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