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 "oops/valueKlass.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 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 unsigned long 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 // Generate a subtype check: branch to ok_is_subtype if sub_klass is a 293 // subtype of super_klass. 294 // 295 // Args: 296 // r0: superklass 297 // Rsub_klass: subklass 298 // 299 // Kills: 300 // r2, r5 301 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass, 302 Label& ok_is_subtype) { 303 assert(Rsub_klass != r0, "r0 holds superklass"); 304 assert(Rsub_klass != r2, "r2 holds 2ndary super array length"); 305 assert(Rsub_klass != r5, "r5 holds 2ndary super array scan ptr"); 306 307 // Profile the not-null value's klass. 308 profile_typecheck(r2, Rsub_klass, r5); // blows r2, reloads r5 309 310 // Do the check. 311 check_klass_subtype(Rsub_klass, r0, r2, ok_is_subtype); // blows r2 312 313 // Profile the failure of the check. 314 profile_typecheck_failed(r2); // blows r2 315 } 316 317 // Java Expression Stack 318 319 void InterpreterMacroAssembler::pop_ptr(Register r) { 320 ldr(r, post(esp, wordSize)); 321 } 322 323 void InterpreterMacroAssembler::pop_i(Register r) { 324 ldrw(r, post(esp, wordSize)); 325 } 326 327 void InterpreterMacroAssembler::pop_l(Register r) { 328 ldr(r, post(esp, 2 * Interpreter::stackElementSize)); 329 } 330 331 void InterpreterMacroAssembler::push_ptr(Register r) { 332 str(r, pre(esp, -wordSize)); 333 } 334 335 void InterpreterMacroAssembler::push_i(Register r) { 336 str(r, pre(esp, -wordSize)); 337 } 338 339 void InterpreterMacroAssembler::push_l(Register r) { 340 str(zr, pre(esp, -wordSize)); 341 str(r, pre(esp, - wordSize)); 342 } 343 344 void InterpreterMacroAssembler::pop_f(FloatRegister r) { 345 ldrs(r, post(esp, wordSize)); 346 } 347 348 void InterpreterMacroAssembler::pop_d(FloatRegister r) { 349 ldrd(r, post(esp, 2 * Interpreter::stackElementSize)); 350 } 351 352 void InterpreterMacroAssembler::push_f(FloatRegister r) { 353 strs(r, pre(esp, -wordSize)); 354 } 355 356 void InterpreterMacroAssembler::push_d(FloatRegister r) { 357 strd(r, pre(esp, 2* -wordSize)); 358 } 359 360 void InterpreterMacroAssembler::pop(TosState state) { 361 switch (state) { 362 case atos: pop_ptr(); break; 363 case btos: 364 case ztos: 365 case ctos: 366 case stos: 367 case itos: pop_i(); break; 368 case ltos: pop_l(); break; 369 case ftos: pop_f(); break; 370 case dtos: pop_d(); break; 371 case vtos: /* nothing to do */ break; 372 default: ShouldNotReachHere(); 373 } 374 verify_oop(r0, state); 375 } 376 377 void InterpreterMacroAssembler::push(TosState state) { 378 verify_oop(r0, state); 379 switch (state) { 380 case atos: push_ptr(); break; 381 case btos: 382 case ztos: 383 case ctos: 384 case stos: 385 case itos: push_i(); break; 386 case ltos: push_l(); break; 387 case ftos: push_f(); break; 388 case dtos: push_d(); break; 389 case vtos: /* nothing to do */ break; 390 default : ShouldNotReachHere(); 391 } 392 } 393 394 // Helpers for swap and dup 395 void InterpreterMacroAssembler::load_ptr(int n, Register val) { 396 ldr(val, Address(esp, Interpreter::expr_offset_in_bytes(n))); 397 } 398 399 void InterpreterMacroAssembler::store_ptr(int n, Register val) { 400 str(val, Address(esp, Interpreter::expr_offset_in_bytes(n))); 401 } 402 403 void InterpreterMacroAssembler::load_float(Address src) { 404 ldrs(v0, src); 405 } 406 407 void InterpreterMacroAssembler::load_double(Address src) { 408 ldrd(v0, src); 409 } 410 411 void InterpreterMacroAssembler::prepare_to_jump_from_interpreted() { 412 // set sender sp 413 mov(r13, sp); 414 // record last_sp 415 str(esp, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize)); 416 } 417 418 // Jump to from_interpreted entry of a call unless single stepping is possible 419 // in this thread in which case we must call the i2i entry 420 void InterpreterMacroAssembler::jump_from_interpreted(Register method, Register temp) { 421 prepare_to_jump_from_interpreted(); 422 423 if (JvmtiExport::can_post_interpreter_events()) { 424 Label run_compiled_code; 425 // JVMTI events, such as single-stepping, are implemented partly by avoiding running 426 // compiled code in threads for which the event is enabled. Check here for 427 // interp_only_mode if these events CAN be enabled. 428 ldrw(rscratch1, Address(rthread, JavaThread::interp_only_mode_offset())); 429 cbzw(rscratch1, run_compiled_code); 430 ldr(rscratch1, Address(method, Method::interpreter_entry_offset())); 431 br(rscratch1); 432 bind(run_compiled_code); 433 } 434 435 ldr(rscratch1, Address(method, Method::from_interpreted_offset())); 436 br(rscratch1); 437 } 438 439 // The following two routines provide a hook so that an implementation 440 // can schedule the dispatch in two parts. amd64 does not do this. 441 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int step) { 442 } 443 444 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int step) { 445 dispatch_next(state, step); 446 } 447 448 void InterpreterMacroAssembler::dispatch_base(TosState state, 449 address* table, 450 bool verifyoop, 451 bool generate_poll) { 452 if (VerifyActivationFrameSize) { 453 Unimplemented(); 454 } 455 if (verifyoop) { 456 verify_oop(r0, state); 457 } 458 459 Label safepoint; 460 address* const safepoint_table = Interpreter::safept_table(state); 461 bool needs_thread_local_poll = generate_poll && 462 SafepointMechanism::uses_thread_local_poll() && table != safepoint_table; 463 464 if (needs_thread_local_poll) { 465 NOT_PRODUCT(block_comment("Thread-local Safepoint poll")); 466 ldr(rscratch2, Address(rthread, Thread::polling_page_offset())); 467 tbnz(rscratch2, exact_log2(SafepointMechanism::poll_bit()), safepoint); 468 } 469 470 if (table == Interpreter::dispatch_table(state)) { 471 addw(rscratch2, rscratch1, Interpreter::distance_from_dispatch_table(state)); 472 ldr(rscratch2, Address(rdispatch, rscratch2, Address::uxtw(3))); 473 } else { 474 mov(rscratch2, (address)table); 475 ldr(rscratch2, Address(rscratch2, rscratch1, Address::uxtw(3))); 476 } 477 br(rscratch2); 478 479 if (needs_thread_local_poll) { 480 bind(safepoint); 481 lea(rscratch2, ExternalAddress((address)safepoint_table)); 482 ldr(rscratch2, Address(rscratch2, rscratch1, Address::uxtw(3))); 483 br(rscratch2); 484 } 485 } 486 487 void InterpreterMacroAssembler::dispatch_only(TosState state, bool generate_poll) { 488 dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll); 489 } 490 491 void InterpreterMacroAssembler::dispatch_only_normal(TosState state) { 492 dispatch_base(state, Interpreter::normal_table(state)); 493 } 494 495 void InterpreterMacroAssembler::dispatch_only_noverify(TosState state) { 496 dispatch_base(state, Interpreter::normal_table(state), false); 497 } 498 499 500 void InterpreterMacroAssembler::dispatch_next(TosState state, int step, bool generate_poll) { 501 // load next bytecode 502 ldrb(rscratch1, Address(pre(rbcp, step))); 503 dispatch_base(state, Interpreter::dispatch_table(state), generate_poll); 504 } 505 506 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) { 507 // load current bytecode 508 ldrb(rscratch1, Address(rbcp, 0)); 509 dispatch_base(state, table); 510 } 511 512 // remove activation 513 // 514 // Unlock the receiver if this is a synchronized method. 515 // Unlock any Java monitors from syncronized blocks. 516 // Remove the activation from the stack. 517 // 518 // If there are locked Java monitors 519 // If throw_monitor_exception 520 // throws IllegalMonitorStateException 521 // Else if install_monitor_exception 522 // installs IllegalMonitorStateException 523 // Else 524 // no error processing 525 void InterpreterMacroAssembler::remove_activation( 526 TosState state, 527 bool throw_monitor_exception, 528 bool install_monitor_exception, 529 bool notify_jvmdi) { 530 // Note: Registers r3 xmm0 may be in use for the 531 // result check if synchronized method 532 Label unlocked, unlock, no_unlock; 533 534 // get the value of _do_not_unlock_if_synchronized into r3 535 const Address do_not_unlock_if_synchronized(rthread, 536 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset())); 537 ldrb(r3, do_not_unlock_if_synchronized); 538 strb(zr, do_not_unlock_if_synchronized); // reset the flag 539 540 // get method access flags 541 ldr(r1, Address(rfp, frame::interpreter_frame_method_offset * wordSize)); 542 ldr(r2, Address(r1, Method::access_flags_offset())); 543 tbz(r2, exact_log2(JVM_ACC_SYNCHRONIZED), unlocked); 544 545 // Don't unlock anything if the _do_not_unlock_if_synchronized flag 546 // is set. 547 cbnz(r3, no_unlock); 548 549 // unlock monitor 550 push(state); // save result 551 552 // BasicObjectLock will be first in list, since this is a 553 // synchronized method. However, need to check that the object has 554 // not been unlocked by an explicit monitorexit bytecode. 555 const Address monitor(rfp, frame::interpreter_frame_initial_sp_offset * 556 wordSize - (int) sizeof(BasicObjectLock)); 557 // We use c_rarg1 so that if we go slow path it will be the correct 558 // register for unlock_object to pass to VM directly 559 lea(c_rarg1, monitor); // address of first monitor 560 561 ldr(r0, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes())); 562 cbnz(r0, unlock); 563 564 pop(state); 565 if (throw_monitor_exception) { 566 // Entry already unlocked, need to throw exception 567 call_VM(noreg, CAST_FROM_FN_PTR(address, 568 InterpreterRuntime::throw_illegal_monitor_state_exception)); 569 should_not_reach_here(); 570 } else { 571 // Monitor already unlocked during a stack unroll. If requested, 572 // install an illegal_monitor_state_exception. Continue with 573 // stack unrolling. 574 if (install_monitor_exception) { 575 call_VM(noreg, CAST_FROM_FN_PTR(address, 576 InterpreterRuntime::new_illegal_monitor_state_exception)); 577 } 578 b(unlocked); 579 } 580 581 bind(unlock); 582 unlock_object(c_rarg1); 583 pop(state); 584 585 // Check that for block-structured locking (i.e., that all locked 586 // objects has been unlocked) 587 bind(unlocked); 588 589 // r0: Might contain return value 590 591 // Check that all monitors are unlocked 592 { 593 Label loop, exception, entry, restart; 594 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize; 595 const Address monitor_block_top( 596 rfp, frame::interpreter_frame_monitor_block_top_offset * wordSize); 597 const Address monitor_block_bot( 598 rfp, frame::interpreter_frame_initial_sp_offset * wordSize); 599 600 bind(restart); 601 // We use c_rarg1 so that if we go slow path it will be the correct 602 // register for unlock_object to pass to VM directly 603 ldr(c_rarg1, monitor_block_top); // points to current entry, starting 604 // with top-most entry 605 lea(r19, monitor_block_bot); // points to word before bottom of 606 // monitor block 607 b(entry); 608 609 // Entry already locked, need to throw exception 610 bind(exception); 611 612 if (throw_monitor_exception) { 613 // Throw exception 614 MacroAssembler::call_VM(noreg, 615 CAST_FROM_FN_PTR(address, InterpreterRuntime:: 616 throw_illegal_monitor_state_exception)); 617 should_not_reach_here(); 618 } else { 619 // Stack unrolling. Unlock object and install illegal_monitor_exception. 620 // Unlock does not block, so don't have to worry about the frame. 621 // We don't have to preserve c_rarg1 since we are going to throw an exception. 622 623 push(state); 624 unlock_object(c_rarg1); 625 pop(state); 626 627 if (install_monitor_exception) { 628 call_VM(noreg, CAST_FROM_FN_PTR(address, 629 InterpreterRuntime:: 630 new_illegal_monitor_state_exception)); 631 } 632 633 b(restart); 634 } 635 636 bind(loop); 637 // check if current entry is used 638 ldr(rscratch1, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes())); 639 cbnz(rscratch1, exception); 640 641 add(c_rarg1, c_rarg1, entry_size); // otherwise advance to next entry 642 bind(entry); 643 cmp(c_rarg1, r19); // check if bottom reached 644 br(Assembler::NE, loop); // if not at bottom then check this entry 645 } 646 647 bind(no_unlock); 648 649 // jvmti support 650 if (notify_jvmdi) { 651 notify_method_exit(state, NotifyJVMTI); // preserve TOSCA 652 } else { 653 notify_method_exit(state, SkipNotifyJVMTI); // preserve TOSCA 654 } 655 656 // remove activation 657 // get sender esp 658 ldr(esp, 659 Address(rfp, frame::interpreter_frame_sender_sp_offset * wordSize)); 660 if (StackReservedPages > 0) { 661 // testing if reserved zone needs to be re-enabled 662 Label no_reserved_zone_enabling; 663 664 ldr(rscratch1, Address(rthread, JavaThread::reserved_stack_activation_offset())); 665 cmp(esp, rscratch1); 666 br(Assembler::LS, no_reserved_zone_enabling); 667 668 call_VM_leaf( 669 CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), rthread); 670 call_VM(noreg, CAST_FROM_FN_PTR(address, 671 InterpreterRuntime::throw_delayed_StackOverflowError)); 672 should_not_reach_here(); 673 674 bind(no_reserved_zone_enabling); 675 } 676 677 // DMS CHECK: ValueTypeReturnedAsFields support should be here 678 // remove frame anchor 679 leave(); 680 // If we're returning to interpreted code we will shortly be 681 // adjusting SP to allow some space for ESP. If we're returning to 682 // compiled code the saved sender SP was saved in sender_sp, so this 683 // restores it. 684 andr(sp, esp, -16); 685 } 686 687 // Lock object 688 // 689 // Args: 690 // c_rarg1: BasicObjectLock to be used for locking 691 // 692 // Kills: 693 // r0 694 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, .. (param regs) 695 // rscratch1, rscratch2 (scratch regs) 696 void InterpreterMacroAssembler::lock_object(Register lock_reg) 697 { 698 assert(lock_reg == c_rarg1, "The argument is only for looks. It must be c_rarg1"); 699 if (UseHeavyMonitors) { 700 call_VM(noreg, 701 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), 702 lock_reg); 703 } else { 704 Label done; 705 706 const Register swap_reg = r0; 707 const Register tmp = c_rarg2; 708 const Register obj_reg = c_rarg3; // Will contain the oop 709 710 const int obj_offset = BasicObjectLock::obj_offset_in_bytes(); 711 const int lock_offset = BasicObjectLock::lock_offset_in_bytes (); 712 const int mark_offset = lock_offset + 713 BasicLock::displaced_header_offset_in_bytes(); 714 715 Label slow_case; 716 717 // Load object pointer into obj_reg %c_rarg3 718 ldr(obj_reg, Address(lock_reg, obj_offset)); 719 720 if (UseBiasedLocking) { 721 biased_locking_enter(lock_reg, obj_reg, swap_reg, tmp, false, done, &slow_case); 722 } 723 724 // Load (object->mark() | 1) into swap_reg 725 ldr(rscratch1, Address(obj_reg, oopDesc::mark_offset_in_bytes())); 726 orr(swap_reg, rscratch1, 1); 727 728 // Save (object->mark() | 1) into BasicLock's displaced header 729 str(swap_reg, Address(lock_reg, mark_offset)); 730 731 if (EnableValhalla && !UseBiasedLocking) { // DMS CHECK 732 // For slow path is_always_locked, using biased, which is never natural for !UseBiasLocking 733 andr(swap_reg, swap_reg, ~markOopDesc::biased_lock_bit_in_place); 734 } 735 736 assert(lock_offset == 0, 737 "displached header must be first word in BasicObjectLock"); 738 739 Label fail; 740 if (PrintBiasedLockingStatistics) { 741 Label fast; 742 cmpxchg_obj_header(swap_reg, lock_reg, obj_reg, rscratch1, fast, &fail); 743 bind(fast); 744 atomic_incw(Address((address)BiasedLocking::fast_path_entry_count_addr()), 745 rscratch2, rscratch1, tmp); 746 b(done); 747 bind(fail); 748 } else { 749 cmpxchg_obj_header(swap_reg, lock_reg, obj_reg, rscratch1, done, /*fallthrough*/NULL); 750 } 751 752 // Test if the oopMark is an obvious stack pointer, i.e., 753 // 1) (mark & 7) == 0, and 754 // 2) rsp <= mark < mark + os::pagesize() 755 // 756 // These 3 tests can be done by evaluating the following 757 // expression: ((mark - rsp) & (7 - os::vm_page_size())), 758 // assuming both stack pointer and pagesize have their 759 // least significant 3 bits clear. 760 // NOTE: the oopMark is in swap_reg %r0 as the result of cmpxchg 761 // NOTE2: aarch64 does not like to subtract sp from rn so take a 762 // copy 763 mov(rscratch1, sp); 764 sub(swap_reg, swap_reg, rscratch1); 765 ands(swap_reg, swap_reg, (unsigned long)(7 - os::vm_page_size())); 766 767 // Save the test result, for recursive case, the result is zero 768 str(swap_reg, Address(lock_reg, mark_offset)); 769 770 if (PrintBiasedLockingStatistics) { 771 br(Assembler::NE, slow_case); 772 atomic_incw(Address((address)BiasedLocking::fast_path_entry_count_addr()), 773 rscratch2, rscratch1, tmp); 774 } 775 br(Assembler::EQ, done); 776 777 bind(slow_case); 778 779 // Call the runtime routine for slow case 780 call_VM(noreg, 781 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), 782 lock_reg); 783 784 bind(done); 785 } 786 } 787 788 789 // Unlocks an object. Used in monitorexit bytecode and 790 // remove_activation. Throws an IllegalMonitorException if object is 791 // not locked by current thread. 792 // 793 // Args: 794 // c_rarg1: BasicObjectLock for lock 795 // 796 // Kills: 797 // r0 798 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ... (param regs) 799 // rscratch1, rscratch2 (scratch regs) 800 void InterpreterMacroAssembler::unlock_object(Register lock_reg) 801 { 802 assert(lock_reg == c_rarg1, "The argument is only for looks. It must be rarg1"); 803 804 if (UseHeavyMonitors) { 805 call_VM(noreg, 806 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), 807 lock_reg); 808 } else { 809 Label done; 810 811 const Register swap_reg = r0; 812 const Register header_reg = c_rarg2; // Will contain the old oopMark 813 const Register obj_reg = c_rarg3; // Will contain the oop 814 815 save_bcp(); // Save in case of exception 816 817 // Convert from BasicObjectLock structure to object and BasicLock 818 // structure Store the BasicLock address into %r0 819 lea(swap_reg, Address(lock_reg, BasicObjectLock::lock_offset_in_bytes())); 820 821 // Load oop into obj_reg(%c_rarg3) 822 ldr(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset_in_bytes())); 823 824 // Free entry 825 str(zr, Address(lock_reg, BasicObjectLock::obj_offset_in_bytes())); 826 827 if (UseBiasedLocking) { 828 biased_locking_exit(obj_reg, header_reg, done); 829 } 830 831 // Load the old header from BasicLock structure 832 ldr(header_reg, Address(swap_reg, 833 BasicLock::displaced_header_offset_in_bytes())); 834 835 // Test for recursion 836 cbz(header_reg, done); 837 838 // Atomic swap back the old header 839 cmpxchg_obj_header(swap_reg, header_reg, obj_reg, rscratch1, done, /*fallthrough*/NULL); 840 841 // Call the runtime routine for slow case. 842 str(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset_in_bytes())); // restore obj 843 call_VM(noreg, 844 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), 845 lock_reg); 846 847 bind(done); 848 849 restore_bcp(); 850 } 851 } 852 853 void InterpreterMacroAssembler::test_method_data_pointer(Register mdp, 854 Label& zero_continue) { 855 assert(ProfileInterpreter, "must be profiling interpreter"); 856 ldr(mdp, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize)); 857 cbz(mdp, zero_continue); 858 } 859 860 // Set the method data pointer for the current bcp. 861 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() { 862 assert(ProfileInterpreter, "must be profiling interpreter"); 863 Label set_mdp; 864 stp(r0, r1, Address(pre(sp, -2 * wordSize))); 865 866 // Test MDO to avoid the call if it is NULL. 867 ldr(r0, Address(rmethod, in_bytes(Method::method_data_offset()))); 868 cbz(r0, set_mdp); 869 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), rmethod, rbcp); 870 // r0: mdi 871 // mdo is guaranteed to be non-zero here, we checked for it before the call. 872 ldr(r1, Address(rmethod, in_bytes(Method::method_data_offset()))); 873 lea(r1, Address(r1, in_bytes(MethodData::data_offset()))); 874 add(r0, r1, r0); 875 str(r0, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize)); 876 bind(set_mdp); 877 ldp(r0, r1, Address(post(sp, 2 * wordSize))); 878 } 879 880 void InterpreterMacroAssembler::verify_method_data_pointer() { 881 assert(ProfileInterpreter, "must be profiling interpreter"); 882 #ifdef ASSERT 883 Label verify_continue; 884 stp(r0, r1, Address(pre(sp, -2 * wordSize))); 885 stp(r2, r3, Address(pre(sp, -2 * wordSize))); 886 test_method_data_pointer(r3, verify_continue); // If mdp is zero, continue 887 get_method(r1); 888 889 // If the mdp is valid, it will point to a DataLayout header which is 890 // consistent with the bcp. The converse is highly probable also. 891 ldrsh(r2, Address(r3, in_bytes(DataLayout::bci_offset()))); 892 ldr(rscratch1, Address(r1, Method::const_offset())); 893 add(r2, r2, rscratch1, Assembler::LSL); 894 lea(r2, Address(r2, ConstMethod::codes_offset())); 895 cmp(r2, rbcp); 896 br(Assembler::EQ, verify_continue); 897 // r1: method 898 // rbcp: bcp // rbcp == 22 899 // r3: mdp 900 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp), 901 r1, rbcp, r3); 902 bind(verify_continue); 903 ldp(r2, r3, Address(post(sp, 2 * wordSize))); 904 ldp(r0, r1, Address(post(sp, 2 * wordSize))); 905 #endif // ASSERT 906 } 907 908 909 void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in, 910 int constant, 911 Register value) { 912 assert(ProfileInterpreter, "must be profiling interpreter"); 913 Address data(mdp_in, constant); 914 str(value, data); 915 } 916 917 918 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in, 919 int constant, 920 bool decrement) { 921 increment_mdp_data_at(mdp_in, noreg, constant, decrement); 922 } 923 924 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in, 925 Register reg, 926 int constant, 927 bool decrement) { 928 assert(ProfileInterpreter, "must be profiling interpreter"); 929 // %%% this does 64bit counters at best it is wasting space 930 // at worst it is a rare bug when counters overflow 931 932 assert_different_registers(rscratch2, rscratch1, mdp_in, reg); 933 934 Address addr1(mdp_in, constant); 935 Address addr2(rscratch2, reg, Address::lsl(0)); 936 Address &addr = addr1; 937 if (reg != noreg) { 938 lea(rscratch2, addr1); 939 addr = addr2; 940 } 941 942 if (decrement) { 943 // Decrement the register. Set condition codes. 944 // Intel does this 945 // addptr(data, (int32_t) -DataLayout::counter_increment); 946 // If the decrement causes the counter to overflow, stay negative 947 // Label L; 948 // jcc(Assembler::negative, L); 949 // addptr(data, (int32_t) DataLayout::counter_increment); 950 // so we do this 951 ldr(rscratch1, addr); 952 subs(rscratch1, rscratch1, (unsigned)DataLayout::counter_increment); 953 Label L; 954 br(Assembler::LO, L); // skip store if counter underflow 955 str(rscratch1, addr); 956 bind(L); 957 } else { 958 assert(DataLayout::counter_increment == 1, 959 "flow-free idiom only works with 1"); 960 // Intel does this 961 // Increment the register. Set carry flag. 962 // addptr(data, DataLayout::counter_increment); 963 // If the increment causes the counter to overflow, pull back by 1. 964 // sbbptr(data, (int32_t)0); 965 // so we do this 966 ldr(rscratch1, addr); 967 adds(rscratch1, rscratch1, DataLayout::counter_increment); 968 Label L; 969 br(Assembler::CS, L); // skip store if counter overflow 970 str(rscratch1, addr); 971 bind(L); 972 } 973 } 974 975 void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in, 976 int flag_byte_constant) { 977 assert(ProfileInterpreter, "must be profiling interpreter"); 978 int flags_offset = in_bytes(DataLayout::flags_offset()); 979 // Set the flag 980 ldrb(rscratch1, Address(mdp_in, flags_offset)); 981 orr(rscratch1, rscratch1, flag_byte_constant); 982 strb(rscratch1, Address(mdp_in, flags_offset)); 983 } 984 985 986 void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in, 987 int offset, 988 Register value, 989 Register test_value_out, 990 Label& not_equal_continue) { 991 assert(ProfileInterpreter, "must be profiling interpreter"); 992 if (test_value_out == noreg) { 993 ldr(rscratch1, Address(mdp_in, offset)); 994 cmp(value, rscratch1); 995 } else { 996 // Put the test value into a register, so caller can use it: 997 ldr(test_value_out, Address(mdp_in, offset)); 998 cmp(value, test_value_out); 999 } 1000 br(Assembler::NE, not_equal_continue); 1001 } 1002 1003 1004 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in, 1005 int offset_of_disp) { 1006 assert(ProfileInterpreter, "must be profiling interpreter"); 1007 ldr(rscratch1, Address(mdp_in, offset_of_disp)); 1008 add(mdp_in, mdp_in, rscratch1, LSL); 1009 str(mdp_in, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize)); 1010 } 1011 1012 1013 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in, 1014 Register reg, 1015 int offset_of_disp) { 1016 assert(ProfileInterpreter, "must be profiling interpreter"); 1017 lea(rscratch1, Address(mdp_in, offset_of_disp)); 1018 ldr(rscratch1, Address(rscratch1, reg, Address::lsl(0))); 1019 add(mdp_in, mdp_in, rscratch1, LSL); 1020 str(mdp_in, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize)); 1021 } 1022 1023 1024 void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in, 1025 int constant) { 1026 assert(ProfileInterpreter, "must be profiling interpreter"); 1027 add(mdp_in, mdp_in, (unsigned)constant); 1028 str(mdp_in, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize)); 1029 } 1030 1031 1032 void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) { 1033 assert(ProfileInterpreter, "must be profiling interpreter"); 1034 // save/restore across call_VM 1035 stp(zr, return_bci, Address(pre(sp, -2 * wordSize))); 1036 call_VM(noreg, 1037 CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret), 1038 return_bci); 1039 ldp(zr, return_bci, Address(post(sp, 2 * wordSize))); 1040 } 1041 1042 1043 void InterpreterMacroAssembler::profile_taken_branch(Register mdp, 1044 Register bumped_count) { 1045 if (ProfileInterpreter) { 1046 Label profile_continue; 1047 1048 // If no method data exists, go to profile_continue. 1049 // Otherwise, assign to mdp 1050 test_method_data_pointer(mdp, profile_continue); 1051 1052 // We are taking a branch. Increment the taken count. 1053 // We inline increment_mdp_data_at to return bumped_count in a register 1054 //increment_mdp_data_at(mdp, in_bytes(JumpData::taken_offset())); 1055 Address data(mdp, in_bytes(JumpData::taken_offset())); 1056 ldr(bumped_count, data); 1057 assert(DataLayout::counter_increment == 1, 1058 "flow-free idiom only works with 1"); 1059 // Intel does this to catch overflow 1060 // addptr(bumped_count, DataLayout::counter_increment); 1061 // sbbptr(bumped_count, 0); 1062 // so we do this 1063 adds(bumped_count, bumped_count, DataLayout::counter_increment); 1064 Label L; 1065 br(Assembler::CS, L); // skip store if counter overflow 1066 str(bumped_count, data); 1067 bind(L); 1068 // The method data pointer needs to be updated to reflect the new target. 1069 update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset())); 1070 bind(profile_continue); 1071 } 1072 } 1073 1074 1075 void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp) { 1076 if (ProfileInterpreter) { 1077 Label profile_continue; 1078 1079 // If no method data exists, go to profile_continue. 1080 test_method_data_pointer(mdp, profile_continue); 1081 1082 // We are taking a branch. Increment the not taken count. 1083 increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset())); 1084 1085 // The method data pointer needs to be updated to correspond to 1086 // the next bytecode 1087 update_mdp_by_constant(mdp, in_bytes(BranchData::branch_data_size())); 1088 bind(profile_continue); 1089 } 1090 } 1091 1092 1093 void InterpreterMacroAssembler::profile_call(Register mdp) { 1094 if (ProfileInterpreter) { 1095 Label profile_continue; 1096 1097 // If no method data exists, go to profile_continue. 1098 test_method_data_pointer(mdp, profile_continue); 1099 1100 // We are making a call. Increment the count. 1101 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); 1102 1103 // The method data pointer needs to be updated to reflect the new target. 1104 update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size())); 1105 bind(profile_continue); 1106 } 1107 } 1108 1109 void InterpreterMacroAssembler::profile_final_call(Register mdp) { 1110 if (ProfileInterpreter) { 1111 Label profile_continue; 1112 1113 // If no method data exists, go to profile_continue. 1114 test_method_data_pointer(mdp, profile_continue); 1115 1116 // We are making a call. Increment the count. 1117 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); 1118 1119 // The method data pointer needs to be updated to reflect the new target. 1120 update_mdp_by_constant(mdp, 1121 in_bytes(VirtualCallData:: 1122 virtual_call_data_size())); 1123 bind(profile_continue); 1124 } 1125 } 1126 1127 1128 void InterpreterMacroAssembler::profile_virtual_call(Register receiver, 1129 Register mdp, 1130 Register reg2, 1131 bool receiver_can_be_null) { 1132 if (ProfileInterpreter) { 1133 Label profile_continue; 1134 1135 // If no method data exists, go to profile_continue. 1136 test_method_data_pointer(mdp, profile_continue); 1137 1138 Label skip_receiver_profile; 1139 if (receiver_can_be_null) { 1140 Label not_null; 1141 // We are making a call. Increment the count for null receiver. 1142 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); 1143 b(skip_receiver_profile); 1144 bind(not_null); 1145 } 1146 1147 // Record the receiver type. 1148 record_klass_in_profile(receiver, mdp, reg2, true); 1149 bind(skip_receiver_profile); 1150 1151 // The method data pointer needs to be updated to reflect the new target. 1152 #if INCLUDE_JVMCI 1153 if (MethodProfileWidth == 0) { 1154 update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size())); 1155 } 1156 #else // INCLUDE_JVMCI 1157 update_mdp_by_constant(mdp, 1158 in_bytes(VirtualCallData:: 1159 virtual_call_data_size())); 1160 #endif // INCLUDE_JVMCI 1161 bind(profile_continue); 1162 } 1163 } 1164 1165 #if INCLUDE_JVMCI 1166 void InterpreterMacroAssembler::profile_called_method(Register method, Register mdp, Register reg2) { 1167 assert_different_registers(method, mdp, reg2); 1168 if (ProfileInterpreter && MethodProfileWidth > 0) { 1169 Label profile_continue; 1170 1171 // If no method data exists, go to profile_continue. 1172 test_method_data_pointer(mdp, profile_continue); 1173 1174 Label done; 1175 record_item_in_profile_helper(method, mdp, reg2, 0, done, MethodProfileWidth, 1176 &VirtualCallData::method_offset, &VirtualCallData::method_count_offset, in_bytes(VirtualCallData::nonprofiled_receiver_count_offset())); 1177 bind(done); 1178 1179 update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size())); 1180 bind(profile_continue); 1181 } 1182 } 1183 #endif // INCLUDE_JVMCI 1184 1185 // This routine creates a state machine for updating the multi-row 1186 // type profile at a virtual call site (or other type-sensitive bytecode). 1187 // The machine visits each row (of receiver/count) until the receiver type 1188 // is found, or until it runs out of rows. At the same time, it remembers 1189 // the location of the first empty row. (An empty row records null for its 1190 // receiver, and can be allocated for a newly-observed receiver type.) 1191 // Because there are two degrees of freedom in the state, a simple linear 1192 // search will not work; it must be a decision tree. Hence this helper 1193 // function is recursive, to generate the required tree structured code. 1194 // It's the interpreter, so we are trading off code space for speed. 1195 // See below for example code. 1196 void InterpreterMacroAssembler::record_klass_in_profile_helper( 1197 Register receiver, Register mdp, 1198 Register reg2, int start_row, 1199 Label& done, bool is_virtual_call) { 1200 if (TypeProfileWidth == 0) { 1201 if (is_virtual_call) { 1202 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); 1203 } 1204 #if INCLUDE_JVMCI 1205 else if (EnableJVMCI) { 1206 increment_mdp_data_at(mdp, in_bytes(ReceiverTypeData::nonprofiled_receiver_count_offset())); 1207 } 1208 #endif // INCLUDE_JVMCI 1209 } else { 1210 int non_profiled_offset = -1; 1211 if (is_virtual_call) { 1212 non_profiled_offset = in_bytes(CounterData::count_offset()); 1213 } 1214 #if INCLUDE_JVMCI 1215 else if (EnableJVMCI) { 1216 non_profiled_offset = in_bytes(ReceiverTypeData::nonprofiled_receiver_count_offset()); 1217 } 1218 #endif // INCLUDE_JVMCI 1219 1220 record_item_in_profile_helper(receiver, mdp, reg2, 0, done, TypeProfileWidth, 1221 &VirtualCallData::receiver_offset, &VirtualCallData::receiver_count_offset, non_profiled_offset); 1222 } 1223 } 1224 1225 void InterpreterMacroAssembler::record_item_in_profile_helper(Register item, Register mdp, 1226 Register reg2, int start_row, Label& done, int total_rows, 1227 OffsetFunction item_offset_fn, OffsetFunction item_count_offset_fn, 1228 int non_profiled_offset) { 1229 int last_row = total_rows - 1; 1230 assert(start_row <= last_row, "must be work left to do"); 1231 // Test this row for both the item and for null. 1232 // Take any of three different outcomes: 1233 // 1. found item => increment count and goto done 1234 // 2. found null => keep looking for case 1, maybe allocate this cell 1235 // 3. found something else => keep looking for cases 1 and 2 1236 // Case 3 is handled by a recursive call. 1237 for (int row = start_row; row <= last_row; row++) { 1238 Label next_test; 1239 bool test_for_null_also = (row == start_row); 1240 1241 // See if the item is item[n]. 1242 int item_offset = in_bytes(item_offset_fn(row)); 1243 test_mdp_data_at(mdp, item_offset, item, 1244 (test_for_null_also ? reg2 : noreg), 1245 next_test); 1246 // (Reg2 now contains the item from the CallData.) 1247 1248 // The item is item[n]. Increment count[n]. 1249 int count_offset = in_bytes(item_count_offset_fn(row)); 1250 increment_mdp_data_at(mdp, count_offset); 1251 b(done); 1252 bind(next_test); 1253 1254 if (test_for_null_also) { 1255 Label found_null; 1256 // Failed the equality check on item[n]... Test for null. 1257 if (start_row == last_row) { 1258 // The only thing left to do is handle the null case. 1259 if (non_profiled_offset >= 0) { 1260 cbz(reg2, found_null); 1261 // Item did not match any saved item and there is no empty row for it. 1262 // Increment total counter to indicate polymorphic case. 1263 increment_mdp_data_at(mdp, non_profiled_offset); 1264 b(done); 1265 bind(found_null); 1266 } else { 1267 cbnz(reg2, done); 1268 } 1269 break; 1270 } 1271 // Since null is rare, make it be the branch-taken case. 1272 cbz(reg2, found_null); 1273 1274 // Put all the "Case 3" tests here. 1275 record_item_in_profile_helper(item, mdp, reg2, start_row + 1, done, total_rows, 1276 item_offset_fn, item_count_offset_fn, non_profiled_offset); 1277 1278 // Found a null. Keep searching for a matching item, 1279 // but remember that this is an empty (unused) slot. 1280 bind(found_null); 1281 } 1282 } 1283 1284 // In the fall-through case, we found no matching item, but we 1285 // observed the item[start_row] is NULL. 1286 1287 // Fill in the item field and increment the count. 1288 int item_offset = in_bytes(item_offset_fn(start_row)); 1289 set_mdp_data_at(mdp, item_offset, item); 1290 int count_offset = in_bytes(item_count_offset_fn(start_row)); 1291 mov(reg2, DataLayout::counter_increment); 1292 set_mdp_data_at(mdp, count_offset, reg2); 1293 if (start_row > 0) { 1294 b(done); 1295 } 1296 } 1297 1298 // Example state machine code for three profile rows: 1299 // // main copy of decision tree, rooted at row[1] 1300 // if (row[0].rec == rec) { row[0].incr(); goto done; } 1301 // if (row[0].rec != NULL) { 1302 // // inner copy of decision tree, rooted at row[1] 1303 // if (row[1].rec == rec) { row[1].incr(); goto done; } 1304 // if (row[1].rec != NULL) { 1305 // // degenerate decision tree, rooted at row[2] 1306 // if (row[2].rec == rec) { row[2].incr(); goto done; } 1307 // if (row[2].rec != NULL) { count.incr(); goto done; } // overflow 1308 // row[2].init(rec); goto done; 1309 // } else { 1310 // // remember row[1] is empty 1311 // if (row[2].rec == rec) { row[2].incr(); goto done; } 1312 // row[1].init(rec); goto done; 1313 // } 1314 // } else { 1315 // // remember row[0] is empty 1316 // if (row[1].rec == rec) { row[1].incr(); goto done; } 1317 // if (row[2].rec == rec) { row[2].incr(); goto done; } 1318 // row[0].init(rec); goto done; 1319 // } 1320 // done: 1321 1322 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver, 1323 Register mdp, Register reg2, 1324 bool is_virtual_call) { 1325 assert(ProfileInterpreter, "must be profiling"); 1326 Label done; 1327 1328 record_klass_in_profile_helper(receiver, mdp, reg2, 0, done, is_virtual_call); 1329 1330 bind (done); 1331 } 1332 1333 void InterpreterMacroAssembler::profile_ret(Register return_bci, 1334 Register mdp) { 1335 if (ProfileInterpreter) { 1336 Label profile_continue; 1337 uint row; 1338 1339 // If no method data exists, go to profile_continue. 1340 test_method_data_pointer(mdp, profile_continue); 1341 1342 // Update the total ret count. 1343 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); 1344 1345 for (row = 0; row < RetData::row_limit(); row++) { 1346 Label next_test; 1347 1348 // See if return_bci is equal to bci[n]: 1349 test_mdp_data_at(mdp, 1350 in_bytes(RetData::bci_offset(row)), 1351 return_bci, noreg, 1352 next_test); 1353 1354 // return_bci is equal to bci[n]. Increment the count. 1355 increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row))); 1356 1357 // The method data pointer needs to be updated to reflect the new target. 1358 update_mdp_by_offset(mdp, 1359 in_bytes(RetData::bci_displacement_offset(row))); 1360 b(profile_continue); 1361 bind(next_test); 1362 } 1363 1364 update_mdp_for_ret(return_bci); 1365 1366 bind(profile_continue); 1367 } 1368 } 1369 1370 void InterpreterMacroAssembler::profile_null_seen(Register mdp) { 1371 if (ProfileInterpreter) { 1372 Label profile_continue; 1373 1374 // If no method data exists, go to profile_continue. 1375 test_method_data_pointer(mdp, profile_continue); 1376 1377 set_mdp_flag_at(mdp, BitData::null_seen_byte_constant()); 1378 1379 // The method data pointer needs to be updated. 1380 int mdp_delta = in_bytes(BitData::bit_data_size()); 1381 if (TypeProfileCasts) { 1382 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size()); 1383 } 1384 update_mdp_by_constant(mdp, mdp_delta); 1385 1386 bind(profile_continue); 1387 } 1388 } 1389 1390 void InterpreterMacroAssembler::profile_typecheck_failed(Register mdp) { 1391 if (ProfileInterpreter && TypeProfileCasts) { 1392 Label profile_continue; 1393 1394 // If no method data exists, go to profile_continue. 1395 test_method_data_pointer(mdp, profile_continue); 1396 1397 int count_offset = in_bytes(CounterData::count_offset()); 1398 // Back up the address, since we have already bumped the mdp. 1399 count_offset -= in_bytes(VirtualCallData::virtual_call_data_size()); 1400 1401 // *Decrement* the counter. We expect to see zero or small negatives. 1402 increment_mdp_data_at(mdp, count_offset, true); 1403 1404 bind (profile_continue); 1405 } 1406 } 1407 1408 void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) { 1409 if (ProfileInterpreter) { 1410 Label profile_continue; 1411 1412 // If no method data exists, go to profile_continue. 1413 test_method_data_pointer(mdp, profile_continue); 1414 1415 // The method data pointer needs to be updated. 1416 int mdp_delta = in_bytes(BitData::bit_data_size()); 1417 if (TypeProfileCasts) { 1418 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size()); 1419 1420 // Record the object type. 1421 record_klass_in_profile(klass, mdp, reg2, false); 1422 } 1423 update_mdp_by_constant(mdp, mdp_delta); 1424 1425 bind(profile_continue); 1426 } 1427 } 1428 1429 void InterpreterMacroAssembler::profile_switch_default(Register mdp) { 1430 if (ProfileInterpreter) { 1431 Label profile_continue; 1432 1433 // If no method data exists, go to profile_continue. 1434 test_method_data_pointer(mdp, profile_continue); 1435 1436 // Update the default case count 1437 increment_mdp_data_at(mdp, 1438 in_bytes(MultiBranchData::default_count_offset())); 1439 1440 // The method data pointer needs to be updated. 1441 update_mdp_by_offset(mdp, 1442 in_bytes(MultiBranchData:: 1443 default_displacement_offset())); 1444 1445 bind(profile_continue); 1446 } 1447 } 1448 1449 void InterpreterMacroAssembler::profile_switch_case(Register index, 1450 Register mdp, 1451 Register reg2) { 1452 if (ProfileInterpreter) { 1453 Label profile_continue; 1454 1455 // If no method data exists, go to profile_continue. 1456 test_method_data_pointer(mdp, profile_continue); 1457 1458 // Build the base (index * per_case_size_in_bytes()) + 1459 // case_array_offset_in_bytes() 1460 movw(reg2, in_bytes(MultiBranchData::per_case_size())); 1461 movw(rscratch1, in_bytes(MultiBranchData::case_array_offset())); 1462 Assembler::maddw(index, index, reg2, rscratch1); 1463 1464 // Update the case count 1465 increment_mdp_data_at(mdp, 1466 index, 1467 in_bytes(MultiBranchData::relative_count_offset())); 1468 1469 // The method data pointer needs to be updated. 1470 update_mdp_by_offset(mdp, 1471 index, 1472 in_bytes(MultiBranchData:: 1473 relative_displacement_offset())); 1474 1475 bind(profile_continue); 1476 } 1477 } 1478 1479 void InterpreterMacroAssembler::verify_oop(Register reg, TosState state) { 1480 if (state == atos) { 1481 MacroAssembler::verify_oop(reg); 1482 } 1483 } 1484 1485 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) { ; } 1486 1487 1488 void InterpreterMacroAssembler::notify_method_entry() { 1489 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to 1490 // track stack depth. If it is possible to enter interp_only_mode we add 1491 // the code to check if the event should be sent. 1492 if (JvmtiExport::can_post_interpreter_events()) { 1493 Label L; 1494 ldrw(r3, Address(rthread, JavaThread::interp_only_mode_offset())); 1495 cbzw(r3, L); 1496 call_VM(noreg, CAST_FROM_FN_PTR(address, 1497 InterpreterRuntime::post_method_entry)); 1498 bind(L); 1499 } 1500 1501 { 1502 SkipIfEqual skip(this, &DTraceMethodProbes, false); 1503 get_method(c_rarg1); 1504 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry), 1505 rthread, c_rarg1); 1506 } 1507 1508 // RedefineClasses() tracing support for obsolete method entry 1509 if (log_is_enabled(Trace, redefine, class, obsolete)) { 1510 get_method(c_rarg1); 1511 call_VM_leaf( 1512 CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry), 1513 rthread, c_rarg1); 1514 } 1515 1516 } 1517 1518 1519 void InterpreterMacroAssembler::notify_method_exit( 1520 TosState state, NotifyMethodExitMode mode) { 1521 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to 1522 // track stack depth. If it is possible to enter interp_only_mode we add 1523 // the code to check if the event should be sent. 1524 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) { 1525 Label L; 1526 // Note: frame::interpreter_frame_result has a dependency on how the 1527 // method result is saved across the call to post_method_exit. If this 1528 // is changed then the interpreter_frame_result implementation will 1529 // need to be updated too. 1530 1531 // template interpreter will leave the result on the top of the stack. 1532 push(state); 1533 ldrw(r3, Address(rthread, JavaThread::interp_only_mode_offset())); 1534 cbz(r3, L); 1535 call_VM(noreg, 1536 CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit)); 1537 bind(L); 1538 pop(state); 1539 } 1540 1541 { 1542 SkipIfEqual skip(this, &DTraceMethodProbes, false); 1543 push(state); 1544 get_method(c_rarg1); 1545 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit), 1546 rthread, c_rarg1); 1547 pop(state); 1548 } 1549 } 1550 1551 1552 // Jump if ((*counter_addr += increment) & mask) satisfies the condition. 1553 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr, 1554 int increment, Address mask, 1555 Register scratch, Register scratch2, 1556 bool preloaded, Condition cond, 1557 Label* where) { 1558 if (!preloaded) { 1559 ldrw(scratch, counter_addr); 1560 } 1561 add(scratch, scratch, increment); 1562 strw(scratch, counter_addr); 1563 ldrw(scratch2, mask); 1564 ands(scratch, scratch, scratch2); 1565 br(cond, *where); 1566 } 1567 1568 void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point, 1569 int number_of_arguments) { 1570 // interpreter specific 1571 // 1572 // Note: No need to save/restore rbcp & rlocals pointer since these 1573 // are callee saved registers and no blocking/ GC can happen 1574 // in leaf calls. 1575 #ifdef ASSERT 1576 { 1577 Label L; 1578 ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize)); 1579 cbz(rscratch1, L); 1580 stop("InterpreterMacroAssembler::call_VM_leaf_base:" 1581 " last_sp != NULL"); 1582 bind(L); 1583 } 1584 #endif /* ASSERT */ 1585 // super call 1586 MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments); 1587 } 1588 1589 void InterpreterMacroAssembler::call_VM_base(Register oop_result, 1590 Register java_thread, 1591 Register last_java_sp, 1592 address entry_point, 1593 int number_of_arguments, 1594 bool check_exceptions) { 1595 // interpreter specific 1596 // 1597 // Note: Could avoid restoring locals ptr (callee saved) - however doesn't 1598 // really make a difference for these runtime calls, since they are 1599 // slow anyway. Btw., bcp must be saved/restored since it may change 1600 // due to GC. 1601 // assert(java_thread == noreg , "not expecting a precomputed java thread"); 1602 save_bcp(); 1603 #ifdef ASSERT 1604 { 1605 Label L; 1606 ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize)); 1607 cbz(rscratch1, L); 1608 stop("InterpreterMacroAssembler::call_VM_leaf_base:" 1609 " last_sp != NULL"); 1610 bind(L); 1611 } 1612 #endif /* ASSERT */ 1613 // super call 1614 MacroAssembler::call_VM_base(oop_result, noreg, last_java_sp, 1615 entry_point, number_of_arguments, 1616 check_exceptions); 1617 // interpreter specific 1618 restore_bcp(); 1619 restore_locals(); 1620 } 1621 1622 void InterpreterMacroAssembler::profile_obj_type(Register obj, const Address& mdo_addr) { 1623 assert_different_registers(obj, rscratch1); 1624 Label update, next, none; 1625 1626 verify_oop(obj); 1627 1628 cbnz(obj, update); 1629 orptr(mdo_addr, TypeEntries::null_seen); 1630 b(next); 1631 1632 bind(update); 1633 load_klass(obj, obj); 1634 1635 ldr(rscratch1, mdo_addr); 1636 eor(obj, obj, rscratch1); 1637 tst(obj, TypeEntries::type_klass_mask); 1638 br(Assembler::EQ, next); // klass seen before, nothing to 1639 // do. The unknown bit may have been 1640 // set already but no need to check. 1641 1642 tbnz(obj, exact_log2(TypeEntries::type_unknown), next); 1643 // already unknown. Nothing to do anymore. 1644 1645 ldr(rscratch1, mdo_addr); 1646 cbz(rscratch1, none); 1647 cmp(rscratch1, (u1)TypeEntries::null_seen); 1648 br(Assembler::EQ, none); 1649 // There is a chance that the checks above (re-reading profiling 1650 // data from memory) fail if another thread has just set the 1651 // profiling to this obj's klass 1652 ldr(rscratch1, mdo_addr); 1653 eor(obj, obj, rscratch1); 1654 tst(obj, TypeEntries::type_klass_mask); 1655 br(Assembler::EQ, next); 1656 1657 // different than before. Cannot keep accurate profile. 1658 orptr(mdo_addr, TypeEntries::type_unknown); 1659 b(next); 1660 1661 bind(none); 1662 // first time here. Set profile type. 1663 str(obj, mdo_addr); 1664 1665 bind(next); 1666 } 1667 1668 void InterpreterMacroAssembler::profile_arguments_type(Register mdp, Register callee, Register tmp, bool is_virtual) { 1669 if (!ProfileInterpreter) { 1670 return; 1671 } 1672 1673 if (MethodData::profile_arguments() || MethodData::profile_return()) { 1674 Label profile_continue; 1675 1676 test_method_data_pointer(mdp, profile_continue); 1677 1678 int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size()); 1679 1680 ldrb(rscratch1, Address(mdp, in_bytes(DataLayout::tag_offset()) - off_to_start)); 1681 cmp(rscratch1, u1(is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag)); 1682 br(Assembler::NE, profile_continue); 1683 1684 if (MethodData::profile_arguments()) { 1685 Label done; 1686 int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset()); 1687 1688 for (int i = 0; i < TypeProfileArgsLimit; i++) { 1689 if (i > 0 || MethodData::profile_return()) { 1690 // If return value type is profiled we may have no argument to profile 1691 ldr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset()))); 1692 sub(tmp, tmp, i*TypeStackSlotEntries::per_arg_count()); 1693 cmp(tmp, (u1)TypeStackSlotEntries::per_arg_count()); 1694 add(rscratch1, mdp, off_to_args); 1695 br(Assembler::LT, done); 1696 } 1697 ldr(tmp, Address(callee, Method::const_offset())); 1698 load_unsigned_short(tmp, Address(tmp, ConstMethod::size_of_parameters_offset())); 1699 // stack offset o (zero based) from the start of the argument 1700 // list, for n arguments translates into offset n - o - 1 from 1701 // the end of the argument list 1702 ldr(rscratch1, Address(mdp, in_bytes(TypeEntriesAtCall::stack_slot_offset(i)))); 1703 sub(tmp, tmp, rscratch1); 1704 sub(tmp, tmp, 1); 1705 Address arg_addr = argument_address(tmp); 1706 ldr(tmp, arg_addr); 1707 1708 Address mdo_arg_addr(mdp, in_bytes(TypeEntriesAtCall::argument_type_offset(i))); 1709 profile_obj_type(tmp, mdo_arg_addr); 1710 1711 int to_add = in_bytes(TypeStackSlotEntries::per_arg_size()); 1712 off_to_args += to_add; 1713 } 1714 1715 if (MethodData::profile_return()) { 1716 ldr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset()))); 1717 sub(tmp, tmp, TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count()); 1718 } 1719 1720 add(rscratch1, mdp, off_to_args); 1721 bind(done); 1722 mov(mdp, rscratch1); 1723 1724 if (MethodData::profile_return()) { 1725 // We're right after the type profile for the last 1726 // argument. tmp is the number of cells left in the 1727 // CallTypeData/VirtualCallTypeData to reach its end. Non null 1728 // if there's a return to profile. 1729 assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type"); 1730 add(mdp, mdp, tmp, LSL, exact_log2(DataLayout::cell_size)); 1731 } 1732 str(mdp, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize)); 1733 } else { 1734 assert(MethodData::profile_return(), "either profile call args or call ret"); 1735 update_mdp_by_constant(mdp, in_bytes(TypeEntriesAtCall::return_only_size())); 1736 } 1737 1738 // mdp points right after the end of the 1739 // CallTypeData/VirtualCallTypeData, right after the cells for the 1740 // return value type if there's one 1741 1742 bind(profile_continue); 1743 } 1744 } 1745 1746 void InterpreterMacroAssembler::profile_return_type(Register mdp, Register ret, Register tmp) { 1747 assert_different_registers(mdp, ret, tmp, rbcp); 1748 if (ProfileInterpreter && MethodData::profile_return()) { 1749 Label profile_continue, done; 1750 1751 test_method_data_pointer(mdp, profile_continue); 1752 1753 if (MethodData::profile_return_jsr292_only()) { 1754 assert(Method::intrinsic_id_size_in_bytes() == 2, "assuming Method::_intrinsic_id is u2"); 1755 1756 // If we don't profile all invoke bytecodes we must make sure 1757 // it's a bytecode we indeed profile. We can't go back to the 1758 // begining of the ProfileData we intend to update to check its 1759 // type because we're right after it and we don't known its 1760 // length 1761 Label do_profile; 1762 ldrb(rscratch1, Address(rbcp, 0)); 1763 cmp(rscratch1, (u1)Bytecodes::_invokedynamic); 1764 br(Assembler::EQ, do_profile); 1765 cmp(rscratch1, (u1)Bytecodes::_invokehandle); 1766 br(Assembler::EQ, do_profile); 1767 get_method(tmp); 1768 ldrh(rscratch1, Address(tmp, Method::intrinsic_id_offset_in_bytes())); 1769 subs(zr, rscratch1, vmIntrinsics::_compiledLambdaForm); 1770 br(Assembler::NE, profile_continue); 1771 1772 bind(do_profile); 1773 } 1774 1775 Address mdo_ret_addr(mdp, -in_bytes(ReturnTypeEntry::size())); 1776 mov(tmp, ret); 1777 profile_obj_type(tmp, mdo_ret_addr); 1778 1779 bind(profile_continue); 1780 } 1781 } 1782 1783 void InterpreterMacroAssembler::profile_parameters_type(Register mdp, Register tmp1, Register tmp2) { 1784 assert_different_registers(rscratch1, rscratch2, mdp, tmp1, tmp2); 1785 if (ProfileInterpreter && MethodData::profile_parameters()) { 1786 Label profile_continue, done; 1787 1788 test_method_data_pointer(mdp, profile_continue); 1789 1790 // Load the offset of the area within the MDO used for 1791 // parameters. If it's negative we're not profiling any parameters 1792 ldrw(tmp1, Address(mdp, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset()))); 1793 tbnz(tmp1, 31, profile_continue); // i.e. sign bit set 1794 1795 // Compute a pointer to the area for parameters from the offset 1796 // and move the pointer to the slot for the last 1797 // parameters. Collect profiling from last parameter down. 1798 // mdo start + parameters offset + array length - 1 1799 add(mdp, mdp, tmp1); 1800 ldr(tmp1, Address(mdp, ArrayData::array_len_offset())); 1801 sub(tmp1, tmp1, TypeStackSlotEntries::per_arg_count()); 1802 1803 Label loop; 1804 bind(loop); 1805 1806 int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0)); 1807 int type_base = in_bytes(ParametersTypeData::type_offset(0)); 1808 int per_arg_scale = exact_log2(DataLayout::cell_size); 1809 add(rscratch1, mdp, off_base); 1810 add(rscratch2, mdp, type_base); 1811 1812 Address arg_off(rscratch1, tmp1, Address::lsl(per_arg_scale)); 1813 Address arg_type(rscratch2, tmp1, Address::lsl(per_arg_scale)); 1814 1815 // load offset on the stack from the slot for this parameter 1816 ldr(tmp2, arg_off); 1817 neg(tmp2, tmp2); 1818 // read the parameter from the local area 1819 ldr(tmp2, Address(rlocals, tmp2, Address::lsl(Interpreter::logStackElementSize))); 1820 1821 // profile the parameter 1822 profile_obj_type(tmp2, arg_type); 1823 1824 // go to next parameter 1825 subs(tmp1, tmp1, TypeStackSlotEntries::per_arg_count()); 1826 br(Assembler::GE, loop); 1827 1828 bind(profile_continue); 1829 } 1830 }