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 push_cont_fastpath(); 1158 call_VM(noreg, 1159 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), 1160 lock_reg); 1161 pop_cont_fastpath(); 1162 } else { 1163 Label count_locking, done, slow_case; 1164 1165 const Register swap_reg = rax; // Must use rax for cmpxchg instruction 1166 const Register tmp_reg = rbx; 1167 const Register obj_reg = LP64_ONLY(c_rarg3) NOT_LP64(rcx); // Will contain the oop 1168 const Register rklass_decode_tmp = rscratch1; 1169 1170 const int obj_offset = in_bytes(BasicObjectLock::obj_offset()); 1171 const int lock_offset = in_bytes(BasicObjectLock::lock_offset()); 1172 const int mark_offset = lock_offset + 1173 BasicLock::displaced_header_offset_in_bytes(); 1174 1175 // Load object pointer into obj_reg 1176 movptr(obj_reg, Address(lock_reg, obj_offset)); 1177 1178 if (DiagnoseSyncOnValueBasedClasses != 0) { 1179 load_klass(tmp_reg, obj_reg, rklass_decode_tmp); 1180 movl(tmp_reg, Address(tmp_reg, Klass::access_flags_offset())); 1181 testl(tmp_reg, JVM_ACC_IS_VALUE_BASED_CLASS); 1182 jcc(Assembler::notZero, slow_case); 1183 } 1184 1185 if (LockingMode == LM_LIGHTWEIGHT) { 1186 #ifdef _LP64 1187 const Register thread = r15_thread; 1188 #else 1189 const Register thread = lock_reg; 1190 get_thread(thread); 1191 #endif 1192 lightweight_lock(obj_reg, swap_reg, thread, tmp_reg, slow_case); 1193 } else if (LockingMode == LM_LEGACY) { 1194 // Load immediate 1 into swap_reg %rax 1195 movl(swap_reg, 1); 1196 1197 // Load (object->mark() | 1) into swap_reg %rax 1198 orptr(swap_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes())); 1199 1200 // Save (object->mark() | 1) into BasicLock's displaced header 1201 movptr(Address(lock_reg, mark_offset), swap_reg); 1202 1203 assert(lock_offset == 0, 1204 "displaced header must be first word in BasicObjectLock"); 1205 1206 lock(); 1207 cmpxchgptr(lock_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes())); 1208 jcc(Assembler::zero, count_locking); 1209 1210 const int zero_bits = LP64_ONLY(7) NOT_LP64(3); 1211 1212 // Fast check for recursive lock. 1213 // 1214 // Can apply the optimization only if this is a stack lock 1215 // allocated in this thread. For efficiency, we can focus on 1216 // recently allocated stack locks (instead of reading the stack 1217 // base and checking whether 'mark' points inside the current 1218 // thread stack): 1219 // 1) (mark & zero_bits) == 0, and 1220 // 2) rsp <= mark < mark + os::pagesize() 1221 // 1222 // Warning: rsp + os::pagesize can overflow the stack base. We must 1223 // neither apply the optimization for an inflated lock allocated 1224 // just above the thread stack (this is why condition 1 matters) 1225 // nor apply the optimization if the stack lock is inside the stack 1226 // of another thread. The latter is avoided even in case of overflow 1227 // because we have guard pages at the end of all stacks. Hence, if 1228 // we go over the stack base and hit the stack of another thread, 1229 // this should not be in a writeable area that could contain a 1230 // stack lock allocated by that thread. As a consequence, a stack 1231 // lock less than page size away from rsp is guaranteed to be 1232 // owned by the current thread. 1233 // 1234 // These 3 tests can be done by evaluating the following 1235 // expression: ((mark - rsp) & (zero_bits - os::vm_page_size())), 1236 // assuming both stack pointer and pagesize have their 1237 // least significant bits clear. 1238 // NOTE: the mark is in swap_reg %rax as the result of cmpxchg 1239 subptr(swap_reg, rsp); 1240 andptr(swap_reg, zero_bits - (int)os::vm_page_size()); 1241 1242 // Save the test result, for recursive case, the result is zero 1243 movptr(Address(lock_reg, mark_offset), swap_reg); 1244 jcc(Assembler::notZero, slow_case); 1245 } 1246 jmp(done); 1247 1248 bind(count_locking); 1249 inc_held_monitor_count(); 1250 jmp(done); 1251 1252 bind(slow_case); 1253 1254 push_cont_fastpath(); 1255 // Call the runtime routine for slow case 1256 if (LockingMode == LM_LIGHTWEIGHT) { 1257 call_VM(noreg, 1258 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter_obj), 1259 obj_reg); 1260 } else { 1261 call_VM(noreg, 1262 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), 1263 lock_reg); 1264 } 1265 pop_cont_fastpath(); 1266 bind(done); 1267 } 1268 } 1269 1270 1271 // Unlocks an object. Used in monitorexit bytecode and 1272 // remove_activation. Throws an IllegalMonitorException if object is 1273 // not locked by current thread. 1274 // 1275 // Args: 1276 // rdx, c_rarg1: BasicObjectLock for lock 1277 // 1278 // Kills: 1279 // rax 1280 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ... (param regs) 1281 // rscratch1 (scratch reg) 1282 // rax, rbx, rcx, rdx 1283 void InterpreterMacroAssembler::unlock_object(Register lock_reg) { 1284 assert(lock_reg == LP64_ONLY(c_rarg1) NOT_LP64(rdx), 1285 "The argument is only for looks. It must be c_rarg1"); 1286 1287 if (LockingMode == LM_MONITOR) { 1288 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg); 1289 } else { 1290 Label done, slow_case; 1291 1292 const Register swap_reg = rax; // Must use rax for cmpxchg instruction 1293 const Register header_reg = LP64_ONLY(c_rarg2) NOT_LP64(rbx); // Will contain the old oopMark 1294 const Register obj_reg = LP64_ONLY(c_rarg3) NOT_LP64(rcx); // Will contain the oop 1295 1296 save_bcp(); // Save in case of exception 1297 1298 if (LockingMode != LM_LIGHTWEIGHT) { 1299 // Convert from BasicObjectLock structure to object and BasicLock 1300 // structure Store the BasicLock address into %rax 1301 lea(swap_reg, Address(lock_reg, BasicObjectLock::lock_offset())); 1302 } 1303 1304 // Load oop into obj_reg(%c_rarg3) 1305 movptr(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset())); 1306 1307 // Free entry 1308 movptr(Address(lock_reg, BasicObjectLock::obj_offset()), NULL_WORD); 1309 1310 if (LockingMode == LM_LIGHTWEIGHT) { 1311 #ifdef _LP64 1312 lightweight_unlock(obj_reg, swap_reg, r15_thread, header_reg, slow_case); 1313 #else 1314 // This relies on the implementation of lightweight_unlock being able to handle 1315 // that the reg_rax and thread Register parameters may alias each other. 1316 get_thread(swap_reg); 1317 lightweight_unlock(obj_reg, swap_reg, swap_reg, header_reg, slow_case); 1318 #endif 1319 } else if (LockingMode == LM_LEGACY) { 1320 // Load the old header from BasicLock structure 1321 movptr(header_reg, Address(swap_reg, 1322 BasicLock::displaced_header_offset_in_bytes())); 1323 1324 // Test for recursion 1325 testptr(header_reg, header_reg); 1326 1327 // zero for recursive case 1328 jcc(Assembler::zero, done); 1329 1330 // Atomic swap back the old header 1331 lock(); 1332 cmpxchgptr(header_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes())); 1333 1334 // zero for simple unlock of a stack-lock case 1335 jcc(Assembler::notZero, slow_case); 1336 dec_held_monitor_count(); 1337 } 1338 jmp(done); 1339 1340 bind(slow_case); 1341 // Call the runtime routine for slow case. 1342 movptr(Address(lock_reg, BasicObjectLock::obj_offset()), obj_reg); // restore obj 1343 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg); 1344 1345 bind(done); 1346 1347 restore_bcp(); 1348 } 1349 } 1350 1351 void InterpreterMacroAssembler::test_method_data_pointer(Register mdp, 1352 Label& zero_continue) { 1353 assert(ProfileInterpreter, "must be profiling interpreter"); 1354 movptr(mdp, Address(rbp, frame::interpreter_frame_mdp_offset * wordSize)); 1355 testptr(mdp, mdp); 1356 jcc(Assembler::zero, zero_continue); 1357 } 1358 1359 1360 // Set the method data pointer for the current bcp. 1361 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() { 1362 assert(ProfileInterpreter, "must be profiling interpreter"); 1363 Label set_mdp; 1364 push(rax); 1365 push(rbx); 1366 1367 get_method(rbx); 1368 // Test MDO to avoid the call if it is null. 1369 movptr(rax, Address(rbx, in_bytes(Method::method_data_offset()))); 1370 testptr(rax, rax); 1371 jcc(Assembler::zero, set_mdp); 1372 // rbx: method 1373 // _bcp_register: bcp 1374 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), rbx, _bcp_register); 1375 // rax: mdi 1376 // mdo is guaranteed to be non-zero here, we checked for it before the call. 1377 movptr(rbx, Address(rbx, in_bytes(Method::method_data_offset()))); 1378 addptr(rbx, in_bytes(MethodData::data_offset())); 1379 addptr(rax, rbx); 1380 bind(set_mdp); 1381 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), rax); 1382 pop(rbx); 1383 pop(rax); 1384 } 1385 1386 void InterpreterMacroAssembler::verify_method_data_pointer() { 1387 assert(ProfileInterpreter, "must be profiling interpreter"); 1388 #ifdef ASSERT 1389 Label verify_continue; 1390 push(rax); 1391 push(rbx); 1392 Register arg3_reg = LP64_ONLY(c_rarg3) NOT_LP64(rcx); 1393 Register arg2_reg = LP64_ONLY(c_rarg2) NOT_LP64(rdx); 1394 push(arg3_reg); 1395 push(arg2_reg); 1396 test_method_data_pointer(arg3_reg, verify_continue); // If mdp is zero, continue 1397 get_method(rbx); 1398 1399 // If the mdp is valid, it will point to a DataLayout header which is 1400 // consistent with the bcp. The converse is highly probable also. 1401 load_unsigned_short(arg2_reg, 1402 Address(arg3_reg, in_bytes(DataLayout::bci_offset()))); 1403 addptr(arg2_reg, Address(rbx, Method::const_offset())); 1404 lea(arg2_reg, Address(arg2_reg, ConstMethod::codes_offset())); 1405 cmpptr(arg2_reg, _bcp_register); 1406 jcc(Assembler::equal, verify_continue); 1407 // rbx: method 1408 // _bcp_register: bcp 1409 // c_rarg3: mdp 1410 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp), 1411 rbx, _bcp_register, arg3_reg); 1412 bind(verify_continue); 1413 pop(arg2_reg); 1414 pop(arg3_reg); 1415 pop(rbx); 1416 pop(rax); 1417 #endif // ASSERT 1418 } 1419 1420 1421 void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in, 1422 int constant, 1423 Register value) { 1424 assert(ProfileInterpreter, "must be profiling interpreter"); 1425 Address data(mdp_in, constant); 1426 movptr(data, value); 1427 } 1428 1429 1430 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in, 1431 int constant, 1432 bool decrement) { 1433 // Counter address 1434 Address data(mdp_in, constant); 1435 1436 increment_mdp_data_at(data, decrement); 1437 } 1438 1439 void InterpreterMacroAssembler::increment_mdp_data_at(Address data, 1440 bool decrement) { 1441 assert(ProfileInterpreter, "must be profiling interpreter"); 1442 // %%% this does 64bit counters at best it is wasting space 1443 // at worst it is a rare bug when counters overflow 1444 1445 if (decrement) { 1446 // Decrement the register. Set condition codes. 1447 addptr(data, -DataLayout::counter_increment); 1448 // If the decrement causes the counter to overflow, stay negative 1449 Label L; 1450 jcc(Assembler::negative, L); 1451 addptr(data, DataLayout::counter_increment); 1452 bind(L); 1453 } else { 1454 assert(DataLayout::counter_increment == 1, 1455 "flow-free idiom only works with 1"); 1456 // Increment the register. Set carry flag. 1457 addptr(data, DataLayout::counter_increment); 1458 // If the increment causes the counter to overflow, pull back by 1. 1459 sbbptr(data, 0); 1460 } 1461 } 1462 1463 1464 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in, 1465 Register reg, 1466 int constant, 1467 bool decrement) { 1468 Address data(mdp_in, reg, Address::times_1, constant); 1469 1470 increment_mdp_data_at(data, decrement); 1471 } 1472 1473 void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in, 1474 int flag_byte_constant) { 1475 assert(ProfileInterpreter, "must be profiling interpreter"); 1476 int header_offset = in_bytes(DataLayout::flags_offset()); 1477 int header_bits = flag_byte_constant; 1478 // Set the flag 1479 orb(Address(mdp_in, header_offset), header_bits); 1480 } 1481 1482 1483 1484 void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in, 1485 int offset, 1486 Register value, 1487 Register test_value_out, 1488 Label& not_equal_continue) { 1489 assert(ProfileInterpreter, "must be profiling interpreter"); 1490 if (test_value_out == noreg) { 1491 cmpptr(value, Address(mdp_in, offset)); 1492 } else { 1493 // Put the test value into a register, so caller can use it: 1494 movptr(test_value_out, Address(mdp_in, offset)); 1495 cmpptr(test_value_out, value); 1496 } 1497 jcc(Assembler::notEqual, not_equal_continue); 1498 } 1499 1500 1501 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in, 1502 int offset_of_disp) { 1503 assert(ProfileInterpreter, "must be profiling interpreter"); 1504 Address disp_address(mdp_in, offset_of_disp); 1505 addptr(mdp_in, disp_address); 1506 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in); 1507 } 1508 1509 1510 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in, 1511 Register reg, 1512 int offset_of_disp) { 1513 assert(ProfileInterpreter, "must be profiling interpreter"); 1514 Address disp_address(mdp_in, reg, Address::times_1, offset_of_disp); 1515 addptr(mdp_in, disp_address); 1516 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in); 1517 } 1518 1519 1520 void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in, 1521 int constant) { 1522 assert(ProfileInterpreter, "must be profiling interpreter"); 1523 addptr(mdp_in, constant); 1524 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in); 1525 } 1526 1527 1528 void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) { 1529 assert(ProfileInterpreter, "must be profiling interpreter"); 1530 push(return_bci); // save/restore across call_VM 1531 call_VM(noreg, 1532 CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret), 1533 return_bci); 1534 pop(return_bci); 1535 } 1536 1537 1538 void InterpreterMacroAssembler::profile_taken_branch(Register mdp, 1539 Register bumped_count) { 1540 if (ProfileInterpreter) { 1541 Label profile_continue; 1542 1543 // If no method data exists, go to profile_continue. 1544 // Otherwise, assign to mdp 1545 test_method_data_pointer(mdp, profile_continue); 1546 1547 // We are taking a branch. Increment the taken count. 1548 // We inline increment_mdp_data_at to return bumped_count in a register 1549 //increment_mdp_data_at(mdp, in_bytes(JumpData::taken_offset())); 1550 Address data(mdp, in_bytes(JumpData::taken_offset())); 1551 movptr(bumped_count, data); 1552 assert(DataLayout::counter_increment == 1, 1553 "flow-free idiom only works with 1"); 1554 addptr(bumped_count, DataLayout::counter_increment); 1555 sbbptr(bumped_count, 0); 1556 movptr(data, bumped_count); // Store back out 1557 1558 // The method data pointer needs to be updated to reflect the new target. 1559 update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset())); 1560 bind(profile_continue); 1561 } 1562 } 1563 1564 1565 void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp) { 1566 if (ProfileInterpreter) { 1567 Label profile_continue; 1568 1569 // If no method data exists, go to profile_continue. 1570 test_method_data_pointer(mdp, profile_continue); 1571 1572 // We are taking a branch. Increment the not taken count. 1573 increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset())); 1574 1575 // The method data pointer needs to be updated to correspond to 1576 // the next bytecode 1577 update_mdp_by_constant(mdp, in_bytes(BranchData::branch_data_size())); 1578 bind(profile_continue); 1579 } 1580 } 1581 1582 void InterpreterMacroAssembler::profile_call(Register mdp) { 1583 if (ProfileInterpreter) { 1584 Label profile_continue; 1585 1586 // If no method data exists, go to profile_continue. 1587 test_method_data_pointer(mdp, profile_continue); 1588 1589 // We are making a call. Increment the count. 1590 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); 1591 1592 // The method data pointer needs to be updated to reflect the new target. 1593 update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size())); 1594 bind(profile_continue); 1595 } 1596 } 1597 1598 1599 void InterpreterMacroAssembler::profile_final_call(Register mdp) { 1600 if (ProfileInterpreter) { 1601 Label profile_continue; 1602 1603 // If no method data exists, go to profile_continue. 1604 test_method_data_pointer(mdp, profile_continue); 1605 1606 // We are making a call. Increment the count. 1607 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); 1608 1609 // The method data pointer needs to be updated to reflect the new target. 1610 update_mdp_by_constant(mdp, 1611 in_bytes(VirtualCallData:: 1612 virtual_call_data_size())); 1613 bind(profile_continue); 1614 } 1615 } 1616 1617 1618 void InterpreterMacroAssembler::profile_virtual_call(Register receiver, 1619 Register mdp, 1620 Register reg2, 1621 bool receiver_can_be_null) { 1622 if (ProfileInterpreter) { 1623 Label profile_continue; 1624 1625 // If no method data exists, go to profile_continue. 1626 test_method_data_pointer(mdp, profile_continue); 1627 1628 Label skip_receiver_profile; 1629 if (receiver_can_be_null) { 1630 Label not_null; 1631 testptr(receiver, receiver); 1632 jccb(Assembler::notZero, not_null); 1633 // We are making a call. Increment the count for null receiver. 1634 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); 1635 jmp(skip_receiver_profile); 1636 bind(not_null); 1637 } 1638 1639 // Record the receiver type. 1640 record_klass_in_profile(receiver, mdp, reg2, true); 1641 bind(skip_receiver_profile); 1642 1643 // The method data pointer needs to be updated to reflect the new target. 1644 update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size())); 1645 bind(profile_continue); 1646 } 1647 } 1648 1649 // This routine creates a state machine for updating the multi-row 1650 // type profile at a virtual call site (or other type-sensitive bytecode). 1651 // The machine visits each row (of receiver/count) until the receiver type 1652 // is found, or until it runs out of rows. At the same time, it remembers 1653 // the location of the first empty row. (An empty row records null for its 1654 // receiver, and can be allocated for a newly-observed receiver type.) 1655 // Because there are two degrees of freedom in the state, a simple linear 1656 // search will not work; it must be a decision tree. Hence this helper 1657 // function is recursive, to generate the required tree structured code. 1658 // It's the interpreter, so we are trading off code space for speed. 1659 // See below for example code. 1660 void InterpreterMacroAssembler::record_klass_in_profile_helper( 1661 Register receiver, Register mdp, 1662 Register reg2, int start_row, 1663 Label& done, bool is_virtual_call) { 1664 if (TypeProfileWidth == 0) { 1665 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); 1666 } else { 1667 record_item_in_profile_helper(receiver, mdp, reg2, 0, done, TypeProfileWidth, 1668 &VirtualCallData::receiver_offset, &VirtualCallData::receiver_count_offset); 1669 } 1670 } 1671 1672 void InterpreterMacroAssembler::record_item_in_profile_helper(Register item, Register mdp, Register reg2, int start_row, 1673 Label& done, int total_rows, 1674 OffsetFunction item_offset_fn, 1675 OffsetFunction item_count_offset_fn) { 1676 int last_row = total_rows - 1; 1677 assert(start_row <= last_row, "must be work left to do"); 1678 // Test this row for both the item and for null. 1679 // Take any of three different outcomes: 1680 // 1. found item => increment count and goto done 1681 // 2. found null => keep looking for case 1, maybe allocate this cell 1682 // 3. found something else => keep looking for cases 1 and 2 1683 // Case 3 is handled by a recursive call. 1684 for (int row = start_row; row <= last_row; row++) { 1685 Label next_test; 1686 bool test_for_null_also = (row == start_row); 1687 1688 // See if the item is item[n]. 1689 int item_offset = in_bytes(item_offset_fn(row)); 1690 test_mdp_data_at(mdp, item_offset, item, 1691 (test_for_null_also ? reg2 : noreg), 1692 next_test); 1693 // (Reg2 now contains the item from the CallData.) 1694 1695 // The item is item[n]. Increment count[n]. 1696 int count_offset = in_bytes(item_count_offset_fn(row)); 1697 increment_mdp_data_at(mdp, count_offset); 1698 jmp(done); 1699 bind(next_test); 1700 1701 if (test_for_null_also) { 1702 // Failed the equality check on item[n]... Test for null. 1703 testptr(reg2, reg2); 1704 if (start_row == last_row) { 1705 // The only thing left to do is handle the null case. 1706 Label found_null; 1707 jccb(Assembler::zero, found_null); 1708 // Item did not match any saved item and there is no empty row for it. 1709 // Increment total counter to indicate polymorphic case. 1710 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); 1711 jmp(done); 1712 bind(found_null); 1713 break; 1714 } 1715 Label found_null; 1716 // Since null is rare, make it be the branch-taken case. 1717 jcc(Assembler::zero, found_null); 1718 1719 // Put all the "Case 3" tests here. 1720 record_item_in_profile_helper(item, mdp, reg2, start_row + 1, done, total_rows, 1721 item_offset_fn, item_count_offset_fn); 1722 1723 // Found a null. Keep searching for a matching item, 1724 // but remember that this is an empty (unused) slot. 1725 bind(found_null); 1726 } 1727 } 1728 1729 // In the fall-through case, we found no matching item, but we 1730 // observed the item[start_row] is null. 1731 1732 // Fill in the item field and increment the count. 1733 int item_offset = in_bytes(item_offset_fn(start_row)); 1734 set_mdp_data_at(mdp, item_offset, item); 1735 int count_offset = in_bytes(item_count_offset_fn(start_row)); 1736 movl(reg2, DataLayout::counter_increment); 1737 set_mdp_data_at(mdp, count_offset, reg2); 1738 if (start_row > 0) { 1739 jmp(done); 1740 } 1741 } 1742 1743 // Example state machine code for three profile rows: 1744 // // main copy of decision tree, rooted at row[1] 1745 // if (row[0].rec == rec) { row[0].incr(); goto done; } 1746 // if (row[0].rec != nullptr) { 1747 // // inner copy of decision tree, rooted at row[1] 1748 // if (row[1].rec == rec) { row[1].incr(); goto done; } 1749 // if (row[1].rec != nullptr) { 1750 // // degenerate decision tree, rooted at row[2] 1751 // if (row[2].rec == rec) { row[2].incr(); goto done; } 1752 // if (row[2].rec != nullptr) { count.incr(); goto done; } // overflow 1753 // row[2].init(rec); goto done; 1754 // } else { 1755 // // remember row[1] is empty 1756 // if (row[2].rec == rec) { row[2].incr(); goto done; } 1757 // row[1].init(rec); goto done; 1758 // } 1759 // } else { 1760 // // remember row[0] is empty 1761 // if (row[1].rec == rec) { row[1].incr(); goto done; } 1762 // if (row[2].rec == rec) { row[2].incr(); goto done; } 1763 // row[0].init(rec); goto done; 1764 // } 1765 // done: 1766 1767 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver, 1768 Register mdp, Register reg2, 1769 bool is_virtual_call) { 1770 assert(ProfileInterpreter, "must be profiling"); 1771 Label done; 1772 1773 record_klass_in_profile_helper(receiver, mdp, reg2, 0, done, is_virtual_call); 1774 1775 bind (done); 1776 } 1777 1778 void InterpreterMacroAssembler::profile_ret(Register return_bci, 1779 Register mdp) { 1780 if (ProfileInterpreter) { 1781 Label profile_continue; 1782 uint row; 1783 1784 // If no method data exists, go to profile_continue. 1785 test_method_data_pointer(mdp, profile_continue); 1786 1787 // Update the total ret count. 1788 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); 1789 1790 for (row = 0; row < RetData::row_limit(); row++) { 1791 Label next_test; 1792 1793 // See if return_bci is equal to bci[n]: 1794 test_mdp_data_at(mdp, 1795 in_bytes(RetData::bci_offset(row)), 1796 return_bci, noreg, 1797 next_test); 1798 1799 // return_bci is equal to bci[n]. Increment the count. 1800 increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row))); 1801 1802 // The method data pointer needs to be updated to reflect the new target. 1803 update_mdp_by_offset(mdp, 1804 in_bytes(RetData::bci_displacement_offset(row))); 1805 jmp(profile_continue); 1806 bind(next_test); 1807 } 1808 1809 update_mdp_for_ret(return_bci); 1810 1811 bind(profile_continue); 1812 } 1813 } 1814 1815 1816 void InterpreterMacroAssembler::profile_null_seen(Register mdp) { 1817 if (ProfileInterpreter) { 1818 Label profile_continue; 1819 1820 // If no method data exists, go to profile_continue. 1821 test_method_data_pointer(mdp, profile_continue); 1822 1823 set_mdp_flag_at(mdp, BitData::null_seen_byte_constant()); 1824 1825 // The method data pointer needs to be updated. 1826 int mdp_delta = in_bytes(BitData::bit_data_size()); 1827 if (TypeProfileCasts) { 1828 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size()); 1829 } 1830 update_mdp_by_constant(mdp, mdp_delta); 1831 1832 bind(profile_continue); 1833 } 1834 } 1835 1836 1837 void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) { 1838 if (ProfileInterpreter) { 1839 Label profile_continue; 1840 1841 // If no method data exists, go to profile_continue. 1842 test_method_data_pointer(mdp, profile_continue); 1843 1844 // The method data pointer needs to be updated. 1845 int mdp_delta = in_bytes(BitData::bit_data_size()); 1846 if (TypeProfileCasts) { 1847 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size()); 1848 1849 // Record the object type. 1850 record_klass_in_profile(klass, mdp, reg2, false); 1851 NOT_LP64(assert(reg2 == rdi, "we know how to fix this blown reg");) 1852 NOT_LP64(restore_locals();) // Restore EDI 1853 } 1854 update_mdp_by_constant(mdp, mdp_delta); 1855 1856 bind(profile_continue); 1857 } 1858 } 1859 1860 1861 void InterpreterMacroAssembler::profile_switch_default(Register mdp) { 1862 if (ProfileInterpreter) { 1863 Label profile_continue; 1864 1865 // If no method data exists, go to profile_continue. 1866 test_method_data_pointer(mdp, profile_continue); 1867 1868 // Update the default case count 1869 increment_mdp_data_at(mdp, 1870 in_bytes(MultiBranchData::default_count_offset())); 1871 1872 // The method data pointer needs to be updated. 1873 update_mdp_by_offset(mdp, 1874 in_bytes(MultiBranchData:: 1875 default_displacement_offset())); 1876 1877 bind(profile_continue); 1878 } 1879 } 1880 1881 1882 void InterpreterMacroAssembler::profile_switch_case(Register index, 1883 Register mdp, 1884 Register reg2) { 1885 if (ProfileInterpreter) { 1886 Label profile_continue; 1887 1888 // If no method data exists, go to profile_continue. 1889 test_method_data_pointer(mdp, profile_continue); 1890 1891 // Build the base (index * per_case_size_in_bytes()) + 1892 // case_array_offset_in_bytes() 1893 movl(reg2, in_bytes(MultiBranchData::per_case_size())); 1894 imulptr(index, reg2); // XXX l ? 1895 addptr(index, in_bytes(MultiBranchData::case_array_offset())); // XXX l ? 1896 1897 // Update the case count 1898 increment_mdp_data_at(mdp, 1899 index, 1900 in_bytes(MultiBranchData::relative_count_offset())); 1901 1902 // The method data pointer needs to be updated. 1903 update_mdp_by_offset(mdp, 1904 index, 1905 in_bytes(MultiBranchData:: 1906 relative_displacement_offset())); 1907 1908 bind(profile_continue); 1909 } 1910 } 1911 1912 1913 1914 void InterpreterMacroAssembler::_interp_verify_oop(Register reg, TosState state, const char* file, int line) { 1915 if (state == atos) { 1916 MacroAssembler::_verify_oop_checked(reg, "broken oop", file, line); 1917 } 1918 } 1919 1920 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) { 1921 #ifndef _LP64 1922 if ((state == ftos && UseSSE < 1) || 1923 (state == dtos && UseSSE < 2)) { 1924 MacroAssembler::verify_FPU(stack_depth); 1925 } 1926 #endif 1927 } 1928 1929 // Jump if ((*counter_addr += increment) & mask) == 0 1930 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr, Address mask, 1931 Register scratch, Label* where) { 1932 // This update is actually not atomic and can lose a number of updates 1933 // under heavy contention, but the alternative of using the (contended) 1934 // atomic update here penalizes profiling paths too much. 1935 movl(scratch, counter_addr); 1936 incrementl(scratch, InvocationCounter::count_increment); 1937 movl(counter_addr, scratch); 1938 andl(scratch, mask); 1939 if (where != nullptr) { 1940 jcc(Assembler::zero, *where); 1941 } 1942 } 1943 1944 void InterpreterMacroAssembler::notify_method_entry() { 1945 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to 1946 // track stack depth. If it is possible to enter interp_only_mode we add 1947 // the code to check if the event should be sent. 1948 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx); 1949 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rbx); 1950 if (JvmtiExport::can_post_interpreter_events()) { 1951 Label L; 1952 NOT_LP64(get_thread(rthread);) 1953 movl(rdx, Address(rthread, JavaThread::interp_only_mode_offset())); 1954 testl(rdx, rdx); 1955 jcc(Assembler::zero, L); 1956 call_VM(noreg, CAST_FROM_FN_PTR(address, 1957 InterpreterRuntime::post_method_entry)); 1958 bind(L); 1959 } 1960 1961 if (DTraceMethodProbes) { 1962 NOT_LP64(get_thread(rthread);) 1963 get_method(rarg); 1964 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry), 1965 rthread, rarg); 1966 } 1967 1968 // RedefineClasses() tracing support for obsolete method entry 1969 if (log_is_enabled(Trace, redefine, class, obsolete)) { 1970 NOT_LP64(get_thread(rthread);) 1971 get_method(rarg); 1972 call_VM_leaf( 1973 CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry), 1974 rthread, rarg); 1975 } 1976 } 1977 1978 1979 void InterpreterMacroAssembler::notify_method_exit( 1980 TosState state, NotifyMethodExitMode mode) { 1981 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to 1982 // track stack depth. If it is possible to enter interp_only_mode we add 1983 // the code to check if the event should be sent. 1984 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx); 1985 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rbx); 1986 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) { 1987 Label L; 1988 // Note: frame::interpreter_frame_result has a dependency on how the 1989 // method result is saved across the call to post_method_exit. If this 1990 // is changed then the interpreter_frame_result implementation will 1991 // need to be updated too. 1992 1993 // template interpreter will leave the result on the top of the stack. 1994 push(state); 1995 NOT_LP64(get_thread(rthread);) 1996 movl(rdx, Address(rthread, JavaThread::interp_only_mode_offset())); 1997 testl(rdx, rdx); 1998 jcc(Assembler::zero, L); 1999 call_VM(noreg, 2000 CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit)); 2001 bind(L); 2002 pop(state); 2003 } 2004 2005 if (DTraceMethodProbes) { 2006 push(state); 2007 NOT_LP64(get_thread(rthread);) 2008 get_method(rarg); 2009 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit), 2010 rthread, rarg); 2011 pop(state); 2012 } 2013 } 2014 2015 void InterpreterMacroAssembler::load_resolved_indy_entry(Register cache, Register index) { 2016 // Get index out of bytecode pointer 2017 get_cache_index_at_bcp(index, 1, sizeof(u4)); 2018 // Get address of invokedynamic array 2019 movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize)); 2020 movptr(cache, Address(cache, in_bytes(ConstantPoolCache::invokedynamic_entries_offset()))); 2021 if (is_power_of_2(sizeof(ResolvedIndyEntry))) { 2022 shll(index, log2i_exact(sizeof(ResolvedIndyEntry))); // Scale index by power of 2 2023 } else { 2024 imull(index, index, sizeof(ResolvedIndyEntry)); // Scale the index to be the entry index * sizeof(ResolvedIndyEntry) 2025 } 2026 lea(cache, Address(cache, index, Address::times_1, Array<ResolvedIndyEntry>::base_offset_in_bytes())); 2027 } 2028 2029 void InterpreterMacroAssembler::load_field_entry(Register cache, Register index, int bcp_offset) { 2030 // Get index out of bytecode pointer 2031 movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize)); 2032 get_cache_index_at_bcp(index, bcp_offset, sizeof(u2)); 2033 2034 movptr(cache, Address(cache, ConstantPoolCache::field_entries_offset())); 2035 // Take shortcut if the size is a power of 2 2036 if (is_power_of_2(sizeof(ResolvedFieldEntry))) { 2037 shll(index, log2i_exact(sizeof(ResolvedFieldEntry))); // Scale index by power of 2 2038 } else { 2039 imull(index, index, sizeof(ResolvedFieldEntry)); // Scale the index to be the entry index * sizeof(ResolvedFieldEntry) 2040 } 2041 lea(cache, Address(cache, index, Address::times_1, Array<ResolvedFieldEntry>::base_offset_in_bytes())); 2042 } 2043 2044 void InterpreterMacroAssembler::load_method_entry(Register cache, Register index, int bcp_offset) { 2045 // Get index out of bytecode pointer 2046 movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize)); 2047 get_cache_index_at_bcp(index, bcp_offset, sizeof(u2)); 2048 2049 movptr(cache, Address(cache, ConstantPoolCache::method_entries_offset())); 2050 imull(index, index, sizeof(ResolvedMethodEntry)); // Scale the index to be the entry index * sizeof(ResolvedMethodEntry) 2051 lea(cache, Address(cache, index, Address::times_1, Array<ResolvedMethodEntry>::base_offset_in_bytes())); 2052 } --- EOF ---