1 /* 2 * Copyright (c) 1997, 2014, Oracle and/or its affiliates. All rights reserved. 3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. 4 * 5 * This code is free software; you can redistribute it and/or modify it 6 * under the terms of the GNU General Public License version 2 only, as 7 * published by the Free Software Foundation. 8 * 9 * This code is distributed in the hope that it will be useful, but WITHOUT 10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License 12 * version 2 for more details (a copy is included in the LICENSE file that 13 * accompanied this code). 14 * 15 * You should have received a copy of the GNU General Public License version 16 * 2 along with this work; if not, write to the Free Software Foundation, 17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. 18 * 19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA 20 * or visit www.oracle.com if you need additional information or have any 21 * questions. 22 * 23 */ 24 25 #include "precompiled.hpp" 26 #include "classfile/vmSymbols.hpp" 27 #include "interpreter/bytecode.hpp" 28 #include "interpreter/interpreter.hpp" 29 #include "memory/allocation.inline.hpp" 30 #include "memory/resourceArea.hpp" 31 #include "memory/universe.inline.hpp" 32 #include "oops/methodData.hpp" 33 #include "oops/oop.inline.hpp" 34 #include "prims/jvmtiThreadState.hpp" 35 #include "runtime/handles.inline.hpp" 36 #include "runtime/monitorChunk.hpp" 37 #include "runtime/sharedRuntime.hpp" 38 #include "runtime/vframe.hpp" 39 #include "runtime/vframeArray.hpp" 40 #include "runtime/vframe_hp.hpp" 41 #include "utilities/events.hpp" 42 #ifdef COMPILER2 43 #include "opto/runtime.hpp" 44 #endif 45 46 PRAGMA_FORMAT_MUTE_WARNINGS_FOR_GCC 47 48 int vframeArrayElement:: bci(void) const { return (_bci == SynchronizationEntryBCI ? 0 : _bci); } 49 50 void vframeArrayElement::free_monitors(JavaThread* jt) { 51 if (_monitors != NULL) { 52 MonitorChunk* chunk = _monitors; 53 _monitors = NULL; 54 jt->remove_monitor_chunk(chunk); 55 delete chunk; 56 } 57 } 58 59 void vframeArrayElement::fill_in(compiledVFrame* vf) { 60 61 // Copy the information from the compiled vframe to the 62 // interpreter frame we will be creating to replace vf 63 64 _method = vf->method(); 65 _bci = vf->raw_bci(); 66 _reexecute = vf->should_reexecute(); 67 68 int index; 69 70 // Get the monitors off-stack 71 72 GrowableArray<MonitorInfo*>* list = vf->monitors(); 73 if (list->is_empty()) { 74 _monitors = NULL; 75 } else { 76 77 // Allocate monitor chunk 78 _monitors = new MonitorChunk(list->length()); 79 vf->thread()->add_monitor_chunk(_monitors); 80 81 // Migrate the BasicLocks from the stack to the monitor chunk 82 for (index = 0; index < list->length(); index++) { 83 MonitorInfo* monitor = list->at(index); 84 assert(!monitor->owner_is_scalar_replaced(), "object should be reallocated already"); 85 assert(monitor->owner() == NULL || (!monitor->owner()->is_unlocked() && !monitor->owner()->has_bias_pattern()), "object must be null or locked, and unbiased"); 86 BasicObjectLock* dest = _monitors->at(index); 87 dest->set_obj(monitor->owner()); 88 monitor->lock()->move_to(monitor->owner(), dest->lock()); 89 } 90 } 91 92 // Convert the vframe locals and expressions to off stack 93 // values. Because we will not gc all oops can be converted to 94 // intptr_t (i.e. a stack slot) and we are fine. This is 95 // good since we are inside a HandleMark and the oops in our 96 // collection would go away between packing them here and 97 // unpacking them in unpack_on_stack. 98 99 // First the locals go off-stack 100 101 // FIXME this seems silly it creates a StackValueCollection 102 // in order to get the size to then copy them and 103 // convert the types to intptr_t size slots. Seems like it 104 // could do it in place... Still uses less memory than the 105 // old way though 106 107 StackValueCollection *locs = vf->locals(); 108 _locals = new StackValueCollection(locs->size()); 109 for(index = 0; index < locs->size(); index++) { 110 StackValue* value = locs->at(index); 111 switch(value->type()) { 112 case T_OBJECT: 113 assert(!value->obj_is_scalar_replaced(), "object should be reallocated already"); 114 // preserve object type 115 _locals->add( new StackValue(cast_from_oop<intptr_t>((value->get_obj()())), T_OBJECT )); 116 break; 117 case T_CONFLICT: 118 // A dead local. Will be initialized to null/zero. 119 _locals->add( new StackValue()); 120 break; 121 case T_INT: 122 _locals->add( new StackValue(value->get_int())); 123 break; 124 default: 125 ShouldNotReachHere(); 126 } 127 } 128 129 // Now the expressions off-stack 130 // Same silliness as above 131 132 StackValueCollection *exprs = vf->expressions(); 133 _expressions = new StackValueCollection(exprs->size()); 134 for(index = 0; index < exprs->size(); index++) { 135 StackValue* value = exprs->at(index); 136 switch(value->type()) { 137 case T_OBJECT: 138 assert(!value->obj_is_scalar_replaced(), "object should be reallocated already"); 139 // preserve object type 140 _expressions->add( new StackValue(cast_from_oop<intptr_t>((value->get_obj()())), T_OBJECT )); 141 break; 142 case T_CONFLICT: 143 // A dead stack element. Will be initialized to null/zero. 144 // This can occur when the compiler emits a state in which stack 145 // elements are known to be dead (because of an imminent exception). 146 _expressions->add( new StackValue()); 147 break; 148 case T_INT: 149 _expressions->add( new StackValue(value->get_int())); 150 break; 151 default: 152 ShouldNotReachHere(); 153 } 154 } 155 } 156 157 int unpack_counter = 0; 158 159 void vframeArrayElement::unpack_on_stack(int caller_actual_parameters, 160 int callee_parameters, 161 int callee_locals, 162 frame* caller, 163 bool is_top_frame, 164 bool is_bottom_frame, 165 int exec_mode) { 166 JavaThread* thread = (JavaThread*) Thread::current(); 167 168 // Look at bci and decide on bcp and continuation pc 169 address bcp; 170 // C++ interpreter doesn't need a pc since it will figure out what to do when it 171 // begins execution 172 address pc; 173 bool use_next_mdp = false; // true if we should use the mdp associated with the next bci 174 // rather than the one associated with bcp 175 if (raw_bci() == SynchronizationEntryBCI) { 176 // We are deoptimizing while hanging in prologue code for synchronized method 177 bcp = method()->bcp_from(0); // first byte code 178 pc = Interpreter::deopt_entry(vtos, 0); // step = 0 since we don't skip current bytecode 179 } else if (should_reexecute()) { //reexecute this bytecode 180 assert(is_top_frame, "reexecute allowed only for the top frame"); 181 bcp = method()->bcp_from(bci()); 182 pc = Interpreter::deopt_reexecute_entry(method(), bcp); 183 } else { 184 bcp = method()->bcp_from(bci()); 185 pc = Interpreter::deopt_continue_after_entry(method(), bcp, callee_parameters, is_top_frame); 186 use_next_mdp = true; 187 } 188 assert(Bytecodes::is_defined(*bcp), "must be a valid bytecode"); 189 190 // Monitorenter and pending exceptions: 191 // 192 // For Compiler2, there should be no pending exception when deoptimizing at monitorenter 193 // because there is no safepoint at the null pointer check (it is either handled explicitly 194 // or prior to the monitorenter) and asynchronous exceptions are not made "pending" by the 195 // runtime interface for the slow case (see JRT_ENTRY_FOR_MONITORENTER). If an asynchronous 196 // exception was processed, the bytecode pointer would have to be extended one bytecode beyond 197 // the monitorenter to place it in the proper exception range. 198 // 199 // For Compiler1, deoptimization can occur while throwing a NullPointerException at monitorenter, 200 // in which case bcp should point to the monitorenter since it is within the exception's range. 201 202 assert(*bcp != Bytecodes::_monitorenter || is_top_frame, "a _monitorenter must be a top frame"); 203 assert(thread->deopt_nmethod() != NULL, "nmethod should be known"); 204 guarantee(!(thread->deopt_nmethod()->is_compiled_by_c2() && 205 *bcp == Bytecodes::_monitorenter && 206 exec_mode == Deoptimization::Unpack_exception), 207 "shouldn't get exception during monitorenter"); 208 209 int popframe_preserved_args_size_in_bytes = 0; 210 int popframe_preserved_args_size_in_words = 0; 211 if (is_top_frame) { 212 JvmtiThreadState *state = thread->jvmti_thread_state(); 213 if (JvmtiExport::can_pop_frame() && 214 (thread->has_pending_popframe() || thread->popframe_forcing_deopt_reexecution())) { 215 if (thread->has_pending_popframe()) { 216 // Pop top frame after deoptimization 217 #ifndef CC_INTERP 218 pc = Interpreter::remove_activation_preserving_args_entry(); 219 #else 220 // Do an uncommon trap type entry. c++ interpreter will know 221 // to pop frame and preserve the args 222 pc = Interpreter::deopt_entry(vtos, 0); 223 use_next_mdp = false; 224 #endif 225 } else { 226 // Reexecute invoke in top frame 227 pc = Interpreter::deopt_entry(vtos, 0); 228 use_next_mdp = false; 229 popframe_preserved_args_size_in_bytes = in_bytes(thread->popframe_preserved_args_size()); 230 // Note: the PopFrame-related extension of the expression stack size is done in 231 // Deoptimization::fetch_unroll_info_helper 232 popframe_preserved_args_size_in_words = in_words(thread->popframe_preserved_args_size_in_words()); 233 } 234 } else if (JvmtiExport::can_force_early_return() && state != NULL && state->is_earlyret_pending()) { 235 // Force early return from top frame after deoptimization 236 #ifndef CC_INTERP 237 pc = Interpreter::remove_activation_early_entry(state->earlyret_tos()); 238 #endif 239 } else { 240 // Possibly override the previous pc computation of the top (youngest) frame 241 switch (exec_mode) { 242 case Deoptimization::Unpack_deopt: 243 // use what we've got 244 break; 245 case Deoptimization::Unpack_exception: 246 // exception is pending 247 pc = SharedRuntime::raw_exception_handler_for_return_address(thread, pc); 248 // [phh] We're going to end up in some handler or other, so it doesn't 249 // matter what mdp we point to. See exception_handler_for_exception() 250 // in interpreterRuntime.cpp. 251 break; 252 case Deoptimization::Unpack_uncommon_trap: 253 case Deoptimization::Unpack_reexecute: 254 // redo last byte code 255 pc = Interpreter::deopt_entry(vtos, 0); 256 use_next_mdp = false; 257 break; 258 default: 259 ShouldNotReachHere(); 260 } 261 } 262 } 263 264 // Setup the interpreter frame 265 266 assert(method() != NULL, "method must exist"); 267 int temps = expressions()->size(); 268 269 int locks = monitors() == NULL ? 0 : monitors()->number_of_monitors(); 270 271 Interpreter::layout_activation(method(), 272 temps + callee_parameters, 273 popframe_preserved_args_size_in_words, 274 locks, 275 caller_actual_parameters, 276 callee_parameters, 277 callee_locals, 278 caller, 279 iframe(), 280 is_top_frame, 281 is_bottom_frame); 282 283 // Update the pc in the frame object and overwrite the temporary pc 284 // we placed in the skeletal frame now that we finally know the 285 // exact interpreter address we should use. 286 287 _frame.patch_pc(thread, pc); 288 289 assert (!method()->is_synchronized() || locks > 0, "synchronized methods must have monitors"); 290 291 BasicObjectLock* top = iframe()->interpreter_frame_monitor_begin(); 292 for (int index = 0; index < locks; index++) { 293 top = iframe()->previous_monitor_in_interpreter_frame(top); 294 BasicObjectLock* src = _monitors->at(index); 295 top->set_obj(src->obj()); 296 src->lock()->move_to(src->obj(), top->lock()); 297 } 298 if (ProfileInterpreter) { 299 iframe()->interpreter_frame_set_mdx(0); // clear out the mdp. 300 } 301 iframe()->interpreter_frame_set_bcx((intptr_t)bcp); // cannot use bcp because frame is not initialized yet 302 if (ProfileInterpreter) { 303 MethodData* mdo = method()->method_data(); 304 if (mdo != NULL) { 305 int bci = iframe()->interpreter_frame_bci(); 306 if (use_next_mdp) ++bci; 307 address mdp = mdo->bci_to_dp(bci); 308 iframe()->interpreter_frame_set_mdp(mdp); 309 } 310 } 311 312 // Unpack expression stack 313 // If this is an intermediate frame (i.e. not top frame) then this 314 // only unpacks the part of the expression stack not used by callee 315 // as parameters. The callee parameters are unpacked as part of the 316 // callee locals. 317 int i; 318 for(i = 0; i < expressions()->size(); i++) { 319 StackValue *value = expressions()->at(i); 320 intptr_t* addr = iframe()->interpreter_frame_expression_stack_at(i); 321 switch(value->type()) { 322 case T_INT: 323 *addr = value->get_int(); 324 break; 325 case T_OBJECT: 326 *addr = value->get_int(T_OBJECT); 327 break; 328 case T_CONFLICT: 329 // A dead stack slot. Initialize to null in case it is an oop. 330 *addr = NULL_WORD; 331 break; 332 default: 333 ShouldNotReachHere(); 334 } 335 } 336 337 338 // Unpack the locals 339 for(i = 0; i < locals()->size(); i++) { 340 StackValue *value = locals()->at(i); 341 intptr_t* addr = iframe()->interpreter_frame_local_at(i); 342 switch(value->type()) { 343 case T_INT: 344 *addr = value->get_int(); 345 break; 346 case T_OBJECT: 347 *addr = value->get_int(T_OBJECT); 348 break; 349 case T_CONFLICT: 350 // A dead location. If it is an oop then we need a NULL to prevent GC from following it 351 *addr = NULL_WORD; 352 break; 353 default: 354 ShouldNotReachHere(); 355 } 356 } 357 358 if (is_top_frame && JvmtiExport::can_pop_frame() && thread->popframe_forcing_deopt_reexecution()) { 359 // An interpreted frame was popped but it returns to a deoptimized 360 // frame. The incoming arguments to the interpreted activation 361 // were preserved in thread-local storage by the 362 // remove_activation_preserving_args_entry in the interpreter; now 363 // we put them back into the just-unpacked interpreter frame. 364 // Note that this assumes that the locals arena grows toward lower 365 // addresses. 366 if (popframe_preserved_args_size_in_words != 0) { 367 void* saved_args = thread->popframe_preserved_args(); 368 assert(saved_args != NULL, "must have been saved by interpreter"); 369 #ifdef ASSERT 370 assert(popframe_preserved_args_size_in_words <= 371 iframe()->interpreter_frame_expression_stack_size()*Interpreter::stackElementWords, 372 "expression stack size should have been extended"); 373 #endif // ASSERT 374 int top_element = iframe()->interpreter_frame_expression_stack_size()-1; 375 intptr_t* base; 376 if (frame::interpreter_frame_expression_stack_direction() < 0) { 377 base = iframe()->interpreter_frame_expression_stack_at(top_element); 378 } else { 379 base = iframe()->interpreter_frame_expression_stack(); 380 } 381 Copy::conjoint_jbytes(saved_args, 382 base, 383 popframe_preserved_args_size_in_bytes); 384 thread->popframe_free_preserved_args(); 385 } 386 } 387 388 #ifndef PRODUCT 389 if (TraceDeoptimization && Verbose) { 390 ttyLocker ttyl; 391 tty->print_cr("[%d Interpreted Frame]", ++unpack_counter); 392 iframe()->print_on(tty); 393 RegisterMap map(thread); 394 vframe* f = vframe::new_vframe(iframe(), &map, thread); 395 f->print(); 396 397 tty->print_cr("locals size %d", locals()->size()); 398 tty->print_cr("expression size %d", expressions()->size()); 399 400 method()->print_value(); 401 tty->cr(); 402 // method()->print_codes(); 403 } else if (TraceDeoptimization) { 404 tty->print(" "); 405 method()->print_value(); 406 Bytecodes::Code code = Bytecodes::java_code_at(method(), bcp); 407 int bci = method()->bci_from(bcp); 408 tty->print(" - %s", Bytecodes::name(code)); 409 tty->print(" @ bci %d ", bci); 410 tty->print_cr("sp = " PTR_FORMAT, iframe()->sp()); 411 } 412 #endif // PRODUCT 413 414 // The expression stack and locals are in the resource area don't leave 415 // a dangling pointer in the vframeArray we leave around for debug 416 // purposes 417 418 _locals = _expressions = NULL; 419 420 } 421 422 int vframeArrayElement::on_stack_size(int callee_parameters, 423 int callee_locals, 424 bool is_top_frame, 425 int popframe_extra_stack_expression_els) const { 426 assert(method()->max_locals() == locals()->size(), "just checking"); 427 int locks = monitors() == NULL ? 0 : monitors()->number_of_monitors(); 428 int temps = expressions()->size(); 429 return Interpreter::size_activation(method()->max_stack(), 430 temps + callee_parameters, 431 popframe_extra_stack_expression_els, 432 locks, 433 callee_parameters, 434 callee_locals, 435 is_top_frame); 436 } 437 438 439 440 vframeArray* vframeArray::allocate(JavaThread* thread, int frame_size, GrowableArray<compiledVFrame*>* chunk, 441 RegisterMap *reg_map, frame sender, frame caller, frame self) { 442 443 // Allocate the vframeArray 444 vframeArray * result = (vframeArray*) AllocateHeap(sizeof(vframeArray) + // fixed part 445 sizeof(vframeArrayElement) * (chunk->length() - 1), // variable part 446 mtCompiler); 447 result->_frames = chunk->length(); 448 result->_owner_thread = thread; 449 result->_sender = sender; 450 result->_caller = caller; 451 result->_original = self; 452 result->set_unroll_block(NULL); // initialize it 453 result->fill_in(thread, frame_size, chunk, reg_map); 454 return result; 455 } 456 457 void vframeArray::fill_in(JavaThread* thread, 458 int frame_size, 459 GrowableArray<compiledVFrame*>* chunk, 460 const RegisterMap *reg_map) { 461 // Set owner first, it is used when adding monitor chunks 462 463 _frame_size = frame_size; 464 for(int i = 0; i < chunk->length(); i++) { 465 element(i)->fill_in(chunk->at(i)); 466 } 467 468 // Copy registers for callee-saved registers 469 if (reg_map != NULL) { 470 for(int i = 0; i < RegisterMap::reg_count; i++) { 471 #ifdef AMD64 472 // The register map has one entry for every int (32-bit value), so 473 // 64-bit physical registers have two entries in the map, one for 474 // each half. Ignore the high halves of 64-bit registers, just like 475 // frame::oopmapreg_to_location does. 476 // 477 // [phh] FIXME: this is a temporary hack! This code *should* work 478 // correctly w/o this hack, possibly by changing RegisterMap::pd_location 479 // in frame_amd64.cpp and the values of the phantom high half registers 480 // in amd64.ad. 481 // if (VMReg::Name(i) < SharedInfo::stack0 && is_even(i)) { 482 intptr_t* src = (intptr_t*) reg_map->location(VMRegImpl::as_VMReg(i)); 483 _callee_registers[i] = src != NULL ? *src : NULL_WORD; 484 // } else { 485 // jint* src = (jint*) reg_map->location(VMReg::Name(i)); 486 // _callee_registers[i] = src != NULL ? *src : NULL_WORD; 487 // } 488 #else 489 jint* src = (jint*) reg_map->location(VMRegImpl::as_VMReg(i)); 490 _callee_registers[i] = src != NULL ? *src : NULL_WORD; 491 #endif 492 if (src == NULL) { 493 set_location_valid(i, false); 494 } else { 495 set_location_valid(i, true); 496 jint* dst = (jint*) register_location(i); 497 *dst = *src; 498 } 499 } 500 } 501 } 502 503 void vframeArray::unpack_to_stack(frame &unpack_frame, int exec_mode, int caller_actual_parameters) { 504 // stack picture 505 // unpack_frame 506 // [new interpreter frames ] (frames are skeletal but walkable) 507 // caller_frame 508 // 509 // This routine fills in the missing data for the skeletal interpreter frames 510 // in the above picture. 511 512 // Find the skeletal interpreter frames to unpack into 513 JavaThread* THREAD = JavaThread::current(); 514 RegisterMap map(THREAD, false); 515 // Get the youngest frame we will unpack (last to be unpacked) 516 frame me = unpack_frame.sender(&map); 517 int index; 518 for (index = 0; index < frames(); index++ ) { 519 *element(index)->iframe() = me; 520 // Get the caller frame (possibly skeletal) 521 me = me.sender(&map); 522 } 523 524 // Do the unpacking of interpreter frames; the frame at index 0 represents the top activation, so it has no callee 525 // Unpack the frames from the oldest (frames() -1) to the youngest (0) 526 frame* caller_frame = &me; 527 for (index = frames() - 1; index >= 0 ; index--) { 528 vframeArrayElement* elem = element(index); // caller 529 int callee_parameters, callee_locals; 530 if (index == 0) { 531 callee_parameters = callee_locals = 0; 532 } else { 533 methodHandle caller = elem->method(); 534 methodHandle callee = element(index - 1)->method(); 535 Bytecode_invoke inv(caller, elem->bci()); 536 // invokedynamic instructions don't have a class but obviously don't have a MemberName appendix. 537 // NOTE: Use machinery here that avoids resolving of any kind. 538 const bool has_member_arg = 539 !inv.is_invokedynamic() && MethodHandles::has_member_arg(inv.klass(), inv.name()); 540 callee_parameters = callee->size_of_parameters() + (has_member_arg ? 1 : 0); 541 callee_locals = callee->max_locals(); 542 } 543 elem->unpack_on_stack(caller_actual_parameters, 544 callee_parameters, 545 callee_locals, 546 caller_frame, 547 index == 0, 548 index == frames() - 1, 549 exec_mode); 550 if (index == frames() - 1) { 551 Deoptimization::unwind_callee_save_values(elem->iframe(), this); 552 } 553 caller_frame = elem->iframe(); 554 caller_actual_parameters = callee_parameters; 555 } 556 deallocate_monitor_chunks(); 557 } 558 559 void vframeArray::deallocate_monitor_chunks() { 560 JavaThread* jt = JavaThread::current(); 561 for (int index = 0; index < frames(); index++ ) { 562 element(index)->free_monitors(jt); 563 } 564 } 565 566 #ifndef PRODUCT 567 568 bool vframeArray::structural_compare(JavaThread* thread, GrowableArray<compiledVFrame*>* chunk) { 569 if (owner_thread() != thread) return false; 570 int index = 0; 571 #if 0 // FIXME can't do this comparison 572 573 // Compare only within vframe array. 574 for (deoptimizedVFrame* vf = deoptimizedVFrame::cast(vframe_at(first_index())); vf; vf = vf->deoptimized_sender_or_null()) { 575 if (index >= chunk->length() || !vf->structural_compare(chunk->at(index))) return false; 576 index++; 577 } 578 if (index != chunk->length()) return false; 579 #endif 580 581 return true; 582 } 583 584 #endif 585 586 address vframeArray::register_location(int i) const { 587 assert(0 <= i && i < RegisterMap::reg_count, "index out of bounds"); 588 return (address) & _callee_registers[i]; 589 } 590 591 592 #ifndef PRODUCT 593 594 // Printing 595 596 // Note: we cannot have print_on as const, as we allocate inside the method 597 void vframeArray::print_on_2(outputStream* st) { 598 st->print_cr(" - sp: " INTPTR_FORMAT, sp()); 599 st->print(" - thread: "); 600 Thread::current()->print(); 601 st->print_cr(" - frame size: %d", frame_size()); 602 for (int index = 0; index < frames() ; index++ ) { 603 element(index)->print(st); 604 } 605 } 606 607 void vframeArrayElement::print(outputStream* st) { 608 st->print_cr(" - interpreter_frame -> sp: " INTPTR_FORMAT, iframe()->sp()); 609 } 610 611 void vframeArray::print_value_on(outputStream* st) const { 612 st->print_cr("vframeArray [%d] ", frames()); 613 } 614 615 616 #endif