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