Sun, 10 Feb 2013 22:35:38 -0800
8006430: TraceTypeProfile is a product flag while it should be a diagnostic flag
Summary: make sure all diagnostic and experimental flag kinds are checked in Flag::is_unlocked()
Reviewed-by: kvn
1 /*
2 * Copyright (c) 1997, 2012, 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 */
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
47 int vframeArrayElement:: bci(void) const { return (_bci == SynchronizationEntryBCI ? 0 : _bci); }
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 }
58 void vframeArrayElement::fill_in(compiledVFrame* vf) {
60 // Copy the information from the compiled vframe to the
61 // interpreter frame we will be creating to replace vf
63 _method = vf->method();
64 _bci = vf->raw_bci();
65 _reexecute = vf->should_reexecute();
67 int index;
69 // Get the monitors off-stack
71 GrowableArray<MonitorInfo*>* list = vf->monitors();
72 if (list->is_empty()) {
73 _monitors = NULL;
74 } else {
76 // Allocate monitor chunk
77 _monitors = new MonitorChunk(list->length());
78 vf->thread()->add_monitor_chunk(_monitors);
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 }
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.
98 // First the locals go off-stack
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
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((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 }
128 // Now the expressions off-stack
129 // Same silliness as above
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((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 }
156 int unpack_counter = 0;
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 int exec_mode) {
164 JavaThread* thread = (JavaThread*) Thread::current();
166 // Look at bci and decide on bcp and continuation pc
167 address bcp;
168 // C++ interpreter doesn't need a pc since it will figure out what to do when it
169 // begins execution
170 address pc;
171 bool use_next_mdp = false; // true if we should use the mdp associated with the next bci
172 // rather than the one associated with bcp
173 if (raw_bci() == SynchronizationEntryBCI) {
174 // We are deoptimizing while hanging in prologue code for synchronized method
175 bcp = method()->bcp_from(0); // first byte code
176 pc = Interpreter::deopt_entry(vtos, 0); // step = 0 since we don't skip current bytecode
177 } else if (should_reexecute()) { //reexecute this bytecode
178 assert(is_top_frame, "reexecute allowed only for the top frame");
179 bcp = method()->bcp_from(bci());
180 pc = Interpreter::deopt_reexecute_entry(method(), bcp);
181 } else {
182 bcp = method()->bcp_from(bci());
183 pc = Interpreter::deopt_continue_after_entry(method(), bcp, callee_parameters, is_top_frame);
184 use_next_mdp = true;
185 }
186 assert(Bytecodes::is_defined(*bcp), "must be a valid bytecode");
188 // Monitorenter and pending exceptions:
189 //
190 // For Compiler2, there should be no pending exception when deoptimizing at monitorenter
191 // because there is no safepoint at the null pointer check (it is either handled explicitly
192 // or prior to the monitorenter) and asynchronous exceptions are not made "pending" by the
193 // runtime interface for the slow case (see JRT_ENTRY_FOR_MONITORENTER). If an asynchronous
194 // exception was processed, the bytecode pointer would have to be extended one bytecode beyond
195 // the monitorenter to place it in the proper exception range.
196 //
197 // For Compiler1, deoptimization can occur while throwing a NullPointerException at monitorenter,
198 // in which case bcp should point to the monitorenter since it is within the exception's range.
200 assert(*bcp != Bytecodes::_monitorenter || is_top_frame, "a _monitorenter must be a top frame");
201 assert(thread->deopt_nmethod() != NULL, "nmethod should be known");
202 guarantee(!(thread->deopt_nmethod()->is_compiled_by_c2() &&
203 *bcp == Bytecodes::_monitorenter &&
204 exec_mode == Deoptimization::Unpack_exception),
205 "shouldn't get exception during monitorenter");
207 int popframe_preserved_args_size_in_bytes = 0;
208 int popframe_preserved_args_size_in_words = 0;
209 if (is_top_frame) {
210 JvmtiThreadState *state = thread->jvmti_thread_state();
211 if (JvmtiExport::can_pop_frame() &&
212 (thread->has_pending_popframe() || thread->popframe_forcing_deopt_reexecution())) {
213 if (thread->has_pending_popframe()) {
214 // Pop top frame after deoptimization
215 #ifndef CC_INTERP
216 pc = Interpreter::remove_activation_preserving_args_entry();
217 #else
218 // Do an uncommon trap type entry. c++ interpreter will know
219 // to pop frame and preserve the args
220 pc = Interpreter::deopt_entry(vtos, 0);
221 use_next_mdp = false;
222 #endif
223 } else {
224 // Reexecute invoke in top frame
225 pc = Interpreter::deopt_entry(vtos, 0);
226 use_next_mdp = false;
227 popframe_preserved_args_size_in_bytes = in_bytes(thread->popframe_preserved_args_size());
228 // Note: the PopFrame-related extension of the expression stack size is done in
229 // Deoptimization::fetch_unroll_info_helper
230 popframe_preserved_args_size_in_words = in_words(thread->popframe_preserved_args_size_in_words());
231 }
232 } else if (JvmtiExport::can_force_early_return() && state != NULL && state->is_earlyret_pending()) {
233 // Force early return from top frame after deoptimization
234 #ifndef CC_INTERP
235 pc = Interpreter::remove_activation_early_entry(state->earlyret_tos());
236 #endif
237 } else {
238 // Possibly override the previous pc computation of the top (youngest) frame
239 switch (exec_mode) {
240 case Deoptimization::Unpack_deopt:
241 // use what we've got
242 break;
243 case Deoptimization::Unpack_exception:
244 // exception is pending
245 pc = SharedRuntime::raw_exception_handler_for_return_address(thread, pc);
246 // [phh] We're going to end up in some handler or other, so it doesn't
247 // matter what mdp we point to. See exception_handler_for_exception()
248 // in interpreterRuntime.cpp.
249 break;
250 case Deoptimization::Unpack_uncommon_trap:
251 case Deoptimization::Unpack_reexecute:
252 // redo last byte code
253 pc = Interpreter::deopt_entry(vtos, 0);
254 use_next_mdp = false;
255 break;
256 default:
257 ShouldNotReachHere();
258 }
259 }
260 }
262 // Setup the interpreter frame
264 assert(method() != NULL, "method must exist");
265 int temps = expressions()->size();
267 int locks = monitors() == NULL ? 0 : monitors()->number_of_monitors();
269 Interpreter::layout_activation(method(),
270 temps + callee_parameters,
271 popframe_preserved_args_size_in_words,
272 locks,
273 caller_actual_parameters,
274 callee_parameters,
275 callee_locals,
276 caller,
277 iframe(),
278 is_top_frame);
280 // Update the pc in the frame object and overwrite the temporary pc
281 // we placed in the skeletal frame now that we finally know the
282 // exact interpreter address we should use.
284 _frame.patch_pc(thread, pc);
286 assert (!method()->is_synchronized() || locks > 0, "synchronized methods must have monitors");
288 BasicObjectLock* top = iframe()->interpreter_frame_monitor_begin();
289 for (int index = 0; index < locks; index++) {
290 top = iframe()->previous_monitor_in_interpreter_frame(top);
291 BasicObjectLock* src = _monitors->at(index);
292 top->set_obj(src->obj());
293 src->lock()->move_to(src->obj(), top->lock());
294 }
295 if (ProfileInterpreter) {
296 iframe()->interpreter_frame_set_mdx(0); // clear out the mdp.
297 }
298 iframe()->interpreter_frame_set_bcx((intptr_t)bcp); // cannot use bcp because frame is not initialized yet
299 if (ProfileInterpreter) {
300 MethodData* mdo = method()->method_data();
301 if (mdo != NULL) {
302 int bci = iframe()->interpreter_frame_bci();
303 if (use_next_mdp) ++bci;
304 address mdp = mdo->bci_to_dp(bci);
305 iframe()->interpreter_frame_set_mdp(mdp);
306 }
307 }
309 // Unpack expression stack
310 // If this is an intermediate frame (i.e. not top frame) then this
311 // only unpacks the part of the expression stack not used by callee
312 // as parameters. The callee parameters are unpacked as part of the
313 // callee locals.
314 int i;
315 for(i = 0; i < expressions()->size(); i++) {
316 StackValue *value = expressions()->at(i);
317 intptr_t* addr = iframe()->interpreter_frame_expression_stack_at(i);
318 switch(value->type()) {
319 case T_INT:
320 *addr = value->get_int();
321 break;
322 case T_OBJECT:
323 *addr = value->get_int(T_OBJECT);
324 break;
325 case T_CONFLICT:
326 // A dead stack slot. Initialize to null in case it is an oop.
327 *addr = NULL_WORD;
328 break;
329 default:
330 ShouldNotReachHere();
331 }
332 }
335 // Unpack the locals
336 for(i = 0; i < locals()->size(); i++) {
337 StackValue *value = locals()->at(i);
338 intptr_t* addr = iframe()->interpreter_frame_local_at(i);
339 switch(value->type()) {
340 case T_INT:
341 *addr = value->get_int();
342 break;
343 case T_OBJECT:
344 *addr = value->get_int(T_OBJECT);
345 break;
346 case T_CONFLICT:
347 // A dead location. If it is an oop then we need a NULL to prevent GC from following it
348 *addr = NULL_WORD;
349 break;
350 default:
351 ShouldNotReachHere();
352 }
353 }
355 if (is_top_frame && JvmtiExport::can_pop_frame() && thread->popframe_forcing_deopt_reexecution()) {
356 // An interpreted frame was popped but it returns to a deoptimized
357 // frame. The incoming arguments to the interpreted activation
358 // were preserved in thread-local storage by the
359 // remove_activation_preserving_args_entry in the interpreter; now
360 // we put them back into the just-unpacked interpreter frame.
361 // Note that this assumes that the locals arena grows toward lower
362 // addresses.
363 if (popframe_preserved_args_size_in_words != 0) {
364 void* saved_args = thread->popframe_preserved_args();
365 assert(saved_args != NULL, "must have been saved by interpreter");
366 #ifdef ASSERT
367 assert(popframe_preserved_args_size_in_words <=
368 iframe()->interpreter_frame_expression_stack_size()*Interpreter::stackElementWords,
369 "expression stack size should have been extended");
370 #endif // ASSERT
371 int top_element = iframe()->interpreter_frame_expression_stack_size()-1;
372 intptr_t* base;
373 if (frame::interpreter_frame_expression_stack_direction() < 0) {
374 base = iframe()->interpreter_frame_expression_stack_at(top_element);
375 } else {
376 base = iframe()->interpreter_frame_expression_stack();
377 }
378 Copy::conjoint_jbytes(saved_args,
379 base,
380 popframe_preserved_args_size_in_bytes);
381 thread->popframe_free_preserved_args();
382 }
383 }
385 #ifndef PRODUCT
386 if (TraceDeoptimization && Verbose) {
387 ttyLocker ttyl;
388 tty->print_cr("[%d Interpreted Frame]", ++unpack_counter);
389 iframe()->print_on(tty);
390 RegisterMap map(thread);
391 vframe* f = vframe::new_vframe(iframe(), &map, thread);
392 f->print();
394 tty->print_cr("locals size %d", locals()->size());
395 tty->print_cr("expression size %d", expressions()->size());
397 method()->print_value();
398 tty->cr();
399 // method()->print_codes();
400 } else if (TraceDeoptimization) {
401 tty->print(" ");
402 method()->print_value();
403 Bytecodes::Code code = Bytecodes::java_code_at(method(), bcp);
404 int bci = method()->bci_from(bcp);
405 tty->print(" - %s", Bytecodes::name(code));
406 tty->print(" @ bci %d ", bci);
407 tty->print_cr("sp = " PTR_FORMAT, iframe()->sp());
408 }
409 #endif // PRODUCT
411 // The expression stack and locals are in the resource area don't leave
412 // a dangling pointer in the vframeArray we leave around for debug
413 // purposes
415 _locals = _expressions = NULL;
417 }
419 int vframeArrayElement::on_stack_size(int caller_actual_parameters,
420 int callee_parameters,
421 int callee_locals,
422 bool is_top_frame,
423 int popframe_extra_stack_expression_els) const {
424 assert(method()->max_locals() == locals()->size(), "just checking");
425 int locks = monitors() == NULL ? 0 : monitors()->number_of_monitors();
426 int temps = expressions()->size();
427 return Interpreter::size_activation(method(),
428 temps + callee_parameters,
429 popframe_extra_stack_expression_els,
430 locks,
431 caller_actual_parameters,
432 callee_parameters,
433 callee_locals,
434 is_top_frame);
435 }
439 vframeArray* vframeArray::allocate(JavaThread* thread, int frame_size, GrowableArray<compiledVFrame*>* chunk,
440 RegisterMap *reg_map, frame sender, frame caller, frame self) {
442 // Allocate the vframeArray
443 vframeArray * result = (vframeArray*) AllocateHeap(sizeof(vframeArray) + // fixed part
444 sizeof(vframeArrayElement) * (chunk->length() - 1), // variable part
445 mtCompiler);
446 result->_frames = chunk->length();
447 result->_owner_thread = thread;
448 result->_sender = sender;
449 result->_caller = caller;
450 result->_original = self;
451 result->set_unroll_block(NULL); // initialize it
452 result->fill_in(thread, frame_size, chunk, reg_map);
453 return result;
454 }
456 void vframeArray::fill_in(JavaThread* thread,
457 int frame_size,
458 GrowableArray<compiledVFrame*>* chunk,
459 const RegisterMap *reg_map) {
460 // Set owner first, it is used when adding monitor chunks
462 _frame_size = frame_size;
463 for(int i = 0; i < chunk->length(); i++) {
464 element(i)->fill_in(chunk->at(i));
465 }
467 // Copy registers for callee-saved registers
468 if (reg_map != NULL) {
469 for(int i = 0; i < RegisterMap::reg_count; i++) {
470 #ifdef AMD64
471 // The register map has one entry for every int (32-bit value), so
472 // 64-bit physical registers have two entries in the map, one for
473 // each half. Ignore the high halves of 64-bit registers, just like
474 // frame::oopmapreg_to_location does.
475 //
476 // [phh] FIXME: this is a temporary hack! This code *should* work
477 // correctly w/o this hack, possibly by changing RegisterMap::pd_location
478 // in frame_amd64.cpp and the values of the phantom high half registers
479 // in amd64.ad.
480 // if (VMReg::Name(i) < SharedInfo::stack0 && is_even(i)) {
481 intptr_t* src = (intptr_t*) reg_map->location(VMRegImpl::as_VMReg(i));
482 _callee_registers[i] = src != NULL ? *src : NULL_WORD;
483 // } else {
484 // jint* src = (jint*) reg_map->location(VMReg::Name(i));
485 // _callee_registers[i] = src != NULL ? *src : NULL_WORD;
486 // }
487 #else
488 jint* src = (jint*) reg_map->location(VMRegImpl::as_VMReg(i));
489 _callee_registers[i] = src != NULL ? *src : NULL_WORD;
490 #endif
491 if (src == NULL) {
492 set_location_valid(i, false);
493 } else {
494 set_location_valid(i, true);
495 jint* dst = (jint*) register_location(i);
496 *dst = *src;
497 }
498 }
499 }
500 }
502 void vframeArray::unpack_to_stack(frame &unpack_frame, int exec_mode, int caller_actual_parameters) {
503 // stack picture
504 // unpack_frame
505 // [new interpreter frames ] (frames are skeletal but walkable)
506 // caller_frame
507 //
508 // This routine fills in the missing data for the skeletal interpreter frames
509 // in the above picture.
511 // Find the skeletal interpreter frames to unpack into
512 JavaThread* THREAD = JavaThread::current();
513 RegisterMap map(THREAD, false);
514 // Get the youngest frame we will unpack (last to be unpacked)
515 frame me = unpack_frame.sender(&map);
516 int index;
517 for (index = 0; index < frames(); index++ ) {
518 *element(index)->iframe() = me;
519 // Get the caller frame (possibly skeletal)
520 me = me.sender(&map);
521 }
523 // Do the unpacking of interpreter frames; the frame at index 0 represents the top activation, so it has no callee
524 // Unpack the frames from the oldest (frames() -1) to the youngest (0)
525 frame caller_frame = me;
526 for (index = frames() - 1; index >= 0 ; index--) {
527 vframeArrayElement* elem = element(index); // caller
528 int callee_parameters, callee_locals;
529 if (index == 0) {
530 callee_parameters = callee_locals = 0;
531 } else {
532 methodHandle caller = elem->method();
533 methodHandle callee = element(index - 1)->method();
534 Bytecode_invoke inv(caller, elem->bci());
535 // invokedynamic instructions don't have a class but obviously don't have a MemberName appendix.
536 // NOTE: Use machinery here that avoids resolving of any kind.
537 const bool has_member_arg =
538 !inv.is_invokedynamic() && MethodHandles::has_member_arg(inv.klass(), inv.name());
539 callee_parameters = callee->size_of_parameters() + (has_member_arg ? 1 : 0);
540 callee_locals = callee->max_locals();
541 }
542 elem->unpack_on_stack(caller_actual_parameters,
543 callee_parameters,
544 callee_locals,
545 &caller_frame,
546 index == 0,
547 exec_mode);
548 if (index == frames() - 1) {
549 Deoptimization::unwind_callee_save_values(elem->iframe(), this);
550 }
551 caller_frame = *elem->iframe();
552 caller_actual_parameters = callee_parameters;
553 }
554 deallocate_monitor_chunks();
555 }
557 void vframeArray::deallocate_monitor_chunks() {
558 JavaThread* jt = JavaThread::current();
559 for (int index = 0; index < frames(); index++ ) {
560 element(index)->free_monitors(jt);
561 }
562 }
564 #ifndef PRODUCT
566 bool vframeArray::structural_compare(JavaThread* thread, GrowableArray<compiledVFrame*>* chunk) {
567 if (owner_thread() != thread) return false;
568 int index = 0;
569 #if 0 // FIXME can't do this comparison
571 // Compare only within vframe array.
572 for (deoptimizedVFrame* vf = deoptimizedVFrame::cast(vframe_at(first_index())); vf; vf = vf->deoptimized_sender_or_null()) {
573 if (index >= chunk->length() || !vf->structural_compare(chunk->at(index))) return false;
574 index++;
575 }
576 if (index != chunk->length()) return false;
577 #endif
579 return true;
580 }
582 #endif
584 address vframeArray::register_location(int i) const {
585 assert(0 <= i && i < RegisterMap::reg_count, "index out of bounds");
586 return (address) & _callee_registers[i];
587 }
590 #ifndef PRODUCT
592 // Printing
594 // Note: we cannot have print_on as const, as we allocate inside the method
595 void vframeArray::print_on_2(outputStream* st) {
596 st->print_cr(" - sp: " INTPTR_FORMAT, sp());
597 st->print(" - thread: ");
598 Thread::current()->print();
599 st->print_cr(" - frame size: %d", frame_size());
600 for (int index = 0; index < frames() ; index++ ) {
601 element(index)->print(st);
602 }
603 }
605 void vframeArrayElement::print(outputStream* st) {
606 st->print_cr(" - interpreter_frame -> sp: " INTPTR_FORMAT, iframe()->sp());
607 }
609 void vframeArray::print_value_on(outputStream* st) const {
610 st->print_cr("vframeArray [%d] ", frames());
611 }
614 #endif