1 /* 2 * Copyright (c) 1997, 2013, 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 47 int vframeArrayElement:: bci(void) const { return (_bci == SynchronizationEntryBCI ? 0 : _bci); } 48 49 void vframeArrayElement::free_monitors(JavaThread* jt) { 50 if (_monitors != NULL) { 51 MonitorChunk* chunk = _monitors; 52 _monitors = NULL; 53 jt->remove_monitor_chunk(chunk); 54 delete chunk; 55 } 56 } 57 58 void vframeArrayElement::fill_in(compiledVFrame* vf) { 59 60 // Copy the information from the compiled vframe to the 61 // interpreter frame we will be creating to replace vf 62 63 _method = vf->method(); 64 _bci = vf->raw_bci(); 65 _reexecute = vf->should_reexecute(); 66 67 int index; 68 69 // Get the monitors off-stack 70 71 GrowableArray<MonitorInfo*>* list = vf->monitors(); 72 if (list->is_empty()) { 73 _monitors = NULL; 74 } else { 75 76 // Allocate monitor chunk 77 _monitors = new MonitorChunk(list->length()); 78 vf->thread()->add_monitor_chunk(_monitors); 79 80 // Migrate the BasicLocks from the stack to the monitor chunk 81 for (index = 0; index < list->length(); index++) { 82 MonitorInfo* monitor = list->at(index); 83 assert(!monitor->owner_is_scalar_replaced(), "object should be reallocated already"); 84 assert(monitor->owner() == NULL || (!monitor->owner()->is_unlocked() && !monitor->owner()->has_bias_pattern()), "object must be null or locked, and unbiased"); 85 BasicObjectLock* dest = _monitors->at(index); 86 dest->set_obj(monitor->owner()); 87 monitor->lock()->move_to(monitor->owner(), dest->lock()); 88 } 89 } 90 91 // Convert the vframe locals and expressions to off stack 92 // values. Because we will not gc all oops can be converted to 93 // intptr_t (i.e. a stack slot) and we are fine. This is 94 // good since we are inside a HandleMark and the oops in our 95 // collection would go away between packing them here and 96 // unpacking them in unpack_on_stack. 97 98 // First the locals go off-stack 99 100 // FIXME this seems silly it creates a StackValueCollection 101 // in order to get the size to then copy them and 102 // convert the types to intptr_t size slots. Seems like it 103 // could do it in place... Still uses less memory than the 104 // old way though 105 106 StackValueCollection *locs = vf->locals(); 107 _locals = new StackValueCollection(locs->size()); 108 for(index = 0; index < locs->size(); index++) { 109 StackValue* value = locs->at(index); 110 switch(value->type()) { 111 case T_OBJECT: 112 assert(!value->obj_is_scalar_replaced(), "object should be reallocated already"); 113 // preserve object type 114 _locals->add( new StackValue(cast_from_oop<intptr_t>((value->get_obj()())), T_OBJECT )); 115 break; 116 case T_CONFLICT: 117 // A dead local. Will be initialized to null/zero. 118 _locals->add( new StackValue()); 119 break; 120 case T_INT: 121 _locals->add( new StackValue(value->get_int())); 122 break; 123 default: 124 ShouldNotReachHere(); 125 } 126 } 127 128 // Now the expressions off-stack 129 // Same silliness as above 130 131 StackValueCollection *exprs = vf->expressions(); 132 _expressions = new StackValueCollection(exprs->size()); 133 for(index = 0; index < exprs->size(); index++) { 134 StackValue* value = exprs->at(index); 135 switch(value->type()) { 136 case T_OBJECT: 137 assert(!value->obj_is_scalar_replaced(), "object should be reallocated already"); 138 // preserve object type 139 _expressions->add( new StackValue(cast_from_oop<intptr_t>((value->get_obj()())), T_OBJECT )); 140 break; 141 case T_CONFLICT: 142 // A dead stack element. Will be initialized to null/zero. 143 // This can occur when the compiler emits a state in which stack 144 // elements are known to be dead (because of an imminent exception). 145 _expressions->add( new StackValue()); 146 break; 147 case T_INT: 148 _expressions->add( new StackValue(value->get_int())); 149 break; 150 default: 151 ShouldNotReachHere(); 152 } 153 } 154 } 155 156 int unpack_counter = 0; 157 158 void vframeArrayElement::unpack_on_stack(int caller_actual_parameters, 159 int callee_parameters, 160 int callee_locals, 161 frame* caller, 162 bool is_top_frame, 163 bool is_bottom_frame, 164 int exec_mode) { 165 JavaThread* thread = (JavaThread*) Thread::current(); 166 167 // Look at bci and decide on bcp and continuation pc 168 address bcp; 169 // C++ interpreter doesn't need a pc since it will figure out what to do when it 170 // begins execution 171 address pc; 172 bool use_next_mdp = false; // true if we should use the mdp associated with the next bci 173 // rather than the one associated with bcp 174 if (raw_bci() == SynchronizationEntryBCI) { 175 // We are deoptimizing while hanging in prologue code for synchronized method 176 bcp = method()->bcp_from(0); // first byte code 177 pc = Interpreter::deopt_entry(vtos, 0); // step = 0 since we don't skip current bytecode 178 } else if (should_reexecute()) { //reexecute this bytecode 179 assert(is_top_frame, "reexecute allowed only for the top frame"); 180 bcp = method()->bcp_from(bci()); 181 pc = Interpreter::deopt_reexecute_entry(method(), bcp); 182 } else { 183 bcp = method()->bcp_from(bci()); 184 pc = Interpreter::deopt_continue_after_entry(method(), bcp, callee_parameters, is_top_frame); 185 use_next_mdp = true; 186 } 187 assert(Bytecodes::is_defined(*bcp), "must be a valid bytecode"); 188 189 // Monitorenter and pending exceptions: 190 // 191 // For Compiler2, there should be no pending exception when deoptimizing at monitorenter 192 // because there is no safepoint at the null pointer check (it is either handled explicitly 193 // or prior to the monitorenter) and asynchronous exceptions are not made "pending" by the 194 // runtime interface for the slow case (see JRT_ENTRY_FOR_MONITORENTER). If an asynchronous 195 // exception was processed, the bytecode pointer would have to be extended one bytecode beyond 196 // the monitorenter to place it in the proper exception range. 197 // 198 // For Compiler1, deoptimization can occur while throwing a NullPointerException at monitorenter, 199 // in which case bcp should point to the monitorenter since it is within the exception's range. 200 201 assert(*bcp != Bytecodes::_monitorenter || is_top_frame, "a _monitorenter must be a top frame"); 202 assert(thread->deopt_nmethod() != NULL, "nmethod should be known"); 203 guarantee(!(thread->deopt_nmethod()->is_compiled_by_c2() && 204 *bcp == Bytecodes::_monitorenter && 205 exec_mode == Deoptimization::Unpack_exception), 206 "shouldn't get exception during monitorenter"); 207 208 int popframe_preserved_args_size_in_bytes = 0; 209 int popframe_preserved_args_size_in_words = 0; 210 if (is_top_frame) { 211 JvmtiThreadState *state = thread->jvmti_thread_state(); 212 if (JvmtiExport::can_pop_frame() && 213 (thread->has_pending_popframe() || thread->popframe_forcing_deopt_reexecution())) { 214 if (thread->has_pending_popframe()) { 215 // Pop top frame after deoptimization 216 #ifndef CC_INTERP 217 pc = Interpreter::remove_activation_preserving_args_entry(); 218 #else 219 // Do an uncommon trap type entry. c++ interpreter will know 220 // to pop frame and preserve the args 221 pc = Interpreter::deopt_entry(vtos, 0); 222 use_next_mdp = false; 223 #endif 224 } else { 225 // Reexecute invoke in top frame 226 pc = Interpreter::deopt_entry(vtos, 0); 227 use_next_mdp = false; 228 popframe_preserved_args_size_in_bytes = in_bytes(thread->popframe_preserved_args_size()); 229 // Note: the PopFrame-related extension of the expression stack size is done in 230 // Deoptimization::fetch_unroll_info_helper 231 popframe_preserved_args_size_in_words = in_words(thread->popframe_preserved_args_size_in_words()); 232 } 233 } else if (JvmtiExport::can_force_early_return() && state != NULL && state->is_earlyret_pending()) { 234 // Force early return from top frame after deoptimization 235 #ifndef CC_INTERP 236 pc = Interpreter::remove_activation_early_entry(state->earlyret_tos()); 237 #endif 238 } else { 239 // Possibly override the previous pc computation of the top (youngest) frame 240 switch (exec_mode) { 241 case Deoptimization::Unpack_deopt: 242 // use what we've got 243 break; 244 case Deoptimization::Unpack_exception: 245 // exception is pending 246 pc = SharedRuntime::raw_exception_handler_for_return_address(thread, pc); 247 // [phh] We're going to end up in some handler or other, so it doesn't 248 // matter what mdp we point to. See exception_handler_for_exception() 249 // in interpreterRuntime.cpp. 250 break; 251 case Deoptimization::Unpack_uncommon_trap: 252 case Deoptimization::Unpack_reexecute: 253 // redo last byte code 254 pc = Interpreter::deopt_entry(vtos, 0); 255 use_next_mdp = false; 256 break; 257 default: 258 ShouldNotReachHere(); 259 } 260 } 261 } 262 263 // Setup the interpreter frame 264 265 assert(method() != NULL, "method must exist"); 266 int temps = expressions()->size(); 267 268 int locks = monitors() == NULL ? 0 : monitors()->number_of_monitors(); 269 270 Interpreter::layout_activation(method(), 271 temps + callee_parameters, 272 popframe_preserved_args_size_in_words, 273 locks, 274 caller_actual_parameters, 275 callee_parameters, 276 callee_locals, 277 caller, 278 iframe(), 279 is_top_frame, 280 is_bottom_frame); 281 282 // Update the pc in the frame object and overwrite the temporary pc 283 // we placed in the skeletal frame now that we finally know the 284 // exact interpreter address we should use. 285 286 _frame.patch_pc(thread, pc); 287 288 assert (!method()->is_synchronized() || locks > 0, "synchronized methods must have monitors"); 289 290 BasicObjectLock* top = iframe()->interpreter_frame_monitor_begin(); 291 for (int index = 0; index < locks; index++) { 292 top = iframe()->previous_monitor_in_interpreter_frame(top); 293 BasicObjectLock* src = _monitors->at(index); 294 top->set_obj(src->obj()); 295 src->lock()->move_to(src->obj(), top->lock()); 296 } 297 if (ProfileInterpreter) { 298 iframe()->interpreter_frame_set_mdx(0); // clear out the mdp. 299 } 300 iframe()->interpreter_frame_set_bcx((intptr_t)bcp); // cannot use bcp because frame is not initialized yet 301 if (ProfileInterpreter) { 302 MethodData* mdo = method()->method_data(); 303 if (mdo != NULL) { 304 int bci = iframe()->interpreter_frame_bci(); 305 if (use_next_mdp) ++bci; 306 address mdp = mdo->bci_to_dp(bci); 307 iframe()->interpreter_frame_set_mdp(mdp); 308 } 309 } 310 311 // Unpack expression stack 312 // If this is an intermediate frame (i.e. not top frame) then this 313 // only unpacks the part of the expression stack not used by callee 314 // as parameters. The callee parameters are unpacked as part of the 315 // callee locals. 316 int i; 317 for(i = 0; i < expressions()->size(); i++) { 318 StackValue *value = expressions()->at(i); 319 intptr_t* addr = iframe()->interpreter_frame_expression_stack_at(i); 320 switch(value->type()) { 321 case T_INT: 322 *addr = value->get_int(); 323 break; 324 case T_OBJECT: 325 *addr = value->get_int(T_OBJECT); 326 break; 327 case T_CONFLICT: 328 // A dead stack slot. Initialize to null in case it is an oop. 329 *addr = NULL_WORD; 330 break; 331 default: 332 ShouldNotReachHere(); 333 } 334 } 335 336 337 // Unpack the locals 338 for(i = 0; i < locals()->size(); i++) { 339 StackValue *value = locals()->at(i); 340 intptr_t* addr = iframe()->interpreter_frame_local_at(i); 341 switch(value->type()) { 342 case T_INT: 343 *addr = value->get_int(); 344 break; 345 case T_OBJECT: 346 *addr = value->get_int(T_OBJECT); 347 break; 348 case T_CONFLICT: 349 // A dead location. If it is an oop then we need a NULL to prevent GC from following it 350 *addr = NULL_WORD; 351 break; 352 default: 353 ShouldNotReachHere(); 354 } 355 } 356 357 if (is_top_frame && JvmtiExport::can_pop_frame() && thread->popframe_forcing_deopt_reexecution()) { 358 // An interpreted frame was popped but it returns to a deoptimized 359 // frame. The incoming arguments to the interpreted activation 360 // were preserved in thread-local storage by the 361 // remove_activation_preserving_args_entry in the interpreter; now 362 // we put them back into the just-unpacked interpreter frame. 363 // Note that this assumes that the locals arena grows toward lower 364 // addresses. 365 if (popframe_preserved_args_size_in_words != 0) { 366 void* saved_args = thread->popframe_preserved_args(); 367 assert(saved_args != NULL, "must have been saved by interpreter"); 368 #ifdef ASSERT 369 assert(popframe_preserved_args_size_in_words <= 370 iframe()->interpreter_frame_expression_stack_size()*Interpreter::stackElementWords, 371 "expression stack size should have been extended"); 372 #endif // ASSERT 373 int top_element = iframe()->interpreter_frame_expression_stack_size()-1; 374 intptr_t* base; 375 if (frame::interpreter_frame_expression_stack_direction() < 0) { 376 base = iframe()->interpreter_frame_expression_stack_at(top_element); 377 } else { 378 base = iframe()->interpreter_frame_expression_stack(); 379 } 380 Copy::conjoint_jbytes(saved_args, 381 base, 382 popframe_preserved_args_size_in_bytes); 383 thread->popframe_free_preserved_args(); 384 } 385 } 386 387 #ifndef PRODUCT 388 if (TraceDeoptimization && Verbose) { 389 ttyLocker ttyl; 390 tty->print_cr("[%d Interpreted Frame]", ++unpack_counter); 391 iframe()->print_on(tty); 392 RegisterMap map(thread); 393 vframe* f = vframe::new_vframe(iframe(), &map, thread); 394 f->print(); 395 396 tty->print_cr("locals size %d", locals()->size()); 397 tty->print_cr("expression size %d", expressions()->size()); 398 399 method()->print_value(); 400 tty->cr(); 401 // method()->print_codes(); 402 } else if (TraceDeoptimization) { 403 tty->print(" "); 404 method()->print_value(); 405 Bytecodes::Code code = Bytecodes::java_code_at(method(), bcp); 406 int bci = method()->bci_from(bcp); 407 tty->print(" - %s", Bytecodes::name(code)); 408 tty->print(" @ bci %d ", bci); 409 tty->print_cr("sp = " PTR_FORMAT, iframe()->sp()); 410 } 411 #endif // PRODUCT 412 413 // The expression stack and locals are in the resource area don't leave 414 // a dangling pointer in the vframeArray we leave around for debug 415 // purposes 416 417 _locals = _expressions = NULL; 418 419 } 420 421 int vframeArrayElement::on_stack_size(int caller_actual_parameters, 422 int callee_parameters, 423 int callee_locals, 424 bool is_top_frame, 425 bool is_bottom_frame, 426 int popframe_extra_stack_expression_els) const { 427 assert(method()->max_locals() == locals()->size(), "just checking"); 428 int locks = monitors() == NULL ? 0 : monitors()->number_of_monitors(); 429 int temps = expressions()->size(); 430 return Interpreter::size_activation(method(), 431 temps + callee_parameters, 432 popframe_extra_stack_expression_els, 433 locks, 434 caller_actual_parameters, 435 callee_parameters, 436 callee_locals, 437 is_top_frame, 438 is_bottom_frame); 439 } 440 441 442 443 vframeArray* vframeArray::allocate(JavaThread* thread, int frame_size, GrowableArray<compiledVFrame*>* chunk, 444 RegisterMap *reg_map, frame sender, frame caller, frame self) { 445 446 // Allocate the vframeArray 447 vframeArray * result = (vframeArray*) AllocateHeap(sizeof(vframeArray) + // fixed part 448 sizeof(vframeArrayElement) * (chunk->length() - 1), // variable part 449 mtCompiler); 450 result->_frames = chunk->length(); 451 result->_owner_thread = thread; 452 result->_sender = sender; 453 result->_caller = caller; 454 result->_original = self; 455 result->set_unroll_block(NULL); // initialize it 456 result->fill_in(thread, frame_size, chunk, reg_map); 457 return result; 458 } 459 460 void vframeArray::fill_in(JavaThread* thread, 461 int frame_size, 462 GrowableArray<compiledVFrame*>* chunk, 463 const RegisterMap *reg_map) { 464 // Set owner first, it is used when adding monitor chunks 465 466 _frame_size = frame_size; 467 for(int i = 0; i < chunk->length(); i++) { 468 element(i)->fill_in(chunk->at(i)); 469 } 470 471 // Copy registers for callee-saved registers 472 if (reg_map != NULL) { 473 for(int i = 0; i < RegisterMap::reg_count; i++) { 474 #ifdef AMD64 475 // The register map has one entry for every int (32-bit value), so 476 // 64-bit physical registers have two entries in the map, one for 477 // each half. Ignore the high halves of 64-bit registers, just like 478 // frame::oopmapreg_to_location does. 479 // 480 // [phh] FIXME: this is a temporary hack! This code *should* work 481 // correctly w/o this hack, possibly by changing RegisterMap::pd_location 482 // in frame_amd64.cpp and the values of the phantom high half registers 483 // in amd64.ad. 484 // if (VMReg::Name(i) < SharedInfo::stack0 && is_even(i)) { 485 intptr_t* src = (intptr_t*) reg_map->location(VMRegImpl::as_VMReg(i)); 486 _callee_registers[i] = src != NULL ? *src : NULL_WORD; 487 // } else { 488 // jint* src = (jint*) reg_map->location(VMReg::Name(i)); 489 // _callee_registers[i] = src != NULL ? *src : NULL_WORD; 490 // } 491 #else 492 jint* src = (jint*) reg_map->location(VMRegImpl::as_VMReg(i)); 493 _callee_registers[i] = src != NULL ? *src : NULL_WORD; 494 #endif 495 if (src == NULL) { 496 set_location_valid(i, false); 497 } else { 498 set_location_valid(i, true); 499 jint* dst = (jint*) register_location(i); 500 *dst = *src; 501 } 502 } 503 } 504 } 505 506 void vframeArray::unpack_to_stack(frame &unpack_frame, int exec_mode, int caller_actual_parameters) { 507 // stack picture 508 // unpack_frame 509 // [new interpreter frames ] (frames are skeletal but walkable) 510 // caller_frame 511 // 512 // This routine fills in the missing data for the skeletal interpreter frames 513 // in the above picture. 514 515 // Find the skeletal interpreter frames to unpack into 516 JavaThread* THREAD = JavaThread::current(); 517 RegisterMap map(THREAD, false); 518 // Get the youngest frame we will unpack (last to be unpacked) 519 frame me = unpack_frame.sender(&map); 520 int index; 521 for (index = 0; index < frames(); index++ ) { 522 *element(index)->iframe() = me; 523 // Get the caller frame (possibly skeletal) 524 me = me.sender(&map); 525 } 526 527 // Do the unpacking of interpreter frames; the frame at index 0 represents the top activation, so it has no callee 528 // Unpack the frames from the oldest (frames() -1) to the youngest (0) 529 frame* caller_frame = &me; 530 for (index = frames() - 1; index >= 0 ; index--) { 531 vframeArrayElement* elem = element(index); // caller 532 int callee_parameters, callee_locals; 533 if (index == 0) { 534 callee_parameters = callee_locals = 0; 535 } else { 536 methodHandle caller = elem->method(); 537 methodHandle callee = element(index - 1)->method(); 538 Bytecode_invoke inv(caller, elem->bci()); 539 // invokedynamic instructions don't have a class but obviously don't have a MemberName appendix. 540 // NOTE: Use machinery here that avoids resolving of any kind. 541 const bool has_member_arg = 542 !inv.is_invokedynamic() && MethodHandles::has_member_arg(inv.klass(), inv.name()); 543 callee_parameters = callee->size_of_parameters() + (has_member_arg ? 1 : 0); 544 callee_locals = callee->max_locals(); 545 } 546 elem->unpack_on_stack(caller_actual_parameters, 547 callee_parameters, 548 callee_locals, 549 caller_frame, 550 index == 0, 551 index == frames() - 1, 552 exec_mode); 553 if (index == frames() - 1) { 554 Deoptimization::unwind_callee_save_values(elem->iframe(), this); 555 } 556 caller_frame = elem->iframe(); 557 caller_actual_parameters = callee_parameters; 558 } 559 deallocate_monitor_chunks(); 560 } 561 562 void vframeArray::deallocate_monitor_chunks() { 563 JavaThread* jt = JavaThread::current(); 564 for (int index = 0; index < frames(); index++ ) { 565 element(index)->free_monitors(jt); 566 } 567 } 568 569 #ifndef PRODUCT 570 571 bool vframeArray::structural_compare(JavaThread* thread, GrowableArray<compiledVFrame*>* chunk) { 572 if (owner_thread() != thread) return false; 573 int index = 0; 574 #if 0 // FIXME can't do this comparison 575 576 // Compare only within vframe array. 577 for (deoptimizedVFrame* vf = deoptimizedVFrame::cast(vframe_at(first_index())); vf; vf = vf->deoptimized_sender_or_null()) { 578 if (index >= chunk->length() || !vf->structural_compare(chunk->at(index))) return false; 579 index++; 580 } 581 if (index != chunk->length()) return false; 582 #endif 583 584 return true; 585 } 586 587 #endif 588 589 address vframeArray::register_location(int i) const { 590 assert(0 <= i && i < RegisterMap::reg_count, "index out of bounds"); 591 return (address) & _callee_registers[i]; 592 } 593 594 595 #ifndef PRODUCT 596 597 // Printing 598 599 // Note: we cannot have print_on as const, as we allocate inside the method 600 void vframeArray::print_on_2(outputStream* st) { 601 st->print_cr(" - sp: " INTPTR_FORMAT, sp()); 602 st->print(" - thread: "); 603 Thread::current()->print(); 604 st->print_cr(" - frame size: %d", frame_size()); 605 for (int index = 0; index < frames() ; index++ ) { 606 element(index)->print(st); 607 } 608 } 609 610 void vframeArrayElement::print(outputStream* st) { 611 st->print_cr(" - interpreter_frame -> sp: " INTPTR_FORMAT, iframe()->sp()); 612 } 613 614 void vframeArray::print_value_on(outputStream* st) const { 615 st->print_cr("vframeArray [%d] ", frames()); 616 } 617 618 619 #endif