Thu, 07 Feb 2013 16:05:48 -0500
Merge
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 #else
237 // TBD: Need to implement ForceEarlyReturn for CC_INTERP (ia64)
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 }
264 // Setup the interpreter frame
266 assert(method() != NULL, "method must exist");
267 int temps = expressions()->size();
269 int locks = monitors() == NULL ? 0 : monitors()->number_of_monitors();
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);
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.
286 _frame.patch_pc(thread, pc);
288 assert (!method()->is_synchronized() || locks > 0, "synchronized methods must have monitors");
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 }
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 }
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 }
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 }
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();
396 tty->print_cr("locals size %d", locals()->size());
397 tty->print_cr("expression size %d", expressions()->size());
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
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
417 _locals = _expressions = NULL;
419 }
421 int vframeArrayElement::on_stack_size(int caller_actual_parameters,
422 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(),
430 temps + callee_parameters,
431 popframe_extra_stack_expression_els,
432 locks,
433 caller_actual_parameters,
434 callee_parameters,
435 callee_locals,
436 is_top_frame);
437 }
441 vframeArray* vframeArray::allocate(JavaThread* thread, int frame_size, GrowableArray<compiledVFrame*>* chunk,
442 RegisterMap *reg_map, frame sender, frame caller, frame self) {
444 // Allocate the vframeArray
445 vframeArray * result = (vframeArray*) AllocateHeap(sizeof(vframeArray) + // fixed part
446 sizeof(vframeArrayElement) * (chunk->length() - 1), // variable part
447 mtCompiler);
448 result->_frames = chunk->length();
449 result->_owner_thread = thread;
450 result->_sender = sender;
451 result->_caller = caller;
452 result->_original = self;
453 result->set_unroll_block(NULL); // initialize it
454 result->fill_in(thread, frame_size, chunk, reg_map);
455 return result;
456 }
458 void vframeArray::fill_in(JavaThread* thread,
459 int frame_size,
460 GrowableArray<compiledVFrame*>* chunk,
461 const RegisterMap *reg_map) {
462 // Set owner first, it is used when adding monitor chunks
464 _frame_size = frame_size;
465 for(int i = 0; i < chunk->length(); i++) {
466 element(i)->fill_in(chunk->at(i));
467 }
469 // Copy registers for callee-saved registers
470 if (reg_map != NULL) {
471 for(int i = 0; i < RegisterMap::reg_count; i++) {
472 #ifdef AMD64
473 // The register map has one entry for every int (32-bit value), so
474 // 64-bit physical registers have two entries in the map, one for
475 // each half. Ignore the high halves of 64-bit registers, just like
476 // frame::oopmapreg_to_location does.
477 //
478 // [phh] FIXME: this is a temporary hack! This code *should* work
479 // correctly w/o this hack, possibly by changing RegisterMap::pd_location
480 // in frame_amd64.cpp and the values of the phantom high half registers
481 // in amd64.ad.
482 // if (VMReg::Name(i) < SharedInfo::stack0 && is_even(i)) {
483 intptr_t* src = (intptr_t*) reg_map->location(VMRegImpl::as_VMReg(i));
484 _callee_registers[i] = src != NULL ? *src : NULL_WORD;
485 // } else {
486 // jint* src = (jint*) reg_map->location(VMReg::Name(i));
487 // _callee_registers[i] = src != NULL ? *src : NULL_WORD;
488 // }
489 #else
490 jint* src = (jint*) reg_map->location(VMRegImpl::as_VMReg(i));
491 _callee_registers[i] = src != NULL ? *src : NULL_WORD;
492 #endif
493 if (src == NULL) {
494 set_location_valid(i, false);
495 } else {
496 set_location_valid(i, true);
497 jint* dst = (jint*) register_location(i);
498 *dst = *src;
499 }
500 }
501 }
502 }
504 void vframeArray::unpack_to_stack(frame &unpack_frame, int exec_mode, int caller_actual_parameters) {
505 // stack picture
506 // unpack_frame
507 // [new interpreter frames ] (frames are skeletal but walkable)
508 // caller_frame
509 //
510 // This routine fills in the missing data for the skeletal interpreter frames
511 // in the above picture.
513 // Find the skeletal interpreter frames to unpack into
514 JavaThread* THREAD = JavaThread::current();
515 RegisterMap map(THREAD, false);
516 // Get the youngest frame we will unpack (last to be unpacked)
517 frame me = unpack_frame.sender(&map);
518 int index;
519 for (index = 0; index < frames(); index++ ) {
520 *element(index)->iframe() = me;
521 // Get the caller frame (possibly skeletal)
522 me = me.sender(&map);
523 }
525 // Do the unpacking of interpreter frames; the frame at index 0 represents the top activation, so it has no callee
526 // Unpack the frames from the oldest (frames() -1) to the youngest (0)
527 frame caller_frame = me;
528 for (index = frames() - 1; index >= 0 ; index--) {
529 vframeArrayElement* elem = element(index); // caller
530 int callee_parameters, callee_locals;
531 if (index == 0) {
532 callee_parameters = callee_locals = 0;
533 } else {
534 methodHandle caller = elem->method();
535 methodHandle callee = element(index - 1)->method();
536 Bytecode_invoke inv(caller, elem->bci());
537 // invokedynamic instructions don't have a class but obviously don't have a MemberName appendix.
538 // NOTE: Use machinery here that avoids resolving of any kind.
539 const bool has_member_arg =
540 !inv.is_invokedynamic() && MethodHandles::has_member_arg(inv.klass(), inv.name());
541 callee_parameters = callee->size_of_parameters() + (has_member_arg ? 1 : 0);
542 callee_locals = callee->max_locals();
543 }
544 elem->unpack_on_stack(caller_actual_parameters,
545 callee_parameters,
546 callee_locals,
547 &caller_frame,
548 index == 0,
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 }
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 }
566 #ifndef PRODUCT
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
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
581 return true;
582 }
584 #endif
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 }
592 #ifndef PRODUCT
594 // Printing
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 }
607 void vframeArrayElement::print(outputStream* st) {
608 st->print_cr(" - interpreter_frame -> sp: " INTPTR_FORMAT, iframe()->sp());
609 }
611 void vframeArray::print_value_on(outputStream* st) const {
612 st->print_cr("vframeArray [%d] ", frames());
613 }
616 #endif