1 /* 2 * Copyright (c) 2003, 2018, Oracle and/or its affiliates. All rights reserved. 3 * Copyright (c) 2014, 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 "gc/shared/barrierSet.hpp" 29 #include "gc/shared/barrierSetAssembler.hpp" 30 #include "interp_masm_aarch64.hpp" 31 #include "interpreter/interpreter.hpp" 32 #include "interpreter/interpreterRuntime.hpp" 33 #include "logging/log.hpp" 34 #include "oops/arrayOop.hpp" 35 #include "oops/markOop.hpp" 36 #include "oops/method.hpp" 37 #include "oops/methodData.hpp" 38 #include "prims/jvmtiExport.hpp" 39 #include "prims/jvmtiThreadState.hpp" 40 #include "runtime/basicLock.hpp" 41 #include "runtime/biasedLocking.hpp" 42 #include "runtime/frame.inline.hpp" 43 #include "runtime/safepointMechanism.hpp" 44 #include "runtime/sharedRuntime.hpp" 45 #include "runtime/thread.inline.hpp" 46 47 48 void InterpreterMacroAssembler::narrow(Register result) { 49 50 // Get method->_constMethod->_result_type 51 ldr(rscratch1, Address(rfp, frame::interpreter_frame_method_offset * wordSize)); 52 ldr(rscratch1, Address(rscratch1, Method::const_offset())); 53 ldrb(rscratch1, Address(rscratch1, ConstMethod::result_type_offset())); 54 55 Label done, notBool, notByte, notChar; 56 57 // common case first 58 cmpw(rscratch1, T_INT); 59 br(Assembler::EQ, done); 60 61 // mask integer result to narrower return type. 62 cmpw(rscratch1, T_BOOLEAN); 63 br(Assembler::NE, notBool); 64 andw(result, result, 0x1); 65 b(done); 66 67 bind(notBool); 68 cmpw(rscratch1, T_BYTE); 69 br(Assembler::NE, notByte); 70 sbfx(result, result, 0, 8); 71 b(done); 72 73 bind(notByte); 74 cmpw(rscratch1, T_CHAR); 75 br(Assembler::NE, notChar); 76 ubfx(result, result, 0, 16); // truncate upper 16 bits 77 b(done); 78 79 bind(notChar); 80 sbfx(result, result, 0, 16); // sign-extend short 81 82 // Nothing to do for T_INT 83 bind(done); 84 } 85 86 void InterpreterMacroAssembler::jump_to_entry(address entry) { 87 assert(entry, "Entry must have been generated by now"); 88 b(entry); 89 } 90 91 void InterpreterMacroAssembler::check_and_handle_popframe(Register java_thread) { 92 if (JvmtiExport::can_pop_frame()) { 93 Label L; 94 // Initiate popframe handling only if it is not already being 95 // processed. If the flag has the popframe_processing bit set, it 96 // means that this code is called *during* popframe handling - we 97 // don't want to reenter. 98 // This method is only called just after the call into the vm in 99 // call_VM_base, so the arg registers are available. 100 ldrw(rscratch1, Address(rthread, JavaThread::popframe_condition_offset())); 101 tbz(rscratch1, exact_log2(JavaThread::popframe_pending_bit), L); 102 tbnz(rscratch1, exact_log2(JavaThread::popframe_processing_bit), L); 103 // Call Interpreter::remove_activation_preserving_args_entry() to get the 104 // address of the same-named entrypoint in the generated interpreter code. 105 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry)); 106 br(r0); 107 bind(L); 108 } 109 } 110 111 112 void InterpreterMacroAssembler::load_earlyret_value(TosState state) { 113 ldr(r2, Address(rthread, JavaThread::jvmti_thread_state_offset())); 114 const Address tos_addr(r2, JvmtiThreadState::earlyret_tos_offset()); 115 const Address oop_addr(r2, JvmtiThreadState::earlyret_oop_offset()); 116 const Address val_addr(r2, JvmtiThreadState::earlyret_value_offset()); 117 switch (state) { 118 case atos: ldr(r0, oop_addr); 119 str(zr, oop_addr); 120 verify_oop(r0, state); break; 121 case ltos: ldr(r0, val_addr); break; 122 case btos: // fall through 123 case ztos: // fall through 124 case ctos: // fall through 125 case stos: // fall through 126 case itos: ldrw(r0, val_addr); break; 127 case ftos: ldrs(v0, val_addr); break; 128 case dtos: ldrd(v0, val_addr); break; 129 case vtos: /* nothing to do */ break; 130 default : ShouldNotReachHere(); 131 } 132 // Clean up tos value in the thread object 133 movw(rscratch1, (int) ilgl); 134 strw(rscratch1, tos_addr); 135 strw(zr, val_addr); 136 } 137 138 139 void InterpreterMacroAssembler::check_and_handle_earlyret(Register java_thread) { 140 if (JvmtiExport::can_force_early_return()) { 141 Label L; 142 ldr(rscratch1, Address(rthread, JavaThread::jvmti_thread_state_offset())); 143 cbz(rscratch1, L); // if (thread->jvmti_thread_state() == NULL) exit; 144 145 // Initiate earlyret handling only if it is not already being processed. 146 // If the flag has the earlyret_processing bit set, it means that this code 147 // is called *during* earlyret handling - we don't want to reenter. 148 ldrw(rscratch1, Address(rscratch1, JvmtiThreadState::earlyret_state_offset())); 149 cmpw(rscratch1, JvmtiThreadState::earlyret_pending); 150 br(Assembler::NE, L); 151 152 // Call Interpreter::remove_activation_early_entry() to get the address of the 153 // same-named entrypoint in the generated interpreter code. 154 ldr(rscratch1, Address(rthread, JavaThread::jvmti_thread_state_offset())); 155 ldrw(rscratch1, Address(rscratch1, JvmtiThreadState::earlyret_tos_offset())); 156 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), rscratch1); 157 br(r0); 158 bind(L); 159 } 160 } 161 162 void InterpreterMacroAssembler::get_unsigned_2_byte_index_at_bcp( 163 Register reg, 164 int bcp_offset) { 165 assert(bcp_offset >= 0, "bcp is still pointing to start of bytecode"); 166 ldrh(reg, Address(rbcp, bcp_offset)); 167 rev16(reg, reg); 168 } 169 170 void InterpreterMacroAssembler::get_dispatch() { 171 unsigned long offset; 172 adrp(rdispatch, ExternalAddress((address)Interpreter::dispatch_table()), offset); 173 lea(rdispatch, Address(rdispatch, offset)); 174 } 175 176 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register index, 177 int bcp_offset, 178 size_t index_size) { 179 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode"); 180 if (index_size == sizeof(u2)) { 181 load_unsigned_short(index, Address(rbcp, bcp_offset)); 182 } else if (index_size == sizeof(u4)) { 183 // assert(EnableInvokeDynamic, "giant index used only for JSR 292"); 184 ldrw(index, Address(rbcp, bcp_offset)); 185 // Check if the secondary index definition is still ~x, otherwise 186 // we have to change the following assembler code to calculate the 187 // plain index. 188 assert(ConstantPool::decode_invokedynamic_index(~123) == 123, "else change next line"); 189 eonw(index, index, zr); // convert to plain index 190 } else if (index_size == sizeof(u1)) { 191 load_unsigned_byte(index, Address(rbcp, bcp_offset)); 192 } else { 193 ShouldNotReachHere(); 194 } 195 } 196 197 // Return 198 // Rindex: index into constant pool 199 // Rcache: address of cache entry - ConstantPoolCache::base_offset() 200 // 201 // A caller must add ConstantPoolCache::base_offset() to Rcache to get 202 // the true address of the cache entry. 203 // 204 void InterpreterMacroAssembler::get_cache_and_index_at_bcp(Register cache, 205 Register index, 206 int bcp_offset, 207 size_t index_size) { 208 assert_different_registers(cache, index); 209 assert_different_registers(cache, rcpool); 210 get_cache_index_at_bcp(index, bcp_offset, index_size); 211 assert(sizeof(ConstantPoolCacheEntry) == 4 * wordSize, "adjust code below"); 212 // convert from field index to ConstantPoolCacheEntry 213 // aarch64 already has the cache in rcpool so there is no need to 214 // install it in cache. instead we pre-add the indexed offset to 215 // rcpool and return it in cache. All clients of this method need to 216 // be modified accordingly. 217 add(cache, rcpool, index, Assembler::LSL, 5); 218 } 219 220 221 void InterpreterMacroAssembler::get_cache_and_index_and_bytecode_at_bcp(Register cache, 222 Register index, 223 Register bytecode, 224 int byte_no, 225 int bcp_offset, 226 size_t index_size) { 227 get_cache_and_index_at_bcp(cache, index, bcp_offset, index_size); 228 // We use a 32-bit load here since the layout of 64-bit words on 229 // little-endian machines allow us that. 230 // n.b. unlike x86 cache already includes the index offset 231 lea(bytecode, Address(cache, 232 ConstantPoolCache::base_offset() 233 + ConstantPoolCacheEntry::indices_offset())); 234 ldarw(bytecode, bytecode); 235 const int shift_count = (1 + byte_no) * BitsPerByte; 236 ubfx(bytecode, bytecode, shift_count, BitsPerByte); 237 } 238 239 void InterpreterMacroAssembler::get_cache_entry_pointer_at_bcp(Register cache, 240 Register tmp, 241 int bcp_offset, 242 size_t index_size) { 243 assert(cache != tmp, "must use different register"); 244 get_cache_index_at_bcp(tmp, bcp_offset, index_size); 245 assert(sizeof(ConstantPoolCacheEntry) == 4 * wordSize, "adjust code below"); 246 // convert from field index to ConstantPoolCacheEntry index 247 // and from word offset to byte offset 248 assert(exact_log2(in_bytes(ConstantPoolCacheEntry::size_in_bytes())) == 2 + LogBytesPerWord, "else change next line"); 249 ldr(cache, Address(rfp, frame::interpreter_frame_cache_offset * wordSize)); 250 // skip past the header 251 add(cache, cache, in_bytes(ConstantPoolCache::base_offset())); 252 add(cache, cache, tmp, Assembler::LSL, 2 + LogBytesPerWord); // construct pointer to cache entry 253 } 254 255 void InterpreterMacroAssembler::get_method_counters(Register method, 256 Register mcs, Label& skip) { 257 Label has_counters; 258 ldr(mcs, Address(method, Method::method_counters_offset())); 259 cbnz(mcs, has_counters); 260 call_VM(noreg, CAST_FROM_FN_PTR(address, 261 InterpreterRuntime::build_method_counters), method); 262 ldr(mcs, Address(method, Method::method_counters_offset())); 263 cbz(mcs, skip); // No MethodCounters allocated, OutOfMemory 264 bind(has_counters); 265 } 266 267 // Load object from cpool->resolved_references(index) 268 void InterpreterMacroAssembler::load_resolved_reference_at_index( 269 Register result, Register index, Register tmp) { 270 assert_different_registers(result, index); 271 272 get_constant_pool(result); 273 // load pointer for resolved_references[] objArray 274 ldr(result, Address(result, ConstantPool::cache_offset_in_bytes())); 275 ldr(result, Address(result, ConstantPoolCache::resolved_references_offset_in_bytes())); 276 resolve_oop_handle(result, tmp); 277 // Add in the index 278 add(index, index, arrayOopDesc::base_offset_in_bytes(T_OBJECT) >> LogBytesPerHeapOop); 279 load_heap_oop(result, Address(result, index, Address::uxtw(LogBytesPerHeapOop))); 280 } 281 282 void InterpreterMacroAssembler::load_resolved_klass_at_offset( 283 Register cpool, Register index, Register klass, Register temp) { 284 add(temp, cpool, index, LSL, LogBytesPerWord); 285 ldrh(temp, Address(temp, sizeof(ConstantPool))); // temp = resolved_klass_index 286 ldr(klass, Address(cpool, ConstantPool::resolved_klasses_offset_in_bytes())); // klass = cpool->_resolved_klasses 287 add(klass, klass, temp, LSL, LogBytesPerWord); 288 ldr(klass, Address(klass, Array<Klass*>::base_offset_in_bytes())); 289 } 290 291 // Generate a subtype check: branch to ok_is_subtype if sub_klass is a 292 // subtype of super_klass. 293 // 294 // Args: 295 // r0: superklass 296 // Rsub_klass: subklass 297 // 298 // Kills: 299 // r2, r5 300 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass, 301 Label& ok_is_subtype) { 302 assert(Rsub_klass != r0, "r0 holds superklass"); 303 assert(Rsub_klass != r2, "r2 holds 2ndary super array length"); 304 assert(Rsub_klass != r5, "r5 holds 2ndary super array scan ptr"); 305 306 // Profile the not-null value's klass. 307 profile_typecheck(r2, Rsub_klass, r5); // blows r2, reloads r5 308 309 // Do the check. 310 check_klass_subtype(Rsub_klass, r0, r2, ok_is_subtype); // blows r2 311 312 // Profile the failure of the check. 313 profile_typecheck_failed(r2); // blows r2 314 } 315 316 // Java Expression Stack 317 318 void InterpreterMacroAssembler::pop_ptr(Register r) { 319 ldr(r, post(esp, wordSize)); 320 } 321 322 void InterpreterMacroAssembler::pop_i(Register r) { 323 ldrw(r, post(esp, wordSize)); 324 } 325 326 void InterpreterMacroAssembler::pop_l(Register r) { 327 ldr(r, post(esp, 2 * Interpreter::stackElementSize)); 328 } 329 330 void InterpreterMacroAssembler::push_ptr(Register r) { 331 str(r, pre(esp, -wordSize)); 332 } 333 334 void InterpreterMacroAssembler::push_i(Register r) { 335 str(r, pre(esp, -wordSize)); 336 } 337 338 void InterpreterMacroAssembler::push_l(Register r) { 339 str(zr, pre(esp, -wordSize)); 340 str(r, pre(esp, - wordSize)); 341 } 342 343 void InterpreterMacroAssembler::pop_f(FloatRegister r) { 344 ldrs(r, post(esp, wordSize)); 345 } 346 347 void InterpreterMacroAssembler::pop_d(FloatRegister r) { 348 ldrd(r, post(esp, 2 * Interpreter::stackElementSize)); 349 } 350 351 void InterpreterMacroAssembler::push_f(FloatRegister r) { 352 strs(r, pre(esp, -wordSize)); 353 } 354 355 void InterpreterMacroAssembler::push_d(FloatRegister r) { 356 strd(r, pre(esp, 2* -wordSize)); 357 } 358 359 void InterpreterMacroAssembler::pop(TosState state) { 360 switch (state) { 361 case atos: pop_ptr(); break; 362 case btos: 363 case ztos: 364 case ctos: 365 case stos: 366 case itos: pop_i(); break; 367 case ltos: pop_l(); break; 368 case ftos: pop_f(); break; 369 case dtos: pop_d(); break; 370 case vtos: /* nothing to do */ break; 371 default: ShouldNotReachHere(); 372 } 373 verify_oop(r0, state); 374 } 375 376 void InterpreterMacroAssembler::push(TosState state) { 377 verify_oop(r0, state); 378 switch (state) { 379 case atos: push_ptr(); break; 380 case btos: 381 case ztos: 382 case ctos: 383 case stos: 384 case itos: push_i(); break; 385 case ltos: push_l(); break; 386 case ftos: push_f(); break; 387 case dtos: push_d(); break; 388 case vtos: /* nothing to do */ break; 389 default : ShouldNotReachHere(); 390 } 391 } 392 393 // Helpers for swap and dup 394 void InterpreterMacroAssembler::load_ptr(int n, Register val) { 395 ldr(val, Address(esp, Interpreter::expr_offset_in_bytes(n))); 396 } 397 398 void InterpreterMacroAssembler::store_ptr(int n, Register val) { 399 str(val, Address(esp, Interpreter::expr_offset_in_bytes(n))); 400 } 401 402 void InterpreterMacroAssembler::load_float(Address src) { 403 ldrs(v0, src); 404 } 405 406 void InterpreterMacroAssembler::load_double(Address src) { 407 ldrd(v0, src); 408 } 409 410 void InterpreterMacroAssembler::prepare_to_jump_from_interpreted() { 411 // set sender sp 412 mov(r13, sp); 413 // record last_sp 414 str(esp, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize)); 415 } 416 417 // Jump to from_interpreted entry of a call unless single stepping is possible 418 // in this thread in which case we must call the i2i entry 419 void InterpreterMacroAssembler::jump_from_interpreted(Register method, Register temp) { 420 prepare_to_jump_from_interpreted(); 421 422 if (JvmtiExport::can_post_interpreter_events()) { 423 Label run_compiled_code; 424 // JVMTI events, such as single-stepping, are implemented partly by avoiding running 425 // compiled code in threads for which the event is enabled. Check here for 426 // interp_only_mode if these events CAN be enabled. 427 ldrw(rscratch1, Address(rthread, JavaThread::interp_only_mode_offset())); 428 cbzw(rscratch1, run_compiled_code); 429 ldr(rscratch1, Address(method, Method::interpreter_entry_offset())); 430 br(rscratch1); 431 bind(run_compiled_code); 432 } 433 434 ldr(rscratch1, Address(method, Method::from_interpreted_offset())); 435 br(rscratch1); 436 } 437 438 // The following two routines provide a hook so that an implementation 439 // can schedule the dispatch in two parts. amd64 does not do this. 440 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int step) { 441 } 442 443 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int step) { 444 dispatch_next(state, step); 445 } 446 447 void InterpreterMacroAssembler::dispatch_base(TosState state, 448 address* table, 449 bool verifyoop, 450 bool generate_poll) { 451 if (VerifyActivationFrameSize) { 452 Unimplemented(); 453 } 454 if (verifyoop) { 455 verify_oop(r0, state); 456 } 457 458 Label safepoint; 459 address* const safepoint_table = Interpreter::safept_table(state); 460 bool needs_thread_local_poll = generate_poll && 461 SafepointMechanism::uses_thread_local_poll() && table != safepoint_table; 462 463 if (needs_thread_local_poll) { 464 NOT_PRODUCT(block_comment("Thread-local Safepoint poll")); 465 ldr(rscratch2, Address(rthread, Thread::polling_page_offset())); 466 tbnz(rscratch2, exact_log2(SafepointMechanism::poll_bit()), safepoint); 467 } 468 469 if (table == Interpreter::dispatch_table(state)) { 470 addw(rscratch2, rscratch1, Interpreter::distance_from_dispatch_table(state)); 471 ldr(rscratch2, Address(rdispatch, rscratch2, Address::uxtw(3))); 472 } else { 473 mov(rscratch2, (address)table); 474 ldr(rscratch2, Address(rscratch2, rscratch1, Address::uxtw(3))); 475 } 476 br(rscratch2); 477 478 if (needs_thread_local_poll) { 479 bind(safepoint); 480 lea(rscratch2, ExternalAddress((address)safepoint_table)); 481 ldr(rscratch2, Address(rscratch2, rscratch1, Address::uxtw(3))); 482 br(rscratch2); 483 } 484 } 485 486 void InterpreterMacroAssembler::dispatch_only(TosState state, bool generate_poll) { 487 dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll); 488 } 489 490 void InterpreterMacroAssembler::dispatch_only_normal(TosState state) { 491 dispatch_base(state, Interpreter::normal_table(state)); 492 } 493 494 void InterpreterMacroAssembler::dispatch_only_noverify(TosState state) { 495 dispatch_base(state, Interpreter::normal_table(state), false); 496 } 497 498 499 void InterpreterMacroAssembler::dispatch_next(TosState state, int step, bool generate_poll) { 500 // load next bytecode 501 ldrb(rscratch1, Address(pre(rbcp, step))); 502 dispatch_base(state, Interpreter::dispatch_table(state), generate_poll); 503 } 504 505 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) { 506 // load current bytecode 507 ldrb(rscratch1, Address(rbcp, 0)); 508 dispatch_base(state, table); 509 } 510 511 // remove activation 512 // 513 // Unlock the receiver if this is a synchronized method. 514 // Unlock any Java monitors from syncronized blocks. 515 // Remove the activation from the stack. 516 // 517 // If there are locked Java monitors 518 // If throw_monitor_exception 519 // throws IllegalMonitorStateException 520 // Else if install_monitor_exception 521 // installs IllegalMonitorStateException 522 // Else 523 // no error processing 524 void InterpreterMacroAssembler::remove_activation( 525 TosState state, 526 bool throw_monitor_exception, 527 bool install_monitor_exception, 528 bool notify_jvmdi) { 529 // Note: Registers r3 xmm0 may be in use for the 530 // result check if synchronized method 531 Label unlocked, unlock, no_unlock; 532 533 // get the value of _do_not_unlock_if_synchronized into r3 534 const Address do_not_unlock_if_synchronized(rthread, 535 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset())); 536 ldrb(r3, do_not_unlock_if_synchronized); 537 strb(zr, do_not_unlock_if_synchronized); // reset the flag 538 539 // get method access flags 540 ldr(r1, Address(rfp, frame::interpreter_frame_method_offset * wordSize)); 541 ldr(r2, Address(r1, Method::access_flags_offset())); 542 tbz(r2, exact_log2(JVM_ACC_SYNCHRONIZED), unlocked); 543 544 // Don't unlock anything if the _do_not_unlock_if_synchronized flag 545 // is set. 546 cbnz(r3, no_unlock); 547 548 // unlock monitor 549 push(state); // save result 550 551 // BasicObjectLock will be first in list, since this is a 552 // synchronized method. However, need to check that the object has 553 // not been unlocked by an explicit monitorexit bytecode. 554 const Address monitor(rfp, frame::interpreter_frame_initial_sp_offset * 555 wordSize - (int) sizeof(BasicObjectLock)); 556 // We use c_rarg1 so that if we go slow path it will be the correct 557 // register for unlock_object to pass to VM directly 558 lea(c_rarg1, monitor); // address of first monitor 559 560 ldr(r0, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes())); 561 cbnz(r0, unlock); 562 563 pop(state); 564 if (throw_monitor_exception) { 565 // Entry already unlocked, need to throw exception 566 call_VM(noreg, CAST_FROM_FN_PTR(address, 567 InterpreterRuntime::throw_illegal_monitor_state_exception)); 568 should_not_reach_here(); 569 } else { 570 // Monitor already unlocked during a stack unroll. If requested, 571 // install an illegal_monitor_state_exception. Continue with 572 // stack unrolling. 573 if (install_monitor_exception) { 574 call_VM(noreg, CAST_FROM_FN_PTR(address, 575 InterpreterRuntime::new_illegal_monitor_state_exception)); 576 } 577 b(unlocked); 578 } 579 580 bind(unlock); 581 unlock_object(c_rarg1); 582 pop(state); 583 584 // Check that for block-structured locking (i.e., that all locked 585 // objects has been unlocked) 586 bind(unlocked); 587 588 // r0: Might contain return value 589 590 // Check that all monitors are unlocked 591 { 592 Label loop, exception, entry, restart; 593 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize; 594 const Address monitor_block_top( 595 rfp, frame::interpreter_frame_monitor_block_top_offset * wordSize); 596 const Address monitor_block_bot( 597 rfp, frame::interpreter_frame_initial_sp_offset * wordSize); 598 599 bind(restart); 600 // We use c_rarg1 so that if we go slow path it will be the correct 601 // register for unlock_object to pass to VM directly 602 ldr(c_rarg1, monitor_block_top); // points to current entry, starting 603 // with top-most entry 604 lea(r19, monitor_block_bot); // points to word before bottom of 605 // monitor block 606 b(entry); 607 608 // Entry already locked, need to throw exception 609 bind(exception); 610 611 if (throw_monitor_exception) { 612 // Throw exception 613 MacroAssembler::call_VM(noreg, 614 CAST_FROM_FN_PTR(address, InterpreterRuntime:: 615 throw_illegal_monitor_state_exception)); 616 should_not_reach_here(); 617 } else { 618 // Stack unrolling. Unlock object and install illegal_monitor_exception. 619 // Unlock does not block, so don't have to worry about the frame. 620 // We don't have to preserve c_rarg1 since we are going to throw an exception. 621 622 push(state); 623 unlock_object(c_rarg1); 624 pop(state); 625 626 if (install_monitor_exception) { 627 call_VM(noreg, CAST_FROM_FN_PTR(address, 628 InterpreterRuntime:: 629 new_illegal_monitor_state_exception)); 630 } 631 632 b(restart); 633 } 634 635 bind(loop); 636 // check if current entry is used 637 ldr(rscratch1, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes())); 638 cbnz(rscratch1, exception); 639 640 add(c_rarg1, c_rarg1, entry_size); // otherwise advance to next entry 641 bind(entry); 642 cmp(c_rarg1, r19); // check if bottom reached 643 br(Assembler::NE, loop); // if not at bottom then check this entry 644 } 645 646 bind(no_unlock); 647 648 // jvmti support 649 if (notify_jvmdi) { 650 notify_method_exit(state, NotifyJVMTI); // preserve TOSCA 651 } else { 652 notify_method_exit(state, SkipNotifyJVMTI); // preserve TOSCA 653 } 654 655 // remove activation 656 // get sender esp 657 ldr(esp, 658 Address(rfp, frame::interpreter_frame_sender_sp_offset * wordSize)); 659 if (StackReservedPages > 0) { 660 // testing if reserved zone needs to be re-enabled 661 Label no_reserved_zone_enabling; 662 663 ldr(rscratch1, Address(rthread, JavaThread::reserved_stack_activation_offset())); 664 cmp(esp, rscratch1); 665 br(Assembler::LS, no_reserved_zone_enabling); 666 667 call_VM_leaf( 668 CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), rthread); 669 call_VM(noreg, CAST_FROM_FN_PTR(address, 670 InterpreterRuntime::throw_delayed_StackOverflowError)); 671 should_not_reach_here(); 672 673 bind(no_reserved_zone_enabling); 674 } 675 // remove frame anchor 676 leave(); 677 // If we're returning to interpreted code we will shortly be 678 // adjusting SP to allow some space for ESP. If we're returning to 679 // compiled code the saved sender SP was saved in sender_sp, so this 680 // restores it. 681 andr(sp, esp, -16); 682 } 683 684 // Lock object 685 // 686 // Args: 687 // c_rarg1: BasicObjectLock to be used for locking 688 // 689 // Kills: 690 // r0 691 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, .. (param regs) 692 // rscratch1, rscratch2 (scratch regs) 693 void InterpreterMacroAssembler::lock_object(Register lock_reg) 694 { 695 assert(lock_reg == c_rarg1, "The argument is only for looks. It must be c_rarg1"); 696 if (UseHeavyMonitors) { 697 call_VM(noreg, 698 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), 699 lock_reg); 700 } else { 701 Label done; 702 703 const Register swap_reg = r0; 704 const Register tmp = c_rarg2; 705 const Register obj_reg = c_rarg3; // Will contain the oop 706 707 const int obj_offset = BasicObjectLock::obj_offset_in_bytes(); 708 const int lock_offset = BasicObjectLock::lock_offset_in_bytes (); 709 const int mark_offset = lock_offset + 710 BasicLock::displaced_header_offset_in_bytes(); 711 712 Label slow_case; 713 714 // Load object pointer into obj_reg %c_rarg3 715 ldr(obj_reg, Address(lock_reg, obj_offset)); 716 717 if (UseBiasedLocking) { 718 biased_locking_enter(lock_reg, obj_reg, swap_reg, tmp, false, done, &slow_case); 719 } 720 721 // Load (object->mark() | 1) into swap_reg 722 ldr(rscratch1, Address(obj_reg, oopDesc::mark_offset_in_bytes())); 723 orr(swap_reg, rscratch1, 1); 724 725 // Save (object->mark() | 1) into BasicLock's displaced header 726 str(swap_reg, Address(lock_reg, mark_offset)); 727 728 assert(lock_offset == 0, 729 "displached header must be first word in BasicObjectLock"); 730 731 Label fail; 732 if (PrintBiasedLockingStatistics) { 733 Label fast; 734 cmpxchg_obj_header(swap_reg, lock_reg, obj_reg, rscratch1, fast, &fail); 735 bind(fast); 736 atomic_incw(Address((address)BiasedLocking::fast_path_entry_count_addr()), 737 rscratch2, rscratch1, tmp); 738 b(done); 739 bind(fail); 740 } else { 741 cmpxchg_obj_header(swap_reg, lock_reg, obj_reg, rscratch1, done, /*fallthrough*/NULL); 742 } 743 744 // Test if the oopMark is an obvious stack pointer, i.e., 745 // 1) (mark & 7) == 0, and 746 // 2) rsp <= mark < mark + os::pagesize() 747 // 748 // These 3 tests can be done by evaluating the following 749 // expression: ((mark - rsp) & (7 - os::vm_page_size())), 750 // assuming both stack pointer and pagesize have their 751 // least significant 3 bits clear. 752 // NOTE: the oopMark is in swap_reg %r0 as the result of cmpxchg 753 // NOTE2: aarch64 does not like to subtract sp from rn so take a 754 // copy 755 mov(rscratch1, sp); 756 sub(swap_reg, swap_reg, rscratch1); 757 ands(swap_reg, swap_reg, (unsigned long)(7 - os::vm_page_size())); 758 759 // Save the test result, for recursive case, the result is zero 760 str(swap_reg, Address(lock_reg, mark_offset)); 761 762 if (PrintBiasedLockingStatistics) { 763 br(Assembler::NE, slow_case); 764 atomic_incw(Address((address)BiasedLocking::fast_path_entry_count_addr()), 765 rscratch2, rscratch1, tmp); 766 } 767 br(Assembler::EQ, done); 768 769 bind(slow_case); 770 771 // Call the runtime routine for slow case 772 call_VM(noreg, 773 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), 774 lock_reg); 775 776 bind(done); 777 } 778 } 779 780 781 // Unlocks an object. Used in monitorexit bytecode and 782 // remove_activation. Throws an IllegalMonitorException if object is 783 // not locked by current thread. 784 // 785 // Args: 786 // c_rarg1: BasicObjectLock for lock 787 // 788 // Kills: 789 // r0 790 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ... (param regs) 791 // rscratch1, rscratch2 (scratch regs) 792 void InterpreterMacroAssembler::unlock_object(Register lock_reg) 793 { 794 assert(lock_reg == c_rarg1, "The argument is only for looks. It must be rarg1"); 795 796 if (UseHeavyMonitors) { 797 call_VM(noreg, 798 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), 799 lock_reg); 800 } else { 801 Label done; 802 803 const Register swap_reg = r0; 804 const Register header_reg = c_rarg2; // Will contain the old oopMark 805 const Register obj_reg = c_rarg3; // Will contain the oop 806 807 save_bcp(); // Save in case of exception 808 809 // Convert from BasicObjectLock structure to object and BasicLock 810 // structure Store the BasicLock address into %r0 811 lea(swap_reg, Address(lock_reg, BasicObjectLock::lock_offset_in_bytes())); 812 813 // Load oop into obj_reg(%c_rarg3) 814 ldr(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset_in_bytes())); 815 816 // Free entry 817 str(zr, Address(lock_reg, BasicObjectLock::obj_offset_in_bytes())); 818 819 if (UseBiasedLocking) { 820 biased_locking_exit(obj_reg, header_reg, done); 821 } 822 823 // Load the old header from BasicLock structure 824 ldr(header_reg, Address(swap_reg, 825 BasicLock::displaced_header_offset_in_bytes())); 826 827 // Test for recursion 828 cbz(header_reg, done); 829 830 // Atomic swap back the old header 831 cmpxchg_obj_header(swap_reg, header_reg, obj_reg, rscratch1, done, /*fallthrough*/NULL); 832 833 // Call the runtime routine for slow case. 834 str(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset_in_bytes())); // restore obj 835 call_VM(noreg, 836 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), 837 lock_reg); 838 839 bind(done); 840 841 restore_bcp(); 842 } 843 } 844 845 void InterpreterMacroAssembler::test_method_data_pointer(Register mdp, 846 Label& zero_continue) { 847 assert(ProfileInterpreter, "must be profiling interpreter"); 848 ldr(mdp, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize)); 849 cbz(mdp, zero_continue); 850 } 851 852 // Set the method data pointer for the current bcp. 853 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() { 854 assert(ProfileInterpreter, "must be profiling interpreter"); 855 Label set_mdp; 856 stp(r0, r1, Address(pre(sp, -2 * wordSize))); 857 858 // Test MDO to avoid the call if it is NULL. 859 ldr(r0, Address(rmethod, in_bytes(Method::method_data_offset()))); 860 cbz(r0, set_mdp); 861 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), rmethod, rbcp); 862 // r0: mdi 863 // mdo is guaranteed to be non-zero here, we checked for it before the call. 864 ldr(r1, Address(rmethod, in_bytes(Method::method_data_offset()))); 865 lea(r1, Address(r1, in_bytes(MethodData::data_offset()))); 866 add(r0, r1, r0); 867 str(r0, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize)); 868 bind(set_mdp); 869 ldp(r0, r1, Address(post(sp, 2 * wordSize))); 870 } 871 872 void InterpreterMacroAssembler::verify_method_data_pointer() { 873 assert(ProfileInterpreter, "must be profiling interpreter"); 874 #ifdef ASSERT 875 Label verify_continue; 876 stp(r0, r1, Address(pre(sp, -2 * wordSize))); 877 stp(r2, r3, Address(pre(sp, -2 * wordSize))); 878 test_method_data_pointer(r3, verify_continue); // If mdp is zero, continue 879 get_method(r1); 880 881 // If the mdp is valid, it will point to a DataLayout header which is 882 // consistent with the bcp. The converse is highly probable also. 883 ldrsh(r2, Address(r3, in_bytes(DataLayout::bci_offset()))); 884 ldr(rscratch1, Address(r1, Method::const_offset())); 885 add(r2, r2, rscratch1, Assembler::LSL); 886 lea(r2, Address(r2, ConstMethod::codes_offset())); 887 cmp(r2, rbcp); 888 br(Assembler::EQ, verify_continue); 889 // r1: method 890 // rbcp: bcp // rbcp == 22 891 // r3: mdp 892 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp), 893 r1, rbcp, r3); 894 bind(verify_continue); 895 ldp(r2, r3, Address(post(sp, 2 * wordSize))); 896 ldp(r0, r1, Address(post(sp, 2 * wordSize))); 897 #endif // ASSERT 898 } 899 900 901 void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in, 902 int constant, 903 Register value) { 904 assert(ProfileInterpreter, "must be profiling interpreter"); 905 Address data(mdp_in, constant); 906 str(value, data); 907 } 908 909 910 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in, 911 int constant, 912 bool decrement) { 913 increment_mdp_data_at(mdp_in, noreg, constant, decrement); 914 } 915 916 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in, 917 Register reg, 918 int constant, 919 bool decrement) { 920 assert(ProfileInterpreter, "must be profiling interpreter"); 921 // %%% this does 64bit counters at best it is wasting space 922 // at worst it is a rare bug when counters overflow 923 924 assert_different_registers(rscratch2, rscratch1, mdp_in, reg); 925 926 Address addr1(mdp_in, constant); 927 Address addr2(rscratch2, reg, Address::lsl(0)); 928 Address &addr = addr1; 929 if (reg != noreg) { 930 lea(rscratch2, addr1); 931 addr = addr2; 932 } 933 934 if (decrement) { 935 // Decrement the register. Set condition codes. 936 // Intel does this 937 // addptr(data, (int32_t) -DataLayout::counter_increment); 938 // If the decrement causes the counter to overflow, stay negative 939 // Label L; 940 // jcc(Assembler::negative, L); 941 // addptr(data, (int32_t) DataLayout::counter_increment); 942 // so we do this 943 ldr(rscratch1, addr); 944 subs(rscratch1, rscratch1, (unsigned)DataLayout::counter_increment); 945 Label L; 946 br(Assembler::LO, L); // skip store if counter underflow 947 str(rscratch1, addr); 948 bind(L); 949 } else { 950 assert(DataLayout::counter_increment == 1, 951 "flow-free idiom only works with 1"); 952 // Intel does this 953 // Increment the register. Set carry flag. 954 // addptr(data, DataLayout::counter_increment); 955 // If the increment causes the counter to overflow, pull back by 1. 956 // sbbptr(data, (int32_t)0); 957 // so we do this 958 ldr(rscratch1, addr); 959 adds(rscratch1, rscratch1, DataLayout::counter_increment); 960 Label L; 961 br(Assembler::CS, L); // skip store if counter overflow 962 str(rscratch1, addr); 963 bind(L); 964 } 965 } 966 967 void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in, 968 int flag_byte_constant) { 969 assert(ProfileInterpreter, "must be profiling interpreter"); 970 int flags_offset = in_bytes(DataLayout::flags_offset()); 971 // Set the flag 972 ldrb(rscratch1, Address(mdp_in, flags_offset)); 973 orr(rscratch1, rscratch1, flag_byte_constant); 974 strb(rscratch1, Address(mdp_in, flags_offset)); 975 } 976 977 978 void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in, 979 int offset, 980 Register value, 981 Register test_value_out, 982 Label& not_equal_continue) { 983 assert(ProfileInterpreter, "must be profiling interpreter"); 984 if (test_value_out == noreg) { 985 ldr(rscratch1, Address(mdp_in, offset)); 986 cmp(value, rscratch1); 987 } else { 988 // Put the test value into a register, so caller can use it: 989 ldr(test_value_out, Address(mdp_in, offset)); 990 cmp(value, test_value_out); 991 } 992 br(Assembler::NE, not_equal_continue); 993 } 994 995 996 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in, 997 int offset_of_disp) { 998 assert(ProfileInterpreter, "must be profiling interpreter"); 999 ldr(rscratch1, Address(mdp_in, offset_of_disp)); 1000 add(mdp_in, mdp_in, rscratch1, LSL); 1001 str(mdp_in, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize)); 1002 } 1003 1004 1005 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in, 1006 Register reg, 1007 int offset_of_disp) { 1008 assert(ProfileInterpreter, "must be profiling interpreter"); 1009 lea(rscratch1, Address(mdp_in, offset_of_disp)); 1010 ldr(rscratch1, Address(rscratch1, reg, Address::lsl(0))); 1011 add(mdp_in, mdp_in, rscratch1, LSL); 1012 str(mdp_in, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize)); 1013 } 1014 1015 1016 void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in, 1017 int constant) { 1018 assert(ProfileInterpreter, "must be profiling interpreter"); 1019 add(mdp_in, mdp_in, (unsigned)constant); 1020 str(mdp_in, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize)); 1021 } 1022 1023 1024 void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) { 1025 assert(ProfileInterpreter, "must be profiling interpreter"); 1026 // save/restore across call_VM 1027 stp(zr, return_bci, Address(pre(sp, -2 * wordSize))); 1028 call_VM(noreg, 1029 CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret), 1030 return_bci); 1031 ldp(zr, return_bci, Address(post(sp, 2 * wordSize))); 1032 } 1033 1034 1035 void InterpreterMacroAssembler::profile_taken_branch(Register mdp, 1036 Register bumped_count) { 1037 if (ProfileInterpreter) { 1038 Label profile_continue; 1039 1040 // If no method data exists, go to profile_continue. 1041 // Otherwise, assign to mdp 1042 test_method_data_pointer(mdp, profile_continue); 1043 1044 // We are taking a branch. Increment the taken count. 1045 // We inline increment_mdp_data_at to return bumped_count in a register 1046 //increment_mdp_data_at(mdp, in_bytes(JumpData::taken_offset())); 1047 Address data(mdp, in_bytes(JumpData::taken_offset())); 1048 ldr(bumped_count, data); 1049 assert(DataLayout::counter_increment == 1, 1050 "flow-free idiom only works with 1"); 1051 // Intel does this to catch overflow 1052 // addptr(bumped_count, DataLayout::counter_increment); 1053 // sbbptr(bumped_count, 0); 1054 // so we do this 1055 adds(bumped_count, bumped_count, DataLayout::counter_increment); 1056 Label L; 1057 br(Assembler::CS, L); // skip store if counter overflow 1058 str(bumped_count, data); 1059 bind(L); 1060 // The method data pointer needs to be updated to reflect the new target. 1061 update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset())); 1062 bind(profile_continue); 1063 } 1064 } 1065 1066 1067 void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp) { 1068 if (ProfileInterpreter) { 1069 Label profile_continue; 1070 1071 // If no method data exists, go to profile_continue. 1072 test_method_data_pointer(mdp, profile_continue); 1073 1074 // We are taking a branch. Increment the not taken count. 1075 increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset())); 1076 1077 // The method data pointer needs to be updated to correspond to 1078 // the next bytecode 1079 update_mdp_by_constant(mdp, in_bytes(BranchData::branch_data_size())); 1080 bind(profile_continue); 1081 } 1082 } 1083 1084 1085 void InterpreterMacroAssembler::profile_call(Register mdp) { 1086 if (ProfileInterpreter) { 1087 Label profile_continue; 1088 1089 // If no method data exists, go to profile_continue. 1090 test_method_data_pointer(mdp, profile_continue); 1091 1092 // We are making a call. Increment the count. 1093 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); 1094 1095 // The method data pointer needs to be updated to reflect the new target. 1096 update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size())); 1097 bind(profile_continue); 1098 } 1099 } 1100 1101 void InterpreterMacroAssembler::profile_final_call(Register mdp) { 1102 if (ProfileInterpreter) { 1103 Label profile_continue; 1104 1105 // If no method data exists, go to profile_continue. 1106 test_method_data_pointer(mdp, profile_continue); 1107 1108 // We are making a call. Increment the count. 1109 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); 1110 1111 // The method data pointer needs to be updated to reflect the new target. 1112 update_mdp_by_constant(mdp, 1113 in_bytes(VirtualCallData:: 1114 virtual_call_data_size())); 1115 bind(profile_continue); 1116 } 1117 } 1118 1119 1120 void InterpreterMacroAssembler::profile_virtual_call(Register receiver, 1121 Register mdp, 1122 Register reg2, 1123 bool receiver_can_be_null) { 1124 if (ProfileInterpreter) { 1125 Label profile_continue; 1126 1127 // If no method data exists, go to profile_continue. 1128 test_method_data_pointer(mdp, profile_continue); 1129 1130 Label skip_receiver_profile; 1131 if (receiver_can_be_null) { 1132 Label not_null; 1133 // We are making a call. Increment the count for null receiver. 1134 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); 1135 b(skip_receiver_profile); 1136 bind(not_null); 1137 } 1138 1139 // Record the receiver type. 1140 record_klass_in_profile(receiver, mdp, reg2, true); 1141 bind(skip_receiver_profile); 1142 1143 // The method data pointer needs to be updated to reflect the new target. 1144 #if INCLUDE_JVMCI 1145 if (MethodProfileWidth == 0) { 1146 update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size())); 1147 } 1148 #else // INCLUDE_JVMCI 1149 update_mdp_by_constant(mdp, 1150 in_bytes(VirtualCallData:: 1151 virtual_call_data_size())); 1152 #endif // INCLUDE_JVMCI 1153 bind(profile_continue); 1154 } 1155 } 1156 1157 #if INCLUDE_JVMCI 1158 void InterpreterMacroAssembler::profile_called_method(Register method, Register mdp, Register reg2) { 1159 assert_different_registers(method, mdp, reg2); 1160 if (ProfileInterpreter && MethodProfileWidth > 0) { 1161 Label profile_continue; 1162 1163 // If no method data exists, go to profile_continue. 1164 test_method_data_pointer(mdp, profile_continue); 1165 1166 Label done; 1167 record_item_in_profile_helper(method, mdp, reg2, 0, done, MethodProfileWidth, 1168 &VirtualCallData::method_offset, &VirtualCallData::method_count_offset, in_bytes(VirtualCallData::nonprofiled_receiver_count_offset())); 1169 bind(done); 1170 1171 update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size())); 1172 bind(profile_continue); 1173 } 1174 } 1175 #endif // INCLUDE_JVMCI 1176 1177 // This routine creates a state machine for updating the multi-row 1178 // type profile at a virtual call site (or other type-sensitive bytecode). 1179 // The machine visits each row (of receiver/count) until the receiver type 1180 // is found, or until it runs out of rows. At the same time, it remembers 1181 // the location of the first empty row. (An empty row records null for its 1182 // receiver, and can be allocated for a newly-observed receiver type.) 1183 // Because there are two degrees of freedom in the state, a simple linear 1184 // search will not work; it must be a decision tree. Hence this helper 1185 // function is recursive, to generate the required tree structured code. 1186 // It's the interpreter, so we are trading off code space for speed. 1187 // See below for example code. 1188 void InterpreterMacroAssembler::record_klass_in_profile_helper( 1189 Register receiver, Register mdp, 1190 Register reg2, int start_row, 1191 Label& done, bool is_virtual_call) { 1192 if (TypeProfileWidth == 0) { 1193 if (is_virtual_call) { 1194 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); 1195 } 1196 #if INCLUDE_JVMCI 1197 else if (EnableJVMCI) { 1198 increment_mdp_data_at(mdp, in_bytes(ReceiverTypeData::nonprofiled_receiver_count_offset())); 1199 } 1200 #endif // INCLUDE_JVMCI 1201 } else { 1202 int non_profiled_offset = -1; 1203 if (is_virtual_call) { 1204 non_profiled_offset = in_bytes(CounterData::count_offset()); 1205 } 1206 #if INCLUDE_JVMCI 1207 else if (EnableJVMCI) { 1208 non_profiled_offset = in_bytes(ReceiverTypeData::nonprofiled_receiver_count_offset()); 1209 } 1210 #endif // INCLUDE_JVMCI 1211 1212 record_item_in_profile_helper(receiver, mdp, reg2, 0, done, TypeProfileWidth, 1213 &VirtualCallData::receiver_offset, &VirtualCallData::receiver_count_offset, non_profiled_offset); 1214 } 1215 } 1216 1217 void InterpreterMacroAssembler::record_item_in_profile_helper(Register item, Register mdp, 1218 Register reg2, int start_row, Label& done, int total_rows, 1219 OffsetFunction item_offset_fn, OffsetFunction item_count_offset_fn, 1220 int non_profiled_offset) { 1221 int last_row = total_rows - 1; 1222 assert(start_row <= last_row, "must be work left to do"); 1223 // Test this row for both the item and for null. 1224 // Take any of three different outcomes: 1225 // 1. found item => increment count and goto done 1226 // 2. found null => keep looking for case 1, maybe allocate this cell 1227 // 3. found something else => keep looking for cases 1 and 2 1228 // Case 3 is handled by a recursive call. 1229 for (int row = start_row; row <= last_row; row++) { 1230 Label next_test; 1231 bool test_for_null_also = (row == start_row); 1232 1233 // See if the item is item[n]. 1234 int item_offset = in_bytes(item_offset_fn(row)); 1235 test_mdp_data_at(mdp, item_offset, item, 1236 (test_for_null_also ? reg2 : noreg), 1237 next_test); 1238 // (Reg2 now contains the item from the CallData.) 1239 1240 // The item is item[n]. Increment count[n]. 1241 int count_offset = in_bytes(item_count_offset_fn(row)); 1242 increment_mdp_data_at(mdp, count_offset); 1243 b(done); 1244 bind(next_test); 1245 1246 if (test_for_null_also) { 1247 Label found_null; 1248 // Failed the equality check on item[n]... Test for null. 1249 if (start_row == last_row) { 1250 // The only thing left to do is handle the null case. 1251 if (non_profiled_offset >= 0) { 1252 cbz(reg2, found_null); 1253 // Item did not match any saved item and there is no empty row for it. 1254 // Increment total counter to indicate polymorphic case. 1255 increment_mdp_data_at(mdp, non_profiled_offset); 1256 b(done); 1257 bind(found_null); 1258 } else { 1259 cbnz(reg2, done); 1260 } 1261 break; 1262 } 1263 // Since null is rare, make it be the branch-taken case. 1264 cbz(reg2, found_null); 1265 1266 // Put all the "Case 3" tests here. 1267 record_item_in_profile_helper(item, mdp, reg2, start_row + 1, done, total_rows, 1268 item_offset_fn, item_count_offset_fn, non_profiled_offset); 1269 1270 // Found a null. Keep searching for a matching item, 1271 // but remember that this is an empty (unused) slot. 1272 bind(found_null); 1273 } 1274 } 1275 1276 // In the fall-through case, we found no matching item, but we 1277 // observed the item[start_row] is NULL. 1278 1279 // Fill in the item field and increment the count. 1280 int item_offset = in_bytes(item_offset_fn(start_row)); 1281 set_mdp_data_at(mdp, item_offset, item); 1282 int count_offset = in_bytes(item_count_offset_fn(start_row)); 1283 mov(reg2, DataLayout::counter_increment); 1284 set_mdp_data_at(mdp, count_offset, reg2); 1285 if (start_row > 0) { 1286 b(done); 1287 } 1288 } 1289 1290 // Example state machine code for three profile rows: 1291 // // main copy of decision tree, rooted at row[1] 1292 // if (row[0].rec == rec) { row[0].incr(); goto done; } 1293 // if (row[0].rec != NULL) { 1294 // // inner copy of decision tree, rooted at row[1] 1295 // if (row[1].rec == rec) { row[1].incr(); goto done; } 1296 // if (row[1].rec != NULL) { 1297 // // degenerate decision tree, rooted at row[2] 1298 // if (row[2].rec == rec) { row[2].incr(); goto done; } 1299 // if (row[2].rec != NULL) { count.incr(); goto done; } // overflow 1300 // row[2].init(rec); goto done; 1301 // } else { 1302 // // remember row[1] is empty 1303 // if (row[2].rec == rec) { row[2].incr(); goto done; } 1304 // row[1].init(rec); goto done; 1305 // } 1306 // } else { 1307 // // remember row[0] is empty 1308 // if (row[1].rec == rec) { row[1].incr(); goto done; } 1309 // if (row[2].rec == rec) { row[2].incr(); goto done; } 1310 // row[0].init(rec); goto done; 1311 // } 1312 // done: 1313 1314 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver, 1315 Register mdp, Register reg2, 1316 bool is_virtual_call) { 1317 assert(ProfileInterpreter, "must be profiling"); 1318 Label done; 1319 1320 record_klass_in_profile_helper(receiver, mdp, reg2, 0, done, is_virtual_call); 1321 1322 bind (done); 1323 } 1324 1325 void InterpreterMacroAssembler::profile_ret(Register return_bci, 1326 Register mdp) { 1327 if (ProfileInterpreter) { 1328 Label profile_continue; 1329 uint row; 1330 1331 // If no method data exists, go to profile_continue. 1332 test_method_data_pointer(mdp, profile_continue); 1333 1334 // Update the total ret count. 1335 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); 1336 1337 for (row = 0; row < RetData::row_limit(); row++) { 1338 Label next_test; 1339 1340 // See if return_bci is equal to bci[n]: 1341 test_mdp_data_at(mdp, 1342 in_bytes(RetData::bci_offset(row)), 1343 return_bci, noreg, 1344 next_test); 1345 1346 // return_bci is equal to bci[n]. Increment the count. 1347 increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row))); 1348 1349 // The method data pointer needs to be updated to reflect the new target. 1350 update_mdp_by_offset(mdp, 1351 in_bytes(RetData::bci_displacement_offset(row))); 1352 b(profile_continue); 1353 bind(next_test); 1354 } 1355 1356 update_mdp_for_ret(return_bci); 1357 1358 bind(profile_continue); 1359 } 1360 } 1361 1362 void InterpreterMacroAssembler::profile_null_seen(Register mdp) { 1363 if (ProfileInterpreter) { 1364 Label profile_continue; 1365 1366 // If no method data exists, go to profile_continue. 1367 test_method_data_pointer(mdp, profile_continue); 1368 1369 set_mdp_flag_at(mdp, BitData::null_seen_byte_constant()); 1370 1371 // The method data pointer needs to be updated. 1372 int mdp_delta = in_bytes(BitData::bit_data_size()); 1373 if (TypeProfileCasts) { 1374 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size()); 1375 } 1376 update_mdp_by_constant(mdp, mdp_delta); 1377 1378 bind(profile_continue); 1379 } 1380 } 1381 1382 void InterpreterMacroAssembler::profile_typecheck_failed(Register mdp) { 1383 if (ProfileInterpreter && TypeProfileCasts) { 1384 Label profile_continue; 1385 1386 // If no method data exists, go to profile_continue. 1387 test_method_data_pointer(mdp, profile_continue); 1388 1389 int count_offset = in_bytes(CounterData::count_offset()); 1390 // Back up the address, since we have already bumped the mdp. 1391 count_offset -= in_bytes(VirtualCallData::virtual_call_data_size()); 1392 1393 // *Decrement* the counter. We expect to see zero or small negatives. 1394 increment_mdp_data_at(mdp, count_offset, true); 1395 1396 bind (profile_continue); 1397 } 1398 } 1399 1400 void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) { 1401 if (ProfileInterpreter) { 1402 Label profile_continue; 1403 1404 // If no method data exists, go to profile_continue. 1405 test_method_data_pointer(mdp, profile_continue); 1406 1407 // The method data pointer needs to be updated. 1408 int mdp_delta = in_bytes(BitData::bit_data_size()); 1409 if (TypeProfileCasts) { 1410 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size()); 1411 1412 // Record the object type. 1413 record_klass_in_profile(klass, mdp, reg2, false); 1414 } 1415 update_mdp_by_constant(mdp, mdp_delta); 1416 1417 bind(profile_continue); 1418 } 1419 } 1420 1421 void InterpreterMacroAssembler::profile_switch_default(Register mdp) { 1422 if (ProfileInterpreter) { 1423 Label profile_continue; 1424 1425 // If no method data exists, go to profile_continue. 1426 test_method_data_pointer(mdp, profile_continue); 1427 1428 // Update the default case count 1429 increment_mdp_data_at(mdp, 1430 in_bytes(MultiBranchData::default_count_offset())); 1431 1432 // The method data pointer needs to be updated. 1433 update_mdp_by_offset(mdp, 1434 in_bytes(MultiBranchData:: 1435 default_displacement_offset())); 1436 1437 bind(profile_continue); 1438 } 1439 } 1440 1441 void InterpreterMacroAssembler::profile_switch_case(Register index, 1442 Register mdp, 1443 Register reg2) { 1444 if (ProfileInterpreter) { 1445 Label profile_continue; 1446 1447 // If no method data exists, go to profile_continue. 1448 test_method_data_pointer(mdp, profile_continue); 1449 1450 // Build the base (index * per_case_size_in_bytes()) + 1451 // case_array_offset_in_bytes() 1452 movw(reg2, in_bytes(MultiBranchData::per_case_size())); 1453 movw(rscratch1, in_bytes(MultiBranchData::case_array_offset())); 1454 Assembler::maddw(index, index, reg2, rscratch1); 1455 1456 // Update the case count 1457 increment_mdp_data_at(mdp, 1458 index, 1459 in_bytes(MultiBranchData::relative_count_offset())); 1460 1461 // The method data pointer needs to be updated. 1462 update_mdp_by_offset(mdp, 1463 index, 1464 in_bytes(MultiBranchData:: 1465 relative_displacement_offset())); 1466 1467 bind(profile_continue); 1468 } 1469 } 1470 1471 void InterpreterMacroAssembler::verify_oop(Register reg, TosState state) { 1472 if (state == atos) { 1473 MacroAssembler::verify_oop(reg); 1474 } 1475 } 1476 1477 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) { ; } 1478 1479 1480 void InterpreterMacroAssembler::notify_method_entry() { 1481 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to 1482 // track stack depth. If it is possible to enter interp_only_mode we add 1483 // the code to check if the event should be sent. 1484 if (JvmtiExport::can_post_interpreter_events()) { 1485 Label L; 1486 ldrw(r3, Address(rthread, JavaThread::interp_only_mode_offset())); 1487 cbzw(r3, L); 1488 call_VM(noreg, CAST_FROM_FN_PTR(address, 1489 InterpreterRuntime::post_method_entry)); 1490 bind(L); 1491 } 1492 1493 { 1494 SkipIfEqual skip(this, &DTraceMethodProbes, false); 1495 get_method(c_rarg1); 1496 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry), 1497 rthread, c_rarg1); 1498 } 1499 1500 // RedefineClasses() tracing support for obsolete method entry 1501 if (log_is_enabled(Trace, redefine, class, obsolete)) { 1502 get_method(c_rarg1); 1503 call_VM_leaf( 1504 CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry), 1505 rthread, c_rarg1); 1506 } 1507 1508 } 1509 1510 1511 void InterpreterMacroAssembler::notify_method_exit( 1512 TosState state, NotifyMethodExitMode mode) { 1513 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to 1514 // track stack depth. If it is possible to enter interp_only_mode we add 1515 // the code to check if the event should be sent. 1516 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) { 1517 Label L; 1518 // Note: frame::interpreter_frame_result has a dependency on how the 1519 // method result is saved across the call to post_method_exit. If this 1520 // is changed then the interpreter_frame_result implementation will 1521 // need to be updated too. 1522 1523 // template interpreter will leave the result on the top of the stack. 1524 push(state); 1525 ldrw(r3, Address(rthread, JavaThread::interp_only_mode_offset())); 1526 cbz(r3, L); 1527 call_VM(noreg, 1528 CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit)); 1529 bind(L); 1530 pop(state); 1531 } 1532 1533 { 1534 SkipIfEqual skip(this, &DTraceMethodProbes, false); 1535 push(state); 1536 get_method(c_rarg1); 1537 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit), 1538 rthread, c_rarg1); 1539 pop(state); 1540 } 1541 } 1542 1543 1544 // Jump if ((*counter_addr += increment) & mask) satisfies the condition. 1545 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr, 1546 int increment, Address mask, 1547 Register scratch, Register scratch2, 1548 bool preloaded, Condition cond, 1549 Label* where) { 1550 if (!preloaded) { 1551 ldrw(scratch, counter_addr); 1552 } 1553 add(scratch, scratch, increment); 1554 strw(scratch, counter_addr); 1555 ldrw(scratch2, mask); 1556 ands(scratch, scratch, scratch2); 1557 br(cond, *where); 1558 } 1559 1560 void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point, 1561 int number_of_arguments) { 1562 // interpreter specific 1563 // 1564 // Note: No need to save/restore rbcp & rlocals pointer since these 1565 // are callee saved registers and no blocking/ GC can happen 1566 // in leaf calls. 1567 #ifdef ASSERT 1568 { 1569 Label L; 1570 ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize)); 1571 cbz(rscratch1, L); 1572 stop("InterpreterMacroAssembler::call_VM_leaf_base:" 1573 " last_sp != NULL"); 1574 bind(L); 1575 } 1576 #endif /* ASSERT */ 1577 // super call 1578 MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments); 1579 } 1580 1581 void InterpreterMacroAssembler::call_VM_base(Register oop_result, 1582 Register java_thread, 1583 Register last_java_sp, 1584 address entry_point, 1585 int number_of_arguments, 1586 bool check_exceptions) { 1587 // interpreter specific 1588 // 1589 // Note: Could avoid restoring locals ptr (callee saved) - however doesn't 1590 // really make a difference for these runtime calls, since they are 1591 // slow anyway. Btw., bcp must be saved/restored since it may change 1592 // due to GC. 1593 // assert(java_thread == noreg , "not expecting a precomputed java thread"); 1594 save_bcp(); 1595 #ifdef ASSERT 1596 { 1597 Label L; 1598 ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize)); 1599 cbz(rscratch1, L); 1600 stop("InterpreterMacroAssembler::call_VM_leaf_base:" 1601 " last_sp != NULL"); 1602 bind(L); 1603 } 1604 #endif /* ASSERT */ 1605 // super call 1606 MacroAssembler::call_VM_base(oop_result, noreg, last_java_sp, 1607 entry_point, number_of_arguments, 1608 check_exceptions); 1609 // interpreter specific 1610 restore_bcp(); 1611 restore_locals(); 1612 } 1613 1614 void InterpreterMacroAssembler::profile_obj_type(Register obj, const Address& mdo_addr) { 1615 assert_different_registers(obj, rscratch1); 1616 Label update, next, none; 1617 1618 verify_oop(obj); 1619 1620 cbnz(obj, update); 1621 orptr(mdo_addr, TypeEntries::null_seen); 1622 b(next); 1623 1624 bind(update); 1625 load_klass(obj, obj); 1626 1627 ldr(rscratch1, mdo_addr); 1628 eor(obj, obj, rscratch1); 1629 tst(obj, TypeEntries::type_klass_mask); 1630 br(Assembler::EQ, next); // klass seen before, nothing to 1631 // do. The unknown bit may have been 1632 // set already but no need to check. 1633 1634 tbnz(obj, exact_log2(TypeEntries::type_unknown), next); 1635 // already unknown. Nothing to do anymore. 1636 1637 ldr(rscratch1, mdo_addr); 1638 cbz(rscratch1, none); 1639 cmp(rscratch1, (u1)TypeEntries::null_seen); 1640 br(Assembler::EQ, none); 1641 // There is a chance that the checks above (re-reading profiling 1642 // data from memory) fail if another thread has just set the 1643 // profiling to this obj's klass 1644 ldr(rscratch1, mdo_addr); 1645 eor(obj, obj, rscratch1); 1646 tst(obj, TypeEntries::type_klass_mask); 1647 br(Assembler::EQ, next); 1648 1649 // different than before. Cannot keep accurate profile. 1650 orptr(mdo_addr, TypeEntries::type_unknown); 1651 b(next); 1652 1653 bind(none); 1654 // first time here. Set profile type. 1655 str(obj, mdo_addr); 1656 1657 bind(next); 1658 } 1659 1660 void InterpreterMacroAssembler::profile_arguments_type(Register mdp, Register callee, Register tmp, bool is_virtual) { 1661 if (!ProfileInterpreter) { 1662 return; 1663 } 1664 1665 if (MethodData::profile_arguments() || MethodData::profile_return()) { 1666 Label profile_continue; 1667 1668 test_method_data_pointer(mdp, profile_continue); 1669 1670 int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size()); 1671 1672 ldrb(rscratch1, Address(mdp, in_bytes(DataLayout::tag_offset()) - off_to_start)); 1673 cmp(rscratch1, u1(is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag)); 1674 br(Assembler::NE, profile_continue); 1675 1676 if (MethodData::profile_arguments()) { 1677 Label done; 1678 int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset()); 1679 1680 for (int i = 0; i < TypeProfileArgsLimit; i++) { 1681 if (i > 0 || MethodData::profile_return()) { 1682 // If return value type is profiled we may have no argument to profile 1683 ldr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset()))); 1684 sub(tmp, tmp, i*TypeStackSlotEntries::per_arg_count()); 1685 cmp(tmp, (u1)TypeStackSlotEntries::per_arg_count()); 1686 add(rscratch1, mdp, off_to_args); 1687 br(Assembler::LT, done); 1688 } 1689 ldr(tmp, Address(callee, Method::const_offset())); 1690 load_unsigned_short(tmp, Address(tmp, ConstMethod::size_of_parameters_offset())); 1691 // stack offset o (zero based) from the start of the argument 1692 // list, for n arguments translates into offset n - o - 1 from 1693 // the end of the argument list 1694 ldr(rscratch1, Address(mdp, in_bytes(TypeEntriesAtCall::stack_slot_offset(i)))); 1695 sub(tmp, tmp, rscratch1); 1696 sub(tmp, tmp, 1); 1697 Address arg_addr = argument_address(tmp); 1698 ldr(tmp, arg_addr); 1699 1700 Address mdo_arg_addr(mdp, in_bytes(TypeEntriesAtCall::argument_type_offset(i))); 1701 profile_obj_type(tmp, mdo_arg_addr); 1702 1703 int to_add = in_bytes(TypeStackSlotEntries::per_arg_size()); 1704 off_to_args += to_add; 1705 } 1706 1707 if (MethodData::profile_return()) { 1708 ldr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset()))); 1709 sub(tmp, tmp, TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count()); 1710 } 1711 1712 add(rscratch1, mdp, off_to_args); 1713 bind(done); 1714 mov(mdp, rscratch1); 1715 1716 if (MethodData::profile_return()) { 1717 // We're right after the type profile for the last 1718 // argument. tmp is the number of cells left in the 1719 // CallTypeData/VirtualCallTypeData to reach its end. Non null 1720 // if there's a return to profile. 1721 assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type"); 1722 add(mdp, mdp, tmp, LSL, exact_log2(DataLayout::cell_size)); 1723 } 1724 str(mdp, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize)); 1725 } else { 1726 assert(MethodData::profile_return(), "either profile call args or call ret"); 1727 update_mdp_by_constant(mdp, in_bytes(TypeEntriesAtCall::return_only_size())); 1728 } 1729 1730 // mdp points right after the end of the 1731 // CallTypeData/VirtualCallTypeData, right after the cells for the 1732 // return value type if there's one 1733 1734 bind(profile_continue); 1735 } 1736 } 1737 1738 void InterpreterMacroAssembler::profile_return_type(Register mdp, Register ret, Register tmp) { 1739 assert_different_registers(mdp, ret, tmp, rbcp); 1740 if (ProfileInterpreter && MethodData::profile_return()) { 1741 Label profile_continue, done; 1742 1743 test_method_data_pointer(mdp, profile_continue); 1744 1745 if (MethodData::profile_return_jsr292_only()) { 1746 assert(Method::intrinsic_id_size_in_bytes() == 2, "assuming Method::_intrinsic_id is u2"); 1747 1748 // If we don't profile all invoke bytecodes we must make sure 1749 // it's a bytecode we indeed profile. We can't go back to the 1750 // begining of the ProfileData we intend to update to check its 1751 // type because we're right after it and we don't known its 1752 // length 1753 Label do_profile; 1754 ldrb(rscratch1, Address(rbcp, 0)); 1755 cmp(rscratch1, (u1)Bytecodes::_invokedynamic); 1756 br(Assembler::EQ, do_profile); 1757 cmp(rscratch1, (u1)Bytecodes::_invokehandle); 1758 br(Assembler::EQ, do_profile); 1759 get_method(tmp); 1760 ldrh(rscratch1, Address(tmp, Method::intrinsic_id_offset_in_bytes())); 1761 subs(zr, rscratch1, vmIntrinsics::_compiledLambdaForm); 1762 br(Assembler::NE, profile_continue); 1763 1764 bind(do_profile); 1765 } 1766 1767 Address mdo_ret_addr(mdp, -in_bytes(ReturnTypeEntry::size())); 1768 mov(tmp, ret); 1769 profile_obj_type(tmp, mdo_ret_addr); 1770 1771 bind(profile_continue); 1772 } 1773 } 1774 1775 void InterpreterMacroAssembler::profile_parameters_type(Register mdp, Register tmp1, Register tmp2) { 1776 assert_different_registers(rscratch1, rscratch2, mdp, tmp1, tmp2); 1777 if (ProfileInterpreter && MethodData::profile_parameters()) { 1778 Label profile_continue, done; 1779 1780 test_method_data_pointer(mdp, profile_continue); 1781 1782 // Load the offset of the area within the MDO used for 1783 // parameters. If it's negative we're not profiling any parameters 1784 ldrw(tmp1, Address(mdp, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset()))); 1785 tbnz(tmp1, 31, profile_continue); // i.e. sign bit set 1786 1787 // Compute a pointer to the area for parameters from the offset 1788 // and move the pointer to the slot for the last 1789 // parameters. Collect profiling from last parameter down. 1790 // mdo start + parameters offset + array length - 1 1791 add(mdp, mdp, tmp1); 1792 ldr(tmp1, Address(mdp, ArrayData::array_len_offset())); 1793 sub(tmp1, tmp1, TypeStackSlotEntries::per_arg_count()); 1794 1795 Label loop; 1796 bind(loop); 1797 1798 int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0)); 1799 int type_base = in_bytes(ParametersTypeData::type_offset(0)); 1800 int per_arg_scale = exact_log2(DataLayout::cell_size); 1801 add(rscratch1, mdp, off_base); 1802 add(rscratch2, mdp, type_base); 1803 1804 Address arg_off(rscratch1, tmp1, Address::lsl(per_arg_scale)); 1805 Address arg_type(rscratch2, tmp1, Address::lsl(per_arg_scale)); 1806 1807 // load offset on the stack from the slot for this parameter 1808 ldr(tmp2, arg_off); 1809 neg(tmp2, tmp2); 1810 // read the parameter from the local area 1811 ldr(tmp2, Address(rlocals, tmp2, Address::lsl(Interpreter::logStackElementSize))); 1812 1813 // profile the parameter 1814 profile_obj_type(tmp2, arg_type); 1815 1816 // go to next parameter 1817 subs(tmp1, tmp1, TypeStackSlotEntries::per_arg_count()); 1818 br(Assembler::GE, loop); 1819 1820 bind(profile_continue); 1821 } 1822 }