not scalarizable in returns 997 testptr(rdi, rdi); 998 jcc(Assembler::zero, skip); 999 call(rdi); 1000 #endif 1001 #ifdef ASSERT 1002 // TODO 8284443 Enable 1003 if (StressCallingConvention && false) { 1004 Label skip_stress; 1005 movptr(rscratch1, Address(rbp, frame::interpreter_frame_method_offset * wordSize)); 1006 movl(rscratch1, Address(rscratch1, Method::flags_offset())); 1007 testl(rcx, MethodFlags::has_scalarized_return_flag()); 1008 jcc(Assembler::zero, skip_stress); 1009 load_klass(rax, rax, rscratch1); 1010 orptr(rax, 1); 1011 bind(skip_stress); 1012 } 1013 #endif 1014 // call above kills the value in rbx. Reload it. 1015 movptr(rbx, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); 1016 bind(skip); 1017 } 1018 leave(); // remove frame anchor 1019 pop(ret_addr); // get return address 1020 mov(rsp, rbx); // set sp to sender sp 1021 pop_cont_fastpath(); 1022 } 1023 1024 void InterpreterMacroAssembler::get_method_counters(Register method, 1025 Register mcs, Label& skip) { 1026 Label has_counters; 1027 movptr(mcs, Address(method, Method::method_counters_offset())); 1028 testptr(mcs, mcs); 1029 jcc(Assembler::notZero, has_counters); 1030 call_VM(noreg, CAST_FROM_FN_PTR(address, 1031 InterpreterRuntime::build_method_counters), method); 1032 movptr(mcs, Address(method,Method::method_counters_offset())); 1033 testptr(mcs, mcs); 1034 jcc(Assembler::zero, skip); // No MethodCounters allocated, OutOfMemory 1035 bind(has_counters); 1036 } 1037 1038 void InterpreterMacroAssembler::allocate_instance(Register klass, Register new_obj, 1039 Register t1, Register t2, 1040 bool clear_fields, Label& alloc_failed) { 1041 MacroAssembler::allocate_instance(klass, new_obj, t1, t2, clear_fields, alloc_failed); 1042 if (DTraceMethodProbes) { 1043 // Trigger dtrace event for fastpath 1044 push(atos); 1045 call_VM_leaf(CAST_FROM_FN_PTR(address, static_cast<int (*)(oopDesc*)>(SharedRuntime::dtrace_object_alloc)), new_obj); 1046 pop(atos); 1047 } 1048 } 1049 1050 void InterpreterMacroAssembler::read_flat_field(Register entry, Register tmp1, Register tmp2, Register obj) { 1051 Label alloc_failed, done; 1052 const Register alloc_temp = LP64_ONLY(rscratch1) NOT_LP64(rsi); 1053 const Register dst_temp = LP64_ONLY(rscratch2) NOT_LP64(rdi); 1054 assert_different_registers(obj, entry, tmp1, tmp2, dst_temp, r8, r9); 1055 1056 // FIXME: code below could be re-written to better use InlineLayoutInfo data structure 1057 // see aarch64 version 1058 1059 // Grap the inline field klass 1060 const Register field_klass = tmp1; 1061 load_unsigned_short(tmp2, Address(entry, in_bytes(ResolvedFieldEntry::field_index_offset()))); 1062 movptr(tmp1, Address(entry, ResolvedFieldEntry::field_holder_offset())); 1063 get_inline_type_field_klass(tmp1, tmp2, field_klass); 1064 1065 // allocate buffer 1066 push(obj); // push object being read from // FIXME spilling on stack could probably be avoided by using tmp2 1067 allocate_instance(field_klass, obj, alloc_temp, dst_temp, false, alloc_failed); 1068 1069 // Have an oop instance buffer, copy into it 1070 load_unsigned_short(r9, Address(entry, in_bytes(ResolvedFieldEntry::field_index_offset()))); 1071 movptr(r8, Address(entry, in_bytes(ResolvedFieldEntry::field_holder_offset()))); 1072 inline_layout_info(r8, r9, r8); // holder, index, info => InlineLayoutInfo into r8 1073 1074 payload_addr(obj, dst_temp, field_klass); 1075 pop(alloc_temp); // restore object being read from 1076 load_sized_value(tmp2, Address(entry, in_bytes(ResolvedFieldEntry::field_offset_offset())), sizeof(int), true /*is_signed*/); 1077 lea(tmp2, Address(alloc_temp, tmp2)); 1078 // call_VM_leaf, clobbers a few regs, save restore new obj 1079 push(obj); 1080 // access_value_copy(IS_DEST_UNINITIALIZED, tmp2, dst_temp, field_klass); 1081 flat_field_copy(IS_DEST_UNINITIALIZED, tmp2, dst_temp, r8); 1082 pop(obj); 1083 jmp(done); 1084 1085 bind(alloc_failed); 1086 pop(obj); 1087 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::read_flat_field), 1088 obj, entry); 1089 get_vm_result_oop(obj); 1090 bind(done); 1091 } 1092 1093 // Lock object 1094 // 1095 // Args: 1096 // rdx, c_rarg1: BasicObjectLock to be used for locking 1097 // 1098 // Kills: 1099 // rax, rbx 1100 void InterpreterMacroAssembler::lock_object(Register lock_reg) { 1101 assert(lock_reg == c_rarg1, "The argument is only for looks. It must be c_rarg1"); 1102 1103 if (LockingMode == LM_MONITOR) { 1104 call_VM_preemptable(noreg, 1105 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), 1106 lock_reg); 1107 } else { 1108 Label count_locking, done, slow_case; 1109 1110 const Register swap_reg = rax; // Must use rax for cmpxchg instruction 1111 const Register tmp_reg = rbx; 1112 const Register obj_reg = c_rarg3; // Will contain the oop 1113 const Register rklass_decode_tmp = rscratch1; 1114 1115 const int obj_offset = in_bytes(BasicObjectLock::obj_offset()); 1116 const int lock_offset = in_bytes(BasicObjectLock::lock_offset()); 1117 const int mark_offset = lock_offset + 1118 BasicLock::displaced_header_offset_in_bytes(); 1119 1120 // Load object pointer into obj_reg 1121 movptr(obj_reg, Address(lock_reg, obj_offset)); 1122 1123 if (LockingMode == LM_LIGHTWEIGHT) { 1124 lightweight_lock(lock_reg, obj_reg, swap_reg, tmp_reg, slow_case); 1125 } else if (LockingMode == LM_LEGACY) { 1126 if (DiagnoseSyncOnValueBasedClasses != 0) { 1127 load_klass(tmp_reg, obj_reg, rklass_decode_tmp); 1128 testb(Address(tmp_reg, Klass::misc_flags_offset()), KlassFlags::_misc_is_value_based_class); 1129 jcc(Assembler::notZero, slow_case); 1130 } 1131 1132 // Load immediate 1 into swap_reg %rax 1133 movl(swap_reg, 1); 1134 1135 // Load (object->mark() | 1) into swap_reg %rax 1136 orptr(swap_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes())); 1137 if (EnableValhalla) { 1138 // Mask inline_type bit such that we go to the slow path if object is an inline type 1139 andptr(swap_reg, ~((int) markWord::inline_type_bit_in_place)); 1140 } 1141 1142 // Save (object->mark() | 1) into BasicLock's displaced header 1143 movptr(Address(lock_reg, mark_offset), swap_reg); 1144 1145 assert(lock_offset == 0, 1146 "displaced header must be first word in BasicObjectLock"); 1147 1148 lock(); 1149 cmpxchgptr(lock_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes())); 1150 jcc(Assembler::zero, count_locking); 1151 1152 const int zero_bits = 7; 1153 1154 // Fast check for recursive lock. 1155 // 1156 // Can apply the optimization only if this is a stack lock 1157 // allocated in this thread. For efficiency, we can focus on 1158 // recently allocated stack locks (instead of reading the stack 1159 // base and checking whether 'mark' points inside the current 1160 // thread stack): 1161 // 1) (mark & zero_bits) == 0, and 1162 // 2) rsp <= mark < mark + os::pagesize() 1163 // 1164 // Warning: rsp + os::pagesize can overflow the stack base. We must 1165 // neither apply the optimization for an inflated lock allocated 1166 // just above the thread stack (this is why condition 1 matters) 1167 // nor apply the optimization if the stack lock is inside the stack 1168 // of another thread. The latter is avoided even in case of overflow 1169 // because we have guard pages at the end of all stacks. Hence, if 1170 // we go over the stack base and hit the stack of another thread, 1171 // this should not be in a writeable area that could contain a 1172 // stack lock allocated by that thread. As a consequence, a stack 1173 // lock less than page size away from rsp is guaranteed to be 1174 // owned by the current thread. 1175 // 1176 // These 3 tests can be done by evaluating the following 1177 // expression: ((mark - rsp) & (zero_bits - os::vm_page_size())), 1178 // assuming both stack pointer and pagesize have their 1179 // least significant bits clear. 1180 // NOTE: the mark is in swap_reg %rax as the result of cmpxchg 1181 subptr(swap_reg, rsp); 1182 andptr(swap_reg, zero_bits - (int)os::vm_page_size()); 1183 1184 // Save the test result, for recursive case, the result is zero 1185 movptr(Address(lock_reg, mark_offset), swap_reg); 1186 jcc(Assembler::notZero, slow_case); 1187 1188 bind(count_locking); 1189 inc_held_monitor_count(); 1190 } 1191 jmp(done); 1192 1193 bind(slow_case); 1194 1195 // Call the runtime routine for slow case 1196 call_VM_preemptable(noreg, 1197 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), 1198 lock_reg); 1199 bind(done); 1200 } 1201 } 1202 1203 1204 // Unlocks an object. Used in monitorexit bytecode and 1205 // remove_activation. Throws an IllegalMonitorException if object is 1206 // not locked by current thread. 1207 // 1208 // Args: 1209 // rdx, c_rarg1: BasicObjectLock for lock 1210 // 1211 // Kills: 1212 // rax 1213 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ... (param regs) 1214 // rscratch1 (scratch reg) 1215 // rax, rbx, rcx, rdx 1216 void InterpreterMacroAssembler::unlock_object(Register lock_reg) { 1217 assert(lock_reg == c_rarg1, "The argument is only for looks. It must be c_rarg1"); 1218 1219 if (LockingMode == LM_MONITOR) { 1220 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg); 1221 } else { 1222 Label count_locking, done, slow_case; 1223 1224 const Register swap_reg = rax; // Must use rax for cmpxchg instruction 1225 const Register header_reg = c_rarg2; // Will contain the old oopMark 1226 const Register obj_reg = c_rarg3; // Will contain the oop 1227 1228 save_bcp(); // Save in case of exception 1229 1230 if (LockingMode != LM_LIGHTWEIGHT) { 1231 // Convert from BasicObjectLock structure to object and BasicLock 1232 // structure Store the BasicLock address into %rax 1233 lea(swap_reg, Address(lock_reg, BasicObjectLock::lock_offset())); 1234 } 1235 1236 // Load oop into obj_reg(%c_rarg3) 1237 movptr(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset())); 1238 1239 // Free entry 1240 movptr(Address(lock_reg, BasicObjectLock::obj_offset()), NULL_WORD); 1241 1242 if (LockingMode == LM_LIGHTWEIGHT) { 1243 lightweight_unlock(obj_reg, swap_reg, header_reg, slow_case); 1244 } else if (LockingMode == LM_LEGACY) { 1245 // Load the old header from BasicLock structure 1246 movptr(header_reg, Address(swap_reg, 1247 BasicLock::displaced_header_offset_in_bytes())); 1248 1249 // Test for recursion 1250 testptr(header_reg, header_reg); 1251 1252 // zero for recursive case 1253 jcc(Assembler::zero, count_locking); 1254 1255 // Atomic swap back the old header 1256 lock(); 1257 cmpxchgptr(header_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes())); 1258 1259 // zero for simple unlock of a stack-lock case 1260 jcc(Assembler::notZero, slow_case); 1261 1262 bind(count_locking); 1263 dec_held_monitor_count(); 1264 } 1265 jmp(done); 1266 1267 bind(slow_case); 1268 // Call the runtime routine for slow case. 1269 movptr(Address(lock_reg, BasicObjectLock::obj_offset()), obj_reg); // restore obj 1270 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg); 1271 1272 bind(done); 1273 1274 restore_bcp(); 1275 } 1276 } 1277 1278 void InterpreterMacroAssembler::test_method_data_pointer(Register mdp, 1279 Label& zero_continue) { 1280 assert(ProfileInterpreter, "must be profiling interpreter"); 1281 movptr(mdp, Address(rbp, frame::interpreter_frame_mdp_offset * wordSize)); 1282 testptr(mdp, mdp); 1283 jcc(Assembler::zero, zero_continue); 1284 } 1285 1286 1287 // Set the method data pointer for the current bcp. 1288 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() { 1289 assert(ProfileInterpreter, "must be profiling interpreter"); 1290 Label set_mdp; 1291 push(rax); 1292 push(rbx); 1293 1294 get_method(rbx); 1295 // Test MDO to avoid the call if it is null. 1296 movptr(rax, Address(rbx, in_bytes(Method::method_data_offset()))); 1297 testptr(rax, rax); 1298 jcc(Assembler::zero, set_mdp); 1299 // rbx: method 1300 // _bcp_register: bcp 1301 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), rbx, _bcp_register); 1302 // rax: mdi 1303 // mdo is guaranteed to be non-zero here, we checked for it before the call. 1304 movptr(rbx, Address(rbx, in_bytes(Method::method_data_offset()))); 1305 addptr(rbx, in_bytes(MethodData::data_offset())); 1306 addptr(rax, rbx); 1307 bind(set_mdp); 1308 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), rax); 1309 pop(rbx); 1310 pop(rax); 1311 } 1312 1313 void InterpreterMacroAssembler::verify_method_data_pointer() { 1314 assert(ProfileInterpreter, "must be profiling interpreter"); 1315 #ifdef ASSERT 1316 Label verify_continue; 1317 push(rax); 1318 push(rbx); 1319 Register arg3_reg = c_rarg3; 1320 Register arg2_reg = c_rarg2; 1321 push(arg3_reg); 1322 push(arg2_reg); 1323 test_method_data_pointer(arg3_reg, verify_continue); // If mdp is zero, continue 1324 get_method(rbx); 1325 1326 // If the mdp is valid, it will point to a DataLayout header which is 1327 // consistent with the bcp. The converse is highly probable also. 1328 load_unsigned_short(arg2_reg, 1329 Address(arg3_reg, in_bytes(DataLayout::bci_offset()))); 1330 addptr(arg2_reg, Address(rbx, Method::const_offset())); 1331 lea(arg2_reg, Address(arg2_reg, ConstMethod::codes_offset())); 1332 cmpptr(arg2_reg, _bcp_register); 1333 jcc(Assembler::equal, verify_continue); 1334 // rbx: method 1335 // _bcp_register: bcp 1336 // c_rarg3: mdp 1337 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp), 1338 rbx, _bcp_register, arg3_reg); 1339 bind(verify_continue); 1340 pop(arg2_reg); 1341 pop(arg3_reg); 1342 pop(rbx); 1343 pop(rax); 1344 #endif // ASSERT 1345 } 1346 1347 1348 void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in, 1349 int constant, 1350 Register value) { 1351 assert(ProfileInterpreter, "must be profiling interpreter"); 1352 Address data(mdp_in, constant); 1353 movptr(data, value); 1354 } 1355 1356 1357 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in, 1358 int constant, 1359 bool decrement) { 1360 // Counter address 1361 Address data(mdp_in, constant); 1362 1363 increment_mdp_data_at(data, decrement); 1364 } 1365 1366 void InterpreterMacroAssembler::increment_mdp_data_at(Address data, 1367 bool decrement) { 1368 assert(ProfileInterpreter, "must be profiling interpreter"); 1369 // %%% this does 64bit counters at best it is wasting space 1370 // at worst it is a rare bug when counters overflow 1371 1372 if (decrement) { 1373 // Decrement the register. Set condition codes. 1374 addptr(data, -DataLayout::counter_increment); 1375 // If the decrement causes the counter to overflow, stay negative 1376 Label L; 1377 jcc(Assembler::negative, L); 1378 addptr(data, DataLayout::counter_increment); 1379 bind(L); 1380 } else { 1381 assert(DataLayout::counter_increment == 1, 1382 "flow-free idiom only works with 1"); 1383 // Increment the register. Set carry flag. 1384 addptr(data, DataLayout::counter_increment); 1385 // If the increment causes the counter to overflow, pull back by 1. 1386 sbbptr(data, 0); 1387 } 1388 } 1389 1390 1391 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in, 1392 Register reg, 1393 int constant, 1394 bool decrement) { 1395 Address data(mdp_in, reg, Address::times_1, constant); 1396 1397 increment_mdp_data_at(data, decrement); 1398 } 1399 1400 void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in, 1401 int flag_byte_constant) { 1402 assert(ProfileInterpreter, "must be profiling interpreter"); 1403 int header_offset = in_bytes(DataLayout::flags_offset()); 1404 int header_bits = flag_byte_constant; 1405 // Set the flag 1406 orb(Address(mdp_in, header_offset), header_bits); 1407 } 1408 1409 1410 1411 void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in, 1412 int offset, 1413 Register value, 1414 Register test_value_out, 1415 Label& not_equal_continue) { 1416 assert(ProfileInterpreter, "must be profiling interpreter"); 1417 if (test_value_out == noreg) { 1418 cmpptr(value, Address(mdp_in, offset)); 1419 } else { 1420 // Put the test value into a register, so caller can use it: 1421 movptr(test_value_out, Address(mdp_in, offset)); 1422 cmpptr(test_value_out, value); 1423 } 1424 jcc(Assembler::notEqual, not_equal_continue); 1425 } 1426 1427 1428 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in, 1429 int offset_of_disp) { 1430 assert(ProfileInterpreter, "must be profiling interpreter"); 1431 Address disp_address(mdp_in, offset_of_disp); 1432 addptr(mdp_in, disp_address); 1433 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in); 1434 } 1435 1436 1437 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in, 1438 Register reg, 1439 int offset_of_disp) { 1440 assert(ProfileInterpreter, "must be profiling interpreter"); 1441 Address disp_address(mdp_in, reg, Address::times_1, offset_of_disp); 1442 addptr(mdp_in, disp_address); 1443 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in); 1444 } 1445 1446 1447 void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in, 1448 int constant) { 1449 assert(ProfileInterpreter, "must be profiling interpreter"); 1450 addptr(mdp_in, constant); 1451 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in); 1452 } 1453 1454 1455 void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) { 1456 assert(ProfileInterpreter, "must be profiling interpreter"); 1457 push(return_bci); // save/restore across call_VM 1458 call_VM(noreg, 1459 CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret), 1460 return_bci); 1461 pop(return_bci); 1462 } 1463 1464 1465 void InterpreterMacroAssembler::profile_taken_branch(Register mdp, 1466 Register bumped_count) { 1467 if (ProfileInterpreter) { 1468 Label profile_continue; 1469 1470 // If no method data exists, go to profile_continue. 1471 // Otherwise, assign to mdp 1472 test_method_data_pointer(mdp, profile_continue); 1473 1474 // We are taking a branch. Increment the taken count. 1475 // We inline increment_mdp_data_at to return bumped_count in a register 1476 //increment_mdp_data_at(mdp, in_bytes(JumpData::taken_offset())); 1477 Address data(mdp, in_bytes(JumpData::taken_offset())); 1478 movptr(bumped_count, data); 1479 assert(DataLayout::counter_increment == 1, 1480 "flow-free idiom only works with 1"); 1481 addptr(bumped_count, DataLayout::counter_increment); 1482 sbbptr(bumped_count, 0); 1483 movptr(data, bumped_count); // Store back out 1484 1485 // The method data pointer needs to be updated to reflect the new target. 1486 update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset())); 1487 bind(profile_continue); 1488 } 1489 } 1490 1491 1492 void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp, bool acmp) { 1493 if (ProfileInterpreter) { 1494 Label profile_continue; 1495 1496 // If no method data exists, go to profile_continue. 1497 test_method_data_pointer(mdp, profile_continue); 1498 1499 // We are taking a branch. Increment the not taken count. 1500 increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset())); 1501 1502 // The method data pointer needs to be updated to correspond to 1503 // the next bytecode 1504 update_mdp_by_constant(mdp, acmp ? in_bytes(ACmpData::acmp_data_size()): in_bytes(BranchData::branch_data_size())); 1505 bind(profile_continue); 1506 } 1507 } 1508 1509 void InterpreterMacroAssembler::profile_call(Register mdp) { 1510 if (ProfileInterpreter) { 1511 Label profile_continue; 1512 1513 // If no method data exists, go to profile_continue. 1514 test_method_data_pointer(mdp, profile_continue); 1515 1516 // We are making a call. Increment the count. 1517 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); 1518 1519 // The method data pointer needs to be updated to reflect the new target. 1520 update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size())); 1521 bind(profile_continue); 1522 } 1523 } 1524 1525 1526 void InterpreterMacroAssembler::profile_final_call(Register mdp) { 1527 if (ProfileInterpreter) { 1528 Label profile_continue; 1529 1530 // If no method data exists, go to profile_continue. 1531 test_method_data_pointer(mdp, profile_continue); 1532 1533 // We are making a call. Increment the count. 1534 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); 1535 1536 // The method data pointer needs to be updated to reflect the new target. 1537 update_mdp_by_constant(mdp, 1538 in_bytes(VirtualCallData:: 1539 virtual_call_data_size())); 1540 bind(profile_continue); 1541 } 1542 } 1543 1544 1545 void InterpreterMacroAssembler::profile_virtual_call(Register receiver, 1546 Register mdp, 1547 Register reg2, 1548 bool receiver_can_be_null) { 1549 if (ProfileInterpreter) { 1550 Label profile_continue; 1551 1552 // If no method data exists, go to profile_continue. 1553 test_method_data_pointer(mdp, profile_continue); 1554 1555 Label skip_receiver_profile; 1556 if (receiver_can_be_null) { 1557 Label not_null; 1558 testptr(receiver, receiver); 1559 jccb(Assembler::notZero, not_null); 1560 // We are making a call. Increment the count for null receiver. 1561 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); 1562 jmp(skip_receiver_profile); 1563 bind(not_null); 1564 } 1565 1566 // Record the receiver type. 1567 record_klass_in_profile(receiver, mdp, reg2); 1568 bind(skip_receiver_profile); 1569 1570 // The method data pointer needs to be updated to reflect the new target. 1571 update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size())); 1572 bind(profile_continue); 1573 } 1574 } 1575 1576 // This routine creates a state machine for updating the multi-row 1577 // type profile at a virtual call site (or other type-sensitive bytecode). 1578 // The machine visits each row (of receiver/count) until the receiver type 1579 // is found, or until it runs out of rows. At the same time, it remembers 1580 // the location of the first empty row. (An empty row records null for its 1581 // receiver, and can be allocated for a newly-observed receiver type.) 1582 // Because there are two degrees of freedom in the state, a simple linear 1583 // search will not work; it must be a decision tree. Hence this helper 1584 // function is recursive, to generate the required tree structured code. 1585 // It's the interpreter, so we are trading off code space for speed. 1586 // See below for example code. 1587 void InterpreterMacroAssembler::record_klass_in_profile_helper(Register receiver, Register mdp, 1588 Register reg2, int start_row, 1589 Label& done) { 1590 if (TypeProfileWidth == 0) { 1591 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); 1592 } else { 1593 record_item_in_profile_helper(receiver, mdp, reg2, 0, done, TypeProfileWidth, 1594 &VirtualCallData::receiver_offset, &VirtualCallData::receiver_count_offset); 1595 } 1596 } 1597 1598 void InterpreterMacroAssembler::record_item_in_profile_helper(Register item, Register mdp, Register reg2, int start_row, 1599 Label& done, int total_rows, 1600 OffsetFunction item_offset_fn, 1601 OffsetFunction item_count_offset_fn) { 1602 int last_row = total_rows - 1; 1603 assert(start_row <= last_row, "must be work left to do"); 1604 // Test this row for both the item and for null. 1605 // Take any of three different outcomes: 1606 // 1. found item => increment count and goto done 1607 // 2. found null => keep looking for case 1, maybe allocate this cell 1608 // 3. found something else => keep looking for cases 1 and 2 1609 // Case 3 is handled by a recursive call. 1610 for (int row = start_row; row <= last_row; row++) { 1611 Label next_test; 1612 bool test_for_null_also = (row == start_row); 1613 1614 // See if the item is item[n]. 1615 int item_offset = in_bytes(item_offset_fn(row)); 1616 test_mdp_data_at(mdp, item_offset, item, 1617 (test_for_null_also ? reg2 : noreg), 1618 next_test); 1619 // (Reg2 now contains the item from the CallData.) 1620 1621 // The item is item[n]. Increment count[n]. 1622 int count_offset = in_bytes(item_count_offset_fn(row)); 1623 increment_mdp_data_at(mdp, count_offset); 1624 jmp(done); 1625 bind(next_test); 1626 1627 if (test_for_null_also) { 1628 // Failed the equality check on item[n]... Test for null. 1629 testptr(reg2, reg2); 1630 if (start_row == last_row) { 1631 // The only thing left to do is handle the null case. 1632 Label found_null; 1633 jccb(Assembler::zero, found_null); 1634 // Item did not match any saved item and there is no empty row for it. 1635 // Increment total counter to indicate polymorphic case. 1636 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); 1637 jmp(done); 1638 bind(found_null); 1639 break; 1640 } 1641 Label found_null; 1642 // Since null is rare, make it be the branch-taken case. 1643 jcc(Assembler::zero, found_null); 1644 1645 // Put all the "Case 3" tests here. 1646 record_item_in_profile_helper(item, mdp, reg2, start_row + 1, done, total_rows, 1647 item_offset_fn, item_count_offset_fn); 1648 1649 // Found a null. Keep searching for a matching item, 1650 // but remember that this is an empty (unused) slot. 1651 bind(found_null); 1652 } 1653 } 1654 1655 // In the fall-through case, we found no matching item, but we 1656 // observed the item[start_row] is null. 1657 1658 // Fill in the item field and increment the count. 1659 int item_offset = in_bytes(item_offset_fn(start_row)); 1660 set_mdp_data_at(mdp, item_offset, item); 1661 int count_offset = in_bytes(item_count_offset_fn(start_row)); 1662 movl(reg2, DataLayout::counter_increment); 1663 set_mdp_data_at(mdp, count_offset, reg2); 1664 if (start_row > 0) { 1665 jmp(done); 1666 } 1667 } 1668 1669 // Example state machine code for three profile rows: 1670 // // main copy of decision tree, rooted at row[1] 1671 // if (row[0].rec == rec) { row[0].incr(); goto done; } 1672 // if (row[0].rec != nullptr) { 1673 // // inner copy of decision tree, rooted at row[1] 1674 // if (row[1].rec == rec) { row[1].incr(); goto done; } 1675 // if (row[1].rec != nullptr) { 1676 // // degenerate decision tree, rooted at row[2] 1677 // if (row[2].rec == rec) { row[2].incr(); goto done; } 1678 // if (row[2].rec != nullptr) { count.incr(); goto done; } // overflow 1679 // row[2].init(rec); goto done; 1680 // } else { 1681 // // remember row[1] is empty 1682 // if (row[2].rec == rec) { row[2].incr(); goto done; } 1683 // row[1].init(rec); goto done; 1684 // } 1685 // } else { 1686 // // remember row[0] is empty 1687 // if (row[1].rec == rec) { row[1].incr(); goto done; } 1688 // if (row[2].rec == rec) { row[2].incr(); goto done; } 1689 // row[0].init(rec); goto done; 1690 // } 1691 // done: 1692 1693 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver, Register mdp, Register reg2) { 1694 assert(ProfileInterpreter, "must be profiling"); 1695 Label done; 1696 1697 record_klass_in_profile_helper(receiver, mdp, reg2, 0, done); 1698 1699 bind (done); 1700 } 1701 1702 void InterpreterMacroAssembler::profile_ret(Register return_bci, 1703 Register mdp) { 1704 if (ProfileInterpreter) { 1705 Label profile_continue; 1706 uint row; 1707 1708 // If no method data exists, go to profile_continue. 1709 test_method_data_pointer(mdp, profile_continue); 1710 1711 // Update the total ret count. 1712 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); 1713 1714 for (row = 0; row < RetData::row_limit(); row++) { 1715 Label next_test; 1716 1717 // See if return_bci is equal to bci[n]: 1718 test_mdp_data_at(mdp, 1719 in_bytes(RetData::bci_offset(row)), 1720 return_bci, noreg, 1721 next_test); 1722 1723 // return_bci is equal to bci[n]. Increment the count. 1724 increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row))); 1725 1726 // The method data pointer needs to be updated to reflect the new target. 1727 update_mdp_by_offset(mdp, 1728 in_bytes(RetData::bci_displacement_offset(row))); 1729 jmp(profile_continue); 1730 bind(next_test); 1731 } 1732 1733 update_mdp_for_ret(return_bci); 1734 1735 bind(profile_continue); 1736 } 1737 } 1738 1739 1740 void InterpreterMacroAssembler::profile_null_seen(Register mdp) { 1741 if (ProfileInterpreter) { 1742 Label profile_continue; 1743 1744 // If no method data exists, go to profile_continue. 1745 test_method_data_pointer(mdp, profile_continue); 1746 1747 set_mdp_flag_at(mdp, BitData::null_seen_byte_constant()); 1748 1749 // The method data pointer needs to be updated. 1750 int mdp_delta = in_bytes(BitData::bit_data_size()); 1751 if (TypeProfileCasts) { 1752 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size()); 1753 } 1754 update_mdp_by_constant(mdp, mdp_delta); 1755 1756 bind(profile_continue); 1757 } 1758 } 1759 1760 1761 void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) { 1762 if (ProfileInterpreter) { 1763 Label profile_continue; 1764 1765 // If no method data exists, go to profile_continue. 1766 test_method_data_pointer(mdp, profile_continue); 1767 1768 // The method data pointer needs to be updated. 1769 int mdp_delta = in_bytes(BitData::bit_data_size()); 1770 if (TypeProfileCasts) { 1771 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size()); 1772 1773 // Record the object type. 1774 record_klass_in_profile(klass, mdp, reg2); 1775 } 1776 update_mdp_by_constant(mdp, mdp_delta); 1777 1778 bind(profile_continue); 1779 } 1780 } 1781 1782 1783 void InterpreterMacroAssembler::profile_switch_default(Register mdp) { 1784 if (ProfileInterpreter) { 1785 Label profile_continue; 1786 1787 // If no method data exists, go to profile_continue. 1788 test_method_data_pointer(mdp, profile_continue); 1789 1790 // Update the default case count 1791 increment_mdp_data_at(mdp, 1792 in_bytes(MultiBranchData::default_count_offset())); 1793 1794 // The method data pointer needs to be updated. 1795 update_mdp_by_offset(mdp, 1796 in_bytes(MultiBranchData:: 1797 default_displacement_offset())); 1798 1799 bind(profile_continue); 1800 } 1801 } 1802 1803 1804 void InterpreterMacroAssembler::profile_switch_case(Register index, 1805 Register mdp, 1806 Register reg2) { 1807 if (ProfileInterpreter) { 1808 Label profile_continue; 1809 1810 // If no method data exists, go to profile_continue. 1811 test_method_data_pointer(mdp, profile_continue); 1812 1813 // Build the base (index * per_case_size_in_bytes()) + 1814 // case_array_offset_in_bytes() 1815 movl(reg2, in_bytes(MultiBranchData::per_case_size())); 1816 imulptr(index, reg2); // XXX l ? 1817 addptr(index, in_bytes(MultiBranchData::case_array_offset())); // XXX l ? 1818 1819 // Update the case count 1820 increment_mdp_data_at(mdp, 1821 index, 1822 in_bytes(MultiBranchData::relative_count_offset())); 1823 1824 // The method data pointer needs to be updated. 1825 update_mdp_by_offset(mdp, 1826 index, 1827 in_bytes(MultiBranchData:: 1828 relative_displacement_offset())); 1829 1830 bind(profile_continue); 1831 } 1832 } 1833 1834 template <class ArrayData> void InterpreterMacroAssembler::profile_array_type(Register mdp, 1835 Register array, 1836 Register tmp) { 1837 if (ProfileInterpreter) { 1838 Label profile_continue; 1839 1840 // If no method data exists, go to profile_continue. 1841 test_method_data_pointer(mdp, profile_continue); 1842 1843 mov(tmp, array); 1844 profile_obj_type(tmp, Address(mdp, in_bytes(ArrayData::array_offset()))); 1845 1846 Label not_flat; 1847 test_non_flat_array_oop(array, tmp, not_flat); 1848 1849 set_mdp_flag_at(mdp, ArrayData::flat_array_byte_constant()); 1850 1851 bind(not_flat); 1852 1853 Label not_null_free; 1854 test_non_null_free_array_oop(array, tmp, not_null_free); 1855 1856 set_mdp_flag_at(mdp, ArrayData::null_free_array_byte_constant()); 1857 1858 bind(not_null_free); 1859 1860 bind(profile_continue); 1861 } 1862 } 1863 1864 template void InterpreterMacroAssembler::profile_array_type<ArrayLoadData>(Register mdp, 1865 Register array, 1866 Register tmp); 1867 template void InterpreterMacroAssembler::profile_array_type<ArrayStoreData>(Register mdp, 1868 Register array, 1869 Register tmp); 1870 1871 1872 void InterpreterMacroAssembler::profile_multiple_element_types(Register mdp, Register element, Register tmp, const Register tmp2) { 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 Label done, update; 1880 testptr(element, element); 1881 jccb(Assembler::notZero, update); 1882 set_mdp_flag_at(mdp, BitData::null_seen_byte_constant()); 1883 jmp(done); 1884 1885 bind(update); 1886 load_klass(tmp, element, rscratch1); 1887 1888 // Record the object type. 1889 record_klass_in_profile(tmp, mdp, tmp2); 1890 1891 bind(done); 1892 1893 // The method data pointer needs to be updated. 1894 update_mdp_by_constant(mdp, in_bytes(ArrayStoreData::array_store_data_size())); 1895 1896 bind(profile_continue); 1897 } 1898 } 1899 1900 void InterpreterMacroAssembler::profile_element_type(Register mdp, 1901 Register element, 1902 Register tmp) { 1903 if (ProfileInterpreter) { 1904 Label profile_continue; 1905 1906 // If no method data exists, go to profile_continue. 1907 test_method_data_pointer(mdp, profile_continue); 1908 1909 mov(tmp, element); 1910 profile_obj_type(tmp, Address(mdp, in_bytes(ArrayLoadData::element_offset()))); 1911 1912 // The method data pointer needs to be updated. 1913 update_mdp_by_constant(mdp, in_bytes(ArrayLoadData::array_load_data_size())); 1914 1915 bind(profile_continue); 1916 } 1917 } 1918 1919 void InterpreterMacroAssembler::profile_acmp(Register mdp, 1920 Register left, 1921 Register right, 1922 Register tmp) { 1923 if (ProfileInterpreter) { 1924 Label profile_continue; 1925 1926 // If no method data exists, go to profile_continue. 1927 test_method_data_pointer(mdp, profile_continue); 1928 1929 mov(tmp, left); 1930 profile_obj_type(tmp, Address(mdp, in_bytes(ACmpData::left_offset()))); 1931 1932 Label left_not_inline_type; 1933 test_oop_is_not_inline_type(left, tmp, left_not_inline_type); 1934 set_mdp_flag_at(mdp, ACmpData::left_inline_type_byte_constant()); 1935 bind(left_not_inline_type); 1936 1937 mov(tmp, right); 1938 profile_obj_type(tmp, Address(mdp, in_bytes(ACmpData::right_offset()))); 1939 1940 Label right_not_inline_type; 1941 test_oop_is_not_inline_type(right, tmp, right_not_inline_type); 1942 set_mdp_flag_at(mdp, ACmpData::right_inline_type_byte_constant()); 1943 bind(right_not_inline_type); 1944 1945 bind(profile_continue); 1946 } 1947 } 1948 1949 1950 void InterpreterMacroAssembler::_interp_verify_oop(Register reg, TosState state, const char* file, int line) { 1951 if (state == atos) { 1952 MacroAssembler::_verify_oop_checked(reg, "broken oop", file, line); 1953 } 1954 } 1955 1956 1957 // Jump if ((*counter_addr += increment) & mask) == 0 1958 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr, Address mask, 1959 Register scratch, Label* where) { 1960 // This update is actually not atomic and can lose a number of updates 1961 // under heavy contention, but the alternative of using the (contended) 1962 // atomic update here penalizes profiling paths too much. 1963 movl(scratch, counter_addr); 1964 incrementl(scratch, InvocationCounter::count_increment); 1965 movl(counter_addr, scratch); 1966 andl(scratch, mask); 1967 if (where != nullptr) { 1968 jcc(Assembler::zero, *where); 1969 } 1970 } 1971 1972 void InterpreterMacroAssembler::notify_method_entry() { 1973 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to 1974 // track stack depth. If it is possible to enter interp_only_mode we add 1975 // the code to check if the event should be sent. 1976 Register rthread = r15_thread; 1977 Register rarg = c_rarg1; 1978 if (JvmtiExport::can_post_interpreter_events()) { 1979 Label L; 1980 movl(rdx, Address(rthread, JavaThread::interp_only_mode_offset())); 1981 testl(rdx, rdx); 1982 jcc(Assembler::zero, L); 1983 call_VM(noreg, CAST_FROM_FN_PTR(address, 1984 InterpreterRuntime::post_method_entry)); 1985 bind(L); 1986 } 1987 1988 if (DTraceMethodProbes) { 1989 get_method(rarg); 1990 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry), 1991 rthread, rarg); 1992 } 1993 1994 // RedefineClasses() tracing support for obsolete method entry 1995 if (log_is_enabled(Trace, redefine, class, obsolete)) { 1996 get_method(rarg); 1997 call_VM_leaf( 1998 CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry), 1999 rthread, rarg); 2000 } 2001 } 2002 2003 2004 void InterpreterMacroAssembler::notify_method_exit( 2005 TosState state, NotifyMethodExitMode mode) { 2006 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to 2007 // track stack depth. If it is possible to enter interp_only_mode we add 2008 // the code to check if the event should be sent. 2009 Register rthread = r15_thread; 2010 Register rarg = c_rarg1; 2011 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) { 2012 Label L; 2013 // Note: frame::interpreter_frame_result has a dependency on how the 2014 // method result is saved across the call to post_method_exit. If this 2015 // is changed then the interpreter_frame_result implementation will 2016 // need to be updated too. 2017 2018 // template interpreter will leave the result on the top of the stack. 2019 push(state); 2020 movl(rdx, Address(rthread, JavaThread::interp_only_mode_offset())); 2021 testl(rdx, rdx); 2022 jcc(Assembler::zero, L); 2023 call_VM(noreg, 2024 CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit)); 2025 bind(L); 2026 pop(state); 2027 } 2028 2029 if (DTraceMethodProbes) { 2030 push(state); 2031 get_method(rarg); 2032 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit), 2033 rthread, rarg); 2034 pop(state); 2035 } 2036 } 2037 2038 void InterpreterMacroAssembler::load_resolved_indy_entry(Register cache, Register index) { 2039 // Get index out of bytecode pointer 2040 get_cache_index_at_bcp(index, 1, sizeof(u4)); 2041 // Get address of invokedynamic array 2042 movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize)); 2043 movptr(cache, Address(cache, in_bytes(ConstantPoolCache::invokedynamic_entries_offset()))); 2044 if (is_power_of_2(sizeof(ResolvedIndyEntry))) { 2045 shll(index, log2i_exact(sizeof(ResolvedIndyEntry))); // Scale index by power of 2 2046 } else { 2047 imull(index, index, sizeof(ResolvedIndyEntry)); // Scale the index to be the entry index * sizeof(ResolvedIndyEntry) 2048 } 2049 lea(cache, Address(cache, index, Address::times_1, Array<ResolvedIndyEntry>::base_offset_in_bytes())); 2050 } 2051 2052 void InterpreterMacroAssembler::load_field_entry(Register cache, Register index, int bcp_offset) { 2053 // Get index out of bytecode pointer 2054 movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize)); 2055 get_cache_index_at_bcp(index, bcp_offset, sizeof(u2)); 2056 2057 movptr(cache, Address(cache, ConstantPoolCache::field_entries_offset())); 2058 // Take shortcut if the size is a power of 2 2059 if (is_power_of_2(sizeof(ResolvedFieldEntry))) { 2060 shll(index, log2i_exact(sizeof(ResolvedFieldEntry))); // Scale index by power of 2 2061 } else { 2062 imull(index, index, sizeof(ResolvedFieldEntry)); // Scale the index to be the entry index * sizeof(ResolvedFieldEntry) 2063 } 2064 lea(cache, Address(cache, index, Address::times_1, Array<ResolvedFieldEntry>::base_offset_in_bytes())); 2065 } 2066 2067 void InterpreterMacroAssembler::load_method_entry(Register cache, Register index, int bcp_offset) { 2068 // Get index out of bytecode pointer 2069 movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize)); 2070 get_cache_index_at_bcp(index, bcp_offset, sizeof(u2)); 2071 2072 movptr(cache, Address(cache, ConstantPoolCache::method_entries_offset())); 2073 imull(index, index, sizeof(ResolvedMethodEntry)); // Scale the index to be the entry index * sizeof(ResolvedMethodEntry) 2074 lea(cache, Address(cache, index, Address::times_1, Array<ResolvedMethodEntry>::base_offset_in_bytes())); 2075 }