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 "memory/vtBuffer.hpp" 31 #include "oops/arrayOop.hpp" 32 #include "oops/markOop.hpp" 33 #include "oops/methodData.hpp" 34 #include "oops/method.hpp" 35 #include "oops/valueKlass.hpp" 36 #include "prims/jvmtiExport.hpp" 37 #include "prims/jvmtiThreadState.hpp" 38 #include "runtime/basicLock.hpp" 39 #include "runtime/biasedLocking.hpp" 40 #include "runtime/safepointMechanism.hpp" 41 #include "runtime/sharedRuntime.hpp" 42 #include "runtime/thread.inline.hpp" 43 44 // Implementation of InterpreterMacroAssembler 45 46 void InterpreterMacroAssembler::jump_to_entry(address entry) { 47 assert(entry, "Entry must have been generated by now"); 48 jump(RuntimeAddress(entry)); 49 } 50 51 void InterpreterMacroAssembler::profile_obj_type(Register obj, const Address& mdo_addr) { 52 Label update, next, none; 53 54 verify_oop(obj); 55 56 testptr(obj, obj); 57 jccb(Assembler::notZero, update); 58 orptr(mdo_addr, TypeEntries::null_seen); 59 jmpb(next); 60 61 bind(update); 62 load_klass(obj, obj); 63 64 xorptr(obj, mdo_addr); 65 testptr(obj, TypeEntries::type_klass_mask); 66 jccb(Assembler::zero, next); // klass seen before, nothing to 67 // do. The unknown bit may have been 68 // set already but no need to check. 69 70 testptr(obj, TypeEntries::type_unknown); 71 jccb(Assembler::notZero, next); // already unknown. Nothing to do anymore. 72 73 cmpptr(mdo_addr, 0); 74 jccb(Assembler::equal, none); 75 cmpptr(mdo_addr, TypeEntries::null_seen); 76 jccb(Assembler::equal, none); 77 // There is a chance that the checks above (re-reading profiling 78 // data from memory) fail if another thread has just set the 79 // profiling to this obj's klass 80 xorptr(obj, mdo_addr); 81 testptr(obj, TypeEntries::type_klass_mask); 82 jccb(Assembler::zero, next); 83 84 // different than before. Cannot keep accurate profile. 85 orptr(mdo_addr, TypeEntries::type_unknown); 86 jmpb(next); 87 88 bind(none); 89 // first time here. Set profile type. 90 movptr(mdo_addr, obj); 91 92 bind(next); 93 } 94 95 void InterpreterMacroAssembler::profile_arguments_type(Register mdp, Register callee, Register tmp, bool is_virtual) { 96 if (!ProfileInterpreter) { 97 return; 98 } 99 100 if (MethodData::profile_arguments() || MethodData::profile_return()) { 101 Label profile_continue; 102 103 test_method_data_pointer(mdp, profile_continue); 104 105 int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size()); 106 107 cmpb(Address(mdp, in_bytes(DataLayout::tag_offset()) - off_to_start), is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag); 108 jcc(Assembler::notEqual, profile_continue); 109 110 if (MethodData::profile_arguments()) { 111 Label done; 112 int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset()); 113 addptr(mdp, off_to_args); 114 115 for (int i = 0; i < TypeProfileArgsLimit; i++) { 116 if (i > 0 || MethodData::profile_return()) { 117 // If return value type is profiled we may have no argument to profile 118 movptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args)); 119 subl(tmp, i*TypeStackSlotEntries::per_arg_count()); 120 cmpl(tmp, TypeStackSlotEntries::per_arg_count()); 121 jcc(Assembler::less, done); 122 } 123 movptr(tmp, Address(callee, Method::const_offset())); 124 load_unsigned_short(tmp, Address(tmp, ConstMethod::size_of_parameters_offset())); 125 // stack offset o (zero based) from the start of the argument 126 // list, for n arguments translates into offset n - o - 1 from 127 // the end of the argument list 128 subptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))-off_to_args)); 129 subl(tmp, 1); 130 Address arg_addr = argument_address(tmp); 131 movptr(tmp, arg_addr); 132 133 Address mdo_arg_addr(mdp, in_bytes(TypeEntriesAtCall::argument_type_offset(i))-off_to_args); 134 profile_obj_type(tmp, mdo_arg_addr); 135 136 int to_add = in_bytes(TypeStackSlotEntries::per_arg_size()); 137 addptr(mdp, to_add); 138 off_to_args += to_add; 139 } 140 141 if (MethodData::profile_return()) { 142 movptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args)); 143 subl(tmp, TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count()); 144 } 145 146 bind(done); 147 148 if (MethodData::profile_return()) { 149 // We're right after the type profile for the last 150 // argument. tmp is the number of cells left in the 151 // CallTypeData/VirtualCallTypeData to reach its end. Non null 152 // if there's a return to profile. 153 assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type"); 154 shll(tmp, exact_log2(DataLayout::cell_size)); 155 addptr(mdp, tmp); 156 } 157 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp); 158 } else { 159 assert(MethodData::profile_return(), "either profile call args or call ret"); 160 update_mdp_by_constant(mdp, in_bytes(TypeEntriesAtCall::return_only_size())); 161 } 162 163 // mdp points right after the end of the 164 // CallTypeData/VirtualCallTypeData, right after the cells for the 165 // return value type if there's one 166 167 bind(profile_continue); 168 } 169 } 170 171 void InterpreterMacroAssembler::profile_return_type(Register mdp, Register ret, Register tmp) { 172 assert_different_registers(mdp, ret, tmp, _bcp_register); 173 if (ProfileInterpreter && MethodData::profile_return()) { 174 Label profile_continue, done; 175 176 test_method_data_pointer(mdp, profile_continue); 177 178 if (MethodData::profile_return_jsr292_only()) { 179 assert(Method::intrinsic_id_size_in_bytes() == 2, "assuming Method::_intrinsic_id is u2"); 180 181 // If we don't profile all invoke bytecodes we must make sure 182 // it's a bytecode we indeed profile. We can't go back to the 183 // begining of the ProfileData we intend to update to check its 184 // type because we're right after it and we don't known its 185 // length 186 Label do_profile; 187 cmpb(Address(_bcp_register, 0), Bytecodes::_invokedynamic); 188 jcc(Assembler::equal, do_profile); 189 cmpb(Address(_bcp_register, 0), Bytecodes::_invokehandle); 190 jcc(Assembler::equal, do_profile); 191 get_method(tmp); 192 cmpw(Address(tmp, Method::intrinsic_id_offset_in_bytes()), vmIntrinsics::_compiledLambdaForm); 193 jcc(Assembler::notEqual, profile_continue); 194 195 bind(do_profile); 196 } 197 198 Address mdo_ret_addr(mdp, -in_bytes(ReturnTypeEntry::size())); 199 mov(tmp, ret); 200 profile_obj_type(tmp, mdo_ret_addr); 201 202 bind(profile_continue); 203 } 204 } 205 206 void InterpreterMacroAssembler::profile_parameters_type(Register mdp, Register tmp1, Register tmp2) { 207 if (ProfileInterpreter && MethodData::profile_parameters()) { 208 Label profile_continue, done; 209 210 test_method_data_pointer(mdp, profile_continue); 211 212 // Load the offset of the area within the MDO used for 213 // parameters. If it's negative we're not profiling any parameters 214 movl(tmp1, Address(mdp, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset()))); 215 testl(tmp1, tmp1); 216 jcc(Assembler::negative, profile_continue); 217 218 // Compute a pointer to the area for parameters from the offset 219 // and move the pointer to the slot for the last 220 // parameters. Collect profiling from last parameter down. 221 // mdo start + parameters offset + array length - 1 222 addptr(mdp, tmp1); 223 movptr(tmp1, Address(mdp, ArrayData::array_len_offset())); 224 decrement(tmp1, TypeStackSlotEntries::per_arg_count()); 225 226 Label loop; 227 bind(loop); 228 229 int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0)); 230 int type_base = in_bytes(ParametersTypeData::type_offset(0)); 231 Address::ScaleFactor per_arg_scale = Address::times(DataLayout::cell_size); 232 Address arg_off(mdp, tmp1, per_arg_scale, off_base); 233 Address arg_type(mdp, tmp1, per_arg_scale, type_base); 234 235 // load offset on the stack from the slot for this parameter 236 movptr(tmp2, arg_off); 237 negptr(tmp2); 238 // read the parameter from the local area 239 movptr(tmp2, Address(_locals_register, tmp2, Interpreter::stackElementScale())); 240 241 // profile the parameter 242 profile_obj_type(tmp2, arg_type); 243 244 // go to next parameter 245 decrement(tmp1, TypeStackSlotEntries::per_arg_count()); 246 jcc(Assembler::positive, loop); 247 248 bind(profile_continue); 249 } 250 } 251 252 void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point, 253 int number_of_arguments) { 254 // interpreter specific 255 // 256 // Note: No need to save/restore bcp & locals registers 257 // since these are callee saved registers and no blocking/ 258 // GC can happen in leaf calls. 259 // Further Note: DO NOT save/restore bcp/locals. If a caller has 260 // already saved them so that it can use rsi/rdi as temporaries 261 // then a save/restore here will DESTROY the copy the caller 262 // saved! There used to be a save_bcp() that only happened in 263 // the ASSERT path (no restore_bcp). Which caused bizarre failures 264 // when jvm built with ASSERTs. 265 #ifdef ASSERT 266 { 267 Label L; 268 cmpptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD); 269 jcc(Assembler::equal, L); 270 stop("InterpreterMacroAssembler::call_VM_leaf_base:" 271 " last_sp != NULL"); 272 bind(L); 273 } 274 #endif 275 // super call 276 MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments); 277 // interpreter specific 278 // LP64: Used to ASSERT that r13/r14 were equal to frame's bcp/locals 279 // but since they may not have been saved (and we don't want to 280 // save them here (see note above) the assert is invalid. 281 } 282 283 void InterpreterMacroAssembler::call_VM_base(Register oop_result, 284 Register java_thread, 285 Register last_java_sp, 286 address entry_point, 287 int number_of_arguments, 288 bool check_exceptions) { 289 // interpreter specific 290 // 291 // Note: Could avoid restoring locals ptr (callee saved) - however doesn't 292 // really make a difference for these runtime calls, since they are 293 // slow anyway. Btw., bcp must be saved/restored since it may change 294 // due to GC. 295 NOT_LP64(assert(java_thread == noreg , "not expecting a precomputed java thread");) 296 save_bcp(); 297 #ifdef ASSERT 298 { 299 Label L; 300 cmpptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD); 301 jcc(Assembler::equal, L); 302 stop("InterpreterMacroAssembler::call_VM_base:" 303 " last_sp != NULL"); 304 bind(L); 305 } 306 #endif /* ASSERT */ 307 // super call 308 MacroAssembler::call_VM_base(oop_result, noreg, last_java_sp, 309 entry_point, number_of_arguments, 310 check_exceptions); 311 // interpreter specific 312 restore_bcp(); 313 restore_locals(); 314 } 315 316 void InterpreterMacroAssembler::check_and_handle_popframe(Register java_thread) { 317 if (JvmtiExport::can_pop_frame()) { 318 Label L; 319 // Initiate popframe handling only if it is not already being 320 // processed. If the flag has the popframe_processing bit set, it 321 // means that this code is called *during* popframe handling - we 322 // don't want to reenter. 323 // This method is only called just after the call into the vm in 324 // call_VM_base, so the arg registers are available. 325 Register pop_cond = NOT_LP64(java_thread) // Not clear if any other register is available on 32 bit 326 LP64_ONLY(c_rarg0); 327 movl(pop_cond, Address(java_thread, JavaThread::popframe_condition_offset())); 328 testl(pop_cond, JavaThread::popframe_pending_bit); 329 jcc(Assembler::zero, L); 330 testl(pop_cond, JavaThread::popframe_processing_bit); 331 jcc(Assembler::notZero, L); 332 // Call Interpreter::remove_activation_preserving_args_entry() to get the 333 // address of the same-named entrypoint in the generated interpreter code. 334 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry)); 335 jmp(rax); 336 bind(L); 337 NOT_LP64(get_thread(java_thread);) 338 } 339 } 340 341 void InterpreterMacroAssembler::load_earlyret_value(TosState state) { 342 Register thread = LP64_ONLY(r15_thread) NOT_LP64(rcx); 343 NOT_LP64(get_thread(thread);) 344 movptr(rcx, Address(thread, JavaThread::jvmti_thread_state_offset())); 345 const Address tos_addr(rcx, JvmtiThreadState::earlyret_tos_offset()); 346 const Address oop_addr(rcx, JvmtiThreadState::earlyret_oop_offset()); 347 const Address val_addr(rcx, JvmtiThreadState::earlyret_value_offset()); 348 #ifdef _LP64 349 switch (state) { 350 case atos: movptr(rax, oop_addr); 351 movptr(oop_addr, (int32_t)NULL_WORD); 352 verify_oop(rax, state); break; 353 case ltos: movptr(rax, val_addr); break; 354 case btos: // fall through 355 case ztos: // fall through 356 case ctos: // fall through 357 case stos: // fall through 358 case itos: movl(rax, val_addr); break; 359 case ftos: load_float(val_addr); break; 360 case dtos: load_double(val_addr); break; 361 case vtos: /* nothing to do */ break; 362 default : ShouldNotReachHere(); 363 } 364 // Clean up tos value in the thread object 365 movl(tos_addr, (int) ilgl); 366 movl(val_addr, (int32_t) NULL_WORD); 367 #else 368 const Address val_addr1(rcx, JvmtiThreadState::earlyret_value_offset() 369 + in_ByteSize(wordSize)); 370 switch (state) { 371 case atos: movptr(rax, oop_addr); 372 movptr(oop_addr, NULL_WORD); 373 verify_oop(rax, state); break; 374 case ltos: 375 movl(rdx, val_addr1); // fall through 376 case btos: // fall through 377 case ztos: // fall through 378 case ctos: // fall through 379 case stos: // fall through 380 case itos: movl(rax, val_addr); break; 381 case ftos: load_float(val_addr); break; 382 case dtos: load_double(val_addr); break; 383 case vtos: /* nothing to do */ break; 384 default : ShouldNotReachHere(); 385 } 386 #endif // _LP64 387 // Clean up tos value in the thread object 388 movl(tos_addr, (int32_t) ilgl); 389 movptr(val_addr, NULL_WORD); 390 NOT_LP64(movptr(val_addr1, NULL_WORD);) 391 } 392 393 394 void InterpreterMacroAssembler::check_and_handle_earlyret(Register java_thread) { 395 if (JvmtiExport::can_force_early_return()) { 396 Label L; 397 Register tmp = LP64_ONLY(c_rarg0) NOT_LP64(java_thread); 398 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(java_thread); 399 400 movptr(tmp, Address(rthread, JavaThread::jvmti_thread_state_offset())); 401 testptr(tmp, tmp); 402 jcc(Assembler::zero, L); // if (thread->jvmti_thread_state() == NULL) exit; 403 404 // Initiate earlyret handling only if it is not already being processed. 405 // If the flag has the earlyret_processing bit set, it means that this code 406 // is called *during* earlyret handling - we don't want to reenter. 407 movl(tmp, Address(tmp, JvmtiThreadState::earlyret_state_offset())); 408 cmpl(tmp, JvmtiThreadState::earlyret_pending); 409 jcc(Assembler::notEqual, L); 410 411 // Call Interpreter::remove_activation_early_entry() to get the address of the 412 // same-named entrypoint in the generated interpreter code. 413 NOT_LP64(get_thread(java_thread);) 414 movptr(tmp, Address(rthread, JavaThread::jvmti_thread_state_offset())); 415 #ifdef _LP64 416 movl(tmp, Address(tmp, JvmtiThreadState::earlyret_tos_offset())); 417 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), tmp); 418 #else 419 pushl(Address(tmp, JvmtiThreadState::earlyret_tos_offset())); 420 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), 1); 421 #endif // _LP64 422 jmp(rax); 423 bind(L); 424 NOT_LP64(get_thread(java_thread);) 425 } 426 } 427 428 void InterpreterMacroAssembler::get_unsigned_2_byte_index_at_bcp(Register reg, int bcp_offset) { 429 assert(bcp_offset >= 0, "bcp is still pointing to start of bytecode"); 430 load_unsigned_short(reg, Address(_bcp_register, bcp_offset)); 431 bswapl(reg); 432 shrl(reg, 16); 433 } 434 435 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register index, 436 int bcp_offset, 437 size_t index_size) { 438 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode"); 439 if (index_size == sizeof(u2)) { 440 load_unsigned_short(index, Address(_bcp_register, bcp_offset)); 441 } else if (index_size == sizeof(u4)) { 442 movl(index, Address(_bcp_register, bcp_offset)); 443 // Check if the secondary index definition is still ~x, otherwise 444 // we have to change the following assembler code to calculate the 445 // plain index. 446 assert(ConstantPool::decode_invokedynamic_index(~123) == 123, "else change next line"); 447 notl(index); // convert to plain index 448 } else if (index_size == sizeof(u1)) { 449 load_unsigned_byte(index, Address(_bcp_register, bcp_offset)); 450 } else { 451 ShouldNotReachHere(); 452 } 453 } 454 455 void InterpreterMacroAssembler::get_cache_and_index_at_bcp(Register cache, 456 Register index, 457 int bcp_offset, 458 size_t index_size) { 459 assert_different_registers(cache, index); 460 get_cache_index_at_bcp(index, bcp_offset, index_size); 461 movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize)); 462 assert(sizeof(ConstantPoolCacheEntry) == 4 * wordSize, "adjust code below"); 463 // convert from field index to ConstantPoolCacheEntry index 464 assert(exact_log2(in_words(ConstantPoolCacheEntry::size())) == 2, "else change next line"); 465 shll(index, 2); 466 } 467 468 void InterpreterMacroAssembler::get_cache_and_index_and_bytecode_at_bcp(Register cache, 469 Register index, 470 Register bytecode, 471 int byte_no, 472 int bcp_offset, 473 size_t index_size) { 474 get_cache_and_index_at_bcp(cache, index, bcp_offset, index_size); 475 // We use a 32-bit load here since the layout of 64-bit words on 476 // little-endian machines allow us that. 477 movl(bytecode, Address(cache, index, Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset())); 478 const int shift_count = (1 + byte_no) * BitsPerByte; 479 assert((byte_no == TemplateTable::f1_byte && shift_count == ConstantPoolCacheEntry::bytecode_1_shift) || 480 (byte_no == TemplateTable::f2_byte && shift_count == ConstantPoolCacheEntry::bytecode_2_shift), 481 "correct shift count"); 482 shrl(bytecode, shift_count); 483 assert(ConstantPoolCacheEntry::bytecode_1_mask == ConstantPoolCacheEntry::bytecode_2_mask, "common mask"); 484 andl(bytecode, ConstantPoolCacheEntry::bytecode_1_mask); 485 } 486 487 void InterpreterMacroAssembler::get_cache_entry_pointer_at_bcp(Register cache, 488 Register tmp, 489 int bcp_offset, 490 size_t index_size) { 491 assert(cache != tmp, "must use different register"); 492 get_cache_index_at_bcp(tmp, bcp_offset, index_size); 493 assert(sizeof(ConstantPoolCacheEntry) == 4 * wordSize, "adjust code below"); 494 // convert from field index to ConstantPoolCacheEntry index 495 // and from word offset to byte offset 496 assert(exact_log2(in_bytes(ConstantPoolCacheEntry::size_in_bytes())) == 2 + LogBytesPerWord, "else change next line"); 497 shll(tmp, 2 + LogBytesPerWord); 498 movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize)); 499 // skip past the header 500 addptr(cache, in_bytes(ConstantPoolCache::base_offset())); 501 addptr(cache, tmp); // construct pointer to cache entry 502 } 503 504 // Load object from cpool->resolved_references(index) 505 void InterpreterMacroAssembler::load_resolved_reference_at_index( 506 Register result, Register index) { 507 assert_different_registers(result, index); 508 // convert from field index to resolved_references() index and from 509 // word index to byte offset. Since this is a java object, it can be compressed 510 Register tmp = index; // reuse 511 shll(tmp, LogBytesPerHeapOop); 512 513 get_constant_pool(result); 514 // load pointer for resolved_references[] objArray 515 movptr(result, Address(result, ConstantPool::cache_offset_in_bytes())); 516 movptr(result, Address(result, ConstantPoolCache::resolved_references_offset_in_bytes())); 517 resolve_oop_handle(result); 518 // Add in the index 519 addptr(result, tmp); 520 load_heap_oop(result, Address(result, arrayOopDesc::base_offset_in_bytes(T_OBJECT))); 521 // The resulting oop is null if the reference is not yet resolved. 522 // It is Universe::the_null_sentinel() if the reference resolved to NULL via condy. 523 } 524 525 // load cpool->resolved_klass_at(index) 526 void InterpreterMacroAssembler::load_resolved_klass_at_index(Register cpool, 527 Register index, Register klass) { 528 movw(index, Address(cpool, index, Address::times_ptr, sizeof(ConstantPool))); 529 Register resolved_klasses = cpool; 530 movptr(resolved_klasses, Address(cpool, ConstantPool::resolved_klasses_offset_in_bytes())); 531 movptr(klass, Address(resolved_klasses, index, Address::times_ptr, Array<Klass*>::base_offset_in_bytes())); 532 } 533 534 // Generate a subtype check: branch to ok_is_subtype if sub_klass is a 535 // subtype of super_klass. 536 // 537 // Args: 538 // rax: superklass 539 // Rsub_klass: subklass 540 // 541 // Kills: 542 // rcx, rdi 543 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass, 544 Label& ok_is_subtype) { 545 assert(Rsub_klass != rax, "rax holds superklass"); 546 LP64_ONLY(assert(Rsub_klass != r14, "r14 holds locals");) 547 LP64_ONLY(assert(Rsub_klass != r13, "r13 holds bcp");) 548 assert(Rsub_klass != rcx, "rcx holds 2ndary super array length"); 549 assert(Rsub_klass != rdi, "rdi holds 2ndary super array scan ptr"); 550 551 // Profile the not-null value's klass. 552 profile_typecheck(rcx, Rsub_klass, rdi); // blows rcx, reloads rdi 553 554 // Do the check. 555 check_klass_subtype(Rsub_klass, rax, rcx, ok_is_subtype); // blows rcx 556 557 // Profile the failure of the check. 558 profile_typecheck_failed(rcx); // blows rcx 559 } 560 561 562 #ifndef _LP64 563 void InterpreterMacroAssembler::f2ieee() { 564 if (IEEEPrecision) { 565 fstp_s(Address(rsp, 0)); 566 fld_s(Address(rsp, 0)); 567 } 568 } 569 570 571 void InterpreterMacroAssembler::d2ieee() { 572 if (IEEEPrecision) { 573 fstp_d(Address(rsp, 0)); 574 fld_d(Address(rsp, 0)); 575 } 576 } 577 #endif // _LP64 578 579 // Java Expression Stack 580 581 void InterpreterMacroAssembler::pop_ptr(Register r) { 582 pop(r); 583 } 584 585 void InterpreterMacroAssembler::push_ptr(Register r) { 586 push(r); 587 } 588 589 void InterpreterMacroAssembler::push_i(Register r) { 590 push(r); 591 } 592 593 void InterpreterMacroAssembler::push_f(XMMRegister r) { 594 subptr(rsp, wordSize); 595 movflt(Address(rsp, 0), r); 596 } 597 598 void InterpreterMacroAssembler::pop_f(XMMRegister r) { 599 movflt(r, Address(rsp, 0)); 600 addptr(rsp, wordSize); 601 } 602 603 void InterpreterMacroAssembler::push_d(XMMRegister r) { 604 subptr(rsp, 2 * wordSize); 605 movdbl(Address(rsp, 0), r); 606 } 607 608 void InterpreterMacroAssembler::pop_d(XMMRegister r) { 609 movdbl(r, Address(rsp, 0)); 610 addptr(rsp, 2 * Interpreter::stackElementSize); 611 } 612 613 #ifdef _LP64 614 void InterpreterMacroAssembler::pop_i(Register r) { 615 // XXX can't use pop currently, upper half non clean 616 movl(r, Address(rsp, 0)); 617 addptr(rsp, wordSize); 618 } 619 620 void InterpreterMacroAssembler::pop_l(Register r) { 621 movq(r, Address(rsp, 0)); 622 addptr(rsp, 2 * Interpreter::stackElementSize); 623 } 624 625 void InterpreterMacroAssembler::push_l(Register r) { 626 subptr(rsp, 2 * wordSize); 627 movptr(Address(rsp, Interpreter::expr_offset_in_bytes(0)), r ); 628 movptr(Address(rsp, Interpreter::expr_offset_in_bytes(1)), NULL_WORD ); 629 } 630 631 void InterpreterMacroAssembler::pop(TosState state) { 632 switch (state) { 633 case atos: pop_ptr(); break; 634 case btos: 635 case ztos: 636 case ctos: 637 case stos: 638 case itos: pop_i(); break; 639 case ltos: pop_l(); break; 640 case ftos: pop_f(xmm0); break; 641 case dtos: pop_d(xmm0); break; 642 case vtos: /* nothing to do */ break; 643 default: ShouldNotReachHere(); 644 } 645 verify_oop(rax, state); 646 } 647 648 void InterpreterMacroAssembler::push(TosState state) { 649 verify_oop(rax, state); 650 switch (state) { 651 case atos: push_ptr(); break; 652 case btos: 653 case ztos: 654 case ctos: 655 case stos: 656 case itos: push_i(); break; 657 case ltos: push_l(); break; 658 case ftos: push_f(xmm0); break; 659 case dtos: push_d(xmm0); break; 660 case vtos: /* nothing to do */ break; 661 default : ShouldNotReachHere(); 662 } 663 } 664 #else 665 void InterpreterMacroAssembler::pop_i(Register r) { 666 pop(r); 667 } 668 669 void InterpreterMacroAssembler::pop_l(Register lo, Register hi) { 670 pop(lo); 671 pop(hi); 672 } 673 674 void InterpreterMacroAssembler::pop_f() { 675 fld_s(Address(rsp, 0)); 676 addptr(rsp, 1 * wordSize); 677 } 678 679 void InterpreterMacroAssembler::pop_d() { 680 fld_d(Address(rsp, 0)); 681 addptr(rsp, 2 * wordSize); 682 } 683 684 685 void InterpreterMacroAssembler::pop(TosState state) { 686 switch (state) { 687 case atos: pop_ptr(rax); break; 688 case btos: // fall through 689 case ztos: // fall through 690 case ctos: // fall through 691 case stos: // fall through 692 case itos: pop_i(rax); break; 693 case ltos: pop_l(rax, rdx); break; 694 case ftos: 695 if (UseSSE >= 1) { 696 pop_f(xmm0); 697 } else { 698 pop_f(); 699 } 700 break; 701 case dtos: 702 if (UseSSE >= 2) { 703 pop_d(xmm0); 704 } else { 705 pop_d(); 706 } 707 break; 708 case vtos: /* nothing to do */ break; 709 default : ShouldNotReachHere(); 710 } 711 verify_oop(rax, state); 712 } 713 714 715 void InterpreterMacroAssembler::push_l(Register lo, Register hi) { 716 push(hi); 717 push(lo); 718 } 719 720 void InterpreterMacroAssembler::push_f() { 721 // Do not schedule for no AGI! Never write beyond rsp! 722 subptr(rsp, 1 * wordSize); 723 fstp_s(Address(rsp, 0)); 724 } 725 726 void InterpreterMacroAssembler::push_d() { 727 // Do not schedule for no AGI! Never write beyond rsp! 728 subptr(rsp, 2 * wordSize); 729 fstp_d(Address(rsp, 0)); 730 } 731 732 733 void InterpreterMacroAssembler::push(TosState state) { 734 verify_oop(rax, state); 735 switch (state) { 736 case atos: push_ptr(rax); break; 737 case btos: // fall through 738 case ztos: // fall through 739 case ctos: // fall through 740 case stos: // fall through 741 case itos: push_i(rax); break; 742 case ltos: push_l(rax, rdx); break; 743 case ftos: 744 if (UseSSE >= 1) { 745 push_f(xmm0); 746 } else { 747 push_f(); 748 } 749 break; 750 case dtos: 751 if (UseSSE >= 2) { 752 push_d(xmm0); 753 } else { 754 push_d(); 755 } 756 break; 757 case vtos: /* nothing to do */ break; 758 default : ShouldNotReachHere(); 759 } 760 } 761 #endif // _LP64 762 763 764 // Helpers for swap and dup 765 void InterpreterMacroAssembler::load_ptr(int n, Register val) { 766 movptr(val, Address(rsp, Interpreter::expr_offset_in_bytes(n))); 767 } 768 769 void InterpreterMacroAssembler::store_ptr(int n, Register val) { 770 movptr(Address(rsp, Interpreter::expr_offset_in_bytes(n)), val); 771 } 772 773 774 void InterpreterMacroAssembler::prepare_to_jump_from_interpreted() { 775 // set sender sp 776 lea(_bcp_register, Address(rsp, wordSize)); 777 // record last_sp 778 movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), _bcp_register); 779 } 780 781 782 // Jump to from_interpreted entry of a call unless single stepping is possible 783 // in this thread in which case we must call the i2i entry 784 void InterpreterMacroAssembler::jump_from_interpreted(Register method, Register temp) { 785 prepare_to_jump_from_interpreted(); 786 787 if (JvmtiExport::can_post_interpreter_events()) { 788 Label run_compiled_code; 789 // JVMTI events, such as single-stepping, are implemented partly by avoiding running 790 // compiled code in threads for which the event is enabled. Check here for 791 // interp_only_mode if these events CAN be enabled. 792 // interp_only is an int, on little endian it is sufficient to test the byte only 793 // Is a cmpl faster? 794 LP64_ONLY(temp = r15_thread;) 795 NOT_LP64(get_thread(temp);) 796 cmpb(Address(temp, JavaThread::interp_only_mode_offset()), 0); 797 jccb(Assembler::zero, run_compiled_code); 798 jmp(Address(method, Method::interpreter_entry_offset())); 799 bind(run_compiled_code); 800 } 801 802 jmp(Address(method, Method::from_interpreted_offset())); 803 } 804 805 // The following two routines provide a hook so that an implementation 806 // can schedule the dispatch in two parts. x86 does not do this. 807 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int step) { 808 // Nothing x86 specific to be done here 809 } 810 811 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int step) { 812 dispatch_next(state, step); 813 } 814 815 void InterpreterMacroAssembler::dispatch_base(TosState state, 816 address* table, 817 bool verifyoop, 818 bool generate_poll) { 819 verify_FPU(1, state); 820 if (VerifyActivationFrameSize) { 821 Label L; 822 mov(rcx, rbp); 823 subptr(rcx, rsp); 824 int32_t min_frame_size = 825 (frame::link_offset - frame::interpreter_frame_initial_sp_offset) * 826 wordSize; 827 cmpptr(rcx, (int32_t)min_frame_size); 828 jcc(Assembler::greaterEqual, L); 829 stop("broken stack frame"); 830 bind(L); 831 } 832 if (verifyoop) { 833 verify_oop(rax, state); 834 } 835 836 address* const safepoint_table = Interpreter::safept_table(state); 837 #ifdef _LP64 838 Label no_safepoint, dispatch; 839 if (SafepointMechanism::uses_thread_local_poll() && table != safepoint_table && generate_poll) { 840 NOT_PRODUCT(block_comment("Thread-local Safepoint poll")); 841 testb(Address(r15_thread, Thread::polling_page_offset()), SafepointMechanism::poll_bit()); 842 843 jccb(Assembler::zero, no_safepoint); 844 lea(rscratch1, ExternalAddress((address)safepoint_table)); 845 jmpb(dispatch); 846 } 847 848 bind(no_safepoint); 849 lea(rscratch1, ExternalAddress((address)table)); 850 bind(dispatch); 851 jmp(Address(rscratch1, rbx, Address::times_8)); 852 853 #else 854 Address index(noreg, rbx, Address::times_ptr); 855 if (SafepointMechanism::uses_thread_local_poll() && table != safepoint_table && generate_poll) { 856 NOT_PRODUCT(block_comment("Thread-local Safepoint poll")); 857 Label no_safepoint; 858 const Register thread = rcx; 859 get_thread(thread); 860 testb(Address(thread, Thread::polling_page_offset()), SafepointMechanism::poll_bit()); 861 862 jccb(Assembler::zero, no_safepoint); 863 ArrayAddress dispatch_addr(ExternalAddress((address)safepoint_table), index); 864 jump(dispatch_addr); 865 bind(no_safepoint); 866 } 867 868 { 869 ArrayAddress dispatch_addr(ExternalAddress((address)table), index); 870 jump(dispatch_addr); 871 } 872 #endif // _LP64 873 } 874 875 void InterpreterMacroAssembler::dispatch_only(TosState state, bool generate_poll) { 876 dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll); 877 } 878 879 void InterpreterMacroAssembler::dispatch_only_normal(TosState state) { 880 dispatch_base(state, Interpreter::normal_table(state)); 881 } 882 883 void InterpreterMacroAssembler::dispatch_only_noverify(TosState state) { 884 dispatch_base(state, Interpreter::normal_table(state), false); 885 } 886 887 888 void InterpreterMacroAssembler::dispatch_next(TosState state, int step, bool generate_poll) { 889 // load next bytecode (load before advancing _bcp_register to prevent AGI) 890 load_unsigned_byte(rbx, Address(_bcp_register, step)); 891 // advance _bcp_register 892 increment(_bcp_register, step); 893 dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll); 894 } 895 896 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) { 897 // load current bytecode 898 load_unsigned_byte(rbx, Address(_bcp_register, 0)); 899 dispatch_base(state, table); 900 } 901 902 void InterpreterMacroAssembler::narrow(Register result) { 903 904 // Get method->_constMethod->_result_type 905 movptr(rcx, Address(rbp, frame::interpreter_frame_method_offset * wordSize)); 906 movptr(rcx, Address(rcx, Method::const_offset())); 907 load_unsigned_byte(rcx, Address(rcx, ConstMethod::result_type_offset())); 908 909 Label done, notBool, notByte, notChar; 910 911 // common case first 912 cmpl(rcx, T_INT); 913 jcc(Assembler::equal, done); 914 915 // mask integer result to narrower return type. 916 cmpl(rcx, T_BOOLEAN); 917 jcc(Assembler::notEqual, notBool); 918 andl(result, 0x1); 919 jmp(done); 920 921 bind(notBool); 922 cmpl(rcx, T_BYTE); 923 jcc(Assembler::notEqual, notByte); 924 LP64_ONLY(movsbl(result, result);) 925 NOT_LP64(shll(result, 24);) // truncate upper 24 bits 926 NOT_LP64(sarl(result, 24);) // and sign-extend byte 927 jmp(done); 928 929 bind(notByte); 930 cmpl(rcx, T_CHAR); 931 jcc(Assembler::notEqual, notChar); 932 LP64_ONLY(movzwl(result, result);) 933 NOT_LP64(andl(result, 0xFFFF);) // truncate upper 16 bits 934 jmp(done); 935 936 bind(notChar); 937 // cmpl(rcx, T_SHORT); // all that's left 938 // jcc(Assembler::notEqual, done); 939 LP64_ONLY(movswl(result, result);) 940 NOT_LP64(shll(result, 16);) // truncate upper 16 bits 941 NOT_LP64(sarl(result, 16);) // and sign-extend short 942 943 // Nothing to do for T_INT 944 bind(done); 945 } 946 947 // remove activation 948 // 949 // Unlock the receiver if this is a synchronized method. 950 // Unlock any Java monitors from syncronized blocks. 951 // Remove the activation from the stack. 952 // 953 // If there are locked Java monitors 954 // If throw_monitor_exception 955 // throws IllegalMonitorStateException 956 // Else if install_monitor_exception 957 // installs IllegalMonitorStateException 958 // Else 959 // no error processing 960 void InterpreterMacroAssembler::remove_activation( 961 TosState state, 962 Register ret_addr, 963 bool throw_monitor_exception, 964 bool install_monitor_exception, 965 bool notify_jvmdi, 966 bool load_values) { 967 // Note: Registers rdx xmm0 may be in use for the 968 // result check if synchronized method 969 Label unlocked, unlock, no_unlock; 970 971 const Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx); 972 const Register robj = LP64_ONLY(c_rarg1) NOT_LP64(rdx); 973 const Register rmon = LP64_ONLY(c_rarg1) NOT_LP64(rcx); 974 // monitor pointers need different register 975 // because rdx may have the result in it 976 NOT_LP64(get_thread(rcx);) 977 978 // get the value of _do_not_unlock_if_synchronized into rdx 979 const Address do_not_unlock_if_synchronized(rthread, 980 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset())); 981 movbool(rbx, do_not_unlock_if_synchronized); 982 movbool(do_not_unlock_if_synchronized, false); // reset the flag 983 984 // get method access flags 985 movptr(rcx, Address(rbp, frame::interpreter_frame_method_offset * wordSize)); 986 movl(rcx, Address(rcx, Method::access_flags_offset())); 987 testl(rcx, JVM_ACC_SYNCHRONIZED); 988 jcc(Assembler::zero, unlocked); 989 990 // Don't unlock anything if the _do_not_unlock_if_synchronized flag 991 // is set. 992 testbool(rbx); 993 jcc(Assembler::notZero, no_unlock); 994 995 // unlock monitor 996 push(state); // save result 997 998 // BasicObjectLock will be first in list, since this is a 999 // synchronized method. However, need to check that the object has 1000 // not been unlocked by an explicit monitorexit bytecode. 1001 const Address monitor(rbp, frame::interpreter_frame_initial_sp_offset * 1002 wordSize - (int) sizeof(BasicObjectLock)); 1003 // We use c_rarg1/rdx so that if we go slow path it will be the correct 1004 // register for unlock_object to pass to VM directly 1005 lea(robj, monitor); // address of first monitor 1006 1007 movptr(rax, Address(robj, BasicObjectLock::obj_offset_in_bytes())); 1008 testptr(rax, rax); 1009 jcc(Assembler::notZero, unlock); 1010 1011 pop(state); 1012 if (throw_monitor_exception) { 1013 // Entry already unlocked, need to throw exception 1014 NOT_LP64(empty_FPU_stack();) // remove possible return value from FPU-stack, otherwise stack could overflow 1015 call_VM(noreg, CAST_FROM_FN_PTR(address, 1016 InterpreterRuntime::throw_illegal_monitor_state_exception)); 1017 should_not_reach_here(); 1018 } else { 1019 // Monitor already unlocked during a stack unroll. If requested, 1020 // install an illegal_monitor_state_exception. Continue with 1021 // stack unrolling. 1022 if (install_monitor_exception) { 1023 NOT_LP64(empty_FPU_stack();) 1024 call_VM(noreg, CAST_FROM_FN_PTR(address, 1025 InterpreterRuntime::new_illegal_monitor_state_exception)); 1026 } 1027 jmp(unlocked); 1028 } 1029 1030 bind(unlock); 1031 unlock_object(robj); 1032 pop(state); 1033 1034 // Check that for block-structured locking (i.e., that all locked 1035 // objects has been unlocked) 1036 bind(unlocked); 1037 1038 // rax, rdx: Might contain return value 1039 1040 // Check that all monitors are unlocked 1041 { 1042 Label loop, exception, entry, restart; 1043 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize; 1044 const Address monitor_block_top( 1045 rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize); 1046 const Address monitor_block_bot( 1047 rbp, frame::interpreter_frame_initial_sp_offset * wordSize); 1048 1049 bind(restart); 1050 // We use c_rarg1 so that if we go slow path it will be the correct 1051 // register for unlock_object to pass to VM directly 1052 movptr(rmon, monitor_block_top); // points to current entry, starting 1053 // with top-most entry 1054 lea(rbx, monitor_block_bot); // points to word before bottom of 1055 // monitor block 1056 jmp(entry); 1057 1058 // Entry already locked, need to throw exception 1059 bind(exception); 1060 1061 if (throw_monitor_exception) { 1062 // Throw exception 1063 NOT_LP64(empty_FPU_stack();) 1064 MacroAssembler::call_VM(noreg, 1065 CAST_FROM_FN_PTR(address, InterpreterRuntime:: 1066 throw_illegal_monitor_state_exception)); 1067 should_not_reach_here(); 1068 } else { 1069 // Stack unrolling. Unlock object and install illegal_monitor_exception. 1070 // Unlock does not block, so don't have to worry about the frame. 1071 // We don't have to preserve c_rarg1 since we are going to throw an exception. 1072 1073 push(state); 1074 mov(robj, rmon); // nop if robj and rmon are the same 1075 unlock_object(robj); 1076 pop(state); 1077 1078 if (install_monitor_exception) { 1079 NOT_LP64(empty_FPU_stack();) 1080 call_VM(noreg, CAST_FROM_FN_PTR(address, 1081 InterpreterRuntime:: 1082 new_illegal_monitor_state_exception)); 1083 } 1084 1085 jmp(restart); 1086 } 1087 1088 bind(loop); 1089 // check if current entry is used 1090 cmpptr(Address(rmon, BasicObjectLock::obj_offset_in_bytes()), (int32_t) NULL); 1091 jcc(Assembler::notEqual, exception); 1092 1093 addptr(rmon, entry_size); // otherwise advance to next entry 1094 bind(entry); 1095 cmpptr(rmon, rbx); // check if bottom reached 1096 jcc(Assembler::notEqual, loop); // if not at bottom then check this entry 1097 } 1098 1099 bind(no_unlock); 1100 1101 // jvmti support 1102 if (notify_jvmdi) { 1103 notify_method_exit(state, NotifyJVMTI); // preserve TOSCA 1104 } else { 1105 notify_method_exit(state, SkipNotifyJVMTI); // preserve TOSCA 1106 } 1107 1108 if (StackReservedPages > 0) { 1109 movptr(rbx, 1110 Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); 1111 // testing if reserved zone needs to be re-enabled 1112 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx); 1113 Label no_reserved_zone_enabling; 1114 1115 NOT_LP64(get_thread(rthread);) 1116 1117 cmpl(Address(rthread, JavaThread::stack_guard_state_offset()), JavaThread::stack_guard_enabled); 1118 jcc(Assembler::equal, no_reserved_zone_enabling); 1119 1120 cmpptr(rbx, Address(rthread, JavaThread::reserved_stack_activation_offset())); 1121 jcc(Assembler::lessEqual, no_reserved_zone_enabling); 1122 1123 call_VM_leaf( 1124 CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), rthread); 1125 call_VM(noreg, CAST_FROM_FN_PTR(address, 1126 InterpreterRuntime::throw_delayed_StackOverflowError)); 1127 should_not_reach_here(); 1128 1129 bind(no_reserved_zone_enabling); 1130 } 1131 1132 if (ValueTypesBufferMaxMemory > 0) { 1133 // Code below is taking care of recycling TLVB memory, no safepoint should 1134 // occur between this point and the end of the remove_activation() method 1135 Label vtbuffer_slow, vtbuffer_done, no_buffered_value_returned; 1136 const Register thread1 = NOT_LP64(rcx) LP64_ONLY(r15_thread); 1137 const uintptr_t chunk_mask = VTBufferChunk::chunk_mask(); 1138 NOT_LP64(get_thread(thread1)); 1139 cmpptr(Address(thread1, JavaThread::return_buffered_value_offset()), (intptr_t)NULL_WORD); 1140 jcc(Assembler::equal, no_buffered_value_returned); 1141 movptr(rbx, Address(rbp, frame::interpreter_frame_vt_alloc_ptr_offset * wordSize)); 1142 call_VM_leaf(CAST_FROM_FN_PTR(address, 1143 InterpreterRuntime::return_value_step2), rax, rbx); 1144 NOT_LP64(get_thread(thread1)); 1145 get_vm_result(rax, thread1); 1146 jmp(vtbuffer_done); 1147 bind(no_buffered_value_returned); 1148 movptr(rbx, Address(rbp, frame::interpreter_frame_vt_alloc_ptr_offset * wordSize)); 1149 NOT_LP64(get_thread(thread1)); 1150 movptr(rcx, Address(thread1, JavaThread::vt_alloc_ptr_offset())); 1151 cmpptr(rbx, rcx); 1152 jcc(Assembler::equal, vtbuffer_done); 1153 andptr(rbx, chunk_mask); 1154 andptr(rcx, chunk_mask); 1155 cmpptr(rbx, rcx); 1156 jcc(Assembler::notEqual, vtbuffer_slow); 1157 movptr(rbx, Address(rbp, frame::interpreter_frame_vt_alloc_ptr_offset * wordSize)); 1158 movptr(Address(thread1, JavaThread::vt_alloc_ptr_offset()), rbx); 1159 jmp(vtbuffer_done); 1160 bind(vtbuffer_slow); 1161 push(state); 1162 movptr(rbx, Address(rbp, frame::interpreter_frame_vt_alloc_ptr_offset * wordSize)); 1163 call_VM_leaf(CAST_FROM_FN_PTR(address, 1164 InterpreterRuntime::recycle_vtbuffer), rbx); 1165 pop(state); 1166 bind(vtbuffer_done); 1167 } 1168 1169 // remove activation 1170 // get sender sp 1171 movptr(rbx, 1172 Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); 1173 1174 if (load_values) { 1175 // We are returning a value type, load its fields into registers 1176 #ifndef _LP64 1177 super_call_VM_leaf(StubRoutines::load_value_type_fields_in_regs()); 1178 #else 1179 load_klass(rdi, rax); 1180 movptr(rdi, Address(rdi, ValueKlass::unpack_handler_offset())); 1181 1182 Label skip; 1183 testptr(rdi, rdi); 1184 jcc(Assembler::equal, skip); 1185 1186 // Load fields from a buffered value with a value class specific 1187 // handler 1188 call(rdi); 1189 1190 bind(skip); 1191 #endif 1192 // call above kills the value in rbx. Reload it. 1193 movptr(rbx, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); 1194 } 1195 leave(); // remove frame anchor 1196 pop(ret_addr); // get return address 1197 mov(rsp, rbx); // set sp to sender sp 1198 } 1199 1200 void InterpreterMacroAssembler::get_method_counters(Register method, 1201 Register mcs, Label& skip) { 1202 Label has_counters; 1203 movptr(mcs, Address(method, Method::method_counters_offset())); 1204 testptr(mcs, mcs); 1205 jcc(Assembler::notZero, has_counters); 1206 call_VM(noreg, CAST_FROM_FN_PTR(address, 1207 InterpreterRuntime::build_method_counters), method); 1208 movptr(mcs, Address(method,Method::method_counters_offset())); 1209 testptr(mcs, mcs); 1210 jcc(Assembler::zero, skip); // No MethodCounters allocated, OutOfMemory 1211 bind(has_counters); 1212 } 1213 1214 1215 // Lock object 1216 // 1217 // Args: 1218 // rdx, c_rarg1: BasicObjectLock to be used for locking 1219 // 1220 // Kills: 1221 // rax, rbx 1222 void InterpreterMacroAssembler::lock_object(Register lock_reg) { 1223 assert(lock_reg == LP64_ONLY(c_rarg1) NOT_LP64(rdx), 1224 "The argument is only for looks. It must be c_rarg1"); 1225 1226 if (UseHeavyMonitors) { 1227 call_VM(noreg, 1228 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), 1229 lock_reg); 1230 } else { 1231 Label done; 1232 1233 const Register swap_reg = rax; // Must use rax for cmpxchg instruction 1234 const Register tmp_reg = rbx; // Will be passed to biased_locking_enter to avoid a 1235 // problematic case where tmp_reg = no_reg. 1236 const Register obj_reg = LP64_ONLY(c_rarg3) NOT_LP64(rcx); // Will contain the oop 1237 1238 const int obj_offset = BasicObjectLock::obj_offset_in_bytes(); 1239 const int lock_offset = BasicObjectLock::lock_offset_in_bytes (); 1240 const int mark_offset = lock_offset + 1241 BasicLock::displaced_header_offset_in_bytes(); 1242 1243 Label slow_case; 1244 1245 // Load object pointer into obj_reg 1246 movptr(obj_reg, Address(lock_reg, obj_offset)); 1247 1248 if (UseBiasedLocking) { 1249 biased_locking_enter(lock_reg, obj_reg, swap_reg, tmp_reg, false, done, &slow_case); 1250 } 1251 1252 // Load immediate 1 into swap_reg %rax 1253 movl(swap_reg, (int32_t)1); 1254 1255 // Load (object->mark() | 1) into swap_reg %rax 1256 orptr(swap_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes())); 1257 1258 // Save (object->mark() | 1) into BasicLock's displaced header 1259 movptr(Address(lock_reg, mark_offset), swap_reg); 1260 1261 assert(lock_offset == 0, 1262 "displaced header must be first word in BasicObjectLock"); 1263 1264 if (os::is_MP()) lock(); 1265 cmpxchgptr(lock_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes())); 1266 if (PrintBiasedLockingStatistics) { 1267 cond_inc32(Assembler::zero, 1268 ExternalAddress((address) BiasedLocking::fast_path_entry_count_addr())); 1269 } 1270 jcc(Assembler::zero, done); 1271 1272 const int zero_bits = LP64_ONLY(7) NOT_LP64(3); 1273 1274 // Test if the oopMark is an obvious stack pointer, i.e., 1275 // 1) (mark & zero_bits) == 0, and 1276 // 2) rsp <= mark < mark + os::pagesize() 1277 // 1278 // These 3 tests can be done by evaluating the following 1279 // expression: ((mark - rsp) & (zero_bits - os::vm_page_size())), 1280 // assuming both stack pointer and pagesize have their 1281 // least significant bits clear. 1282 // NOTE: the oopMark is in swap_reg %rax as the result of cmpxchg 1283 subptr(swap_reg, rsp); 1284 andptr(swap_reg, zero_bits - os::vm_page_size()); 1285 1286 // Save the test result, for recursive case, the result is zero 1287 movptr(Address(lock_reg, mark_offset), swap_reg); 1288 1289 if (PrintBiasedLockingStatistics) { 1290 cond_inc32(Assembler::zero, 1291 ExternalAddress((address) BiasedLocking::fast_path_entry_count_addr())); 1292 } 1293 jcc(Assembler::zero, done); 1294 1295 bind(slow_case); 1296 1297 // Call the runtime routine for slow case 1298 call_VM(noreg, 1299 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), 1300 lock_reg); 1301 1302 bind(done); 1303 } 1304 } 1305 1306 1307 // Unlocks an object. Used in monitorexit bytecode and 1308 // remove_activation. Throws an IllegalMonitorException if object is 1309 // not locked by current thread. 1310 // 1311 // Args: 1312 // rdx, c_rarg1: BasicObjectLock for lock 1313 // 1314 // Kills: 1315 // rax 1316 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ... (param regs) 1317 // rscratch1 (scratch reg) 1318 // rax, rbx, rcx, rdx 1319 void InterpreterMacroAssembler::unlock_object(Register lock_reg) { 1320 assert(lock_reg == LP64_ONLY(c_rarg1) NOT_LP64(rdx), 1321 "The argument is only for looks. It must be c_rarg1"); 1322 1323 if (UseHeavyMonitors) { 1324 call_VM(noreg, 1325 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), 1326 lock_reg); 1327 } else { 1328 Label done; 1329 1330 const Register swap_reg = rax; // Must use rax for cmpxchg instruction 1331 const Register header_reg = LP64_ONLY(c_rarg2) NOT_LP64(rbx); // Will contain the old oopMark 1332 const Register obj_reg = LP64_ONLY(c_rarg3) NOT_LP64(rcx); // Will contain the oop 1333 1334 save_bcp(); // Save in case of exception 1335 1336 // Convert from BasicObjectLock structure to object and BasicLock 1337 // structure Store the BasicLock address into %rax 1338 lea(swap_reg, Address(lock_reg, BasicObjectLock::lock_offset_in_bytes())); 1339 1340 // Load oop into obj_reg(%c_rarg3) 1341 movptr(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset_in_bytes())); 1342 1343 // Free entry 1344 movptr(Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()), (int32_t)NULL_WORD); 1345 1346 if (UseBiasedLocking) { 1347 biased_locking_exit(obj_reg, header_reg, done); 1348 } 1349 1350 // Load the old header from BasicLock structure 1351 movptr(header_reg, Address(swap_reg, 1352 BasicLock::displaced_header_offset_in_bytes())); 1353 1354 // Test for recursion 1355 testptr(header_reg, header_reg); 1356 1357 // zero for recursive case 1358 jcc(Assembler::zero, done); 1359 1360 // Atomic swap back the old header 1361 if (os::is_MP()) lock(); 1362 cmpxchgptr(header_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes())); 1363 1364 // zero for simple unlock of a stack-lock case 1365 jcc(Assembler::zero, done); 1366 1367 // Call the runtime routine for slow case. 1368 movptr(Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()), 1369 obj_reg); // restore obj 1370 call_VM(noreg, 1371 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), 1372 lock_reg); 1373 1374 bind(done); 1375 1376 restore_bcp(); 1377 } 1378 } 1379 1380 void InterpreterMacroAssembler::test_method_data_pointer(Register mdp, 1381 Label& zero_continue) { 1382 assert(ProfileInterpreter, "must be profiling interpreter"); 1383 movptr(mdp, Address(rbp, frame::interpreter_frame_mdp_offset * wordSize)); 1384 testptr(mdp, mdp); 1385 jcc(Assembler::zero, zero_continue); 1386 } 1387 1388 1389 // Set the method data pointer for the current bcp. 1390 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() { 1391 assert(ProfileInterpreter, "must be profiling interpreter"); 1392 Label set_mdp; 1393 push(rax); 1394 push(rbx); 1395 1396 get_method(rbx); 1397 // Test MDO to avoid the call if it is NULL. 1398 movptr(rax, Address(rbx, in_bytes(Method::method_data_offset()))); 1399 testptr(rax, rax); 1400 jcc(Assembler::zero, set_mdp); 1401 // rbx: method 1402 // _bcp_register: bcp 1403 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), rbx, _bcp_register); 1404 // rax: mdi 1405 // mdo is guaranteed to be non-zero here, we checked for it before the call. 1406 movptr(rbx, Address(rbx, in_bytes(Method::method_data_offset()))); 1407 addptr(rbx, in_bytes(MethodData::data_offset())); 1408 addptr(rax, rbx); 1409 bind(set_mdp); 1410 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), rax); 1411 pop(rbx); 1412 pop(rax); 1413 } 1414 1415 void InterpreterMacroAssembler::verify_method_data_pointer() { 1416 assert(ProfileInterpreter, "must be profiling interpreter"); 1417 #ifdef ASSERT 1418 Label verify_continue; 1419 push(rax); 1420 push(rbx); 1421 Register arg3_reg = LP64_ONLY(c_rarg3) NOT_LP64(rcx); 1422 Register arg2_reg = LP64_ONLY(c_rarg2) NOT_LP64(rdx); 1423 push(arg3_reg); 1424 push(arg2_reg); 1425 test_method_data_pointer(arg3_reg, verify_continue); // If mdp is zero, continue 1426 get_method(rbx); 1427 1428 // If the mdp is valid, it will point to a DataLayout header which is 1429 // consistent with the bcp. The converse is highly probable also. 1430 load_unsigned_short(arg2_reg, 1431 Address(arg3_reg, in_bytes(DataLayout::bci_offset()))); 1432 addptr(arg2_reg, Address(rbx, Method::const_offset())); 1433 lea(arg2_reg, Address(arg2_reg, ConstMethod::codes_offset())); 1434 cmpptr(arg2_reg, _bcp_register); 1435 jcc(Assembler::equal, verify_continue); 1436 // rbx: method 1437 // _bcp_register: bcp 1438 // c_rarg3: mdp 1439 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp), 1440 rbx, _bcp_register, arg3_reg); 1441 bind(verify_continue); 1442 pop(arg2_reg); 1443 pop(arg3_reg); 1444 pop(rbx); 1445 pop(rax); 1446 #endif // ASSERT 1447 } 1448 1449 1450 void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in, 1451 int constant, 1452 Register value) { 1453 assert(ProfileInterpreter, "must be profiling interpreter"); 1454 Address data(mdp_in, constant); 1455 movptr(data, value); 1456 } 1457 1458 1459 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in, 1460 int constant, 1461 bool decrement) { 1462 // Counter address 1463 Address data(mdp_in, constant); 1464 1465 increment_mdp_data_at(data, decrement); 1466 } 1467 1468 void InterpreterMacroAssembler::increment_mdp_data_at(Address data, 1469 bool decrement) { 1470 assert(ProfileInterpreter, "must be profiling interpreter"); 1471 // %%% this does 64bit counters at best it is wasting space 1472 // at worst it is a rare bug when counters overflow 1473 1474 if (decrement) { 1475 // Decrement the register. Set condition codes. 1476 addptr(data, (int32_t) -DataLayout::counter_increment); 1477 // If the decrement causes the counter to overflow, stay negative 1478 Label L; 1479 jcc(Assembler::negative, L); 1480 addptr(data, (int32_t) DataLayout::counter_increment); 1481 bind(L); 1482 } else { 1483 assert(DataLayout::counter_increment == 1, 1484 "flow-free idiom only works with 1"); 1485 // Increment the register. Set carry flag. 1486 addptr(data, DataLayout::counter_increment); 1487 // If the increment causes the counter to overflow, pull back by 1. 1488 sbbptr(data, (int32_t)0); 1489 } 1490 } 1491 1492 1493 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in, 1494 Register reg, 1495 int constant, 1496 bool decrement) { 1497 Address data(mdp_in, reg, Address::times_1, constant); 1498 1499 increment_mdp_data_at(data, decrement); 1500 } 1501 1502 void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in, 1503 int flag_byte_constant) { 1504 assert(ProfileInterpreter, "must be profiling interpreter"); 1505 int header_offset = in_bytes(DataLayout::header_offset()); 1506 int header_bits = DataLayout::flag_mask_to_header_mask(flag_byte_constant); 1507 // Set the flag 1508 orl(Address(mdp_in, header_offset), header_bits); 1509 } 1510 1511 1512 1513 void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in, 1514 int offset, 1515 Register value, 1516 Register test_value_out, 1517 Label& not_equal_continue) { 1518 assert(ProfileInterpreter, "must be profiling interpreter"); 1519 if (test_value_out == noreg) { 1520 cmpptr(value, Address(mdp_in, offset)); 1521 } else { 1522 // Put the test value into a register, so caller can use it: 1523 movptr(test_value_out, Address(mdp_in, offset)); 1524 cmpptr(test_value_out, value); 1525 } 1526 jcc(Assembler::notEqual, not_equal_continue); 1527 } 1528 1529 1530 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in, 1531 int offset_of_disp) { 1532 assert(ProfileInterpreter, "must be profiling interpreter"); 1533 Address disp_address(mdp_in, offset_of_disp); 1534 addptr(mdp_in, disp_address); 1535 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in); 1536 } 1537 1538 1539 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in, 1540 Register reg, 1541 int offset_of_disp) { 1542 assert(ProfileInterpreter, "must be profiling interpreter"); 1543 Address disp_address(mdp_in, reg, Address::times_1, offset_of_disp); 1544 addptr(mdp_in, disp_address); 1545 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in); 1546 } 1547 1548 1549 void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in, 1550 int constant) { 1551 assert(ProfileInterpreter, "must be profiling interpreter"); 1552 addptr(mdp_in, constant); 1553 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in); 1554 } 1555 1556 1557 void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) { 1558 assert(ProfileInterpreter, "must be profiling interpreter"); 1559 push(return_bci); // save/restore across call_VM 1560 call_VM(noreg, 1561 CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret), 1562 return_bci); 1563 pop(return_bci); 1564 } 1565 1566 1567 void InterpreterMacroAssembler::profile_taken_branch(Register mdp, 1568 Register bumped_count) { 1569 if (ProfileInterpreter) { 1570 Label profile_continue; 1571 1572 // If no method data exists, go to profile_continue. 1573 // Otherwise, assign to mdp 1574 test_method_data_pointer(mdp, profile_continue); 1575 1576 // We are taking a branch. Increment the taken count. 1577 // We inline increment_mdp_data_at to return bumped_count in a register 1578 //increment_mdp_data_at(mdp, in_bytes(JumpData::taken_offset())); 1579 Address data(mdp, in_bytes(JumpData::taken_offset())); 1580 movptr(bumped_count, data); 1581 assert(DataLayout::counter_increment == 1, 1582 "flow-free idiom only works with 1"); 1583 addptr(bumped_count, DataLayout::counter_increment); 1584 sbbptr(bumped_count, 0); 1585 movptr(data, bumped_count); // Store back out 1586 1587 // The method data pointer needs to be updated to reflect the new target. 1588 update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset())); 1589 bind(profile_continue); 1590 } 1591 } 1592 1593 1594 void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp) { 1595 if (ProfileInterpreter) { 1596 Label profile_continue; 1597 1598 // If no method data exists, go to profile_continue. 1599 test_method_data_pointer(mdp, profile_continue); 1600 1601 // We are taking a branch. Increment the not taken count. 1602 increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset())); 1603 1604 // The method data pointer needs to be updated to correspond to 1605 // the next bytecode 1606 update_mdp_by_constant(mdp, in_bytes(BranchData::branch_data_size())); 1607 bind(profile_continue); 1608 } 1609 } 1610 1611 void InterpreterMacroAssembler::profile_call(Register mdp) { 1612 if (ProfileInterpreter) { 1613 Label profile_continue; 1614 1615 // If no method data exists, go to profile_continue. 1616 test_method_data_pointer(mdp, profile_continue); 1617 1618 // We are making a call. Increment the count. 1619 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); 1620 1621 // The method data pointer needs to be updated to reflect the new target. 1622 update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size())); 1623 bind(profile_continue); 1624 } 1625 } 1626 1627 1628 void InterpreterMacroAssembler::profile_final_call(Register mdp) { 1629 if (ProfileInterpreter) { 1630 Label profile_continue; 1631 1632 // If no method data exists, go to profile_continue. 1633 test_method_data_pointer(mdp, profile_continue); 1634 1635 // We are making a call. Increment the count. 1636 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); 1637 1638 // The method data pointer needs to be updated to reflect the new target. 1639 update_mdp_by_constant(mdp, 1640 in_bytes(VirtualCallData:: 1641 virtual_call_data_size())); 1642 bind(profile_continue); 1643 } 1644 } 1645 1646 1647 void InterpreterMacroAssembler::profile_virtual_call(Register receiver, 1648 Register mdp, 1649 Register reg2, 1650 bool receiver_can_be_null) { 1651 if (ProfileInterpreter) { 1652 Label profile_continue; 1653 1654 // If no method data exists, go to profile_continue. 1655 test_method_data_pointer(mdp, profile_continue); 1656 1657 Label skip_receiver_profile; 1658 if (receiver_can_be_null) { 1659 Label not_null; 1660 testptr(receiver, receiver); 1661 jccb(Assembler::notZero, not_null); 1662 // We are making a call. Increment the count for null receiver. 1663 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); 1664 jmp(skip_receiver_profile); 1665 bind(not_null); 1666 } 1667 1668 // Record the receiver type. 1669 record_klass_in_profile(receiver, mdp, reg2, true); 1670 bind(skip_receiver_profile); 1671 1672 // The method data pointer needs to be updated to reflect the new target. 1673 #if INCLUDE_JVMCI 1674 if (MethodProfileWidth == 0) { 1675 update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size())); 1676 } 1677 #else // INCLUDE_JVMCI 1678 update_mdp_by_constant(mdp, 1679 in_bytes(VirtualCallData:: 1680 virtual_call_data_size())); 1681 #endif // INCLUDE_JVMCI 1682 bind(profile_continue); 1683 } 1684 } 1685 1686 #if INCLUDE_JVMCI 1687 void InterpreterMacroAssembler::profile_called_method(Register method, Register mdp, Register reg2) { 1688 assert_different_registers(method, mdp, reg2); 1689 if (ProfileInterpreter && MethodProfileWidth > 0) { 1690 Label profile_continue; 1691 1692 // If no method data exists, go to profile_continue. 1693 test_method_data_pointer(mdp, profile_continue); 1694 1695 Label done; 1696 record_item_in_profile_helper(method, mdp, reg2, 0, done, MethodProfileWidth, 1697 &VirtualCallData::method_offset, &VirtualCallData::method_count_offset, in_bytes(VirtualCallData::nonprofiled_receiver_count_offset())); 1698 bind(done); 1699 1700 update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size())); 1701 bind(profile_continue); 1702 } 1703 } 1704 #endif // INCLUDE_JVMCI 1705 1706 // This routine creates a state machine for updating the multi-row 1707 // type profile at a virtual call site (or other type-sensitive bytecode). 1708 // The machine visits each row (of receiver/count) until the receiver type 1709 // is found, or until it runs out of rows. At the same time, it remembers 1710 // the location of the first empty row. (An empty row records null for its 1711 // receiver, and can be allocated for a newly-observed receiver type.) 1712 // Because there are two degrees of freedom in the state, a simple linear 1713 // search will not work; it must be a decision tree. Hence this helper 1714 // function is recursive, to generate the required tree structured code. 1715 // It's the interpreter, so we are trading off code space for speed. 1716 // See below for example code. 1717 void InterpreterMacroAssembler::record_klass_in_profile_helper( 1718 Register receiver, Register mdp, 1719 Register reg2, int start_row, 1720 Label& done, bool is_virtual_call) { 1721 if (TypeProfileWidth == 0) { 1722 if (is_virtual_call) { 1723 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); 1724 } 1725 #if INCLUDE_JVMCI 1726 else if (EnableJVMCI) { 1727 increment_mdp_data_at(mdp, in_bytes(ReceiverTypeData::nonprofiled_receiver_count_offset())); 1728 } 1729 #endif // INCLUDE_JVMCI 1730 } else { 1731 int non_profiled_offset = -1; 1732 if (is_virtual_call) { 1733 non_profiled_offset = in_bytes(CounterData::count_offset()); 1734 } 1735 #if INCLUDE_JVMCI 1736 else if (EnableJVMCI) { 1737 non_profiled_offset = in_bytes(ReceiverTypeData::nonprofiled_receiver_count_offset()); 1738 } 1739 #endif // INCLUDE_JVMCI 1740 1741 record_item_in_profile_helper(receiver, mdp, reg2, 0, done, TypeProfileWidth, 1742 &VirtualCallData::receiver_offset, &VirtualCallData::receiver_count_offset, non_profiled_offset); 1743 } 1744 } 1745 1746 void InterpreterMacroAssembler::record_item_in_profile_helper(Register item, Register mdp, 1747 Register reg2, int start_row, Label& done, int total_rows, 1748 OffsetFunction item_offset_fn, OffsetFunction item_count_offset_fn, 1749 int non_profiled_offset) { 1750 int last_row = total_rows - 1; 1751 assert(start_row <= last_row, "must be work left to do"); 1752 // Test this row for both the item and for null. 1753 // Take any of three different outcomes: 1754 // 1. found item => increment count and goto done 1755 // 2. found null => keep looking for case 1, maybe allocate this cell 1756 // 3. found something else => keep looking for cases 1 and 2 1757 // Case 3 is handled by a recursive call. 1758 for (int row = start_row; row <= last_row; row++) { 1759 Label next_test; 1760 bool test_for_null_also = (row == start_row); 1761 1762 // See if the item is item[n]. 1763 int item_offset = in_bytes(item_offset_fn(row)); 1764 test_mdp_data_at(mdp, item_offset, item, 1765 (test_for_null_also ? reg2 : noreg), 1766 next_test); 1767 // (Reg2 now contains the item from the CallData.) 1768 1769 // The item is item[n]. Increment count[n]. 1770 int count_offset = in_bytes(item_count_offset_fn(row)); 1771 increment_mdp_data_at(mdp, count_offset); 1772 jmp(done); 1773 bind(next_test); 1774 1775 if (test_for_null_also) { 1776 Label found_null; 1777 // Failed the equality check on item[n]... Test for null. 1778 testptr(reg2, reg2); 1779 if (start_row == last_row) { 1780 // The only thing left to do is handle the null case. 1781 if (non_profiled_offset >= 0) { 1782 jccb(Assembler::zero, found_null); 1783 // Item did not match any saved item and there is no empty row for it. 1784 // Increment total counter to indicate polymorphic case. 1785 increment_mdp_data_at(mdp, non_profiled_offset); 1786 jmp(done); 1787 bind(found_null); 1788 } else { 1789 jcc(Assembler::notZero, done); 1790 } 1791 break; 1792 } 1793 // Since null is rare, make it be the branch-taken case. 1794 jcc(Assembler::zero, found_null); 1795 1796 // Put all the "Case 3" tests here. 1797 record_item_in_profile_helper(item, mdp, reg2, start_row + 1, done, total_rows, 1798 item_offset_fn, item_count_offset_fn, non_profiled_offset); 1799 1800 // Found a null. Keep searching for a matching item, 1801 // but remember that this is an empty (unused) slot. 1802 bind(found_null); 1803 } 1804 } 1805 1806 // In the fall-through case, we found no matching item, but we 1807 // observed the item[start_row] is NULL. 1808 1809 // Fill in the item field and increment the count. 1810 int item_offset = in_bytes(item_offset_fn(start_row)); 1811 set_mdp_data_at(mdp, item_offset, item); 1812 int count_offset = in_bytes(item_count_offset_fn(start_row)); 1813 movl(reg2, DataLayout::counter_increment); 1814 set_mdp_data_at(mdp, count_offset, reg2); 1815 if (start_row > 0) { 1816 jmp(done); 1817 } 1818 } 1819 1820 // Example state machine code for three profile rows: 1821 // // main copy of decision tree, rooted at row[1] 1822 // if (row[0].rec == rec) { row[0].incr(); goto done; } 1823 // if (row[0].rec != NULL) { 1824 // // inner copy of decision tree, rooted at row[1] 1825 // if (row[1].rec == rec) { row[1].incr(); goto done; } 1826 // if (row[1].rec != NULL) { 1827 // // degenerate decision tree, rooted at row[2] 1828 // if (row[2].rec == rec) { row[2].incr(); goto done; } 1829 // if (row[2].rec != NULL) { count.incr(); goto done; } // overflow 1830 // row[2].init(rec); goto done; 1831 // } else { 1832 // // remember row[1] is empty 1833 // if (row[2].rec == rec) { row[2].incr(); goto done; } 1834 // row[1].init(rec); goto done; 1835 // } 1836 // } else { 1837 // // remember row[0] is empty 1838 // if (row[1].rec == rec) { row[1].incr(); goto done; } 1839 // if (row[2].rec == rec) { row[2].incr(); goto done; } 1840 // row[0].init(rec); goto done; 1841 // } 1842 // done: 1843 1844 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver, 1845 Register mdp, Register reg2, 1846 bool is_virtual_call) { 1847 assert(ProfileInterpreter, "must be profiling"); 1848 Label done; 1849 1850 record_klass_in_profile_helper(receiver, mdp, reg2, 0, done, is_virtual_call); 1851 1852 bind (done); 1853 } 1854 1855 void InterpreterMacroAssembler::profile_ret(Register return_bci, 1856 Register mdp) { 1857 if (ProfileInterpreter) { 1858 Label profile_continue; 1859 uint row; 1860 1861 // If no method data exists, go to profile_continue. 1862 test_method_data_pointer(mdp, profile_continue); 1863 1864 // Update the total ret count. 1865 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); 1866 1867 for (row = 0; row < RetData::row_limit(); row++) { 1868 Label next_test; 1869 1870 // See if return_bci is equal to bci[n]: 1871 test_mdp_data_at(mdp, 1872 in_bytes(RetData::bci_offset(row)), 1873 return_bci, noreg, 1874 next_test); 1875 1876 // return_bci is equal to bci[n]. Increment the count. 1877 increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row))); 1878 1879 // The method data pointer needs to be updated to reflect the new target. 1880 update_mdp_by_offset(mdp, 1881 in_bytes(RetData::bci_displacement_offset(row))); 1882 jmp(profile_continue); 1883 bind(next_test); 1884 } 1885 1886 update_mdp_for_ret(return_bci); 1887 1888 bind(profile_continue); 1889 } 1890 } 1891 1892 1893 void InterpreterMacroAssembler::profile_null_seen(Register mdp) { 1894 if (ProfileInterpreter) { 1895 Label profile_continue; 1896 1897 // If no method data exists, go to profile_continue. 1898 test_method_data_pointer(mdp, profile_continue); 1899 1900 set_mdp_flag_at(mdp, BitData::null_seen_byte_constant()); 1901 1902 // The method data pointer needs to be updated. 1903 int mdp_delta = in_bytes(BitData::bit_data_size()); 1904 if (TypeProfileCasts) { 1905 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size()); 1906 } 1907 update_mdp_by_constant(mdp, mdp_delta); 1908 1909 bind(profile_continue); 1910 } 1911 } 1912 1913 1914 void InterpreterMacroAssembler::profile_typecheck_failed(Register mdp) { 1915 if (ProfileInterpreter && TypeProfileCasts) { 1916 Label profile_continue; 1917 1918 // If no method data exists, go to profile_continue. 1919 test_method_data_pointer(mdp, profile_continue); 1920 1921 int count_offset = in_bytes(CounterData::count_offset()); 1922 // Back up the address, since we have already bumped the mdp. 1923 count_offset -= in_bytes(VirtualCallData::virtual_call_data_size()); 1924 1925 // *Decrement* the counter. We expect to see zero or small negatives. 1926 increment_mdp_data_at(mdp, count_offset, true); 1927 1928 bind (profile_continue); 1929 } 1930 } 1931 1932 1933 void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) { 1934 if (ProfileInterpreter) { 1935 Label profile_continue; 1936 1937 // If no method data exists, go to profile_continue. 1938 test_method_data_pointer(mdp, profile_continue); 1939 1940 // The method data pointer needs to be updated. 1941 int mdp_delta = in_bytes(BitData::bit_data_size()); 1942 if (TypeProfileCasts) { 1943 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size()); 1944 1945 // Record the object type. 1946 record_klass_in_profile(klass, mdp, reg2, false); 1947 NOT_LP64(assert(reg2 == rdi, "we know how to fix this blown reg");) 1948 NOT_LP64(restore_locals();) // Restore EDI 1949 } 1950 update_mdp_by_constant(mdp, mdp_delta); 1951 1952 bind(profile_continue); 1953 } 1954 } 1955 1956 1957 void InterpreterMacroAssembler::profile_switch_default(Register mdp) { 1958 if (ProfileInterpreter) { 1959 Label profile_continue; 1960 1961 // If no method data exists, go to profile_continue. 1962 test_method_data_pointer(mdp, profile_continue); 1963 1964 // Update the default case count 1965 increment_mdp_data_at(mdp, 1966 in_bytes(MultiBranchData::default_count_offset())); 1967 1968 // The method data pointer needs to be updated. 1969 update_mdp_by_offset(mdp, 1970 in_bytes(MultiBranchData:: 1971 default_displacement_offset())); 1972 1973 bind(profile_continue); 1974 } 1975 } 1976 1977 1978 void InterpreterMacroAssembler::profile_switch_case(Register index, 1979 Register mdp, 1980 Register reg2) { 1981 if (ProfileInterpreter) { 1982 Label profile_continue; 1983 1984 // If no method data exists, go to profile_continue. 1985 test_method_data_pointer(mdp, profile_continue); 1986 1987 // Build the base (index * per_case_size_in_bytes()) + 1988 // case_array_offset_in_bytes() 1989 movl(reg2, in_bytes(MultiBranchData::per_case_size())); 1990 imulptr(index, reg2); // XXX l ? 1991 addptr(index, in_bytes(MultiBranchData::case_array_offset())); // XXX l ? 1992 1993 // Update the case count 1994 increment_mdp_data_at(mdp, 1995 index, 1996 in_bytes(MultiBranchData::relative_count_offset())); 1997 1998 // The method data pointer needs to be updated. 1999 update_mdp_by_offset(mdp, 2000 index, 2001 in_bytes(MultiBranchData:: 2002 relative_displacement_offset())); 2003 2004 bind(profile_continue); 2005 } 2006 } 2007 2008 2009 2010 void InterpreterMacroAssembler::verify_oop(Register reg, TosState state) { 2011 if (state == atos) { 2012 MacroAssembler::verify_oop(reg); 2013 } 2014 } 2015 2016 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) { 2017 #ifndef _LP64 2018 if ((state == ftos && UseSSE < 1) || 2019 (state == dtos && UseSSE < 2)) { 2020 MacroAssembler::verify_FPU(stack_depth); 2021 } 2022 #endif 2023 } 2024 2025 // Jump if ((*counter_addr += increment) & mask) satisfies the condition. 2026 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr, 2027 int increment, Address mask, 2028 Register scratch, bool preloaded, 2029 Condition cond, Label* where) { 2030 if (!preloaded) { 2031 movl(scratch, counter_addr); 2032 } 2033 incrementl(scratch, increment); 2034 movl(counter_addr, scratch); 2035 andl(scratch, mask); 2036 jcc(cond, *where); 2037 } 2038 2039 void InterpreterMacroAssembler::notify_method_entry() { 2040 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to 2041 // track stack depth. If it is possible to enter interp_only_mode we add 2042 // the code to check if the event should be sent. 2043 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx); 2044 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rbx); 2045 if (JvmtiExport::can_post_interpreter_events()) { 2046 Label L; 2047 NOT_LP64(get_thread(rthread);) 2048 movl(rdx, Address(rthread, JavaThread::interp_only_mode_offset())); 2049 testl(rdx, rdx); 2050 jcc(Assembler::zero, L); 2051 call_VM(noreg, CAST_FROM_FN_PTR(address, 2052 InterpreterRuntime::post_method_entry)); 2053 bind(L); 2054 } 2055 2056 { 2057 SkipIfEqual skip(this, &DTraceMethodProbes, false); 2058 NOT_LP64(get_thread(rthread);) 2059 get_method(rarg); 2060 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry), 2061 rthread, rarg); 2062 } 2063 2064 // RedefineClasses() tracing support for obsolete method entry 2065 if (log_is_enabled(Trace, redefine, class, obsolete)) { 2066 NOT_LP64(get_thread(rthread);) 2067 get_method(rarg); 2068 call_VM_leaf( 2069 CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry), 2070 rthread, rarg); 2071 } 2072 } 2073 2074 2075 void InterpreterMacroAssembler::notify_method_exit( 2076 TosState state, NotifyMethodExitMode mode) { 2077 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to 2078 // track stack depth. If it is possible to enter interp_only_mode we add 2079 // the code to check if the event should be sent. 2080 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx); 2081 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rbx); 2082 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) { 2083 Label L; 2084 // Note: frame::interpreter_frame_result has a dependency on how the 2085 // method result is saved across the call to post_method_exit. If this 2086 // is changed then the interpreter_frame_result implementation will 2087 // need to be updated too. 2088 2089 // template interpreter will leave the result on the top of the stack. 2090 push(state); 2091 NOT_LP64(get_thread(rthread);) 2092 movl(rdx, Address(rthread, JavaThread::interp_only_mode_offset())); 2093 testl(rdx, rdx); 2094 jcc(Assembler::zero, L); 2095 call_VM(noreg, 2096 CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit)); 2097 bind(L); 2098 pop(state); 2099 } 2100 2101 { 2102 SkipIfEqual skip(this, &DTraceMethodProbes, false); 2103 push(state); 2104 NOT_LP64(get_thread(rthread);) 2105 get_method(rarg); 2106 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit), 2107 rthread, rarg); 2108 pop(state); 2109 } 2110 }