1 /* 2 * Copyright (c) 1997, 2018, Oracle and/or its affiliates. All rights reserved. 3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. 4 * 5 * This code is free software; you can redistribute it and/or modify it 6 * under the terms of the GNU General Public License version 2 only, as 7 * published by the Free Software Foundation. 8 * 9 * This code is distributed in the hope that it will be useful, but WITHOUT 10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License 12 * version 2 for more details (a copy is included in the LICENSE file that 13 * accompanied this code). 14 * 15 * You should have received a copy of the GNU General Public License version 16 * 2 along with this work; if not, write to the Free Software Foundation, 17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. 18 * 19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA 20 * or visit www.oracle.com if you need additional information or have any 21 * questions. 22 * 23 */ 24 25 #include "precompiled.hpp" 26 #include "interp_masm_x86.hpp" 27 #include "interpreter/interpreter.hpp" 28 #include "interpreter/interpreterRuntime.hpp" 29 #include "logging/log.hpp" 30 #include "oops/arrayOop.hpp" 31 #include "oops/markWord.hpp" 32 #include "oops/methodData.hpp" 33 #include "oops/method.hpp" 34 #include "oops/valueKlass.hpp" 35 #include "prims/jvmtiExport.hpp" 36 #include "prims/jvmtiThreadState.hpp" 37 #include "runtime/basicLock.hpp" 38 #include "runtime/biasedLocking.hpp" 39 #include "runtime/frame.inline.hpp" 40 #include "runtime/safepointMechanism.hpp" 41 #include "runtime/sharedRuntime.hpp" 42 #include "runtime/thread.inline.hpp" 43 44 // Implementation of InterpreterMacroAssembler 45 46 void InterpreterMacroAssembler::jump_to_entry(address entry) { 47 assert(entry, "Entry must have been generated by now"); 48 jump(RuntimeAddress(entry)); 49 } 50 51 void InterpreterMacroAssembler::profile_obj_type(Register obj, const Address& mdo_addr) { 52 Label update, next, none; 53 54 verify_oop(obj); 55 56 testptr(obj, obj); 57 jccb(Assembler::notZero, update); 58 orptr(mdo_addr, TypeEntries::null_seen); 59 jmpb(next); 60 61 bind(update); 62 load_klass(obj, obj); 63 64 xorptr(obj, mdo_addr); 65 testptr(obj, TypeEntries::type_klass_mask); 66 jccb(Assembler::zero, next); // klass seen before, nothing to 67 // do. The unknown bit may have been 68 // set already but no need to check. 69 70 testptr(obj, TypeEntries::type_unknown); 71 jccb(Assembler::notZero, next); // already unknown. Nothing to do anymore. 72 73 cmpptr(mdo_addr, 0); 74 jccb(Assembler::equal, none); 75 cmpptr(mdo_addr, TypeEntries::null_seen); 76 jccb(Assembler::equal, none); 77 // There is a chance that the checks above (re-reading profiling 78 // data from memory) fail if another thread has just set the 79 // profiling to this obj's klass 80 xorptr(obj, mdo_addr); 81 testptr(obj, TypeEntries::type_klass_mask); 82 jccb(Assembler::zero, next); 83 84 // different than before. Cannot keep accurate profile. 85 orptr(mdo_addr, TypeEntries::type_unknown); 86 jmpb(next); 87 88 bind(none); 89 // first time here. Set profile type. 90 movptr(mdo_addr, obj); 91 92 bind(next); 93 } 94 95 void InterpreterMacroAssembler::profile_arguments_type(Register mdp, Register callee, Register tmp, bool is_virtual) { 96 if (!ProfileInterpreter) { 97 return; 98 } 99 100 if (MethodData::profile_arguments() || MethodData::profile_return()) { 101 Label profile_continue; 102 103 test_method_data_pointer(mdp, profile_continue); 104 105 int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size()); 106 107 cmpb(Address(mdp, in_bytes(DataLayout::tag_offset()) - off_to_start), is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag); 108 jcc(Assembler::notEqual, profile_continue); 109 110 if (MethodData::profile_arguments()) { 111 Label done; 112 int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset()); 113 addptr(mdp, off_to_args); 114 115 for (int i = 0; i < TypeProfileArgsLimit; i++) { 116 if (i > 0 || MethodData::profile_return()) { 117 // If return value type is profiled we may have no argument to profile 118 movptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args)); 119 subl(tmp, i*TypeStackSlotEntries::per_arg_count()); 120 cmpl(tmp, TypeStackSlotEntries::per_arg_count()); 121 jcc(Assembler::less, done); 122 } 123 movptr(tmp, Address(callee, Method::const_offset())); 124 load_unsigned_short(tmp, Address(tmp, ConstMethod::size_of_parameters_offset())); 125 // stack offset o (zero based) from the start of the argument 126 // list, for n arguments translates into offset n - o - 1 from 127 // the end of the argument list 128 subptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))-off_to_args)); 129 subl(tmp, 1); 130 Address arg_addr = argument_address(tmp); 131 movptr(tmp, arg_addr); 132 133 Address mdo_arg_addr(mdp, in_bytes(TypeEntriesAtCall::argument_type_offset(i))-off_to_args); 134 profile_obj_type(tmp, mdo_arg_addr); 135 136 int to_add = in_bytes(TypeStackSlotEntries::per_arg_size()); 137 addptr(mdp, to_add); 138 off_to_args += to_add; 139 } 140 141 if (MethodData::profile_return()) { 142 movptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args)); 143 subl(tmp, TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count()); 144 } 145 146 bind(done); 147 148 if (MethodData::profile_return()) { 149 // We're right after the type profile for the last 150 // argument. tmp is the number of cells left in the 151 // CallTypeData/VirtualCallTypeData to reach its end. Non null 152 // if there's a return to profile. 153 assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type"); 154 shll(tmp, exact_log2(DataLayout::cell_size)); 155 addptr(mdp, tmp); 156 } 157 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp); 158 } else { 159 assert(MethodData::profile_return(), "either profile call args or call ret"); 160 update_mdp_by_constant(mdp, in_bytes(TypeEntriesAtCall::return_only_size())); 161 } 162 163 // mdp points right after the end of the 164 // CallTypeData/VirtualCallTypeData, right after the cells for the 165 // return value type if there's one 166 167 bind(profile_continue); 168 } 169 } 170 171 void InterpreterMacroAssembler::profile_return_type(Register mdp, Register ret, Register tmp) { 172 assert_different_registers(mdp, ret, tmp, _bcp_register); 173 if (ProfileInterpreter && MethodData::profile_return()) { 174 Label profile_continue; 175 176 test_method_data_pointer(mdp, profile_continue); 177 178 if (MethodData::profile_return_jsr292_only()) { 179 assert(Method::intrinsic_id_size_in_bytes() == 2, "assuming Method::_intrinsic_id is u2"); 180 181 // If we don't profile all invoke bytecodes we must make sure 182 // it's a bytecode we indeed profile. We can't go back to the 183 // begining of the ProfileData we intend to update to check its 184 // type because we're right after it and we don't known its 185 // length 186 Label do_profile; 187 cmpb(Address(_bcp_register, 0), Bytecodes::_invokedynamic); 188 jcc(Assembler::equal, do_profile); 189 cmpb(Address(_bcp_register, 0), Bytecodes::_invokehandle); 190 jcc(Assembler::equal, do_profile); 191 get_method(tmp); 192 cmpw(Address(tmp, Method::intrinsic_id_offset_in_bytes()), vmIntrinsics::_compiledLambdaForm); 193 jcc(Assembler::notEqual, profile_continue); 194 195 bind(do_profile); 196 } 197 198 Address mdo_ret_addr(mdp, -in_bytes(ReturnTypeEntry::size())); 199 mov(tmp, ret); 200 profile_obj_type(tmp, mdo_ret_addr); 201 202 bind(profile_continue); 203 } 204 } 205 206 void InterpreterMacroAssembler::profile_parameters_type(Register mdp, Register tmp1, Register tmp2) { 207 if (ProfileInterpreter && MethodData::profile_parameters()) { 208 Label profile_continue; 209 210 test_method_data_pointer(mdp, profile_continue); 211 212 // Load the offset of the area within the MDO used for 213 // parameters. If it's negative we're not profiling any parameters 214 movl(tmp1, Address(mdp, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset()))); 215 testl(tmp1, tmp1); 216 jcc(Assembler::negative, profile_continue); 217 218 // Compute a pointer to the area for parameters from the offset 219 // and move the pointer to the slot for the last 220 // parameters. Collect profiling from last parameter down. 221 // mdo start + parameters offset + array length - 1 222 addptr(mdp, tmp1); 223 movptr(tmp1, Address(mdp, ArrayData::array_len_offset())); 224 decrement(tmp1, TypeStackSlotEntries::per_arg_count()); 225 226 Label loop; 227 bind(loop); 228 229 int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0)); 230 int type_base = in_bytes(ParametersTypeData::type_offset(0)); 231 Address::ScaleFactor per_arg_scale = Address::times(DataLayout::cell_size); 232 Address arg_off(mdp, tmp1, per_arg_scale, off_base); 233 Address arg_type(mdp, tmp1, per_arg_scale, type_base); 234 235 // load offset on the stack from the slot for this parameter 236 movptr(tmp2, arg_off); 237 negptr(tmp2); 238 // read the parameter from the local area 239 movptr(tmp2, Address(_locals_register, tmp2, Interpreter::stackElementScale())); 240 241 // profile the parameter 242 profile_obj_type(tmp2, arg_type); 243 244 // go to next parameter 245 decrement(tmp1, TypeStackSlotEntries::per_arg_count()); 246 jcc(Assembler::positive, loop); 247 248 bind(profile_continue); 249 } 250 } 251 252 void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point, 253 int number_of_arguments) { 254 // interpreter specific 255 // 256 // Note: No need to save/restore bcp & locals registers 257 // since these are callee saved registers and no blocking/ 258 // GC can happen in leaf calls. 259 // Further Note: DO NOT save/restore bcp/locals. If a caller has 260 // already saved them so that it can use rsi/rdi as temporaries 261 // then a save/restore here will DESTROY the copy the caller 262 // saved! There used to be a save_bcp() that only happened in 263 // the ASSERT path (no restore_bcp). Which caused bizarre failures 264 // when jvm built with ASSERTs. 265 #ifdef ASSERT 266 { 267 Label L; 268 cmpptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD); 269 jcc(Assembler::equal, L); 270 stop("InterpreterMacroAssembler::call_VM_leaf_base:" 271 " last_sp != NULL"); 272 bind(L); 273 } 274 #endif 275 // super call 276 MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments); 277 // interpreter specific 278 // LP64: Used to ASSERT that r13/r14 were equal to frame's bcp/locals 279 // but since they may not have been saved (and we don't want to 280 // save them here (see note above) the assert is invalid. 281 } 282 283 void InterpreterMacroAssembler::call_VM_base(Register oop_result, 284 Register java_thread, 285 Register last_java_sp, 286 address entry_point, 287 int number_of_arguments, 288 bool check_exceptions) { 289 // interpreter specific 290 // 291 // Note: Could avoid restoring locals ptr (callee saved) - however doesn't 292 // really make a difference for these runtime calls, since they are 293 // slow anyway. Btw., bcp must be saved/restored since it may change 294 // due to GC. 295 NOT_LP64(assert(java_thread == noreg , "not expecting a precomputed java thread");) 296 save_bcp(); 297 #ifdef ASSERT 298 { 299 Label L; 300 cmpptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD); 301 jcc(Assembler::equal, L); 302 stop("InterpreterMacroAssembler::call_VM_base:" 303 " last_sp != NULL"); 304 bind(L); 305 } 306 #endif /* ASSERT */ 307 // super call 308 MacroAssembler::call_VM_base(oop_result, noreg, last_java_sp, 309 entry_point, number_of_arguments, 310 check_exceptions); 311 // interpreter specific 312 restore_bcp(); 313 restore_locals(); 314 } 315 316 void InterpreterMacroAssembler::check_and_handle_popframe(Register java_thread) { 317 if (JvmtiExport::can_pop_frame()) { 318 Label L; 319 // Initiate popframe handling only if it is not already being 320 // processed. If the flag has the popframe_processing bit set, it 321 // means that this code is called *during* popframe handling - we 322 // don't want to reenter. 323 // This method is only called just after the call into the vm in 324 // call_VM_base, so the arg registers are available. 325 Register pop_cond = NOT_LP64(java_thread) // Not clear if any other register is available on 32 bit 326 LP64_ONLY(c_rarg0); 327 movl(pop_cond, Address(java_thread, JavaThread::popframe_condition_offset())); 328 testl(pop_cond, JavaThread::popframe_pending_bit); 329 jcc(Assembler::zero, L); 330 testl(pop_cond, JavaThread::popframe_processing_bit); 331 jcc(Assembler::notZero, L); 332 // Call Interpreter::remove_activation_preserving_args_entry() to get the 333 // address of the same-named entrypoint in the generated interpreter code. 334 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry)); 335 jmp(rax); 336 bind(L); 337 NOT_LP64(get_thread(java_thread);) 338 } 339 } 340 341 void InterpreterMacroAssembler::load_earlyret_value(TosState state) { 342 Register thread = LP64_ONLY(r15_thread) NOT_LP64(rcx); 343 NOT_LP64(get_thread(thread);) 344 movptr(rcx, Address(thread, JavaThread::jvmti_thread_state_offset())); 345 const Address tos_addr(rcx, JvmtiThreadState::earlyret_tos_offset()); 346 const Address oop_addr(rcx, JvmtiThreadState::earlyret_oop_offset()); 347 const Address val_addr(rcx, JvmtiThreadState::earlyret_value_offset()); 348 #ifdef _LP64 349 switch (state) { 350 case atos: movptr(rax, oop_addr); 351 movptr(oop_addr, (int32_t)NULL_WORD); 352 verify_oop(rax, state); break; 353 case ltos: movptr(rax, val_addr); break; 354 case btos: // fall through 355 case ztos: // fall through 356 case ctos: // fall through 357 case stos: // fall through 358 case itos: movl(rax, val_addr); break; 359 case ftos: load_float(val_addr); break; 360 case dtos: load_double(val_addr); break; 361 case vtos: /* nothing to do */ break; 362 default : ShouldNotReachHere(); 363 } 364 // Clean up tos value in the thread object 365 movl(tos_addr, (int) ilgl); 366 movl(val_addr, (int32_t) NULL_WORD); 367 #else 368 const Address val_addr1(rcx, JvmtiThreadState::earlyret_value_offset() 369 + in_ByteSize(wordSize)); 370 switch (state) { 371 case atos: movptr(rax, oop_addr); 372 movptr(oop_addr, NULL_WORD); 373 verify_oop(rax, state); break; 374 case ltos: 375 movl(rdx, val_addr1); // fall through 376 case btos: // fall through 377 case ztos: // fall through 378 case ctos: // fall through 379 case stos: // fall through 380 case itos: movl(rax, val_addr); break; 381 case ftos: load_float(val_addr); break; 382 case dtos: load_double(val_addr); break; 383 case vtos: /* nothing to do */ break; 384 default : ShouldNotReachHere(); 385 } 386 #endif // _LP64 387 // Clean up tos value in the thread object 388 movl(tos_addr, (int32_t) ilgl); 389 movptr(val_addr, NULL_WORD); 390 NOT_LP64(movptr(val_addr1, NULL_WORD);) 391 } 392 393 394 void InterpreterMacroAssembler::check_and_handle_earlyret(Register java_thread) { 395 if (JvmtiExport::can_force_early_return()) { 396 Label L; 397 Register tmp = LP64_ONLY(c_rarg0) NOT_LP64(java_thread); 398 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(java_thread); 399 400 movptr(tmp, Address(rthread, JavaThread::jvmti_thread_state_offset())); 401 testptr(tmp, tmp); 402 jcc(Assembler::zero, L); // if (thread->jvmti_thread_state() == NULL) exit; 403 404 // Initiate earlyret handling only if it is not already being processed. 405 // If the flag has the earlyret_processing bit set, it means that this code 406 // is called *during* earlyret handling - we don't want to reenter. 407 movl(tmp, Address(tmp, JvmtiThreadState::earlyret_state_offset())); 408 cmpl(tmp, JvmtiThreadState::earlyret_pending); 409 jcc(Assembler::notEqual, L); 410 411 // Call Interpreter::remove_activation_early_entry() to get the address of the 412 // same-named entrypoint in the generated interpreter code. 413 NOT_LP64(get_thread(java_thread);) 414 movptr(tmp, Address(rthread, JavaThread::jvmti_thread_state_offset())); 415 #ifdef _LP64 416 movl(tmp, Address(tmp, JvmtiThreadState::earlyret_tos_offset())); 417 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), tmp); 418 #else 419 pushl(Address(tmp, JvmtiThreadState::earlyret_tos_offset())); 420 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), 1); 421 #endif // _LP64 422 jmp(rax); 423 bind(L); 424 NOT_LP64(get_thread(java_thread);) 425 } 426 } 427 428 void InterpreterMacroAssembler::get_unsigned_2_byte_index_at_bcp(Register reg, int bcp_offset) { 429 assert(bcp_offset >= 0, "bcp is still pointing to start of bytecode"); 430 load_unsigned_short(reg, Address(_bcp_register, bcp_offset)); 431 bswapl(reg); 432 shrl(reg, 16); 433 } 434 435 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register index, 436 int bcp_offset, 437 size_t index_size) { 438 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode"); 439 if (index_size == sizeof(u2)) { 440 load_unsigned_short(index, Address(_bcp_register, bcp_offset)); 441 } else if (index_size == sizeof(u4)) { 442 movl(index, Address(_bcp_register, bcp_offset)); 443 // Check if the secondary index definition is still ~x, otherwise 444 // we have to change the following assembler code to calculate the 445 // plain index. 446 assert(ConstantPool::decode_invokedynamic_index(~123) == 123, "else change next line"); 447 notl(index); // convert to plain index 448 } else if (index_size == sizeof(u1)) { 449 load_unsigned_byte(index, Address(_bcp_register, bcp_offset)); 450 } else { 451 ShouldNotReachHere(); 452 } 453 } 454 455 void InterpreterMacroAssembler::get_cache_and_index_at_bcp(Register cache, 456 Register index, 457 int bcp_offset, 458 size_t index_size) { 459 assert_different_registers(cache, index); 460 get_cache_index_at_bcp(index, bcp_offset, index_size); 461 movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize)); 462 assert(sizeof(ConstantPoolCacheEntry) == 4 * wordSize, "adjust code below"); 463 // convert from field index to ConstantPoolCacheEntry index 464 assert(exact_log2(in_words(ConstantPoolCacheEntry::size())) == 2, "else change next line"); 465 shll(index, 2); 466 } 467 468 void InterpreterMacroAssembler::get_cache_and_index_and_bytecode_at_bcp(Register cache, 469 Register index, 470 Register bytecode, 471 int byte_no, 472 int bcp_offset, 473 size_t index_size) { 474 get_cache_and_index_at_bcp(cache, index, bcp_offset, index_size); 475 // We use a 32-bit load here since the layout of 64-bit words on 476 // little-endian machines allow us that. 477 movl(bytecode, Address(cache, index, Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset())); 478 const int shift_count = (1 + byte_no) * BitsPerByte; 479 assert((byte_no == TemplateTable::f1_byte && shift_count == ConstantPoolCacheEntry::bytecode_1_shift) || 480 (byte_no == TemplateTable::f2_byte && shift_count == ConstantPoolCacheEntry::bytecode_2_shift), 481 "correct shift count"); 482 shrl(bytecode, shift_count); 483 assert(ConstantPoolCacheEntry::bytecode_1_mask == ConstantPoolCacheEntry::bytecode_2_mask, "common mask"); 484 andl(bytecode, ConstantPoolCacheEntry::bytecode_1_mask); 485 } 486 487 void InterpreterMacroAssembler::get_cache_entry_pointer_at_bcp(Register cache, 488 Register tmp, 489 int bcp_offset, 490 size_t index_size) { 491 assert_different_registers(cache, tmp); 492 493 get_cache_index_at_bcp(tmp, bcp_offset, index_size); 494 assert(sizeof(ConstantPoolCacheEntry) == 4 * wordSize, "adjust code below"); 495 // convert from field index to ConstantPoolCacheEntry index 496 // and from word offset to byte offset 497 assert(exact_log2(in_bytes(ConstantPoolCacheEntry::size_in_bytes())) == 2 + LogBytesPerWord, "else change next line"); 498 shll(tmp, 2 + LogBytesPerWord); 499 movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize)); 500 // skip past the header 501 addptr(cache, in_bytes(ConstantPoolCache::base_offset())); 502 addptr(cache, tmp); // construct pointer to cache entry 503 } 504 505 // Load object from cpool->resolved_references(index) 506 void InterpreterMacroAssembler::load_resolved_reference_at_index(Register result, 507 Register index, 508 Register tmp) { 509 assert_different_registers(result, index); 510 511 get_constant_pool(result); 512 // load pointer for resolved_references[] objArray 513 movptr(result, Address(result, ConstantPool::cache_offset_in_bytes())); 514 movptr(result, Address(result, ConstantPoolCache::resolved_references_offset_in_bytes())); 515 resolve_oop_handle(result, tmp); 516 load_heap_oop(result, Address(result, index, 517 UseCompressedOops ? Address::times_4 : Address::times_ptr, 518 arrayOopDesc::base_offset_in_bytes(T_OBJECT)), tmp); 519 } 520 521 // load cpool->resolved_klass_at(index) 522 void InterpreterMacroAssembler::load_resolved_klass_at_index(Register klass, 523 Register cpool, 524 Register index) { 525 assert_different_registers(cpool, index); 526 527 movw(index, Address(cpool, index, Address::times_ptr, sizeof(ConstantPool))); 528 Register resolved_klasses = cpool; 529 movptr(resolved_klasses, Address(cpool, ConstantPool::resolved_klasses_offset_in_bytes())); 530 movptr(klass, Address(resolved_klasses, index, Address::times_ptr, Array<Klass*>::base_offset_in_bytes())); 531 } 532 533 void InterpreterMacroAssembler::load_resolved_method_at_index(int byte_no, 534 Register method, 535 Register cache, 536 Register index) { 537 assert_different_registers(cache, index); 538 539 const int method_offset = in_bytes( 540 ConstantPoolCache::base_offset() + 541 ((byte_no == TemplateTable::f2_byte) 542 ? ConstantPoolCacheEntry::f2_offset() 543 : ConstantPoolCacheEntry::f1_offset())); 544 545 movptr(method, Address(cache, index, Address::times_ptr, method_offset)); // get f1 Method* 546 } 547 548 // Generate a subtype check: branch to ok_is_subtype if sub_klass is a 549 // subtype of super_klass. 550 // 551 // Args: 552 // rax: superklass 553 // Rsub_klass: subklass 554 // 555 // Kills: 556 // rcx, rdi 557 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass, 558 Label& ok_is_subtype) { 559 assert(Rsub_klass != rax, "rax holds superklass"); 560 LP64_ONLY(assert(Rsub_klass != r14, "r14 holds locals");) 561 LP64_ONLY(assert(Rsub_klass != r13, "r13 holds bcp");) 562 assert(Rsub_klass != rcx, "rcx holds 2ndary super array length"); 563 assert(Rsub_klass != rdi, "rdi holds 2ndary super array scan ptr"); 564 565 // Profile the not-null value's klass. 566 profile_typecheck(rcx, Rsub_klass, rdi); // blows rcx, reloads rdi 567 568 // Do the check. 569 check_klass_subtype(Rsub_klass, rax, rcx, ok_is_subtype); // blows rcx 570 571 // Profile the failure of the check. 572 profile_typecheck_failed(rcx); // blows rcx 573 } 574 575 576 #ifndef _LP64 577 void InterpreterMacroAssembler::f2ieee() { 578 if (IEEEPrecision) { 579 fstp_s(Address(rsp, 0)); 580 fld_s(Address(rsp, 0)); 581 } 582 } 583 584 585 void InterpreterMacroAssembler::d2ieee() { 586 if (IEEEPrecision) { 587 fstp_d(Address(rsp, 0)); 588 fld_d(Address(rsp, 0)); 589 } 590 } 591 #endif // _LP64 592 593 // Java Expression Stack 594 595 void InterpreterMacroAssembler::pop_ptr(Register r) { 596 pop(r); 597 } 598 599 void InterpreterMacroAssembler::push_ptr(Register r) { 600 push(r); 601 } 602 603 void InterpreterMacroAssembler::push_i(Register r) { 604 push(r); 605 } 606 607 void InterpreterMacroAssembler::push_f(XMMRegister r) { 608 subptr(rsp, wordSize); 609 movflt(Address(rsp, 0), r); 610 } 611 612 void InterpreterMacroAssembler::pop_f(XMMRegister r) { 613 movflt(r, Address(rsp, 0)); 614 addptr(rsp, wordSize); 615 } 616 617 void InterpreterMacroAssembler::push_d(XMMRegister r) { 618 subptr(rsp, 2 * wordSize); 619 movdbl(Address(rsp, 0), r); 620 } 621 622 void InterpreterMacroAssembler::pop_d(XMMRegister r) { 623 movdbl(r, Address(rsp, 0)); 624 addptr(rsp, 2 * Interpreter::stackElementSize); 625 } 626 627 #ifdef _LP64 628 void InterpreterMacroAssembler::pop_i(Register r) { 629 // XXX can't use pop currently, upper half non clean 630 movl(r, Address(rsp, 0)); 631 addptr(rsp, wordSize); 632 } 633 634 void InterpreterMacroAssembler::pop_l(Register r) { 635 movq(r, Address(rsp, 0)); 636 addptr(rsp, 2 * Interpreter::stackElementSize); 637 } 638 639 void InterpreterMacroAssembler::push_l(Register r) { 640 subptr(rsp, 2 * wordSize); 641 movptr(Address(rsp, Interpreter::expr_offset_in_bytes(0)), r ); 642 movptr(Address(rsp, Interpreter::expr_offset_in_bytes(1)), NULL_WORD ); 643 } 644 645 void InterpreterMacroAssembler::pop(TosState state) { 646 switch (state) { 647 case atos: pop_ptr(); break; 648 case btos: 649 case ztos: 650 case ctos: 651 case stos: 652 case itos: pop_i(); break; 653 case ltos: pop_l(); break; 654 case ftos: pop_f(xmm0); break; 655 case dtos: pop_d(xmm0); break; 656 case vtos: /* nothing to do */ break; 657 default: ShouldNotReachHere(); 658 } 659 verify_oop(rax, state); 660 } 661 662 void InterpreterMacroAssembler::push(TosState state) { 663 verify_oop(rax, state); 664 switch (state) { 665 case atos: push_ptr(); break; 666 case btos: 667 case ztos: 668 case ctos: 669 case stos: 670 case itos: push_i(); break; 671 case ltos: push_l(); break; 672 case ftos: push_f(xmm0); break; 673 case dtos: push_d(xmm0); break; 674 case vtos: /* nothing to do */ break; 675 default : ShouldNotReachHere(); 676 } 677 } 678 #else 679 void InterpreterMacroAssembler::pop_i(Register r) { 680 pop(r); 681 } 682 683 void InterpreterMacroAssembler::pop_l(Register lo, Register hi) { 684 pop(lo); 685 pop(hi); 686 } 687 688 void InterpreterMacroAssembler::pop_f() { 689 fld_s(Address(rsp, 0)); 690 addptr(rsp, 1 * wordSize); 691 } 692 693 void InterpreterMacroAssembler::pop_d() { 694 fld_d(Address(rsp, 0)); 695 addptr(rsp, 2 * wordSize); 696 } 697 698 699 void InterpreterMacroAssembler::pop(TosState state) { 700 switch (state) { 701 case atos: pop_ptr(rax); break; 702 case btos: // fall through 703 case ztos: // fall through 704 case ctos: // fall through 705 case stos: // fall through 706 case itos: pop_i(rax); break; 707 case ltos: pop_l(rax, rdx); break; 708 case ftos: 709 if (UseSSE >= 1) { 710 pop_f(xmm0); 711 } else { 712 pop_f(); 713 } 714 break; 715 case dtos: 716 if (UseSSE >= 2) { 717 pop_d(xmm0); 718 } else { 719 pop_d(); 720 } 721 break; 722 case vtos: /* nothing to do */ break; 723 default : ShouldNotReachHere(); 724 } 725 verify_oop(rax, state); 726 } 727 728 729 void InterpreterMacroAssembler::push_l(Register lo, Register hi) { 730 push(hi); 731 push(lo); 732 } 733 734 void InterpreterMacroAssembler::push_f() { 735 // Do not schedule for no AGI! Never write beyond rsp! 736 subptr(rsp, 1 * wordSize); 737 fstp_s(Address(rsp, 0)); 738 } 739 740 void InterpreterMacroAssembler::push_d() { 741 // Do not schedule for no AGI! Never write beyond rsp! 742 subptr(rsp, 2 * wordSize); 743 fstp_d(Address(rsp, 0)); 744 } 745 746 747 void InterpreterMacroAssembler::push(TosState state) { 748 verify_oop(rax, state); 749 switch (state) { 750 case atos: push_ptr(rax); break; 751 case btos: // fall through 752 case ztos: // fall through 753 case ctos: // fall through 754 case stos: // fall through 755 case itos: push_i(rax); break; 756 case ltos: push_l(rax, rdx); break; 757 case ftos: 758 if (UseSSE >= 1) { 759 push_f(xmm0); 760 } else { 761 push_f(); 762 } 763 break; 764 case dtos: 765 if (UseSSE >= 2) { 766 push_d(xmm0); 767 } else { 768 push_d(); 769 } 770 break; 771 case vtos: /* nothing to do */ break; 772 default : ShouldNotReachHere(); 773 } 774 } 775 #endif // _LP64 776 777 778 // Helpers for swap and dup 779 void InterpreterMacroAssembler::load_ptr(int n, Register val) { 780 movptr(val, Address(rsp, Interpreter::expr_offset_in_bytes(n))); 781 } 782 783 void InterpreterMacroAssembler::store_ptr(int n, Register val) { 784 movptr(Address(rsp, Interpreter::expr_offset_in_bytes(n)), val); 785 } 786 787 788 void InterpreterMacroAssembler::prepare_to_jump_from_interpreted() { 789 // set sender sp 790 lea(_bcp_register, Address(rsp, wordSize)); 791 // record last_sp 792 movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), _bcp_register); 793 } 794 795 796 // Jump to from_interpreted entry of a call unless single stepping is possible 797 // in this thread in which case we must call the i2i entry 798 void InterpreterMacroAssembler::jump_from_interpreted(Register method, Register temp) { 799 prepare_to_jump_from_interpreted(); 800 801 if (JvmtiExport::can_post_interpreter_events()) { 802 Label run_compiled_code; 803 // JVMTI events, such as single-stepping, are implemented partly by avoiding running 804 // compiled code in threads for which the event is enabled. Check here for 805 // interp_only_mode if these events CAN be enabled. 806 // interp_only is an int, on little endian it is sufficient to test the byte only 807 // Is a cmpl faster? 808 LP64_ONLY(temp = r15_thread;) 809 NOT_LP64(get_thread(temp);) 810 cmpb(Address(temp, JavaThread::interp_only_mode_offset()), 0); 811 jccb(Assembler::zero, run_compiled_code); 812 jmp(Address(method, Method::interpreter_entry_offset())); 813 bind(run_compiled_code); 814 } 815 816 jmp(Address(method, Method::from_interpreted_offset())); 817 } 818 819 // The following two routines provide a hook so that an implementation 820 // can schedule the dispatch in two parts. x86 does not do this. 821 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int step) { 822 // Nothing x86 specific to be done here 823 } 824 825 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int step) { 826 dispatch_next(state, step); 827 } 828 829 void InterpreterMacroAssembler::dispatch_base(TosState state, 830 address* table, 831 bool verifyoop, 832 bool generate_poll) { 833 verify_FPU(1, state); 834 if (VerifyActivationFrameSize) { 835 Label L; 836 mov(rcx, rbp); 837 subptr(rcx, rsp); 838 int32_t min_frame_size = 839 (frame::link_offset - frame::interpreter_frame_initial_sp_offset) * 840 wordSize; 841 cmpptr(rcx, (int32_t)min_frame_size); 842 jcc(Assembler::greaterEqual, L); 843 stop("broken stack frame"); 844 bind(L); 845 } 846 if (verifyoop) { 847 verify_oop(rax, state); 848 } 849 850 address* const safepoint_table = Interpreter::safept_table(state); 851 #ifdef _LP64 852 Label no_safepoint, dispatch; 853 if (SafepointMechanism::uses_thread_local_poll() && table != safepoint_table && generate_poll) { 854 NOT_PRODUCT(block_comment("Thread-local Safepoint poll")); 855 testb(Address(r15_thread, Thread::polling_page_offset()), SafepointMechanism::poll_bit()); 856 857 jccb(Assembler::zero, no_safepoint); 858 lea(rscratch1, ExternalAddress((address)safepoint_table)); 859 jmpb(dispatch); 860 } 861 862 bind(no_safepoint); 863 lea(rscratch1, ExternalAddress((address)table)); 864 bind(dispatch); 865 jmp(Address(rscratch1, rbx, Address::times_8)); 866 867 #else 868 Address index(noreg, rbx, Address::times_ptr); 869 if (SafepointMechanism::uses_thread_local_poll() && table != safepoint_table && generate_poll) { 870 NOT_PRODUCT(block_comment("Thread-local Safepoint poll")); 871 Label no_safepoint; 872 const Register thread = rcx; 873 get_thread(thread); 874 testb(Address(thread, Thread::polling_page_offset()), SafepointMechanism::poll_bit()); 875 876 jccb(Assembler::zero, no_safepoint); 877 ArrayAddress dispatch_addr(ExternalAddress((address)safepoint_table), index); 878 jump(dispatch_addr); 879 bind(no_safepoint); 880 } 881 882 { 883 ArrayAddress dispatch_addr(ExternalAddress((address)table), index); 884 jump(dispatch_addr); 885 } 886 #endif // _LP64 887 } 888 889 void InterpreterMacroAssembler::dispatch_only(TosState state, bool generate_poll) { 890 dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll); 891 } 892 893 void InterpreterMacroAssembler::dispatch_only_normal(TosState state) { 894 dispatch_base(state, Interpreter::normal_table(state)); 895 } 896 897 void InterpreterMacroAssembler::dispatch_only_noverify(TosState state) { 898 dispatch_base(state, Interpreter::normal_table(state), false); 899 } 900 901 902 void InterpreterMacroAssembler::dispatch_next(TosState state, int step, bool generate_poll) { 903 // load next bytecode (load before advancing _bcp_register to prevent AGI) 904 load_unsigned_byte(rbx, Address(_bcp_register, step)); 905 // advance _bcp_register 906 increment(_bcp_register, step); 907 dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll); 908 } 909 910 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) { 911 // load current bytecode 912 load_unsigned_byte(rbx, Address(_bcp_register, 0)); 913 dispatch_base(state, table); 914 } 915 916 void InterpreterMacroAssembler::narrow(Register result) { 917 918 // Get method->_constMethod->_result_type 919 movptr(rcx, Address(rbp, frame::interpreter_frame_method_offset * wordSize)); 920 movptr(rcx, Address(rcx, Method::const_offset())); 921 load_unsigned_byte(rcx, Address(rcx, ConstMethod::result_type_offset())); 922 923 Label done, notBool, notByte, notChar; 924 925 // common case first 926 cmpl(rcx, T_INT); 927 jcc(Assembler::equal, done); 928 929 // mask integer result to narrower return type. 930 cmpl(rcx, T_BOOLEAN); 931 jcc(Assembler::notEqual, notBool); 932 andl(result, 0x1); 933 jmp(done); 934 935 bind(notBool); 936 cmpl(rcx, T_BYTE); 937 jcc(Assembler::notEqual, notByte); 938 LP64_ONLY(movsbl(result, result);) 939 NOT_LP64(shll(result, 24);) // truncate upper 24 bits 940 NOT_LP64(sarl(result, 24);) // and sign-extend byte 941 jmp(done); 942 943 bind(notByte); 944 cmpl(rcx, T_CHAR); 945 jcc(Assembler::notEqual, notChar); 946 LP64_ONLY(movzwl(result, result);) 947 NOT_LP64(andl(result, 0xFFFF);) // truncate upper 16 bits 948 jmp(done); 949 950 bind(notChar); 951 // cmpl(rcx, T_SHORT); // all that's left 952 // jcc(Assembler::notEqual, done); 953 LP64_ONLY(movswl(result, result);) 954 NOT_LP64(shll(result, 16);) // truncate upper 16 bits 955 NOT_LP64(sarl(result, 16);) // and sign-extend short 956 957 // Nothing to do for T_INT 958 bind(done); 959 } 960 961 // remove activation 962 // 963 // Unlock the receiver if this is a synchronized method. 964 // Unlock any Java monitors from syncronized blocks. 965 // Remove the activation from the stack. 966 // 967 // If there are locked Java monitors 968 // If throw_monitor_exception 969 // throws IllegalMonitorStateException 970 // Else if install_monitor_exception 971 // installs IllegalMonitorStateException 972 // Else 973 // no error processing 974 void InterpreterMacroAssembler::remove_activation( 975 TosState state, 976 Register ret_addr, 977 bool throw_monitor_exception, 978 bool install_monitor_exception, 979 bool notify_jvmdi) { 980 // Note: Registers rdx xmm0 may be in use for the 981 // result check if synchronized method 982 Label unlocked, unlock, no_unlock; 983 984 const Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx); 985 const Register robj = LP64_ONLY(c_rarg1) NOT_LP64(rdx); 986 const Register rmon = LP64_ONLY(c_rarg1) NOT_LP64(rcx); 987 // monitor pointers need different register 988 // because rdx may have the result in it 989 NOT_LP64(get_thread(rcx);) 990 991 // get the value of _do_not_unlock_if_synchronized into rdx 992 const Address do_not_unlock_if_synchronized(rthread, 993 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset())); 994 movbool(rbx, do_not_unlock_if_synchronized); 995 movbool(do_not_unlock_if_synchronized, false); // reset the flag 996 997 // get method access flags 998 movptr(rcx, Address(rbp, frame::interpreter_frame_method_offset * wordSize)); 999 movl(rcx, Address(rcx, Method::access_flags_offset())); 1000 testl(rcx, JVM_ACC_SYNCHRONIZED); 1001 jcc(Assembler::zero, unlocked); 1002 1003 // Don't unlock anything if the _do_not_unlock_if_synchronized flag 1004 // is set. 1005 testbool(rbx); 1006 jcc(Assembler::notZero, no_unlock); 1007 1008 // unlock monitor 1009 push(state); // save result 1010 1011 // BasicObjectLock will be first in list, since this is a 1012 // synchronized method. However, need to check that the object has 1013 // not been unlocked by an explicit monitorexit bytecode. 1014 const Address monitor(rbp, frame::interpreter_frame_initial_sp_offset * 1015 wordSize - (int) sizeof(BasicObjectLock)); 1016 // We use c_rarg1/rdx so that if we go slow path it will be the correct 1017 // register for unlock_object to pass to VM directly 1018 lea(robj, monitor); // address of first monitor 1019 1020 movptr(rax, Address(robj, BasicObjectLock::obj_offset_in_bytes())); 1021 testptr(rax, rax); 1022 jcc(Assembler::notZero, unlock); 1023 1024 pop(state); 1025 if (throw_monitor_exception) { 1026 // Entry already unlocked, need to throw exception 1027 NOT_LP64(empty_FPU_stack();) // remove possible return value from FPU-stack, otherwise stack could overflow 1028 call_VM(noreg, CAST_FROM_FN_PTR(address, 1029 InterpreterRuntime::throw_illegal_monitor_state_exception)); 1030 should_not_reach_here(); 1031 } else { 1032 // Monitor already unlocked during a stack unroll. If requested, 1033 // install an illegal_monitor_state_exception. Continue with 1034 // stack unrolling. 1035 if (install_monitor_exception) { 1036 NOT_LP64(empty_FPU_stack();) 1037 call_VM(noreg, CAST_FROM_FN_PTR(address, 1038 InterpreterRuntime::new_illegal_monitor_state_exception)); 1039 } 1040 jmp(unlocked); 1041 } 1042 1043 bind(unlock); 1044 unlock_object(robj); 1045 pop(state); 1046 1047 // Check that for block-structured locking (i.e., that all locked 1048 // objects has been unlocked) 1049 bind(unlocked); 1050 1051 // rax, rdx: Might contain return value 1052 1053 // Check that all monitors are unlocked 1054 { 1055 Label loop, exception, entry, restart; 1056 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize; 1057 const Address monitor_block_top( 1058 rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize); 1059 const Address monitor_block_bot( 1060 rbp, frame::interpreter_frame_initial_sp_offset * wordSize); 1061 1062 bind(restart); 1063 // We use c_rarg1 so that if we go slow path it will be the correct 1064 // register for unlock_object to pass to VM directly 1065 movptr(rmon, monitor_block_top); // points to current entry, starting 1066 // with top-most entry 1067 lea(rbx, monitor_block_bot); // points to word before bottom of 1068 // monitor block 1069 jmp(entry); 1070 1071 // Entry already locked, need to throw exception 1072 bind(exception); 1073 1074 if (throw_monitor_exception) { 1075 // Throw exception 1076 NOT_LP64(empty_FPU_stack();) 1077 MacroAssembler::call_VM(noreg, 1078 CAST_FROM_FN_PTR(address, InterpreterRuntime:: 1079 throw_illegal_monitor_state_exception)); 1080 should_not_reach_here(); 1081 } else { 1082 // Stack unrolling. Unlock object and install illegal_monitor_exception. 1083 // Unlock does not block, so don't have to worry about the frame. 1084 // We don't have to preserve c_rarg1 since we are going to throw an exception. 1085 1086 push(state); 1087 mov(robj, rmon); // nop if robj and rmon are the same 1088 unlock_object(robj); 1089 pop(state); 1090 1091 if (install_monitor_exception) { 1092 NOT_LP64(empty_FPU_stack();) 1093 call_VM(noreg, CAST_FROM_FN_PTR(address, 1094 InterpreterRuntime:: 1095 new_illegal_monitor_state_exception)); 1096 } 1097 1098 jmp(restart); 1099 } 1100 1101 bind(loop); 1102 // check if current entry is used 1103 cmpptr(Address(rmon, BasicObjectLock::obj_offset_in_bytes()), (int32_t) NULL); 1104 jcc(Assembler::notEqual, exception); 1105 1106 addptr(rmon, entry_size); // otherwise advance to next entry 1107 bind(entry); 1108 cmpptr(rmon, rbx); // check if bottom reached 1109 jcc(Assembler::notEqual, loop); // if not at bottom then check this entry 1110 } 1111 1112 bind(no_unlock); 1113 1114 // jvmti support 1115 if (notify_jvmdi) { 1116 notify_method_exit(state, NotifyJVMTI); // preserve TOSCA 1117 } else { 1118 notify_method_exit(state, SkipNotifyJVMTI); // preserve TOSCA 1119 } 1120 1121 if (StackReservedPages > 0) { 1122 movptr(rbx, 1123 Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); 1124 // testing if reserved zone needs to be re-enabled 1125 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx); 1126 Label no_reserved_zone_enabling; 1127 1128 NOT_LP64(get_thread(rthread);) 1129 1130 cmpl(Address(rthread, JavaThread::stack_guard_state_offset()), JavaThread::stack_guard_enabled); 1131 jcc(Assembler::equal, no_reserved_zone_enabling); 1132 1133 cmpptr(rbx, Address(rthread, JavaThread::reserved_stack_activation_offset())); 1134 jcc(Assembler::lessEqual, no_reserved_zone_enabling); 1135 1136 call_VM_leaf( 1137 CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), rthread); 1138 call_VM(noreg, CAST_FROM_FN_PTR(address, 1139 InterpreterRuntime::throw_delayed_StackOverflowError)); 1140 should_not_reach_here(); 1141 1142 bind(no_reserved_zone_enabling); 1143 } 1144 1145 // remove activation 1146 // get sender sp 1147 movptr(rbx, 1148 Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); 1149 1150 if (state == atos && ValueTypeReturnedAsFields) { 1151 Label skip; 1152 // Test if the return type is a value type 1153 movptr(rdi, Address(rbp, frame::interpreter_frame_method_offset * wordSize)); 1154 movptr(rdi, Address(rdi, Method::const_offset())); 1155 load_unsigned_byte(rdi, Address(rdi, ConstMethod::result_type_offset())); 1156 cmpl(rdi, T_VALUETYPE); 1157 jcc(Assembler::notEqual, skip); 1158 1159 // We are returning a value type, load its fields into registers 1160 #ifndef _LP64 1161 super_call_VM_leaf(StubRoutines::load_value_type_fields_in_regs()); 1162 #else 1163 // Load fields from a buffered value with a value class specific handler 1164 load_klass(rdi, rax); 1165 movptr(rdi, Address(rdi, InstanceKlass::adr_valueklass_fixed_block_offset())); 1166 movptr(rdi, Address(rdi, ValueKlass::unpack_handler_offset())); 1167 1168 testptr(rdi, rdi); 1169 jcc(Assembler::equal, skip); 1170 1171 call(rdi); 1172 #endif 1173 // call above kills the value in rbx. Reload it. 1174 movptr(rbx, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); 1175 bind(skip); 1176 } 1177 leave(); // remove frame anchor 1178 pop(ret_addr); // get return address 1179 mov(rsp, rbx); // set sp to sender sp 1180 } 1181 1182 void InterpreterMacroAssembler::get_method_counters(Register method, 1183 Register mcs, Label& skip) { 1184 Label has_counters; 1185 movptr(mcs, Address(method, Method::method_counters_offset())); 1186 testptr(mcs, mcs); 1187 jcc(Assembler::notZero, has_counters); 1188 call_VM(noreg, CAST_FROM_FN_PTR(address, 1189 InterpreterRuntime::build_method_counters), method); 1190 movptr(mcs, Address(method,Method::method_counters_offset())); 1191 testptr(mcs, mcs); 1192 jcc(Assembler::zero, skip); // No MethodCounters allocated, OutOfMemory 1193 bind(has_counters); 1194 } 1195 1196 1197 // Lock object 1198 // 1199 // Args: 1200 // rdx, c_rarg1: BasicObjectLock to be used for locking 1201 // 1202 // Kills: 1203 // rax, rbx 1204 void InterpreterMacroAssembler::lock_object(Register lock_reg) { 1205 assert(lock_reg == LP64_ONLY(c_rarg1) NOT_LP64(rdx), 1206 "The argument is only for looks. It must be c_rarg1"); 1207 1208 if (UseHeavyMonitors) { 1209 call_VM(noreg, 1210 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), 1211 lock_reg); 1212 } else { 1213 Label done; 1214 1215 const Register swap_reg = rax; // Must use rax for cmpxchg instruction 1216 const Register tmp_reg = rbx; // Will be passed to biased_locking_enter to avoid a 1217 // problematic case where tmp_reg = no_reg. 1218 const Register obj_reg = LP64_ONLY(c_rarg3) NOT_LP64(rcx); // Will contain the oop 1219 1220 const int obj_offset = BasicObjectLock::obj_offset_in_bytes(); 1221 const int lock_offset = BasicObjectLock::lock_offset_in_bytes (); 1222 const int mark_offset = lock_offset + 1223 BasicLock::displaced_header_offset_in_bytes(); 1224 1225 Label slow_case; 1226 1227 // Load object pointer into obj_reg 1228 movptr(obj_reg, Address(lock_reg, obj_offset)); 1229 1230 if (UseBiasedLocking) { 1231 biased_locking_enter(lock_reg, obj_reg, swap_reg, tmp_reg, false, done, &slow_case); 1232 } 1233 1234 // Load immediate 1 into swap_reg %rax 1235 movl(swap_reg, (int32_t)1); 1236 1237 // Load (object->mark() | 1) into swap_reg %rax 1238 orptr(swap_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes())); 1239 if (EnableValhalla && !UseBiasedLocking) { 1240 // For slow path is_always_locked, using biased, which is never natural for !UseBiasLocking 1241 andptr(swap_reg, ~((int) markWord::biased_lock_bit_in_place)); 1242 } 1243 1244 // Save (object->mark() | 1) into BasicLock's displaced header 1245 movptr(Address(lock_reg, mark_offset), swap_reg); 1246 1247 assert(lock_offset == 0, 1248 "displaced header must be first word in BasicObjectLock"); 1249 1250 lock(); 1251 cmpxchgptr(lock_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes())); 1252 if (PrintBiasedLockingStatistics) { 1253 cond_inc32(Assembler::zero, 1254 ExternalAddress((address) BiasedLocking::fast_path_entry_count_addr())); 1255 } 1256 jcc(Assembler::zero, done); 1257 1258 const int zero_bits = LP64_ONLY(7) NOT_LP64(3); 1259 1260 // Test if the oopMark is an obvious stack pointer, i.e., 1261 // 1) (mark & zero_bits) == 0, and 1262 // 2) rsp <= mark < mark + os::pagesize() 1263 // 1264 // These 3 tests can be done by evaluating the following 1265 // expression: ((mark - rsp) & (zero_bits - os::vm_page_size())), 1266 // assuming both stack pointer and pagesize have their 1267 // least significant bits clear. 1268 // NOTE: the oopMark is in swap_reg %rax as the result of cmpxchg 1269 subptr(swap_reg, rsp); 1270 andptr(swap_reg, zero_bits - os::vm_page_size()); 1271 1272 // Save the test result, for recursive case, the result is zero 1273 movptr(Address(lock_reg, mark_offset), swap_reg); 1274 1275 if (PrintBiasedLockingStatistics) { 1276 cond_inc32(Assembler::zero, 1277 ExternalAddress((address) BiasedLocking::fast_path_entry_count_addr())); 1278 } 1279 jcc(Assembler::zero, done); 1280 1281 bind(slow_case); 1282 1283 // Call the runtime routine for slow case 1284 call_VM(noreg, 1285 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), 1286 lock_reg); 1287 1288 bind(done); 1289 } 1290 } 1291 1292 1293 // Unlocks an object. Used in monitorexit bytecode and 1294 // remove_activation. Throws an IllegalMonitorException if object is 1295 // not locked by current thread. 1296 // 1297 // Args: 1298 // rdx, c_rarg1: BasicObjectLock for lock 1299 // 1300 // Kills: 1301 // rax 1302 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ... (param regs) 1303 // rscratch1 (scratch reg) 1304 // rax, rbx, rcx, rdx 1305 void InterpreterMacroAssembler::unlock_object(Register lock_reg) { 1306 assert(lock_reg == LP64_ONLY(c_rarg1) NOT_LP64(rdx), 1307 "The argument is only for looks. It must be c_rarg1"); 1308 1309 if (UseHeavyMonitors) { 1310 call_VM(noreg, 1311 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), 1312 lock_reg); 1313 } else { 1314 Label done; 1315 1316 const Register swap_reg = rax; // Must use rax for cmpxchg instruction 1317 const Register header_reg = LP64_ONLY(c_rarg2) NOT_LP64(rbx); // Will contain the old oopMark 1318 const Register obj_reg = LP64_ONLY(c_rarg3) NOT_LP64(rcx); // Will contain the oop 1319 1320 save_bcp(); // Save in case of exception 1321 1322 // Convert from BasicObjectLock structure to object and BasicLock 1323 // structure Store the BasicLock address into %rax 1324 lea(swap_reg, Address(lock_reg, BasicObjectLock::lock_offset_in_bytes())); 1325 1326 // Load oop into obj_reg(%c_rarg3) 1327 movptr(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset_in_bytes())); 1328 1329 // Free entry 1330 movptr(Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()), (int32_t)NULL_WORD); 1331 1332 if (UseBiasedLocking) { 1333 biased_locking_exit(obj_reg, header_reg, done); 1334 } 1335 1336 // Load the old header from BasicLock structure 1337 movptr(header_reg, Address(swap_reg, 1338 BasicLock::displaced_header_offset_in_bytes())); 1339 1340 // Test for recursion 1341 testptr(header_reg, header_reg); 1342 1343 // zero for recursive case 1344 jcc(Assembler::zero, done); 1345 1346 // Atomic swap back the old header 1347 lock(); 1348 cmpxchgptr(header_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes())); 1349 1350 // zero for simple unlock of a stack-lock case 1351 jcc(Assembler::zero, done); 1352 1353 // Call the runtime routine for slow case. 1354 movptr(Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()), 1355 obj_reg); // restore obj 1356 call_VM(noreg, 1357 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), 1358 lock_reg); 1359 1360 bind(done); 1361 1362 restore_bcp(); 1363 } 1364 } 1365 1366 void InterpreterMacroAssembler::test_method_data_pointer(Register mdp, 1367 Label& zero_continue) { 1368 assert(ProfileInterpreter, "must be profiling interpreter"); 1369 movptr(mdp, Address(rbp, frame::interpreter_frame_mdp_offset * wordSize)); 1370 testptr(mdp, mdp); 1371 jcc(Assembler::zero, zero_continue); 1372 } 1373 1374 1375 // Set the method data pointer for the current bcp. 1376 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() { 1377 assert(ProfileInterpreter, "must be profiling interpreter"); 1378 Label set_mdp; 1379 push(rax); 1380 push(rbx); 1381 1382 get_method(rbx); 1383 // Test MDO to avoid the call if it is NULL. 1384 movptr(rax, Address(rbx, in_bytes(Method::method_data_offset()))); 1385 testptr(rax, rax); 1386 jcc(Assembler::zero, set_mdp); 1387 // rbx: method 1388 // _bcp_register: bcp 1389 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), rbx, _bcp_register); 1390 // rax: mdi 1391 // mdo is guaranteed to be non-zero here, we checked for it before the call. 1392 movptr(rbx, Address(rbx, in_bytes(Method::method_data_offset()))); 1393 addptr(rbx, in_bytes(MethodData::data_offset())); 1394 addptr(rax, rbx); 1395 bind(set_mdp); 1396 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), rax); 1397 pop(rbx); 1398 pop(rax); 1399 } 1400 1401 void InterpreterMacroAssembler::verify_method_data_pointer() { 1402 assert(ProfileInterpreter, "must be profiling interpreter"); 1403 #ifdef ASSERT 1404 Label verify_continue; 1405 push(rax); 1406 push(rbx); 1407 Register arg3_reg = LP64_ONLY(c_rarg3) NOT_LP64(rcx); 1408 Register arg2_reg = LP64_ONLY(c_rarg2) NOT_LP64(rdx); 1409 push(arg3_reg); 1410 push(arg2_reg); 1411 test_method_data_pointer(arg3_reg, verify_continue); // If mdp is zero, continue 1412 get_method(rbx); 1413 1414 // If the mdp is valid, it will point to a DataLayout header which is 1415 // consistent with the bcp. The converse is highly probable also. 1416 load_unsigned_short(arg2_reg, 1417 Address(arg3_reg, in_bytes(DataLayout::bci_offset()))); 1418 addptr(arg2_reg, Address(rbx, Method::const_offset())); 1419 lea(arg2_reg, Address(arg2_reg, ConstMethod::codes_offset())); 1420 cmpptr(arg2_reg, _bcp_register); 1421 jcc(Assembler::equal, verify_continue); 1422 // rbx: method 1423 // _bcp_register: bcp 1424 // c_rarg3: mdp 1425 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp), 1426 rbx, _bcp_register, arg3_reg); 1427 bind(verify_continue); 1428 pop(arg2_reg); 1429 pop(arg3_reg); 1430 pop(rbx); 1431 pop(rax); 1432 #endif // ASSERT 1433 } 1434 1435 1436 void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in, 1437 int constant, 1438 Register value) { 1439 assert(ProfileInterpreter, "must be profiling interpreter"); 1440 Address data(mdp_in, constant); 1441 movptr(data, value); 1442 } 1443 1444 1445 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in, 1446 int constant, 1447 bool decrement) { 1448 // Counter address 1449 Address data(mdp_in, constant); 1450 1451 increment_mdp_data_at(data, decrement); 1452 } 1453 1454 void InterpreterMacroAssembler::increment_mdp_data_at(Address data, 1455 bool decrement) { 1456 assert(ProfileInterpreter, "must be profiling interpreter"); 1457 // %%% this does 64bit counters at best it is wasting space 1458 // at worst it is a rare bug when counters overflow 1459 1460 if (decrement) { 1461 // Decrement the register. Set condition codes. 1462 addptr(data, (int32_t) -DataLayout::counter_increment); 1463 // If the decrement causes the counter to overflow, stay negative 1464 Label L; 1465 jcc(Assembler::negative, L); 1466 addptr(data, (int32_t) DataLayout::counter_increment); 1467 bind(L); 1468 } else { 1469 assert(DataLayout::counter_increment == 1, 1470 "flow-free idiom only works with 1"); 1471 // Increment the register. Set carry flag. 1472 addptr(data, DataLayout::counter_increment); 1473 // If the increment causes the counter to overflow, pull back by 1. 1474 sbbptr(data, (int32_t)0); 1475 } 1476 } 1477 1478 1479 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in, 1480 Register reg, 1481 int constant, 1482 bool decrement) { 1483 Address data(mdp_in, reg, Address::times_1, constant); 1484 1485 increment_mdp_data_at(data, decrement); 1486 } 1487 1488 void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in, 1489 int flag_byte_constant) { 1490 assert(ProfileInterpreter, "must be profiling interpreter"); 1491 int header_offset = in_bytes(DataLayout::flags_offset()); 1492 int header_bits = flag_byte_constant; 1493 // Set the flag 1494 orb(Address(mdp_in, header_offset), header_bits); 1495 } 1496 1497 1498 1499 void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in, 1500 int offset, 1501 Register value, 1502 Register test_value_out, 1503 Label& not_equal_continue) { 1504 assert(ProfileInterpreter, "must be profiling interpreter"); 1505 if (test_value_out == noreg) { 1506 cmpptr(value, Address(mdp_in, offset)); 1507 } else { 1508 // Put the test value into a register, so caller can use it: 1509 movptr(test_value_out, Address(mdp_in, offset)); 1510 cmpptr(test_value_out, value); 1511 } 1512 jcc(Assembler::notEqual, not_equal_continue); 1513 } 1514 1515 1516 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in, 1517 int offset_of_disp) { 1518 assert(ProfileInterpreter, "must be profiling interpreter"); 1519 Address disp_address(mdp_in, offset_of_disp); 1520 addptr(mdp_in, disp_address); 1521 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in); 1522 } 1523 1524 1525 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in, 1526 Register reg, 1527 int offset_of_disp) { 1528 assert(ProfileInterpreter, "must be profiling interpreter"); 1529 Address disp_address(mdp_in, reg, Address::times_1, offset_of_disp); 1530 addptr(mdp_in, disp_address); 1531 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in); 1532 } 1533 1534 1535 void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in, 1536 int constant) { 1537 assert(ProfileInterpreter, "must be profiling interpreter"); 1538 addptr(mdp_in, constant); 1539 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in); 1540 } 1541 1542 1543 void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) { 1544 assert(ProfileInterpreter, "must be profiling interpreter"); 1545 push(return_bci); // save/restore across call_VM 1546 call_VM(noreg, 1547 CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret), 1548 return_bci); 1549 pop(return_bci); 1550 } 1551 1552 1553 void InterpreterMacroAssembler::profile_taken_branch(Register mdp, 1554 Register bumped_count) { 1555 if (ProfileInterpreter) { 1556 Label profile_continue; 1557 1558 // If no method data exists, go to profile_continue. 1559 // Otherwise, assign to mdp 1560 test_method_data_pointer(mdp, profile_continue); 1561 1562 // We are taking a branch. Increment the taken count. 1563 // We inline increment_mdp_data_at to return bumped_count in a register 1564 //increment_mdp_data_at(mdp, in_bytes(JumpData::taken_offset())); 1565 Address data(mdp, in_bytes(JumpData::taken_offset())); 1566 movptr(bumped_count, data); 1567 assert(DataLayout::counter_increment == 1, 1568 "flow-free idiom only works with 1"); 1569 addptr(bumped_count, DataLayout::counter_increment); 1570 sbbptr(bumped_count, 0); 1571 movptr(data, bumped_count); // Store back out 1572 1573 // The method data pointer needs to be updated to reflect the new target. 1574 update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset())); 1575 bind(profile_continue); 1576 } 1577 } 1578 1579 1580 void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp) { 1581 if (ProfileInterpreter) { 1582 Label profile_continue; 1583 1584 // If no method data exists, go to profile_continue. 1585 test_method_data_pointer(mdp, profile_continue); 1586 1587 // We are taking a branch. Increment the not taken count. 1588 increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset())); 1589 1590 // The method data pointer needs to be updated to correspond to 1591 // the next bytecode 1592 update_mdp_by_constant(mdp, in_bytes(BranchData::branch_data_size())); 1593 bind(profile_continue); 1594 } 1595 } 1596 1597 void InterpreterMacroAssembler::profile_call(Register mdp) { 1598 if (ProfileInterpreter) { 1599 Label profile_continue; 1600 1601 // If no method data exists, go to profile_continue. 1602 test_method_data_pointer(mdp, profile_continue); 1603 1604 // We are making a call. Increment the count. 1605 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); 1606 1607 // The method data pointer needs to be updated to reflect the new target. 1608 update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size())); 1609 bind(profile_continue); 1610 } 1611 } 1612 1613 1614 void InterpreterMacroAssembler::profile_final_call(Register mdp) { 1615 if (ProfileInterpreter) { 1616 Label profile_continue; 1617 1618 // If no method data exists, go to profile_continue. 1619 test_method_data_pointer(mdp, profile_continue); 1620 1621 // We are making a call. Increment the count. 1622 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); 1623 1624 // The method data pointer needs to be updated to reflect the new target. 1625 update_mdp_by_constant(mdp, 1626 in_bytes(VirtualCallData:: 1627 virtual_call_data_size())); 1628 bind(profile_continue); 1629 } 1630 } 1631 1632 1633 void InterpreterMacroAssembler::profile_virtual_call(Register receiver, 1634 Register mdp, 1635 Register reg2, 1636 bool receiver_can_be_null) { 1637 if (ProfileInterpreter) { 1638 Label profile_continue; 1639 1640 // If no method data exists, go to profile_continue. 1641 test_method_data_pointer(mdp, profile_continue); 1642 1643 Label skip_receiver_profile; 1644 if (receiver_can_be_null) { 1645 Label not_null; 1646 testptr(receiver, receiver); 1647 jccb(Assembler::notZero, not_null); 1648 // We are making a call. Increment the count for null receiver. 1649 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); 1650 jmp(skip_receiver_profile); 1651 bind(not_null); 1652 } 1653 1654 // Record the receiver type. 1655 record_klass_in_profile(receiver, mdp, reg2, true); 1656 bind(skip_receiver_profile); 1657 1658 // The method data pointer needs to be updated to reflect the new target. 1659 #if INCLUDE_JVMCI 1660 if (MethodProfileWidth == 0) { 1661 update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size())); 1662 } 1663 #else // INCLUDE_JVMCI 1664 update_mdp_by_constant(mdp, 1665 in_bytes(VirtualCallData:: 1666 virtual_call_data_size())); 1667 #endif // INCLUDE_JVMCI 1668 bind(profile_continue); 1669 } 1670 } 1671 1672 #if INCLUDE_JVMCI 1673 void InterpreterMacroAssembler::profile_called_method(Register method, Register mdp, Register reg2) { 1674 assert_different_registers(method, mdp, reg2); 1675 if (ProfileInterpreter && MethodProfileWidth > 0) { 1676 Label profile_continue; 1677 1678 // If no method data exists, go to profile_continue. 1679 test_method_data_pointer(mdp, profile_continue); 1680 1681 Label done; 1682 record_item_in_profile_helper(method, mdp, reg2, 0, done, MethodProfileWidth, 1683 &VirtualCallData::method_offset, &VirtualCallData::method_count_offset, in_bytes(VirtualCallData::nonprofiled_receiver_count_offset())); 1684 bind(done); 1685 1686 update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size())); 1687 bind(profile_continue); 1688 } 1689 } 1690 #endif // INCLUDE_JVMCI 1691 1692 // This routine creates a state machine for updating the multi-row 1693 // type profile at a virtual call site (or other type-sensitive bytecode). 1694 // The machine visits each row (of receiver/count) until the receiver type 1695 // is found, or until it runs out of rows. At the same time, it remembers 1696 // the location of the first empty row. (An empty row records null for its 1697 // receiver, and can be allocated for a newly-observed receiver type.) 1698 // Because there are two degrees of freedom in the state, a simple linear 1699 // search will not work; it must be a decision tree. Hence this helper 1700 // function is recursive, to generate the required tree structured code. 1701 // It's the interpreter, so we are trading off code space for speed. 1702 // See below for example code. 1703 void InterpreterMacroAssembler::record_klass_in_profile_helper( 1704 Register receiver, Register mdp, 1705 Register reg2, int start_row, 1706 Label& done, bool is_virtual_call) { 1707 if (TypeProfileWidth == 0) { 1708 if (is_virtual_call) { 1709 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); 1710 } 1711 #if INCLUDE_JVMCI 1712 else if (EnableJVMCI) { 1713 increment_mdp_data_at(mdp, in_bytes(ReceiverTypeData::nonprofiled_receiver_count_offset())); 1714 } 1715 #endif // INCLUDE_JVMCI 1716 } else { 1717 int non_profiled_offset = -1; 1718 if (is_virtual_call) { 1719 non_profiled_offset = in_bytes(CounterData::count_offset()); 1720 } 1721 #if INCLUDE_JVMCI 1722 else if (EnableJVMCI) { 1723 non_profiled_offset = in_bytes(ReceiverTypeData::nonprofiled_receiver_count_offset()); 1724 } 1725 #endif // INCLUDE_JVMCI 1726 1727 record_item_in_profile_helper(receiver, mdp, reg2, 0, done, TypeProfileWidth, 1728 &VirtualCallData::receiver_offset, &VirtualCallData::receiver_count_offset, non_profiled_offset); 1729 } 1730 } 1731 1732 void InterpreterMacroAssembler::record_item_in_profile_helper(Register item, Register mdp, 1733 Register reg2, int start_row, Label& done, int total_rows, 1734 OffsetFunction item_offset_fn, OffsetFunction item_count_offset_fn, 1735 int non_profiled_offset) { 1736 int last_row = total_rows - 1; 1737 assert(start_row <= last_row, "must be work left to do"); 1738 // Test this row for both the item and for null. 1739 // Take any of three different outcomes: 1740 // 1. found item => increment count and goto done 1741 // 2. found null => keep looking for case 1, maybe allocate this cell 1742 // 3. found something else => keep looking for cases 1 and 2 1743 // Case 3 is handled by a recursive call. 1744 for (int row = start_row; row <= last_row; row++) { 1745 Label next_test; 1746 bool test_for_null_also = (row == start_row); 1747 1748 // See if the item is item[n]. 1749 int item_offset = in_bytes(item_offset_fn(row)); 1750 test_mdp_data_at(mdp, item_offset, item, 1751 (test_for_null_also ? reg2 : noreg), 1752 next_test); 1753 // (Reg2 now contains the item from the CallData.) 1754 1755 // The item is item[n]. Increment count[n]. 1756 int count_offset = in_bytes(item_count_offset_fn(row)); 1757 increment_mdp_data_at(mdp, count_offset); 1758 jmp(done); 1759 bind(next_test); 1760 1761 if (test_for_null_also) { 1762 // Failed the equality check on item[n]... Test for null. 1763 testptr(reg2, reg2); 1764 if (start_row == last_row) { 1765 // The only thing left to do is handle the null case. 1766 if (non_profiled_offset >= 0) { 1767 Label found_null; 1768 jccb(Assembler::zero, found_null); 1769 // Item did not match any saved item and there is no empty row for it. 1770 // Increment total counter to indicate polymorphic case. 1771 increment_mdp_data_at(mdp, non_profiled_offset); 1772 jmp(done); 1773 bind(found_null); 1774 } else { 1775 jcc(Assembler::notZero, done); 1776 } 1777 break; 1778 } 1779 Label found_null; 1780 // Since null is rare, make it be the branch-taken case. 1781 jcc(Assembler::zero, found_null); 1782 1783 // Put all the "Case 3" tests here. 1784 record_item_in_profile_helper(item, mdp, reg2, start_row + 1, done, total_rows, 1785 item_offset_fn, item_count_offset_fn, non_profiled_offset); 1786 1787 // Found a null. Keep searching for a matching item, 1788 // but remember that this is an empty (unused) slot. 1789 bind(found_null); 1790 } 1791 } 1792 1793 // In the fall-through case, we found no matching item, but we 1794 // observed the item[start_row] is NULL. 1795 1796 // Fill in the item field and increment the count. 1797 int item_offset = in_bytes(item_offset_fn(start_row)); 1798 set_mdp_data_at(mdp, item_offset, item); 1799 int count_offset = in_bytes(item_count_offset_fn(start_row)); 1800 movl(reg2, DataLayout::counter_increment); 1801 set_mdp_data_at(mdp, count_offset, reg2); 1802 if (start_row > 0) { 1803 jmp(done); 1804 } 1805 } 1806 1807 // Example state machine code for three profile rows: 1808 // // main copy of decision tree, rooted at row[1] 1809 // if (row[0].rec == rec) { row[0].incr(); goto done; } 1810 // if (row[0].rec != NULL) { 1811 // // inner copy of decision tree, rooted at row[1] 1812 // if (row[1].rec == rec) { row[1].incr(); goto done; } 1813 // if (row[1].rec != NULL) { 1814 // // degenerate decision tree, rooted at row[2] 1815 // if (row[2].rec == rec) { row[2].incr(); goto done; } 1816 // if (row[2].rec != NULL) { count.incr(); goto done; } // overflow 1817 // row[2].init(rec); goto done; 1818 // } else { 1819 // // remember row[1] is empty 1820 // if (row[2].rec == rec) { row[2].incr(); goto done; } 1821 // row[1].init(rec); goto done; 1822 // } 1823 // } else { 1824 // // remember row[0] is empty 1825 // if (row[1].rec == rec) { row[1].incr(); goto done; } 1826 // if (row[2].rec == rec) { row[2].incr(); goto done; } 1827 // row[0].init(rec); goto done; 1828 // } 1829 // done: 1830 1831 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver, 1832 Register mdp, Register reg2, 1833 bool is_virtual_call) { 1834 assert(ProfileInterpreter, "must be profiling"); 1835 Label done; 1836 1837 record_klass_in_profile_helper(receiver, mdp, reg2, 0, done, is_virtual_call); 1838 1839 bind (done); 1840 } 1841 1842 void InterpreterMacroAssembler::profile_ret(Register return_bci, 1843 Register mdp) { 1844 if (ProfileInterpreter) { 1845 Label profile_continue; 1846 uint row; 1847 1848 // If no method data exists, go to profile_continue. 1849 test_method_data_pointer(mdp, profile_continue); 1850 1851 // Update the total ret count. 1852 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); 1853 1854 for (row = 0; row < RetData::row_limit(); row++) { 1855 Label next_test; 1856 1857 // See if return_bci is equal to bci[n]: 1858 test_mdp_data_at(mdp, 1859 in_bytes(RetData::bci_offset(row)), 1860 return_bci, noreg, 1861 next_test); 1862 1863 // return_bci is equal to bci[n]. Increment the count. 1864 increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row))); 1865 1866 // The method data pointer needs to be updated to reflect the new target. 1867 update_mdp_by_offset(mdp, 1868 in_bytes(RetData::bci_displacement_offset(row))); 1869 jmp(profile_continue); 1870 bind(next_test); 1871 } 1872 1873 update_mdp_for_ret(return_bci); 1874 1875 bind(profile_continue); 1876 } 1877 } 1878 1879 1880 void InterpreterMacroAssembler::profile_null_seen(Register mdp) { 1881 if (ProfileInterpreter) { 1882 Label profile_continue; 1883 1884 // If no method data exists, go to profile_continue. 1885 test_method_data_pointer(mdp, profile_continue); 1886 1887 set_mdp_flag_at(mdp, BitData::null_seen_byte_constant()); 1888 1889 // The method data pointer needs to be updated. 1890 int mdp_delta = in_bytes(BitData::bit_data_size()); 1891 if (TypeProfileCasts) { 1892 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size()); 1893 } 1894 update_mdp_by_constant(mdp, mdp_delta); 1895 1896 bind(profile_continue); 1897 } 1898 } 1899 1900 1901 void InterpreterMacroAssembler::profile_typecheck_failed(Register mdp) { 1902 if (ProfileInterpreter && TypeProfileCasts) { 1903 Label profile_continue; 1904 1905 // If no method data exists, go to profile_continue. 1906 test_method_data_pointer(mdp, profile_continue); 1907 1908 int count_offset = in_bytes(CounterData::count_offset()); 1909 // Back up the address, since we have already bumped the mdp. 1910 count_offset -= in_bytes(VirtualCallData::virtual_call_data_size()); 1911 1912 // *Decrement* the counter. We expect to see zero or small negatives. 1913 increment_mdp_data_at(mdp, count_offset, true); 1914 1915 bind (profile_continue); 1916 } 1917 } 1918 1919 1920 void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) { 1921 if (ProfileInterpreter) { 1922 Label profile_continue; 1923 1924 // If no method data exists, go to profile_continue. 1925 test_method_data_pointer(mdp, profile_continue); 1926 1927 // The method data pointer needs to be updated. 1928 int mdp_delta = in_bytes(BitData::bit_data_size()); 1929 if (TypeProfileCasts) { 1930 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size()); 1931 1932 // Record the object type. 1933 record_klass_in_profile(klass, mdp, reg2, false); 1934 NOT_LP64(assert(reg2 == rdi, "we know how to fix this blown reg");) 1935 NOT_LP64(restore_locals();) // Restore EDI 1936 } 1937 update_mdp_by_constant(mdp, mdp_delta); 1938 1939 bind(profile_continue); 1940 } 1941 } 1942 1943 1944 void InterpreterMacroAssembler::profile_switch_default(Register mdp) { 1945 if (ProfileInterpreter) { 1946 Label profile_continue; 1947 1948 // If no method data exists, go to profile_continue. 1949 test_method_data_pointer(mdp, profile_continue); 1950 1951 // Update the default case count 1952 increment_mdp_data_at(mdp, 1953 in_bytes(MultiBranchData::default_count_offset())); 1954 1955 // The method data pointer needs to be updated. 1956 update_mdp_by_offset(mdp, 1957 in_bytes(MultiBranchData:: 1958 default_displacement_offset())); 1959 1960 bind(profile_continue); 1961 } 1962 } 1963 1964 1965 void InterpreterMacroAssembler::profile_switch_case(Register index, 1966 Register mdp, 1967 Register reg2) { 1968 if (ProfileInterpreter) { 1969 Label profile_continue; 1970 1971 // If no method data exists, go to profile_continue. 1972 test_method_data_pointer(mdp, profile_continue); 1973 1974 // Build the base (index * per_case_size_in_bytes()) + 1975 // case_array_offset_in_bytes() 1976 movl(reg2, in_bytes(MultiBranchData::per_case_size())); 1977 imulptr(index, reg2); // XXX l ? 1978 addptr(index, in_bytes(MultiBranchData::case_array_offset())); // XXX l ? 1979 1980 // Update the case count 1981 increment_mdp_data_at(mdp, 1982 index, 1983 in_bytes(MultiBranchData::relative_count_offset())); 1984 1985 // The method data pointer needs to be updated. 1986 update_mdp_by_offset(mdp, 1987 index, 1988 in_bytes(MultiBranchData:: 1989 relative_displacement_offset())); 1990 1991 bind(profile_continue); 1992 } 1993 } 1994 1995 1996 1997 void InterpreterMacroAssembler::verify_oop(Register reg, TosState state) { 1998 if (state == atos) { 1999 MacroAssembler::verify_oop(reg); 2000 } 2001 } 2002 2003 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) { 2004 #ifndef _LP64 2005 if ((state == ftos && UseSSE < 1) || 2006 (state == dtos && UseSSE < 2)) { 2007 MacroAssembler::verify_FPU(stack_depth); 2008 } 2009 #endif 2010 } 2011 2012 // Jump if ((*counter_addr += increment) & mask) satisfies the condition. 2013 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr, 2014 int increment, Address mask, 2015 Register scratch, bool preloaded, 2016 Condition cond, Label* where) { 2017 if (!preloaded) { 2018 movl(scratch, counter_addr); 2019 } 2020 incrementl(scratch, increment); 2021 movl(counter_addr, scratch); 2022 andl(scratch, mask); 2023 if (where != NULL) { 2024 jcc(cond, *where); 2025 } 2026 } 2027 2028 void InterpreterMacroAssembler::notify_method_entry() { 2029 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to 2030 // track stack depth. If it is possible to enter interp_only_mode we add 2031 // the code to check if the event should be sent. 2032 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx); 2033 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rbx); 2034 if (JvmtiExport::can_post_interpreter_events()) { 2035 Label L; 2036 NOT_LP64(get_thread(rthread);) 2037 movl(rdx, Address(rthread, JavaThread::interp_only_mode_offset())); 2038 testl(rdx, rdx); 2039 jcc(Assembler::zero, L); 2040 call_VM(noreg, CAST_FROM_FN_PTR(address, 2041 InterpreterRuntime::post_method_entry)); 2042 bind(L); 2043 } 2044 2045 { 2046 SkipIfEqual skip(this, &DTraceMethodProbes, false); 2047 NOT_LP64(get_thread(rthread);) 2048 get_method(rarg); 2049 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry), 2050 rthread, rarg); 2051 } 2052 2053 // RedefineClasses() tracing support for obsolete method entry 2054 if (log_is_enabled(Trace, redefine, class, obsolete)) { 2055 NOT_LP64(get_thread(rthread);) 2056 get_method(rarg); 2057 call_VM_leaf( 2058 CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry), 2059 rthread, rarg); 2060 } 2061 } 2062 2063 2064 void InterpreterMacroAssembler::notify_method_exit( 2065 TosState state, NotifyMethodExitMode mode) { 2066 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to 2067 // track stack depth. If it is possible to enter interp_only_mode we add 2068 // the code to check if the event should be sent. 2069 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx); 2070 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rbx); 2071 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) { 2072 Label L; 2073 // Note: frame::interpreter_frame_result has a dependency on how the 2074 // method result is saved across the call to post_method_exit. If this 2075 // is changed then the interpreter_frame_result implementation will 2076 // need to be updated too. 2077 2078 // template interpreter will leave the result on the top of the stack. 2079 push(state); 2080 NOT_LP64(get_thread(rthread);) 2081 movl(rdx, Address(rthread, JavaThread::interp_only_mode_offset())); 2082 testl(rdx, rdx); 2083 jcc(Assembler::zero, L); 2084 call_VM(noreg, 2085 CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit)); 2086 bind(L); 2087 pop(state); 2088 } 2089 2090 { 2091 SkipIfEqual skip(this, &DTraceMethodProbes, false); 2092 push(state); 2093 NOT_LP64(get_thread(rthread);) 2094 get_method(rarg); 2095 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit), 2096 rthread, rarg); 2097 pop(state); 2098 } 2099 }