1 /*
2 * Copyright (c) 1997, 2010, 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/interpreter.hpp"
28 #include "memory/allocation.inline.hpp"
29 #include "memory/resourceArea.hpp"
30 #include "memory/universe.inline.hpp"
31 #include "oops/methodDataOop.hpp"
32 #include "oops/oop.inline.hpp"
33 #include "prims/jvmtiThreadState.hpp"
34 #include "runtime/handles.inline.hpp"
35 #include "runtime/monitorChunk.hpp"
36 #include "runtime/sharedRuntime.hpp"
37 #include "runtime/vframe.hpp"
38 #include "runtime/vframeArray.hpp"
39 #include "runtime/vframe_hp.hpp"
40 #include "utilities/events.hpp"
41 #ifdef COMPILER2
42 #include "opto/runtime.hpp"
43 #endif
44
45
46 int vframeArrayElement:: bci(void) const { return (_bci == SynchronizationEntryBCI ? 0 : _bci); }
47
48 void vframeArrayElement::free_monitors(JavaThread* jt) {
49 if (_monitors != NULL) {
50 MonitorChunk* chunk = _monitors;
51 _monitors = NULL;
52 jt->remove_monitor_chunk(chunk);
53 delete chunk;
54 }
55 }
56
57 void vframeArrayElement::fill_in(compiledVFrame* vf) {
58
59 // Copy the information from the compiled vframe to the
60 // interpreter frame we will be creating to replace vf
61
62 _method = vf->method();
63 _bci = vf->raw_bci();
64 _reexecute = vf->should_reexecute();
65
66 int index;
67
68 // Get the monitors off-stack
69
70 GrowableArray<MonitorInfo*>* list = vf->monitors();
71 if (list->is_empty()) {
72 _monitors = NULL;
73 } else {
74
75 // Allocate monitor chunk
76 _monitors = new MonitorChunk(list->length());
77 vf->thread()->add_monitor_chunk(_monitors);
78
79 // Migrate the BasicLocks from the stack to the monitor chunk
80 for (index = 0; index < list->length(); index++) {
81 MonitorInfo* monitor = list->at(index);
82 assert(!monitor->owner_is_scalar_replaced(), "object should be reallocated already");
83 assert(monitor->owner() == NULL || (!monitor->owner()->is_unlocked() && !monitor->owner()->has_bias_pattern()), "object must be null or locked, and unbiased");
84 BasicObjectLock* dest = _monitors->at(index);
85 dest->set_obj(monitor->owner());
86 monitor->lock()->move_to(monitor->owner(), dest->lock());
87 }
88 }
89
90 // Convert the vframe locals and expressions to off stack
91 // values. Because we will not gc all oops can be converted to
92 // intptr_t (i.e. a stack slot) and we are fine. This is
93 // good since we are inside a HandleMark and the oops in our
94 // collection would go away between packing them here and
95 // unpacking them in unpack_on_stack.
96
97 // First the locals go off-stack
98
99 // FIXME this seems silly it creates a StackValueCollection
100 // in order to get the size to then copy them and
101 // convert the types to intptr_t size slots. Seems like it
102 // could do it in place... Still uses less memory than the
103 // old way though
104
105 StackValueCollection *locs = vf->locals();
106 _locals = new StackValueCollection(locs->size());
107 for(index = 0; index < locs->size(); index++) {
108 StackValue* value = locs->at(index);
109 switch(value->type()) {
110 case T_OBJECT:
111 assert(!value->obj_is_scalar_replaced(), "object should be reallocated already");
112 // preserve object type
113 _locals->add( new StackValue((intptr_t) (value->get_obj()()), T_OBJECT ));
114 break;
115 case T_CONFLICT:
116 // A dead local. Will be initialized to null/zero.
117 _locals->add( new StackValue());
118 break;
119 case T_INT:
120 _locals->add( new StackValue(value->get_int()));
121 break;
122 default:
123 ShouldNotReachHere();
124 }
125 }
126
127 // Now the expressions off-stack
128 // Same silliness as above
129
130 StackValueCollection *exprs = vf->expressions();
131 _expressions = new StackValueCollection(exprs->size());
132 for(index = 0; index < exprs->size(); index++) {
133 StackValue* value = exprs->at(index);
134 switch(value->type()) {
135 case T_OBJECT:
136 assert(!value->obj_is_scalar_replaced(), "object should be reallocated already");
137 // preserve object type
138 _expressions->add( new StackValue((intptr_t) (value->get_obj()()), T_OBJECT ));
139 break;
140 case T_CONFLICT:
141 // A dead stack element. Will be initialized to null/zero.
142 // This can occur when the compiler emits a state in which stack
143 // elements are known to be dead (because of an imminent exception).
144 _expressions->add( new StackValue());
145 break;
146 case T_INT:
147 _expressions->add( new StackValue(value->get_int()));
148 break;
149 default:
150 ShouldNotReachHere();
151 }
152 }
153 }
154
155 int unpack_counter = 0;
156
157 void vframeArrayElement::unpack_on_stack(int callee_parameters,
158 int callee_locals,
159 frame* caller,
160 bool is_top_frame,
161 int exec_mode) {
162 JavaThread* thread = (JavaThread*) Thread::current();
163
164 // Look at bci and decide on bcp and continuation pc
165 address bcp;
166 // C++ interpreter doesn't need a pc since it will figure out what to do when it
167 // begins execution
168 address pc;
169 bool use_next_mdp = false; // true if we should use the mdp associated with the next bci
170 // rather than the one associated with bcp
171 if (raw_bci() == SynchronizationEntryBCI) {
172 // We are deoptimizing while hanging in prologue code for synchronized method
173 bcp = method()->bcp_from(0); // first byte code
174 pc = Interpreter::deopt_entry(vtos, 0); // step = 0 since we don't skip current bytecode
175 } else if (should_reexecute()) { //reexecute this bytecode
176 assert(is_top_frame, "reexecute allowed only for the top frame");
177 bcp = method()->bcp_from(bci());
178 pc = Interpreter::deopt_reexecute_entry(method(), bcp);
179 } else {
180 bcp = method()->bcp_from(bci());
181 pc = Interpreter::deopt_continue_after_entry(method(), bcp, callee_parameters, is_top_frame);
182 use_next_mdp = true;
183 }
184 assert(Bytecodes::is_defined(*bcp), "must be a valid bytecode");
185
186 // Monitorenter and pending exceptions:
187 //
188 // For Compiler2, there should be no pending exception when deoptimizing at monitorenter
189 // because there is no safepoint at the null pointer check (it is either handled explicitly
190 // or prior to the monitorenter) and asynchronous exceptions are not made "pending" by the
191 // runtime interface for the slow case (see JRT_ENTRY_FOR_MONITORENTER). If an asynchronous
192 // exception was processed, the bytecode pointer would have to be extended one bytecode beyond
193 // the monitorenter to place it in the proper exception range.
194 //
195 // For Compiler1, deoptimization can occur while throwing a NullPointerException at monitorenter,
196 // in which case bcp should point to the monitorenter since it is within the exception's range.
197
198 assert(*bcp != Bytecodes::_monitorenter || is_top_frame, "a _monitorenter must be a top frame");
199 // TIERED Must know the compiler of the deoptee QQQ
200 COMPILER2_PRESENT(guarantee(*bcp != Bytecodes::_monitorenter || exec_mode != Deoptimization::Unpack_exception,
201 "shouldn't get exception during monitorenter");)
202
203 int popframe_preserved_args_size_in_bytes = 0;
204 int popframe_preserved_args_size_in_words = 0;
205 if (is_top_frame) {
206 JvmtiThreadState *state = thread->jvmti_thread_state();
207 if (JvmtiExport::can_pop_frame() &&
208 (thread->has_pending_popframe() || thread->popframe_forcing_deopt_reexecution())) {
209 if (thread->has_pending_popframe()) {
210 // Pop top frame after deoptimization
211 #ifndef CC_INTERP
212 pc = Interpreter::remove_activation_preserving_args_entry();
213 #else
214 // Do an uncommon trap type entry. c++ interpreter will know
215 // to pop frame and preserve the args
216 pc = Interpreter::deopt_entry(vtos, 0);
217 use_next_mdp = false;
218 #endif
219 } else {
220 // Reexecute invoke in top frame
221 pc = Interpreter::deopt_entry(vtos, 0);
222 use_next_mdp = false;
223 popframe_preserved_args_size_in_bytes = in_bytes(thread->popframe_preserved_args_size());
224 // Note: the PopFrame-related extension of the expression stack size is done in
225 // Deoptimization::fetch_unroll_info_helper
226 popframe_preserved_args_size_in_words = in_words(thread->popframe_preserved_args_size_in_words());
227 }
228 } else if (JvmtiExport::can_force_early_return() && state != NULL && state->is_earlyret_pending()) {
229 // Force early return from top frame after deoptimization
230 #ifndef CC_INTERP
231 pc = Interpreter::remove_activation_early_entry(state->earlyret_tos());
232 #else
233 // TBD: Need to implement ForceEarlyReturn for CC_INTERP (ia64)
234 #endif
235 } else {
236 // Possibly override the previous pc computation of the top (youngest) frame
237 switch (exec_mode) {
238 case Deoptimization::Unpack_deopt:
239 // use what we've got
240 break;
241 case Deoptimization::Unpack_exception:
242 // exception is pending
243 pc = SharedRuntime::raw_exception_handler_for_return_address(thread, pc);
244 // [phh] We're going to end up in some handler or other, so it doesn't
245 // matter what mdp we point to. See exception_handler_for_exception()
246 // in interpreterRuntime.cpp.
247 break;
248 case Deoptimization::Unpack_uncommon_trap:
249 case Deoptimization::Unpack_reexecute:
250 // redo last byte code
251 pc = Interpreter::deopt_entry(vtos, 0);
252 use_next_mdp = false;
253 break;
254 default:
255 ShouldNotReachHere();
256 }
257 }
258 }
259
260 // Setup the interpreter frame
261
262 assert(method() != NULL, "method must exist");
263 int temps = expressions()->size();
264
265 int locks = monitors() == NULL ? 0 : monitors()->number_of_monitors();
266
267 Interpreter::layout_activation(method(),
268 temps + callee_parameters,
269 popframe_preserved_args_size_in_words,
270 locks,
271 callee_parameters,
272 callee_locals,
273 caller,
274 iframe(),
275 is_top_frame);
276
277 // Update the pc in the frame object and overwrite the temporary pc
278 // we placed in the skeletal frame now that we finally know the
279 // exact interpreter address we should use.
280
281 _frame.patch_pc(thread, pc);
282
283 assert (!method()->is_synchronized() || locks > 0, "synchronized methods must have monitors");
284
285 BasicObjectLock* top = iframe()->interpreter_frame_monitor_begin();
286 for (int index = 0; index < locks; index++) {
287 top = iframe()->previous_monitor_in_interpreter_frame(top);
288 BasicObjectLock* src = _monitors->at(index);
289 top->set_obj(src->obj());
290 src->lock()->move_to(src->obj(), top->lock());
291 }
292 if (ProfileInterpreter) {
293 iframe()->interpreter_frame_set_mdx(0); // clear out the mdp.
294 }
295 iframe()->interpreter_frame_set_bcx((intptr_t)bcp); // cannot use bcp because frame is not initialized yet
296 if (ProfileInterpreter) {
297 methodDataOop mdo = method()->method_data();
298 if (mdo != NULL) {
299 int bci = iframe()->interpreter_frame_bci();
300 if (use_next_mdp) ++bci;
301 address mdp = mdo->bci_to_dp(bci);
302 iframe()->interpreter_frame_set_mdp(mdp);
303 }
304 }
305
306 // Unpack expression stack
307 // If this is an intermediate frame (i.e. not top frame) then this
308 // only unpacks the part of the expression stack not used by callee
309 // as parameters. The callee parameters are unpacked as part of the
310 // callee locals.
311 int i;
312 for(i = 0; i < expressions()->size(); i++) {
313 StackValue *value = expressions()->at(i);
314 intptr_t* addr = iframe()->interpreter_frame_expression_stack_at(i);
315 switch(value->type()) {
316 case T_INT:
317 *addr = value->get_int();
318 break;
319 case T_OBJECT:
320 *addr = value->get_int(T_OBJECT);
321 break;
322 case T_CONFLICT:
323 // A dead stack slot. Initialize to null in case it is an oop.
324 *addr = NULL_WORD;
325 break;
326 default:
327 ShouldNotReachHere();
328 }
329 }
330
331
332 // Unpack the locals
333 for(i = 0; i < locals()->size(); i++) {
334 StackValue *value = locals()->at(i);
335 intptr_t* addr = iframe()->interpreter_frame_local_at(i);
336 switch(value->type()) {
337 case T_INT:
338 *addr = value->get_int();
339 break;
340 case T_OBJECT:
341 *addr = value->get_int(T_OBJECT);
342 break;
343 case T_CONFLICT:
344 // A dead location. If it is an oop then we need a NULL to prevent GC from following it
345 *addr = NULL_WORD;
346 break;
347 default:
348 ShouldNotReachHere();
349 }
350 }
351
352 if (is_top_frame && JvmtiExport::can_pop_frame() && thread->popframe_forcing_deopt_reexecution()) {
353 // An interpreted frame was popped but it returns to a deoptimized
354 // frame. The incoming arguments to the interpreted activation
355 // were preserved in thread-local storage by the
356 // remove_activation_preserving_args_entry in the interpreter; now
357 // we put them back into the just-unpacked interpreter frame.
358 // Note that this assumes that the locals arena grows toward lower
359 // addresses.
360 if (popframe_preserved_args_size_in_words != 0) {
361 void* saved_args = thread->popframe_preserved_args();
362 assert(saved_args != NULL, "must have been saved by interpreter");
363 #ifdef ASSERT
364 assert(popframe_preserved_args_size_in_words <=
365 iframe()->interpreter_frame_expression_stack_size()*Interpreter::stackElementWords,
366 "expression stack size should have been extended");
367 #endif // ASSERT
368 int top_element = iframe()->interpreter_frame_expression_stack_size()-1;
369 intptr_t* base;
370 if (frame::interpreter_frame_expression_stack_direction() < 0) {
371 base = iframe()->interpreter_frame_expression_stack_at(top_element);
372 } else {
373 base = iframe()->interpreter_frame_expression_stack();
374 }
375 Copy::conjoint_jbytes(saved_args,
376 base,
377 popframe_preserved_args_size_in_bytes);
378 thread->popframe_free_preserved_args();
379 }
380 }
381
382 #ifndef PRODUCT
383 if (TraceDeoptimization && Verbose) {
384 ttyLocker ttyl;
385 tty->print_cr("[%d Interpreted Frame]", ++unpack_counter);
386 iframe()->print_on(tty);
387 RegisterMap map(thread);
388 vframe* f = vframe::new_vframe(iframe(), &map, thread);
389 f->print();
390
391 tty->print_cr("locals size %d", locals()->size());
392 tty->print_cr("expression size %d", expressions()->size());
393
394 method()->print_value();
395 tty->cr();
396 // method()->print_codes();
397 } else if (TraceDeoptimization) {
398 tty->print(" ");
399 method()->print_value();
400 Bytecodes::Code code = Bytecodes::java_code_at(bcp);
401 int bci = method()->bci_from(bcp);
402 tty->print(" - %s", Bytecodes::name(code));
403 tty->print(" @ bci %d ", bci);
404 tty->print_cr("sp = " PTR_FORMAT, iframe()->sp());
405 }
406 #endif // PRODUCT
407
408 // The expression stack and locals are in the resource area don't leave
409 // a dangling pointer in the vframeArray we leave around for debug
410 // purposes
411
412 _locals = _expressions = NULL;
413
414 }
415
416 int vframeArrayElement::on_stack_size(int callee_parameters,
417 int callee_locals,
418 bool is_top_frame,
419 int popframe_extra_stack_expression_els) const {
420 assert(method()->max_locals() == locals()->size(), "just checking");
421 int locks = monitors() == NULL ? 0 : monitors()->number_of_monitors();
422 int temps = expressions()->size();
423 return Interpreter::size_activation(method(),
424 temps + callee_parameters,
425 popframe_extra_stack_expression_els,
426 locks,
427 callee_parameters,
428 callee_locals,
429 is_top_frame);
430 }
431
432
433
434 vframeArray* vframeArray::allocate(JavaThread* thread, int frame_size, GrowableArray<compiledVFrame*>* chunk,
435 RegisterMap *reg_map, frame sender, frame caller, frame self) {
436
437 // Allocate the vframeArray
438 vframeArray * result = (vframeArray*) AllocateHeap(sizeof(vframeArray) + // fixed part
439 sizeof(vframeArrayElement) * (chunk->length() - 1), // variable part
440 "vframeArray::allocate");
441 result->_frames = chunk->length();
442 result->_owner_thread = thread;
443 result->_sender = sender;
444 result->_caller = caller;
445 result->_original = self;
446 result->set_unroll_block(NULL); // initialize it
447 result->fill_in(thread, frame_size, chunk, reg_map);
448 return result;
449 }
450
451 void vframeArray::fill_in(JavaThread* thread,
452 int frame_size,
453 GrowableArray<compiledVFrame*>* chunk,
454 const RegisterMap *reg_map) {
455 // Set owner first, it is used when adding monitor chunks
456
457 _frame_size = frame_size;
458 for(int i = 0; i < chunk->length(); i++) {
459 element(i)->fill_in(chunk->at(i));
460 }
461
462 // Copy registers for callee-saved registers
463 if (reg_map != NULL) {
464 for(int i = 0; i < RegisterMap::reg_count; i++) {
465 #ifdef AMD64
466 // The register map has one entry for every int (32-bit value), so
467 // 64-bit physical registers have two entries in the map, one for
468 // each half. Ignore the high halves of 64-bit registers, just like
469 // frame::oopmapreg_to_location does.
470 //
471 // [phh] FIXME: this is a temporary hack! This code *should* work
472 // correctly w/o this hack, possibly by changing RegisterMap::pd_location
473 // in frame_amd64.cpp and the values of the phantom high half registers
474 // in amd64.ad.
475 // if (VMReg::Name(i) < SharedInfo::stack0 && is_even(i)) {
476 intptr_t* src = (intptr_t*) reg_map->location(VMRegImpl::as_VMReg(i));
477 _callee_registers[i] = src != NULL ? *src : NULL_WORD;
478 // } else {
479 // jint* src = (jint*) reg_map->location(VMReg::Name(i));
480 // _callee_registers[i] = src != NULL ? *src : NULL_WORD;
481 // }
482 #else
483 jint* src = (jint*) reg_map->location(VMRegImpl::as_VMReg(i));
484 _callee_registers[i] = src != NULL ? *src : NULL_WORD;
485 #endif
486 if (src == NULL) {
487 set_location_valid(i, false);
488 } else {
489 set_location_valid(i, true);
490 jint* dst = (jint*) register_location(i);
491 *dst = *src;
492 }
493 }
494 }
495 }
496
497 void vframeArray::unpack_to_stack(frame &unpack_frame, int exec_mode) {
498 // stack picture
499 // unpack_frame
500 // [new interpreter frames ] (frames are skeletal but walkable)
501 // caller_frame
502 //
503 // This routine fills in the missing data for the skeletal interpreter frames
504 // in the above picture.
505
506 // Find the skeletal interpreter frames to unpack into
507 RegisterMap map(JavaThread::current(), false);
508 // Get the youngest frame we will unpack (last to be unpacked)
509 frame me = unpack_frame.sender(&map);
510 int index;
511 for (index = 0; index < frames(); index++ ) {
512 *element(index)->iframe() = me;
513 // Get the caller frame (possibly skeletal)
514 me = me.sender(&map);
515 }
516
517 frame caller_frame = me;
518
519 // Do the unpacking of interpreter frames; the frame at index 0 represents the top activation, so it has no callee
520
521 // Unpack the frames from the oldest (frames() -1) to the youngest (0)
522
523 for (index = frames() - 1; index >= 0 ; index--) {
524 int callee_parameters = index == 0 ? 0 : element(index-1)->method()->size_of_parameters();
525 int callee_locals = index == 0 ? 0 : element(index-1)->method()->max_locals();
526 element(index)->unpack_on_stack(callee_parameters,
527 callee_locals,
528 &caller_frame,
529 index == 0,
530 exec_mode);
531 if (index == frames() - 1) {
532 Deoptimization::unwind_callee_save_values(element(index)->iframe(), this);
533 }
534 caller_frame = *element(index)->iframe();
535 }
536
537
538 deallocate_monitor_chunks();
539 }
540
541 void vframeArray::deallocate_monitor_chunks() {
542 JavaThread* jt = JavaThread::current();
543 for (int index = 0; index < frames(); index++ ) {
544 element(index)->free_monitors(jt);
545 }
546 }
547
548 #ifndef PRODUCT
549
550 bool vframeArray::structural_compare(JavaThread* thread, GrowableArray<compiledVFrame*>* chunk) {
551 if (owner_thread() != thread) return false;
552 int index = 0;
553 #if 0 // FIXME can't do this comparison
554
555 // Compare only within vframe array.
556 for (deoptimizedVFrame* vf = deoptimizedVFrame::cast(vframe_at(first_index())); vf; vf = vf->deoptimized_sender_or_null()) {
557 if (index >= chunk->length() || !vf->structural_compare(chunk->at(index))) return false;
558 index++;
559 }
560 if (index != chunk->length()) return false;
561 #endif
562
563 return true;
564 }
565
566 #endif
567
568 address vframeArray::register_location(int i) const {
569 assert(0 <= i && i < RegisterMap::reg_count, "index out of bounds");
570 return (address) & _callee_registers[i];
571 }
572
573
574 #ifndef PRODUCT
575
576 // Printing
577
578 // Note: we cannot have print_on as const, as we allocate inside the method
579 void vframeArray::print_on_2(outputStream* st) {
580 st->print_cr(" - sp: " INTPTR_FORMAT, sp());
581 st->print(" - thread: ");
582 Thread::current()->print();
583 st->print_cr(" - frame size: %d", frame_size());
584 for (int index = 0; index < frames() ; index++ ) {
585 element(index)->print(st);
586 }
587 }
588
589 void vframeArrayElement::print(outputStream* st) {
590 st->print_cr(" - interpreter_frame -> sp: " INTPTR_FORMAT, iframe()->sp());
591 }
592
593 void vframeArray::print_value_on(outputStream* st) const {
594 st->print_cr("vframeArray [%d] ", frames());
595 }
596
597
598 #endif