1 /* 2 * Copyright (c) 2003, 2021, Oracle and/or its affiliates. All rights reserved. 3 * Copyright (c) 2014, 2020, Red Hat Inc. All rights reserved. 4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. 5 * 6 * This code is free software; you can redistribute it and/or modify it 7 * under the terms of the GNU General Public License version 2 only, as 8 * published by the Free Software Foundation. 9 * 10 * This code is distributed in the hope that it will be useful, but WITHOUT 11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License 13 * version 2 for more details (a copy is included in the LICENSE file that 14 * accompanied this code). 15 * 16 * You should have received a copy of the GNU General Public License version 17 * 2 along with this work; if not, write to the Free Software Foundation, 18 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. 19 * 20 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA 21 * or visit www.oracle.com if you need additional information or have any 22 * questions. 23 * 24 */ 25 26 #include "precompiled.hpp" 27 #include "asm/macroAssembler.inline.hpp" 28 #include "compiler/compiler_globals.hpp" 29 #include "gc/shared/barrierSet.hpp" 30 #include "gc/shared/barrierSetAssembler.hpp" 31 #include "interp_masm_aarch64.hpp" 32 #include "interpreter/interpreter.hpp" 33 #include "interpreter/interpreterRuntime.hpp" 34 #include "logging/log.hpp" 35 #include "oops/arrayOop.hpp" 36 #include "oops/markWord.hpp" 37 #include "oops/method.hpp" 38 #include "oops/methodData.hpp" 39 #include "prims/jvmtiExport.hpp" 40 #include "prims/jvmtiThreadState.hpp" 41 #include "runtime/basicLock.hpp" 42 #include "runtime/biasedLocking.hpp" 43 #include "runtime/frame.inline.hpp" 44 #include "runtime/safepointMechanism.hpp" 45 #include "runtime/sharedRuntime.hpp" 46 #include "runtime/thread.inline.hpp" 47 #include "utilities/powerOfTwo.hpp" 48 49 void InterpreterMacroAssembler::narrow(Register result) { 50 51 // Get method->_constMethod->_result_type 52 ldr(rscratch1, Address(rfp, frame::interpreter_frame_method_offset * wordSize)); 53 ldr(rscratch1, Address(rscratch1, Method::const_offset())); 54 ldrb(rscratch1, Address(rscratch1, ConstMethod::result_type_offset())); 55 56 Label done, notBool, notByte, notChar; 57 58 // common case first 59 cmpw(rscratch1, T_INT); 60 br(Assembler::EQ, done); 61 62 // mask integer result to narrower return type. 63 cmpw(rscratch1, T_BOOLEAN); 64 br(Assembler::NE, notBool); 65 andw(result, result, 0x1); 66 b(done); 67 68 bind(notBool); 69 cmpw(rscratch1, T_BYTE); 70 br(Assembler::NE, notByte); 71 sbfx(result, result, 0, 8); 72 b(done); 73 74 bind(notByte); 75 cmpw(rscratch1, T_CHAR); 76 br(Assembler::NE, notChar); 77 ubfx(result, result, 0, 16); // truncate upper 16 bits 78 b(done); 79 80 bind(notChar); 81 sbfx(result, result, 0, 16); // sign-extend short 82 83 // Nothing to do for T_INT 84 bind(done); 85 } 86 87 void InterpreterMacroAssembler::jump_to_entry(address entry) { 88 assert(entry, "Entry must have been generated by now"); 89 b(entry); 90 } 91 92 void InterpreterMacroAssembler::check_and_handle_popframe(Register java_thread) { 93 if (JvmtiExport::can_pop_frame()) { 94 Label L; 95 // Initiate popframe handling only if it is not already being 96 // processed. If the flag has the popframe_processing bit set, it 97 // means that this code is called *during* popframe handling - we 98 // don't want to reenter. 99 // This method is only called just after the call into the vm in 100 // call_VM_base, so the arg registers are available. 101 ldrw(rscratch1, Address(rthread, JavaThread::popframe_condition_offset())); 102 tbz(rscratch1, exact_log2(JavaThread::popframe_pending_bit), L); 103 tbnz(rscratch1, exact_log2(JavaThread::popframe_processing_bit), L); 104 // Call Interpreter::remove_activation_preserving_args_entry() to get the 105 // address of the same-named entrypoint in the generated interpreter code. 106 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry)); 107 br(r0); 108 bind(L); 109 } 110 } 111 112 113 void InterpreterMacroAssembler::load_earlyret_value(TosState state) { 114 ldr(r2, Address(rthread, JavaThread::jvmti_thread_state_offset())); 115 const Address tos_addr(r2, JvmtiThreadState::earlyret_tos_offset()); 116 const Address oop_addr(r2, JvmtiThreadState::earlyret_oop_offset()); 117 const Address val_addr(r2, JvmtiThreadState::earlyret_value_offset()); 118 switch (state) { 119 case atos: ldr(r0, oop_addr); 120 str(zr, oop_addr); 121 verify_oop(r0, state); break; 122 case ltos: ldr(r0, val_addr); break; 123 case btos: // fall through 124 case ztos: // fall through 125 case ctos: // fall through 126 case stos: // fall through 127 case itos: ldrw(r0, val_addr); break; 128 case ftos: ldrs(v0, val_addr); break; 129 case dtos: ldrd(v0, val_addr); break; 130 case vtos: /* nothing to do */ break; 131 default : ShouldNotReachHere(); 132 } 133 // Clean up tos value in the thread object 134 movw(rscratch1, (int) ilgl); 135 strw(rscratch1, tos_addr); 136 strw(zr, val_addr); 137 } 138 139 140 void InterpreterMacroAssembler::check_and_handle_earlyret(Register java_thread) { 141 if (JvmtiExport::can_force_early_return()) { 142 Label L; 143 ldr(rscratch1, Address(rthread, JavaThread::jvmti_thread_state_offset())); 144 cbz(rscratch1, L); // if (thread->jvmti_thread_state() == NULL) exit; 145 146 // Initiate earlyret handling only if it is not already being processed. 147 // If the flag has the earlyret_processing bit set, it means that this code 148 // is called *during* earlyret handling - we don't want to reenter. 149 ldrw(rscratch1, Address(rscratch1, JvmtiThreadState::earlyret_state_offset())); 150 cmpw(rscratch1, JvmtiThreadState::earlyret_pending); 151 br(Assembler::NE, L); 152 153 // Call Interpreter::remove_activation_early_entry() to get the address of the 154 // same-named entrypoint in the generated interpreter code. 155 ldr(rscratch1, Address(rthread, JavaThread::jvmti_thread_state_offset())); 156 ldrw(rscratch1, Address(rscratch1, JvmtiThreadState::earlyret_tos_offset())); 157 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), rscratch1); 158 br(r0); 159 bind(L); 160 } 161 } 162 163 void InterpreterMacroAssembler::get_unsigned_2_byte_index_at_bcp( 164 Register reg, 165 int bcp_offset) { 166 assert(bcp_offset >= 0, "bcp is still pointing to start of bytecode"); 167 ldrh(reg, Address(rbcp, bcp_offset)); 168 rev16(reg, reg); 169 } 170 171 void InterpreterMacroAssembler::get_dispatch() { 172 uint64_t offset; 173 adrp(rdispatch, ExternalAddress((address)Interpreter::dispatch_table()), offset); 174 lea(rdispatch, Address(rdispatch, offset)); 175 } 176 177 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register index, 178 int bcp_offset, 179 size_t index_size) { 180 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode"); 181 if (index_size == sizeof(u2)) { 182 load_unsigned_short(index, Address(rbcp, bcp_offset)); 183 } else if (index_size == sizeof(u4)) { 184 // assert(EnableInvokeDynamic, "giant index used only for JSR 292"); 185 ldrw(index, Address(rbcp, bcp_offset)); 186 // Check if the secondary index definition is still ~x, otherwise 187 // we have to change the following assembler code to calculate the 188 // plain index. 189 assert(ConstantPool::decode_invokedynamic_index(~123) == 123, "else change next line"); 190 eonw(index, index, zr); // convert to plain index 191 } else if (index_size == sizeof(u1)) { 192 load_unsigned_byte(index, Address(rbcp, bcp_offset)); 193 } else { 194 ShouldNotReachHere(); 195 } 196 } 197 198 // Return 199 // Rindex: index into constant pool 200 // Rcache: address of cache entry - ConstantPoolCache::base_offset() 201 // 202 // A caller must add ConstantPoolCache::base_offset() to Rcache to get 203 // the true address of the cache entry. 204 // 205 void InterpreterMacroAssembler::get_cache_and_index_at_bcp(Register cache, 206 Register index, 207 int bcp_offset, 208 size_t index_size) { 209 assert_different_registers(cache, index); 210 assert_different_registers(cache, rcpool); 211 get_cache_index_at_bcp(index, bcp_offset, index_size); 212 assert(sizeof(ConstantPoolCacheEntry) == 4 * wordSize, "adjust code below"); 213 // convert from field index to ConstantPoolCacheEntry 214 // aarch64 already has the cache in rcpool so there is no need to 215 // install it in cache. instead we pre-add the indexed offset to 216 // rcpool and return it in cache. All clients of this method need to 217 // be modified accordingly. 218 add(cache, rcpool, index, Assembler::LSL, 5); 219 } 220 221 222 void InterpreterMacroAssembler::get_cache_and_index_and_bytecode_at_bcp(Register cache, 223 Register index, 224 Register bytecode, 225 int byte_no, 226 int bcp_offset, 227 size_t index_size) { 228 get_cache_and_index_at_bcp(cache, index, bcp_offset, index_size); 229 // We use a 32-bit load here since the layout of 64-bit words on 230 // little-endian machines allow us that. 231 // n.b. unlike x86 cache already includes the index offset 232 lea(bytecode, Address(cache, 233 ConstantPoolCache::base_offset() 234 + ConstantPoolCacheEntry::indices_offset())); 235 ldarw(bytecode, bytecode); 236 const int shift_count = (1 + byte_no) * BitsPerByte; 237 ubfx(bytecode, bytecode, shift_count, BitsPerByte); 238 } 239 240 void InterpreterMacroAssembler::get_cache_entry_pointer_at_bcp(Register cache, 241 Register tmp, 242 int bcp_offset, 243 size_t index_size) { 244 assert(cache != tmp, "must use different register"); 245 get_cache_index_at_bcp(tmp, bcp_offset, index_size); 246 assert(sizeof(ConstantPoolCacheEntry) == 4 * wordSize, "adjust code below"); 247 // convert from field index to ConstantPoolCacheEntry index 248 // and from word offset to byte offset 249 assert(exact_log2(in_bytes(ConstantPoolCacheEntry::size_in_bytes())) == 2 + LogBytesPerWord, "else change next line"); 250 ldr(cache, Address(rfp, frame::interpreter_frame_cache_offset * wordSize)); 251 // skip past the header 252 add(cache, cache, in_bytes(ConstantPoolCache::base_offset())); 253 add(cache, cache, tmp, Assembler::LSL, 2 + LogBytesPerWord); // construct pointer to cache entry 254 } 255 256 void InterpreterMacroAssembler::get_method_counters(Register method, 257 Register mcs, Label& skip) { 258 Label has_counters; 259 ldr(mcs, Address(method, Method::method_counters_offset())); 260 cbnz(mcs, has_counters); 261 call_VM(noreg, CAST_FROM_FN_PTR(address, 262 InterpreterRuntime::build_method_counters), method); 263 ldr(mcs, Address(method, Method::method_counters_offset())); 264 cbz(mcs, skip); // No MethodCounters allocated, OutOfMemory 265 bind(has_counters); 266 } 267 268 // Load object from cpool->resolved_references(index) 269 void InterpreterMacroAssembler::load_resolved_reference_at_index( 270 Register result, Register index, Register tmp) { 271 assert_different_registers(result, index); 272 273 get_constant_pool(result); 274 // load pointer for resolved_references[] objArray 275 ldr(result, Address(result, ConstantPool::cache_offset_in_bytes())); 276 ldr(result, Address(result, ConstantPoolCache::resolved_references_offset_in_bytes())); 277 resolve_oop_handle(result, tmp); 278 // Add in the index 279 add(index, index, arrayOopDesc::base_offset_in_bytes(T_OBJECT) >> LogBytesPerHeapOop); 280 load_heap_oop(result, Address(result, index, Address::uxtw(LogBytesPerHeapOop))); 281 } 282 283 void InterpreterMacroAssembler::load_resolved_klass_at_offset( 284 Register cpool, Register index, Register klass, Register temp) { 285 add(temp, cpool, index, LSL, LogBytesPerWord); 286 ldrh(temp, Address(temp, sizeof(ConstantPool))); // temp = resolved_klass_index 287 ldr(klass, Address(cpool, ConstantPool::resolved_klasses_offset_in_bytes())); // klass = cpool->_resolved_klasses 288 add(klass, klass, temp, LSL, LogBytesPerWord); 289 ldr(klass, Address(klass, Array<Klass*>::base_offset_in_bytes())); 290 } 291 292 void InterpreterMacroAssembler::load_resolved_method_at_index(int byte_no, 293 Register method, 294 Register cache) { 295 const int method_offset = in_bytes( 296 ConstantPoolCache::base_offset() + 297 ((byte_no == TemplateTable::f2_byte) 298 ? ConstantPoolCacheEntry::f2_offset() 299 : ConstantPoolCacheEntry::f1_offset())); 300 301 ldr(method, Address(cache, method_offset)); // get f1 Method* 302 } 303 304 // Generate a subtype check: branch to ok_is_subtype if sub_klass is a 305 // subtype of super_klass. 306 // 307 // Args: 308 // r0: superklass 309 // Rsub_klass: subklass 310 // 311 // Kills: 312 // r2, r5 313 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass, 314 Label& ok_is_subtype) { 315 assert(Rsub_klass != r0, "r0 holds superklass"); 316 assert(Rsub_klass != r2, "r2 holds 2ndary super array length"); 317 assert(Rsub_klass != r5, "r5 holds 2ndary super array scan ptr"); 318 319 // Profile the not-null value's klass. 320 profile_typecheck(r2, Rsub_klass, r5); // blows r2, reloads r5 321 322 // Do the check. 323 check_klass_subtype(Rsub_klass, r0, r2, ok_is_subtype); // blows r2 324 325 // Profile the failure of the check. 326 profile_typecheck_failed(r2); // blows r2 327 } 328 329 // Java Expression Stack 330 331 void InterpreterMacroAssembler::pop_ptr(Register r) { 332 ldr(r, post(esp, wordSize)); 333 } 334 335 void InterpreterMacroAssembler::pop_i(Register r) { 336 ldrw(r, post(esp, wordSize)); 337 } 338 339 void InterpreterMacroAssembler::pop_l(Register r) { 340 ldr(r, post(esp, 2 * Interpreter::stackElementSize)); 341 } 342 343 void InterpreterMacroAssembler::push_ptr(Register r) { 344 str(r, pre(esp, -wordSize)); 345 } 346 347 void InterpreterMacroAssembler::push_i(Register r) { 348 str(r, pre(esp, -wordSize)); 349 } 350 351 void InterpreterMacroAssembler::push_l(Register r) { 352 str(zr, pre(esp, -wordSize)); 353 str(r, pre(esp, - wordSize)); 354 } 355 356 void InterpreterMacroAssembler::pop_f(FloatRegister r) { 357 ldrs(r, post(esp, wordSize)); 358 } 359 360 void InterpreterMacroAssembler::pop_d(FloatRegister r) { 361 ldrd(r, post(esp, 2 * Interpreter::stackElementSize)); 362 } 363 364 void InterpreterMacroAssembler::push_f(FloatRegister r) { 365 strs(r, pre(esp, -wordSize)); 366 } 367 368 void InterpreterMacroAssembler::push_d(FloatRegister r) { 369 strd(r, pre(esp, 2* -wordSize)); 370 } 371 372 void InterpreterMacroAssembler::pop(TosState state) { 373 switch (state) { 374 case atos: pop_ptr(); break; 375 case btos: 376 case ztos: 377 case ctos: 378 case stos: 379 case itos: pop_i(); break; 380 case ltos: pop_l(); break; 381 case ftos: pop_f(); break; 382 case dtos: pop_d(); break; 383 case vtos: /* nothing to do */ break; 384 default: ShouldNotReachHere(); 385 } 386 verify_oop(r0, state); 387 } 388 389 void InterpreterMacroAssembler::push(TosState state) { 390 verify_oop(r0, state); 391 switch (state) { 392 case atos: push_ptr(); break; 393 case btos: 394 case ztos: 395 case ctos: 396 case stos: 397 case itos: push_i(); break; 398 case ltos: push_l(); break; 399 case ftos: push_f(); break; 400 case dtos: push_d(); break; 401 case vtos: /* nothing to do */ break; 402 default : ShouldNotReachHere(); 403 } 404 } 405 406 // Helpers for swap and dup 407 void InterpreterMacroAssembler::load_ptr(int n, Register val) { 408 ldr(val, Address(esp, Interpreter::expr_offset_in_bytes(n))); 409 } 410 411 void InterpreterMacroAssembler::store_ptr(int n, Register val) { 412 str(val, Address(esp, Interpreter::expr_offset_in_bytes(n))); 413 } 414 415 void InterpreterMacroAssembler::load_float(Address src) { 416 ldrs(v0, src); 417 } 418 419 void InterpreterMacroAssembler::load_double(Address src) { 420 ldrd(v0, src); 421 } 422 423 void InterpreterMacroAssembler::prepare_to_jump_from_interpreted() { 424 // set sender sp 425 mov(r13, sp); 426 // record last_sp 427 str(esp, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize)); 428 } 429 430 // Jump to from_interpreted entry of a call unless single stepping is possible 431 // in this thread in which case we must call the i2i entry 432 void InterpreterMacroAssembler::jump_from_interpreted(Register method, Register temp) { 433 prepare_to_jump_from_interpreted(); 434 435 if (JvmtiExport::can_post_interpreter_events()) { 436 Label run_compiled_code; 437 // JVMTI events, such as single-stepping, are implemented partly by avoiding running 438 // compiled code in threads for which the event is enabled. Check here for 439 // interp_only_mode if these events CAN be enabled. 440 ldrw(rscratch1, Address(rthread, JavaThread::interp_only_mode_offset())); 441 cbzw(rscratch1, run_compiled_code); 442 ldr(rscratch1, Address(method, Method::interpreter_entry_offset())); 443 br(rscratch1); 444 bind(run_compiled_code); 445 } 446 447 ldr(rscratch1, Address(method, Method::from_interpreted_offset())); 448 br(rscratch1); 449 } 450 451 // The following two routines provide a hook so that an implementation 452 // can schedule the dispatch in two parts. amd64 does not do this. 453 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int step) { 454 } 455 456 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int step) { 457 dispatch_next(state, step); 458 } 459 460 void InterpreterMacroAssembler::dispatch_base(TosState state, 461 address* table, 462 bool verifyoop, 463 bool generate_poll) { 464 if (VerifyActivationFrameSize) { 465 Unimplemented(); 466 } 467 if (verifyoop) { 468 verify_oop(r0, state); 469 } 470 471 Label safepoint; 472 address* const safepoint_table = Interpreter::safept_table(state); 473 bool needs_thread_local_poll = generate_poll && table != safepoint_table; 474 475 if (needs_thread_local_poll) { 476 NOT_PRODUCT(block_comment("Thread-local Safepoint poll")); 477 ldr(rscratch2, Address(rthread, JavaThread::polling_word_offset())); 478 tbnz(rscratch2, exact_log2(SafepointMechanism::poll_bit()), safepoint); 479 } 480 481 if (table == Interpreter::dispatch_table(state)) { 482 addw(rscratch2, rscratch1, Interpreter::distance_from_dispatch_table(state)); 483 ldr(rscratch2, Address(rdispatch, rscratch2, Address::uxtw(3))); 484 } else { 485 mov(rscratch2, (address)table); 486 ldr(rscratch2, Address(rscratch2, rscratch1, Address::uxtw(3))); 487 } 488 br(rscratch2); 489 490 if (needs_thread_local_poll) { 491 bind(safepoint); 492 lea(rscratch2, ExternalAddress((address)safepoint_table)); 493 ldr(rscratch2, Address(rscratch2, rscratch1, Address::uxtw(3))); 494 br(rscratch2); 495 } 496 } 497 498 void InterpreterMacroAssembler::dispatch_only(TosState state, bool generate_poll) { 499 dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll); 500 } 501 502 void InterpreterMacroAssembler::dispatch_only_normal(TosState state) { 503 dispatch_base(state, Interpreter::normal_table(state)); 504 } 505 506 void InterpreterMacroAssembler::dispatch_only_noverify(TosState state) { 507 dispatch_base(state, Interpreter::normal_table(state), false); 508 } 509 510 511 void InterpreterMacroAssembler::dispatch_next(TosState state, int step, bool generate_poll) { 512 // load next bytecode 513 ldrb(rscratch1, Address(pre(rbcp, step))); 514 dispatch_base(state, Interpreter::dispatch_table(state), generate_poll); 515 } 516 517 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) { 518 // load current bytecode 519 ldrb(rscratch1, Address(rbcp, 0)); 520 dispatch_base(state, table); 521 } 522 523 // remove activation 524 // 525 // Apply stack watermark barrier. 526 // Unlock the receiver if this is a synchronized method. 527 // Unlock any Java monitors from syncronized blocks. 528 // Remove the activation from the stack. 529 // 530 // If there are locked Java monitors 531 // If throw_monitor_exception 532 // throws IllegalMonitorStateException 533 // Else if install_monitor_exception 534 // installs IllegalMonitorStateException 535 // Else 536 // no error processing 537 void InterpreterMacroAssembler::remove_activation( 538 TosState state, 539 bool throw_monitor_exception, 540 bool install_monitor_exception, 541 bool notify_jvmdi) { 542 // Note: Registers r3 xmm0 may be in use for the 543 // result check if synchronized method 544 Label unlocked, unlock, no_unlock; 545 546 // The below poll is for the stack watermark barrier. It allows fixing up frames lazily, 547 // that would normally not be safe to use. Such bad returns into unsafe territory of 548 // the stack, will call InterpreterRuntime::at_unwind. 549 Label slow_path; 550 Label fast_path; 551 safepoint_poll(slow_path, true /* at_return */, false /* acquire */, false /* in_nmethod */); 552 br(Assembler::AL, fast_path); 553 bind(slow_path); 554 push(state); 555 set_last_Java_frame(esp, rfp, (address)pc(), rscratch1); 556 super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::at_unwind), rthread); 557 reset_last_Java_frame(true); 558 pop(state); 559 bind(fast_path); 560 561 // get the value of _do_not_unlock_if_synchronized into r3 562 const Address do_not_unlock_if_synchronized(rthread, 563 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset())); 564 ldrb(r3, do_not_unlock_if_synchronized); 565 strb(zr, do_not_unlock_if_synchronized); // reset the flag 566 567 // get method access flags 568 ldr(r1, Address(rfp, frame::interpreter_frame_method_offset * wordSize)); 569 ldr(r2, Address(r1, Method::access_flags_offset())); 570 tbz(r2, exact_log2(JVM_ACC_SYNCHRONIZED), unlocked); 571 572 // Don't unlock anything if the _do_not_unlock_if_synchronized flag 573 // is set. 574 cbnz(r3, no_unlock); 575 576 // unlock monitor 577 push(state); // save result 578 579 // BasicObjectLock will be first in list, since this is a 580 // synchronized method. However, need to check that the object has 581 // not been unlocked by an explicit monitorexit bytecode. 582 const Address monitor(rfp, frame::interpreter_frame_initial_sp_offset * 583 wordSize - (int) sizeof(BasicObjectLock)); 584 // We use c_rarg1 so that if we go slow path it will be the correct 585 // register for unlock_object to pass to VM directly 586 lea(c_rarg1, monitor); // address of first monitor 587 588 ldr(r0, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes())); 589 cbnz(r0, unlock); 590 591 pop(state); 592 if (throw_monitor_exception) { 593 // Entry already unlocked, need to throw exception 594 call_VM(noreg, CAST_FROM_FN_PTR(address, 595 InterpreterRuntime::throw_illegal_monitor_state_exception)); 596 should_not_reach_here(); 597 } else { 598 // Monitor already unlocked during a stack unroll. If requested, 599 // install an illegal_monitor_state_exception. Continue with 600 // stack unrolling. 601 if (install_monitor_exception) { 602 call_VM(noreg, CAST_FROM_FN_PTR(address, 603 InterpreterRuntime::new_illegal_monitor_state_exception)); 604 } 605 b(unlocked); 606 } 607 608 bind(unlock); 609 unlock_object(c_rarg1); 610 pop(state); 611 612 // Check that for block-structured locking (i.e., that all locked 613 // objects has been unlocked) 614 bind(unlocked); 615 616 // r0: Might contain return value 617 618 // Check that all monitors are unlocked 619 { 620 Label loop, exception, entry, restart; 621 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize; 622 const Address monitor_block_top( 623 rfp, frame::interpreter_frame_monitor_block_top_offset * wordSize); 624 const Address monitor_block_bot( 625 rfp, frame::interpreter_frame_initial_sp_offset * wordSize); 626 627 bind(restart); 628 // We use c_rarg1 so that if we go slow path it will be the correct 629 // register for unlock_object to pass to VM directly 630 ldr(c_rarg1, monitor_block_top); // points to current entry, starting 631 // with top-most entry 632 lea(r19, monitor_block_bot); // points to word before bottom of 633 // monitor block 634 b(entry); 635 636 // Entry already locked, need to throw exception 637 bind(exception); 638 639 if (throw_monitor_exception) { 640 // Throw exception 641 MacroAssembler::call_VM(noreg, 642 CAST_FROM_FN_PTR(address, InterpreterRuntime:: 643 throw_illegal_monitor_state_exception)); 644 should_not_reach_here(); 645 } else { 646 // Stack unrolling. Unlock object and install illegal_monitor_exception. 647 // Unlock does not block, so don't have to worry about the frame. 648 // We don't have to preserve c_rarg1 since we are going to throw an exception. 649 650 push(state); 651 unlock_object(c_rarg1); 652 pop(state); 653 654 if (install_monitor_exception) { 655 call_VM(noreg, CAST_FROM_FN_PTR(address, 656 InterpreterRuntime:: 657 new_illegal_monitor_state_exception)); 658 } 659 660 b(restart); 661 } 662 663 bind(loop); 664 // check if current entry is used 665 ldr(rscratch1, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes())); 666 cbnz(rscratch1, exception); 667 668 add(c_rarg1, c_rarg1, entry_size); // otherwise advance to next entry 669 bind(entry); 670 cmp(c_rarg1, r19); // check if bottom reached 671 br(Assembler::NE, loop); // if not at bottom then check this entry 672 } 673 674 bind(no_unlock); 675 676 // jvmti support 677 if (notify_jvmdi) { 678 notify_method_exit(state, NotifyJVMTI); // preserve TOSCA 679 } else { 680 notify_method_exit(state, SkipNotifyJVMTI); // preserve TOSCA 681 } 682 683 // remove activation 684 // get sender esp 685 ldr(rscratch2, 686 Address(rfp, frame::interpreter_frame_sender_sp_offset * wordSize)); 687 if (StackReservedPages > 0) { 688 // testing if reserved zone needs to be re-enabled 689 Label no_reserved_zone_enabling; 690 691 // look for an overflow into the stack reserved zone, i.e. 692 // interpreter_frame_sender_sp <= JavaThread::reserved_stack_activation 693 ldr(rscratch1, Address(rthread, JavaThread::reserved_stack_activation_offset())); 694 cmp(rscratch2, rscratch1); 695 br(Assembler::LS, no_reserved_zone_enabling); 696 697 call_VM_leaf( 698 CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), rthread); 699 call_VM(noreg, CAST_FROM_FN_PTR(address, 700 InterpreterRuntime::throw_delayed_StackOverflowError)); 701 should_not_reach_here(); 702 703 bind(no_reserved_zone_enabling); 704 } 705 706 // restore sender esp 707 mov(esp, rscratch2); 708 // remove frame anchor 709 leave(); 710 // If we're returning to interpreted code we will shortly be 711 // adjusting SP to allow some space for ESP. If we're returning to 712 // compiled code the saved sender SP was saved in sender_sp, so this 713 // restores it. 714 andr(sp, esp, -16); 715 } 716 717 // Lock object 718 // 719 // Args: 720 // c_rarg1: BasicObjectLock to be used for locking 721 // 722 // Kills: 723 // r0 724 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, .. (param regs) 725 // rscratch1, rscratch2 (scratch regs) 726 void InterpreterMacroAssembler::lock_object(Register lock_reg) 727 { 728 assert(lock_reg == c_rarg1, "The argument is only for looks. It must be c_rarg1"); 729 if (UseHeavyMonitors) { 730 call_VM(noreg, 731 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), 732 lock_reg); 733 } else { 734 Label done; 735 736 const Register swap_reg = r0; 737 const Register tmp = c_rarg2; 738 const Register obj_reg = c_rarg3; // Will contain the oop 739 740 const int obj_offset = BasicObjectLock::obj_offset_in_bytes(); 741 const int lock_offset = BasicObjectLock::lock_offset_in_bytes (); 742 const int mark_offset = lock_offset + 743 BasicLock::displaced_header_offset_in_bytes(); 744 745 Label slow_case; 746 747 // Load object pointer into obj_reg %c_rarg3 748 ldr(obj_reg, Address(lock_reg, obj_offset)); 749 750 if (DiagnoseSyncOnValueBasedClasses != 0) { 751 load_klass(tmp, obj_reg); 752 ldrw(tmp, Address(tmp, Klass::access_flags_offset())); 753 tstw(tmp, JVM_ACC_IS_VALUE_BASED_CLASS); 754 br(Assembler::NE, slow_case); 755 } 756 757 if (UseFastLocking) { 758 ldr(tmp, Address(obj_reg, oopDesc::mark_offset_in_bytes())); 759 fast_lock(obj_reg, tmp, rscratch1, rscratch2, slow_case); 760 b(done); 761 } else { 762 if (UseBiasedLocking) { 763 biased_locking_enter(lock_reg, obj_reg, swap_reg, tmp, false, done, &slow_case); 764 } 765 766 // Load (object->mark() | 1) into swap_reg 767 ldr(rscratch1, Address(obj_reg, oopDesc::mark_offset_in_bytes())); 768 orr(swap_reg, rscratch1, 1); 769 770 // Save (object->mark() | 1) into BasicLock's displaced header 771 str(swap_reg, Address(lock_reg, mark_offset)); 772 773 assert(lock_offset == 0, 774 "displached header must be first word in BasicObjectLock"); 775 776 Label fail; 777 if (PrintBiasedLockingStatistics) { 778 Label fast; 779 cmpxchg_obj_header(swap_reg, lock_reg, obj_reg, rscratch1, fast, &fail); 780 bind(fast); 781 atomic_incw(Address((address)BiasedLocking::fast_path_entry_count_addr()), 782 rscratch2, rscratch1, tmp); 783 b(done); 784 bind(fail); 785 } else { 786 cmpxchg_obj_header(swap_reg, lock_reg, obj_reg, rscratch1, done, /*fallthrough*/NULL); 787 } 788 789 // Fast check for recursive lock. 790 // 791 // Can apply the optimization only if this is a stack lock 792 // allocated in this thread. For efficiency, we can focus on 793 // recently allocated stack locks (instead of reading the stack 794 // base and checking whether 'mark' points inside the current 795 // thread stack): 796 // 1) (mark & 7) == 0, and 797 // 2) sp <= mark < mark + os::pagesize() 798 // 799 // Warning: sp + os::pagesize can overflow the stack base. We must 800 // neither apply the optimization for an inflated lock allocated 801 // just above the thread stack (this is why condition 1 matters) 802 // nor apply the optimization if the stack lock is inside the stack 803 // of another thread. The latter is avoided even in case of overflow 804 // because we have guard pages at the end of all stacks. Hence, if 805 // we go over the stack base and hit the stack of another thread, 806 // this should not be in a writeable area that could contain a 807 // stack lock allocated by that thread. As a consequence, a stack 808 // lock less than page size away from sp is guaranteed to be 809 // owned by the current thread. 810 // 811 // These 3 tests can be done by evaluating the following 812 // expression: ((mark - sp) & (7 - os::vm_page_size())), 813 // assuming both stack pointer and pagesize have their 814 // least significant 3 bits clear. 815 // NOTE: the mark is in swap_reg %r0 as the result of cmpxchg 816 // NOTE2: aarch64 does not like to subtract sp from rn so take a 817 // copy 818 mov(rscratch1, sp); 819 sub(swap_reg, swap_reg, rscratch1); 820 ands(swap_reg, swap_reg, (uint64_t)(7 - os::vm_page_size())); 821 822 // Save the test result, for recursive case, the result is zero 823 str(swap_reg, Address(lock_reg, mark_offset)); 824 825 if (PrintBiasedLockingStatistics) { 826 br(Assembler::NE, slow_case); 827 atomic_incw(Address((address)BiasedLocking::fast_path_entry_count_addr()), 828 rscratch2, rscratch1, tmp); 829 } 830 br(Assembler::EQ, done); 831 } 832 bind(slow_case); 833 834 // Call the runtime routine for slow case 835 call_VM(noreg, 836 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), 837 UseFastLocking ? obj_reg : lock_reg); 838 839 bind(done); 840 } 841 } 842 843 844 // Unlocks an object. Used in monitorexit bytecode and 845 // remove_activation. Throws an IllegalMonitorException if object is 846 // not locked by current thread. 847 // 848 // Args: 849 // c_rarg1: BasicObjectLock for lock 850 // 851 // Kills: 852 // r0 853 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ... (param regs) 854 // rscratch1, rscratch2 (scratch regs) 855 void InterpreterMacroAssembler::unlock_object(Register lock_reg) 856 { 857 assert(lock_reg == c_rarg1, "The argument is only for looks. It must be rarg1"); 858 859 if (UseHeavyMonitors) { 860 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg); 861 } else { 862 Label done; 863 864 const Register swap_reg = r0; 865 const Register header_reg = c_rarg2; // Will contain the old oopMark 866 const Register obj_reg = c_rarg3; // Will contain the oop 867 868 save_bcp(); // Save in case of exception 869 870 if (UseFastLocking) { 871 Label slow_case; 872 873 // Load oop into obj_reg(%c_rarg3) 874 ldr(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset_in_bytes())); 875 876 // Free entry 877 str(zr, Address(lock_reg, BasicObjectLock::obj_offset_in_bytes())); 878 879 // Check for non-symmetric locking. This is allowed by the spec and the interpreter 880 // must handle it. 881 Register tmp = header_reg; 882 ldr(tmp, Address(rthread, JavaThread::lock_stack_current_offset())); 883 ldr(tmp, Address(tmp, -oopSize)); 884 cmpoop(tmp, obj_reg); 885 br(Assembler::NE, slow_case); 886 887 ldr(header_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes())); 888 fast_unlock(obj_reg, header_reg, swap_reg, rscratch1, slow_case); 889 b(done); 890 bind(slow_case); 891 } else { 892 // Convert from BasicObjectLock structure to object and BasicLock 893 // structure Store the BasicLock address into %r0 894 lea(swap_reg, Address(lock_reg, BasicObjectLock::lock_offset_in_bytes())); 895 896 // Load oop into obj_reg(%c_rarg3) 897 ldr(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset_in_bytes())); 898 899 // Free entry 900 str(zr, Address(lock_reg, BasicObjectLock::obj_offset_in_bytes())); 901 902 if (UseBiasedLocking) { 903 biased_locking_exit(obj_reg, header_reg, done); 904 } 905 906 // Load the old header from BasicLock structure 907 ldr(header_reg, Address(swap_reg, 908 BasicLock::displaced_header_offset_in_bytes())); 909 910 // Test for recursion 911 cbz(header_reg, done); 912 913 // Atomic swap back the old header 914 cmpxchg_obj_header(swap_reg, header_reg, obj_reg, rscratch1, done, /*fallthrough*/NULL); 915 } 916 // Call the runtime routine for slow case. 917 str(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset_in_bytes())); // restore obj 918 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg); 919 920 bind(done); 921 922 restore_bcp(); 923 } 924 } 925 926 void InterpreterMacroAssembler::test_method_data_pointer(Register mdp, 927 Label& zero_continue) { 928 assert(ProfileInterpreter, "must be profiling interpreter"); 929 ldr(mdp, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize)); 930 cbz(mdp, zero_continue); 931 } 932 933 // Set the method data pointer for the current bcp. 934 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() { 935 assert(ProfileInterpreter, "must be profiling interpreter"); 936 Label set_mdp; 937 stp(r0, r1, Address(pre(sp, -2 * wordSize))); 938 939 // Test MDO to avoid the call if it is NULL. 940 ldr(r0, Address(rmethod, in_bytes(Method::method_data_offset()))); 941 cbz(r0, set_mdp); 942 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), rmethod, rbcp); 943 // r0: mdi 944 // mdo is guaranteed to be non-zero here, we checked for it before the call. 945 ldr(r1, Address(rmethod, in_bytes(Method::method_data_offset()))); 946 lea(r1, Address(r1, in_bytes(MethodData::data_offset()))); 947 add(r0, r1, r0); 948 str(r0, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize)); 949 bind(set_mdp); 950 ldp(r0, r1, Address(post(sp, 2 * wordSize))); 951 } 952 953 void InterpreterMacroAssembler::verify_method_data_pointer() { 954 assert(ProfileInterpreter, "must be profiling interpreter"); 955 #ifdef ASSERT 956 Label verify_continue; 957 stp(r0, r1, Address(pre(sp, -2 * wordSize))); 958 stp(r2, r3, Address(pre(sp, -2 * wordSize))); 959 test_method_data_pointer(r3, verify_continue); // If mdp is zero, continue 960 get_method(r1); 961 962 // If the mdp is valid, it will point to a DataLayout header which is 963 // consistent with the bcp. The converse is highly probable also. 964 ldrsh(r2, Address(r3, in_bytes(DataLayout::bci_offset()))); 965 ldr(rscratch1, Address(r1, Method::const_offset())); 966 add(r2, r2, rscratch1, Assembler::LSL); 967 lea(r2, Address(r2, ConstMethod::codes_offset())); 968 cmp(r2, rbcp); 969 br(Assembler::EQ, verify_continue); 970 // r1: method 971 // rbcp: bcp // rbcp == 22 972 // r3: mdp 973 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp), 974 r1, rbcp, r3); 975 bind(verify_continue); 976 ldp(r2, r3, Address(post(sp, 2 * wordSize))); 977 ldp(r0, r1, Address(post(sp, 2 * wordSize))); 978 #endif // ASSERT 979 } 980 981 982 void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in, 983 int constant, 984 Register value) { 985 assert(ProfileInterpreter, "must be profiling interpreter"); 986 Address data(mdp_in, constant); 987 str(value, data); 988 } 989 990 991 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in, 992 int constant, 993 bool decrement) { 994 increment_mdp_data_at(mdp_in, noreg, constant, decrement); 995 } 996 997 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in, 998 Register reg, 999 int constant, 1000 bool decrement) { 1001 assert(ProfileInterpreter, "must be profiling interpreter"); 1002 // %%% this does 64bit counters at best it is wasting space 1003 // at worst it is a rare bug when counters overflow 1004 1005 assert_different_registers(rscratch2, rscratch1, mdp_in, reg); 1006 1007 Address addr1(mdp_in, constant); 1008 Address addr2(rscratch2, reg, Address::lsl(0)); 1009 Address &addr = addr1; 1010 if (reg != noreg) { 1011 lea(rscratch2, addr1); 1012 addr = addr2; 1013 } 1014 1015 if (decrement) { 1016 // Decrement the register. Set condition codes. 1017 // Intel does this 1018 // addptr(data, (int32_t) -DataLayout::counter_increment); 1019 // If the decrement causes the counter to overflow, stay negative 1020 // Label L; 1021 // jcc(Assembler::negative, L); 1022 // addptr(data, (int32_t) DataLayout::counter_increment); 1023 // so we do this 1024 ldr(rscratch1, addr); 1025 subs(rscratch1, rscratch1, (unsigned)DataLayout::counter_increment); 1026 Label L; 1027 br(Assembler::LO, L); // skip store if counter underflow 1028 str(rscratch1, addr); 1029 bind(L); 1030 } else { 1031 assert(DataLayout::counter_increment == 1, 1032 "flow-free idiom only works with 1"); 1033 // Intel does this 1034 // Increment the register. Set carry flag. 1035 // addptr(data, DataLayout::counter_increment); 1036 // If the increment causes the counter to overflow, pull back by 1. 1037 // sbbptr(data, (int32_t)0); 1038 // so we do this 1039 ldr(rscratch1, addr); 1040 adds(rscratch1, rscratch1, DataLayout::counter_increment); 1041 Label L; 1042 br(Assembler::CS, L); // skip store if counter overflow 1043 str(rscratch1, addr); 1044 bind(L); 1045 } 1046 } 1047 1048 void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in, 1049 int flag_byte_constant) { 1050 assert(ProfileInterpreter, "must be profiling interpreter"); 1051 int flags_offset = in_bytes(DataLayout::flags_offset()); 1052 // Set the flag 1053 ldrb(rscratch1, Address(mdp_in, flags_offset)); 1054 orr(rscratch1, rscratch1, flag_byte_constant); 1055 strb(rscratch1, Address(mdp_in, flags_offset)); 1056 } 1057 1058 1059 void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in, 1060 int offset, 1061 Register value, 1062 Register test_value_out, 1063 Label& not_equal_continue) { 1064 assert(ProfileInterpreter, "must be profiling interpreter"); 1065 if (test_value_out == noreg) { 1066 ldr(rscratch1, Address(mdp_in, offset)); 1067 cmp(value, rscratch1); 1068 } else { 1069 // Put the test value into a register, so caller can use it: 1070 ldr(test_value_out, Address(mdp_in, offset)); 1071 cmp(value, test_value_out); 1072 } 1073 br(Assembler::NE, not_equal_continue); 1074 } 1075 1076 1077 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in, 1078 int offset_of_disp) { 1079 assert(ProfileInterpreter, "must be profiling interpreter"); 1080 ldr(rscratch1, Address(mdp_in, offset_of_disp)); 1081 add(mdp_in, mdp_in, rscratch1, LSL); 1082 str(mdp_in, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize)); 1083 } 1084 1085 1086 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in, 1087 Register reg, 1088 int offset_of_disp) { 1089 assert(ProfileInterpreter, "must be profiling interpreter"); 1090 lea(rscratch1, Address(mdp_in, offset_of_disp)); 1091 ldr(rscratch1, Address(rscratch1, reg, Address::lsl(0))); 1092 add(mdp_in, mdp_in, rscratch1, LSL); 1093 str(mdp_in, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize)); 1094 } 1095 1096 1097 void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in, 1098 int constant) { 1099 assert(ProfileInterpreter, "must be profiling interpreter"); 1100 add(mdp_in, mdp_in, (unsigned)constant); 1101 str(mdp_in, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize)); 1102 } 1103 1104 1105 void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) { 1106 assert(ProfileInterpreter, "must be profiling interpreter"); 1107 // save/restore across call_VM 1108 stp(zr, return_bci, Address(pre(sp, -2 * wordSize))); 1109 call_VM(noreg, 1110 CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret), 1111 return_bci); 1112 ldp(zr, return_bci, Address(post(sp, 2 * wordSize))); 1113 } 1114 1115 1116 void InterpreterMacroAssembler::profile_taken_branch(Register mdp, 1117 Register bumped_count) { 1118 if (ProfileInterpreter) { 1119 Label profile_continue; 1120 1121 // If no method data exists, go to profile_continue. 1122 // Otherwise, assign to mdp 1123 test_method_data_pointer(mdp, profile_continue); 1124 1125 // We are taking a branch. Increment the taken count. 1126 // We inline increment_mdp_data_at to return bumped_count in a register 1127 //increment_mdp_data_at(mdp, in_bytes(JumpData::taken_offset())); 1128 Address data(mdp, in_bytes(JumpData::taken_offset())); 1129 ldr(bumped_count, data); 1130 assert(DataLayout::counter_increment == 1, 1131 "flow-free idiom only works with 1"); 1132 // Intel does this to catch overflow 1133 // addptr(bumped_count, DataLayout::counter_increment); 1134 // sbbptr(bumped_count, 0); 1135 // so we do this 1136 adds(bumped_count, bumped_count, DataLayout::counter_increment); 1137 Label L; 1138 br(Assembler::CS, L); // skip store if counter overflow 1139 str(bumped_count, data); 1140 bind(L); 1141 // The method data pointer needs to be updated to reflect the new target. 1142 update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset())); 1143 bind(profile_continue); 1144 } 1145 } 1146 1147 1148 void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp) { 1149 if (ProfileInterpreter) { 1150 Label profile_continue; 1151 1152 // If no method data exists, go to profile_continue. 1153 test_method_data_pointer(mdp, profile_continue); 1154 1155 // We are taking a branch. Increment the not taken count. 1156 increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset())); 1157 1158 // The method data pointer needs to be updated to correspond to 1159 // the next bytecode 1160 update_mdp_by_constant(mdp, in_bytes(BranchData::branch_data_size())); 1161 bind(profile_continue); 1162 } 1163 } 1164 1165 1166 void InterpreterMacroAssembler::profile_call(Register mdp) { 1167 if (ProfileInterpreter) { 1168 Label profile_continue; 1169 1170 // If no method data exists, go to profile_continue. 1171 test_method_data_pointer(mdp, profile_continue); 1172 1173 // We are making a call. Increment the count. 1174 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); 1175 1176 // The method data pointer needs to be updated to reflect the new target. 1177 update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size())); 1178 bind(profile_continue); 1179 } 1180 } 1181 1182 void InterpreterMacroAssembler::profile_final_call(Register mdp) { 1183 if (ProfileInterpreter) { 1184 Label profile_continue; 1185 1186 // If no method data exists, go to profile_continue. 1187 test_method_data_pointer(mdp, profile_continue); 1188 1189 // We are making a call. Increment the count. 1190 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); 1191 1192 // The method data pointer needs to be updated to reflect the new target. 1193 update_mdp_by_constant(mdp, 1194 in_bytes(VirtualCallData:: 1195 virtual_call_data_size())); 1196 bind(profile_continue); 1197 } 1198 } 1199 1200 1201 void InterpreterMacroAssembler::profile_virtual_call(Register receiver, 1202 Register mdp, 1203 Register reg2, 1204 bool receiver_can_be_null) { 1205 if (ProfileInterpreter) { 1206 Label profile_continue; 1207 1208 // If no method data exists, go to profile_continue. 1209 test_method_data_pointer(mdp, profile_continue); 1210 1211 Label skip_receiver_profile; 1212 if (receiver_can_be_null) { 1213 Label not_null; 1214 // We are making a call. Increment the count for null receiver. 1215 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); 1216 b(skip_receiver_profile); 1217 bind(not_null); 1218 } 1219 1220 // Record the receiver type. 1221 record_klass_in_profile(receiver, mdp, reg2, true); 1222 bind(skip_receiver_profile); 1223 1224 // The method data pointer needs to be updated to reflect the new target. 1225 update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size())); 1226 bind(profile_continue); 1227 } 1228 } 1229 1230 // This routine creates a state machine for updating the multi-row 1231 // type profile at a virtual call site (or other type-sensitive bytecode). 1232 // The machine visits each row (of receiver/count) until the receiver type 1233 // is found, or until it runs out of rows. At the same time, it remembers 1234 // the location of the first empty row. (An empty row records null for its 1235 // receiver, and can be allocated for a newly-observed receiver type.) 1236 // Because there are two degrees of freedom in the state, a simple linear 1237 // search will not work; it must be a decision tree. Hence this helper 1238 // function is recursive, to generate the required tree structured code. 1239 // It's the interpreter, so we are trading off code space for speed. 1240 // See below for example code. 1241 void InterpreterMacroAssembler::record_klass_in_profile_helper( 1242 Register receiver, Register mdp, 1243 Register reg2, int start_row, 1244 Label& done, bool is_virtual_call) { 1245 if (TypeProfileWidth == 0) { 1246 if (is_virtual_call) { 1247 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); 1248 } 1249 #if INCLUDE_JVMCI 1250 else if (EnableJVMCI) { 1251 increment_mdp_data_at(mdp, in_bytes(ReceiverTypeData::nonprofiled_receiver_count_offset())); 1252 } 1253 #endif // INCLUDE_JVMCI 1254 } else { 1255 int non_profiled_offset = -1; 1256 if (is_virtual_call) { 1257 non_profiled_offset = in_bytes(CounterData::count_offset()); 1258 } 1259 #if INCLUDE_JVMCI 1260 else if (EnableJVMCI) { 1261 non_profiled_offset = in_bytes(ReceiverTypeData::nonprofiled_receiver_count_offset()); 1262 } 1263 #endif // INCLUDE_JVMCI 1264 1265 record_item_in_profile_helper(receiver, mdp, reg2, 0, done, TypeProfileWidth, 1266 &VirtualCallData::receiver_offset, &VirtualCallData::receiver_count_offset, non_profiled_offset); 1267 } 1268 } 1269 1270 void InterpreterMacroAssembler::record_item_in_profile_helper(Register item, Register mdp, 1271 Register reg2, int start_row, Label& done, int total_rows, 1272 OffsetFunction item_offset_fn, OffsetFunction item_count_offset_fn, 1273 int non_profiled_offset) { 1274 int last_row = total_rows - 1; 1275 assert(start_row <= last_row, "must be work left to do"); 1276 // Test this row for both the item and for null. 1277 // Take any of three different outcomes: 1278 // 1. found item => increment count and goto done 1279 // 2. found null => keep looking for case 1, maybe allocate this cell 1280 // 3. found something else => keep looking for cases 1 and 2 1281 // Case 3 is handled by a recursive call. 1282 for (int row = start_row; row <= last_row; row++) { 1283 Label next_test; 1284 bool test_for_null_also = (row == start_row); 1285 1286 // See if the item is item[n]. 1287 int item_offset = in_bytes(item_offset_fn(row)); 1288 test_mdp_data_at(mdp, item_offset, item, 1289 (test_for_null_also ? reg2 : noreg), 1290 next_test); 1291 // (Reg2 now contains the item from the CallData.) 1292 1293 // The item is item[n]. Increment count[n]. 1294 int count_offset = in_bytes(item_count_offset_fn(row)); 1295 increment_mdp_data_at(mdp, count_offset); 1296 b(done); 1297 bind(next_test); 1298 1299 if (test_for_null_also) { 1300 Label found_null; 1301 // Failed the equality check on item[n]... Test for null. 1302 if (start_row == last_row) { 1303 // The only thing left to do is handle the null case. 1304 if (non_profiled_offset >= 0) { 1305 cbz(reg2, found_null); 1306 // Item did not match any saved item and there is no empty row for it. 1307 // Increment total counter to indicate polymorphic case. 1308 increment_mdp_data_at(mdp, non_profiled_offset); 1309 b(done); 1310 bind(found_null); 1311 } else { 1312 cbnz(reg2, done); 1313 } 1314 break; 1315 } 1316 // Since null is rare, make it be the branch-taken case. 1317 cbz(reg2, found_null); 1318 1319 // Put all the "Case 3" tests here. 1320 record_item_in_profile_helper(item, mdp, reg2, start_row + 1, done, total_rows, 1321 item_offset_fn, item_count_offset_fn, non_profiled_offset); 1322 1323 // Found a null. Keep searching for a matching item, 1324 // but remember that this is an empty (unused) slot. 1325 bind(found_null); 1326 } 1327 } 1328 1329 // In the fall-through case, we found no matching item, but we 1330 // observed the item[start_row] is NULL. 1331 1332 // Fill in the item field and increment the count. 1333 int item_offset = in_bytes(item_offset_fn(start_row)); 1334 set_mdp_data_at(mdp, item_offset, item); 1335 int count_offset = in_bytes(item_count_offset_fn(start_row)); 1336 mov(reg2, DataLayout::counter_increment); 1337 set_mdp_data_at(mdp, count_offset, reg2); 1338 if (start_row > 0) { 1339 b(done); 1340 } 1341 } 1342 1343 // Example state machine code for three profile rows: 1344 // // main copy of decision tree, rooted at row[1] 1345 // if (row[0].rec == rec) { row[0].incr(); goto done; } 1346 // if (row[0].rec != NULL) { 1347 // // inner copy of decision tree, rooted at row[1] 1348 // if (row[1].rec == rec) { row[1].incr(); goto done; } 1349 // if (row[1].rec != NULL) { 1350 // // degenerate decision tree, rooted at row[2] 1351 // if (row[2].rec == rec) { row[2].incr(); goto done; } 1352 // if (row[2].rec != NULL) { count.incr(); goto done; } // overflow 1353 // row[2].init(rec); goto done; 1354 // } else { 1355 // // remember row[1] is empty 1356 // if (row[2].rec == rec) { row[2].incr(); goto done; } 1357 // row[1].init(rec); goto done; 1358 // } 1359 // } else { 1360 // // remember row[0] is empty 1361 // if (row[1].rec == rec) { row[1].incr(); goto done; } 1362 // if (row[2].rec == rec) { row[2].incr(); goto done; } 1363 // row[0].init(rec); goto done; 1364 // } 1365 // done: 1366 1367 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver, 1368 Register mdp, Register reg2, 1369 bool is_virtual_call) { 1370 assert(ProfileInterpreter, "must be profiling"); 1371 Label done; 1372 1373 record_klass_in_profile_helper(receiver, mdp, reg2, 0, done, is_virtual_call); 1374 1375 bind (done); 1376 } 1377 1378 void InterpreterMacroAssembler::profile_ret(Register return_bci, 1379 Register mdp) { 1380 if (ProfileInterpreter) { 1381 Label profile_continue; 1382 uint row; 1383 1384 // If no method data exists, go to profile_continue. 1385 test_method_data_pointer(mdp, profile_continue); 1386 1387 // Update the total ret count. 1388 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); 1389 1390 for (row = 0; row < RetData::row_limit(); row++) { 1391 Label next_test; 1392 1393 // See if return_bci is equal to bci[n]: 1394 test_mdp_data_at(mdp, 1395 in_bytes(RetData::bci_offset(row)), 1396 return_bci, noreg, 1397 next_test); 1398 1399 // return_bci is equal to bci[n]. Increment the count. 1400 increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row))); 1401 1402 // The method data pointer needs to be updated to reflect the new target. 1403 update_mdp_by_offset(mdp, 1404 in_bytes(RetData::bci_displacement_offset(row))); 1405 b(profile_continue); 1406 bind(next_test); 1407 } 1408 1409 update_mdp_for_ret(return_bci); 1410 1411 bind(profile_continue); 1412 } 1413 } 1414 1415 void InterpreterMacroAssembler::profile_null_seen(Register mdp) { 1416 if (ProfileInterpreter) { 1417 Label profile_continue; 1418 1419 // If no method data exists, go to profile_continue. 1420 test_method_data_pointer(mdp, profile_continue); 1421 1422 set_mdp_flag_at(mdp, BitData::null_seen_byte_constant()); 1423 1424 // The method data pointer needs to be updated. 1425 int mdp_delta = in_bytes(BitData::bit_data_size()); 1426 if (TypeProfileCasts) { 1427 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size()); 1428 } 1429 update_mdp_by_constant(mdp, mdp_delta); 1430 1431 bind(profile_continue); 1432 } 1433 } 1434 1435 void InterpreterMacroAssembler::profile_typecheck_failed(Register mdp) { 1436 if (ProfileInterpreter && TypeProfileCasts) { 1437 Label profile_continue; 1438 1439 // If no method data exists, go to profile_continue. 1440 test_method_data_pointer(mdp, profile_continue); 1441 1442 int count_offset = in_bytes(CounterData::count_offset()); 1443 // Back up the address, since we have already bumped the mdp. 1444 count_offset -= in_bytes(VirtualCallData::virtual_call_data_size()); 1445 1446 // *Decrement* the counter. We expect to see zero or small negatives. 1447 increment_mdp_data_at(mdp, count_offset, true); 1448 1449 bind (profile_continue); 1450 } 1451 } 1452 1453 void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) { 1454 if (ProfileInterpreter) { 1455 Label profile_continue; 1456 1457 // If no method data exists, go to profile_continue. 1458 test_method_data_pointer(mdp, profile_continue); 1459 1460 // The method data pointer needs to be updated. 1461 int mdp_delta = in_bytes(BitData::bit_data_size()); 1462 if (TypeProfileCasts) { 1463 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size()); 1464 1465 // Record the object type. 1466 record_klass_in_profile(klass, mdp, reg2, false); 1467 } 1468 update_mdp_by_constant(mdp, mdp_delta); 1469 1470 bind(profile_continue); 1471 } 1472 } 1473 1474 void InterpreterMacroAssembler::profile_switch_default(Register mdp) { 1475 if (ProfileInterpreter) { 1476 Label profile_continue; 1477 1478 // If no method data exists, go to profile_continue. 1479 test_method_data_pointer(mdp, profile_continue); 1480 1481 // Update the default case count 1482 increment_mdp_data_at(mdp, 1483 in_bytes(MultiBranchData::default_count_offset())); 1484 1485 // The method data pointer needs to be updated. 1486 update_mdp_by_offset(mdp, 1487 in_bytes(MultiBranchData:: 1488 default_displacement_offset())); 1489 1490 bind(profile_continue); 1491 } 1492 } 1493 1494 void InterpreterMacroAssembler::profile_switch_case(Register index, 1495 Register mdp, 1496 Register reg2) { 1497 if (ProfileInterpreter) { 1498 Label profile_continue; 1499 1500 // If no method data exists, go to profile_continue. 1501 test_method_data_pointer(mdp, profile_continue); 1502 1503 // Build the base (index * per_case_size_in_bytes()) + 1504 // case_array_offset_in_bytes() 1505 movw(reg2, in_bytes(MultiBranchData::per_case_size())); 1506 movw(rscratch1, in_bytes(MultiBranchData::case_array_offset())); 1507 Assembler::maddw(index, index, reg2, rscratch1); 1508 1509 // Update the case count 1510 increment_mdp_data_at(mdp, 1511 index, 1512 in_bytes(MultiBranchData::relative_count_offset())); 1513 1514 // The method data pointer needs to be updated. 1515 update_mdp_by_offset(mdp, 1516 index, 1517 in_bytes(MultiBranchData:: 1518 relative_displacement_offset())); 1519 1520 bind(profile_continue); 1521 } 1522 } 1523 1524 void InterpreterMacroAssembler::verify_oop(Register reg, TosState state) { 1525 if (state == atos) { 1526 MacroAssembler::verify_oop(reg); 1527 } 1528 } 1529 1530 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) { ; } 1531 1532 1533 void InterpreterMacroAssembler::notify_method_entry() { 1534 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to 1535 // track stack depth. If it is possible to enter interp_only_mode we add 1536 // the code to check if the event should be sent. 1537 if (JvmtiExport::can_post_interpreter_events()) { 1538 Label L; 1539 ldrw(r3, Address(rthread, JavaThread::interp_only_mode_offset())); 1540 cbzw(r3, L); 1541 call_VM(noreg, CAST_FROM_FN_PTR(address, 1542 InterpreterRuntime::post_method_entry)); 1543 bind(L); 1544 } 1545 1546 { 1547 SkipIfEqual skip(this, &DTraceMethodProbes, false); 1548 get_method(c_rarg1); 1549 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry), 1550 rthread, c_rarg1); 1551 } 1552 1553 // RedefineClasses() tracing support for obsolete method entry 1554 if (log_is_enabled(Trace, redefine, class, obsolete)) { 1555 get_method(c_rarg1); 1556 call_VM_leaf( 1557 CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry), 1558 rthread, c_rarg1); 1559 } 1560 1561 } 1562 1563 1564 void InterpreterMacroAssembler::notify_method_exit( 1565 TosState state, NotifyMethodExitMode mode) { 1566 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to 1567 // track stack depth. If it is possible to enter interp_only_mode we add 1568 // the code to check if the event should be sent. 1569 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) { 1570 Label L; 1571 // Note: frame::interpreter_frame_result has a dependency on how the 1572 // method result is saved across the call to post_method_exit. If this 1573 // is changed then the interpreter_frame_result implementation will 1574 // need to be updated too. 1575 1576 // template interpreter will leave the result on the top of the stack. 1577 push(state); 1578 ldrw(r3, Address(rthread, JavaThread::interp_only_mode_offset())); 1579 cbz(r3, L); 1580 call_VM(noreg, 1581 CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit)); 1582 bind(L); 1583 pop(state); 1584 } 1585 1586 { 1587 SkipIfEqual skip(this, &DTraceMethodProbes, false); 1588 push(state); 1589 get_method(c_rarg1); 1590 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit), 1591 rthread, c_rarg1); 1592 pop(state); 1593 } 1594 } 1595 1596 1597 // Jump if ((*counter_addr += increment) & mask) satisfies the condition. 1598 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr, 1599 int increment, Address mask, 1600 Register scratch, Register scratch2, 1601 bool preloaded, Condition cond, 1602 Label* where) { 1603 if (!preloaded) { 1604 ldrw(scratch, counter_addr); 1605 } 1606 add(scratch, scratch, increment); 1607 strw(scratch, counter_addr); 1608 ldrw(scratch2, mask); 1609 ands(scratch, scratch, scratch2); 1610 br(cond, *where); 1611 } 1612 1613 void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point, 1614 int number_of_arguments) { 1615 // interpreter specific 1616 // 1617 // Note: No need to save/restore rbcp & rlocals pointer since these 1618 // are callee saved registers and no blocking/ GC can happen 1619 // in leaf calls. 1620 #ifdef ASSERT 1621 { 1622 Label L; 1623 ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize)); 1624 cbz(rscratch1, L); 1625 stop("InterpreterMacroAssembler::call_VM_leaf_base:" 1626 " last_sp != NULL"); 1627 bind(L); 1628 } 1629 #endif /* ASSERT */ 1630 // super call 1631 MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments); 1632 } 1633 1634 void InterpreterMacroAssembler::call_VM_base(Register oop_result, 1635 Register java_thread, 1636 Register last_java_sp, 1637 address entry_point, 1638 int number_of_arguments, 1639 bool check_exceptions) { 1640 // interpreter specific 1641 // 1642 // Note: Could avoid restoring locals ptr (callee saved) - however doesn't 1643 // really make a difference for these runtime calls, since they are 1644 // slow anyway. Btw., bcp must be saved/restored since it may change 1645 // due to GC. 1646 // assert(java_thread == noreg , "not expecting a precomputed java thread"); 1647 save_bcp(); 1648 #ifdef ASSERT 1649 { 1650 Label L; 1651 ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize)); 1652 cbz(rscratch1, L); 1653 stop("InterpreterMacroAssembler::call_VM_base:" 1654 " last_sp != NULL"); 1655 bind(L); 1656 } 1657 #endif /* ASSERT */ 1658 // super call 1659 MacroAssembler::call_VM_base(oop_result, noreg, last_java_sp, 1660 entry_point, number_of_arguments, 1661 check_exceptions); 1662 // interpreter specific 1663 restore_bcp(); 1664 restore_locals(); 1665 } 1666 1667 void InterpreterMacroAssembler::profile_obj_type(Register obj, const Address& mdo_addr) { 1668 assert_different_registers(obj, rscratch1); 1669 Label update, next, none; 1670 1671 verify_oop(obj); 1672 1673 cbnz(obj, update); 1674 orptr(mdo_addr, TypeEntries::null_seen); 1675 b(next); 1676 1677 bind(update); 1678 load_klass(obj, obj); 1679 1680 ldr(rscratch1, mdo_addr); 1681 eor(obj, obj, rscratch1); 1682 tst(obj, TypeEntries::type_klass_mask); 1683 br(Assembler::EQ, next); // klass seen before, nothing to 1684 // do. The unknown bit may have been 1685 // set already but no need to check. 1686 1687 tbnz(obj, exact_log2(TypeEntries::type_unknown), next); 1688 // already unknown. Nothing to do anymore. 1689 1690 ldr(rscratch1, mdo_addr); 1691 cbz(rscratch1, none); 1692 cmp(rscratch1, (u1)TypeEntries::null_seen); 1693 br(Assembler::EQ, none); 1694 // There is a chance that the checks above (re-reading profiling 1695 // data from memory) fail if another thread has just set the 1696 // profiling to this obj's klass 1697 ldr(rscratch1, mdo_addr); 1698 eor(obj, obj, rscratch1); 1699 tst(obj, TypeEntries::type_klass_mask); 1700 br(Assembler::EQ, next); 1701 1702 // different than before. Cannot keep accurate profile. 1703 orptr(mdo_addr, TypeEntries::type_unknown); 1704 b(next); 1705 1706 bind(none); 1707 // first time here. Set profile type. 1708 str(obj, mdo_addr); 1709 1710 bind(next); 1711 } 1712 1713 void InterpreterMacroAssembler::profile_arguments_type(Register mdp, Register callee, Register tmp, bool is_virtual) { 1714 if (!ProfileInterpreter) { 1715 return; 1716 } 1717 1718 if (MethodData::profile_arguments() || MethodData::profile_return()) { 1719 Label profile_continue; 1720 1721 test_method_data_pointer(mdp, profile_continue); 1722 1723 int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size()); 1724 1725 ldrb(rscratch1, Address(mdp, in_bytes(DataLayout::tag_offset()) - off_to_start)); 1726 cmp(rscratch1, u1(is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag)); 1727 br(Assembler::NE, profile_continue); 1728 1729 if (MethodData::profile_arguments()) { 1730 Label done; 1731 int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset()); 1732 1733 for (int i = 0; i < TypeProfileArgsLimit; i++) { 1734 if (i > 0 || MethodData::profile_return()) { 1735 // If return value type is profiled we may have no argument to profile 1736 ldr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset()))); 1737 sub(tmp, tmp, i*TypeStackSlotEntries::per_arg_count()); 1738 cmp(tmp, (u1)TypeStackSlotEntries::per_arg_count()); 1739 add(rscratch1, mdp, off_to_args); 1740 br(Assembler::LT, done); 1741 } 1742 ldr(tmp, Address(callee, Method::const_offset())); 1743 load_unsigned_short(tmp, Address(tmp, ConstMethod::size_of_parameters_offset())); 1744 // stack offset o (zero based) from the start of the argument 1745 // list, for n arguments translates into offset n - o - 1 from 1746 // the end of the argument list 1747 ldr(rscratch1, Address(mdp, in_bytes(TypeEntriesAtCall::stack_slot_offset(i)))); 1748 sub(tmp, tmp, rscratch1); 1749 sub(tmp, tmp, 1); 1750 Address arg_addr = argument_address(tmp); 1751 ldr(tmp, arg_addr); 1752 1753 Address mdo_arg_addr(mdp, in_bytes(TypeEntriesAtCall::argument_type_offset(i))); 1754 profile_obj_type(tmp, mdo_arg_addr); 1755 1756 int to_add = in_bytes(TypeStackSlotEntries::per_arg_size()); 1757 off_to_args += to_add; 1758 } 1759 1760 if (MethodData::profile_return()) { 1761 ldr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset()))); 1762 sub(tmp, tmp, TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count()); 1763 } 1764 1765 add(rscratch1, mdp, off_to_args); 1766 bind(done); 1767 mov(mdp, rscratch1); 1768 1769 if (MethodData::profile_return()) { 1770 // We're right after the type profile for the last 1771 // argument. tmp is the number of cells left in the 1772 // CallTypeData/VirtualCallTypeData to reach its end. Non null 1773 // if there's a return to profile. 1774 assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type"); 1775 add(mdp, mdp, tmp, LSL, exact_log2(DataLayout::cell_size)); 1776 } 1777 str(mdp, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize)); 1778 } else { 1779 assert(MethodData::profile_return(), "either profile call args or call ret"); 1780 update_mdp_by_constant(mdp, in_bytes(TypeEntriesAtCall::return_only_size())); 1781 } 1782 1783 // mdp points right after the end of the 1784 // CallTypeData/VirtualCallTypeData, right after the cells for the 1785 // return value type if there's one 1786 1787 bind(profile_continue); 1788 } 1789 } 1790 1791 void InterpreterMacroAssembler::profile_return_type(Register mdp, Register ret, Register tmp) { 1792 assert_different_registers(mdp, ret, tmp, rbcp); 1793 if (ProfileInterpreter && MethodData::profile_return()) { 1794 Label profile_continue, done; 1795 1796 test_method_data_pointer(mdp, profile_continue); 1797 1798 if (MethodData::profile_return_jsr292_only()) { 1799 assert(Method::intrinsic_id_size_in_bytes() == 2, "assuming Method::_intrinsic_id is u2"); 1800 1801 // If we don't profile all invoke bytecodes we must make sure 1802 // it's a bytecode we indeed profile. We can't go back to the 1803 // begining of the ProfileData we intend to update to check its 1804 // type because we're right after it and we don't known its 1805 // length 1806 Label do_profile; 1807 ldrb(rscratch1, Address(rbcp, 0)); 1808 cmp(rscratch1, (u1)Bytecodes::_invokedynamic); 1809 br(Assembler::EQ, do_profile); 1810 cmp(rscratch1, (u1)Bytecodes::_invokehandle); 1811 br(Assembler::EQ, do_profile); 1812 get_method(tmp); 1813 ldrh(rscratch1, Address(tmp, Method::intrinsic_id_offset_in_bytes())); 1814 subs(zr, rscratch1, static_cast<int>(vmIntrinsics::_compiledLambdaForm)); 1815 br(Assembler::NE, profile_continue); 1816 1817 bind(do_profile); 1818 } 1819 1820 Address mdo_ret_addr(mdp, -in_bytes(ReturnTypeEntry::size())); 1821 mov(tmp, ret); 1822 profile_obj_type(tmp, mdo_ret_addr); 1823 1824 bind(profile_continue); 1825 } 1826 } 1827 1828 void InterpreterMacroAssembler::profile_parameters_type(Register mdp, Register tmp1, Register tmp2) { 1829 assert_different_registers(rscratch1, rscratch2, mdp, tmp1, tmp2); 1830 if (ProfileInterpreter && MethodData::profile_parameters()) { 1831 Label profile_continue, done; 1832 1833 test_method_data_pointer(mdp, profile_continue); 1834 1835 // Load the offset of the area within the MDO used for 1836 // parameters. If it's negative we're not profiling any parameters 1837 ldrw(tmp1, Address(mdp, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset()))); 1838 tbnz(tmp1, 31, profile_continue); // i.e. sign bit set 1839 1840 // Compute a pointer to the area for parameters from the offset 1841 // and move the pointer to the slot for the last 1842 // parameters. Collect profiling from last parameter down. 1843 // mdo start + parameters offset + array length - 1 1844 add(mdp, mdp, tmp1); 1845 ldr(tmp1, Address(mdp, ArrayData::array_len_offset())); 1846 sub(tmp1, tmp1, TypeStackSlotEntries::per_arg_count()); 1847 1848 Label loop; 1849 bind(loop); 1850 1851 int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0)); 1852 int type_base = in_bytes(ParametersTypeData::type_offset(0)); 1853 int per_arg_scale = exact_log2(DataLayout::cell_size); 1854 add(rscratch1, mdp, off_base); 1855 add(rscratch2, mdp, type_base); 1856 1857 Address arg_off(rscratch1, tmp1, Address::lsl(per_arg_scale)); 1858 Address arg_type(rscratch2, tmp1, Address::lsl(per_arg_scale)); 1859 1860 // load offset on the stack from the slot for this parameter 1861 ldr(tmp2, arg_off); 1862 neg(tmp2, tmp2); 1863 // read the parameter from the local area 1864 ldr(tmp2, Address(rlocals, tmp2, Address::lsl(Interpreter::logStackElementSize))); 1865 1866 // profile the parameter 1867 profile_obj_type(tmp2, arg_type); 1868 1869 // go to next parameter 1870 subs(tmp1, tmp1, TypeStackSlotEntries::per_arg_count()); 1871 br(Assembler::GE, loop); 1872 1873 bind(profile_continue); 1874 } 1875 }