Thu, 27 May 2010 18:01:56 -0700
6916623: Align object to 16 bytes to use Compressed Oops with java heap up to 64Gb
Summary: Added new product ObjectAlignmentInBytes flag to control object alignment.
Reviewed-by: twisti, ysr, iveresov
1 /*
2 * Copyright 1997-2010 Sun Microsystems, Inc. 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 Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
20 * CA 95054 USA or visit www.sun.com if you need additional information or
21 * have any questions.
22 *
23 */
25 # include "incls/_precompiled.incl"
26 # include "incls/_vframeArray.cpp.incl"
29 int vframeArrayElement:: bci(void) const { return (_bci == SynchronizationEntryBCI ? 0 : _bci); }
31 void vframeArrayElement::free_monitors(JavaThread* jt) {
32 if (_monitors != NULL) {
33 MonitorChunk* chunk = _monitors;
34 _monitors = NULL;
35 jt->remove_monitor_chunk(chunk);
36 delete chunk;
37 }
38 }
40 void vframeArrayElement::fill_in(compiledVFrame* vf) {
42 // Copy the information from the compiled vframe to the
43 // interpreter frame we will be creating to replace vf
45 _method = vf->method();
46 _bci = vf->raw_bci();
47 _reexecute = vf->should_reexecute();
49 int index;
51 // Get the monitors off-stack
53 GrowableArray<MonitorInfo*>* list = vf->monitors();
54 if (list->is_empty()) {
55 _monitors = NULL;
56 } else {
58 // Allocate monitor chunk
59 _monitors = new MonitorChunk(list->length());
60 vf->thread()->add_monitor_chunk(_monitors);
62 // Migrate the BasicLocks from the stack to the monitor chunk
63 for (index = 0; index < list->length(); index++) {
64 MonitorInfo* monitor = list->at(index);
65 assert(!monitor->owner_is_scalar_replaced(), "object should be reallocated already");
66 assert(monitor->owner() == NULL || (!monitor->owner()->is_unlocked() && !monitor->owner()->has_bias_pattern()), "object must be null or locked, and unbiased");
67 BasicObjectLock* dest = _monitors->at(index);
68 dest->set_obj(monitor->owner());
69 monitor->lock()->move_to(monitor->owner(), dest->lock());
70 }
71 }
73 // Convert the vframe locals and expressions to off stack
74 // values. Because we will not gc all oops can be converted to
75 // intptr_t (i.e. a stack slot) and we are fine. This is
76 // good since we are inside a HandleMark and the oops in our
77 // collection would go away between packing them here and
78 // unpacking them in unpack_on_stack.
80 // First the locals go off-stack
82 // FIXME this seems silly it creates a StackValueCollection
83 // in order to get the size to then copy them and
84 // convert the types to intptr_t size slots. Seems like it
85 // could do it in place... Still uses less memory than the
86 // old way though
88 StackValueCollection *locs = vf->locals();
89 _locals = new StackValueCollection(locs->size());
90 for(index = 0; index < locs->size(); index++) {
91 StackValue* value = locs->at(index);
92 switch(value->type()) {
93 case T_OBJECT:
94 assert(!value->obj_is_scalar_replaced(), "object should be reallocated already");
95 // preserve object type
96 _locals->add( new StackValue((intptr_t) (value->get_obj()()), T_OBJECT ));
97 break;
98 case T_CONFLICT:
99 // A dead local. Will be initialized to null/zero.
100 _locals->add( new StackValue());
101 break;
102 case T_INT:
103 _locals->add( new StackValue(value->get_int()));
104 break;
105 default:
106 ShouldNotReachHere();
107 }
108 }
110 // Now the expressions off-stack
111 // Same silliness as above
113 StackValueCollection *exprs = vf->expressions();
114 _expressions = new StackValueCollection(exprs->size());
115 for(index = 0; index < exprs->size(); index++) {
116 StackValue* value = exprs->at(index);
117 switch(value->type()) {
118 case T_OBJECT:
119 assert(!value->obj_is_scalar_replaced(), "object should be reallocated already");
120 // preserve object type
121 _expressions->add( new StackValue((intptr_t) (value->get_obj()()), T_OBJECT ));
122 break;
123 case T_CONFLICT:
124 // A dead stack element. Will be initialized to null/zero.
125 // This can occur when the compiler emits a state in which stack
126 // elements are known to be dead (because of an imminent exception).
127 _expressions->add( new StackValue());
128 break;
129 case T_INT:
130 _expressions->add( new StackValue(value->get_int()));
131 break;
132 default:
133 ShouldNotReachHere();
134 }
135 }
136 }
138 int unpack_counter = 0;
140 void vframeArrayElement::unpack_on_stack(int callee_parameters,
141 int callee_locals,
142 frame* caller,
143 bool is_top_frame,
144 int exec_mode) {
145 JavaThread* thread = (JavaThread*) Thread::current();
147 // Look at bci and decide on bcp and continuation pc
148 address bcp;
149 // C++ interpreter doesn't need a pc since it will figure out what to do when it
150 // begins execution
151 address pc;
152 bool use_next_mdp = false; // true if we should use the mdp associated with the next bci
153 // rather than the one associated with bcp
154 if (raw_bci() == SynchronizationEntryBCI) {
155 // We are deoptimizing while hanging in prologue code for synchronized method
156 bcp = method()->bcp_from(0); // first byte code
157 pc = Interpreter::deopt_entry(vtos, 0); // step = 0 since we don't skip current bytecode
158 } else if (should_reexecute()) { //reexecute this bytecode
159 assert(is_top_frame, "reexecute allowed only for the top frame");
160 bcp = method()->bcp_from(bci());
161 pc = Interpreter::deopt_reexecute_entry(method(), bcp);
162 } else {
163 bcp = method()->bcp_from(bci());
164 pc = Interpreter::deopt_continue_after_entry(method(), bcp, callee_parameters, is_top_frame);
165 use_next_mdp = true;
166 }
167 assert(Bytecodes::is_defined(*bcp), "must be a valid bytecode");
169 // Monitorenter and pending exceptions:
170 //
171 // For Compiler2, there should be no pending exception when deoptimizing at monitorenter
172 // because there is no safepoint at the null pointer check (it is either handled explicitly
173 // or prior to the monitorenter) and asynchronous exceptions are not made "pending" by the
174 // runtime interface for the slow case (see JRT_ENTRY_FOR_MONITORENTER). If an asynchronous
175 // exception was processed, the bytecode pointer would have to be extended one bytecode beyond
176 // the monitorenter to place it in the proper exception range.
177 //
178 // For Compiler1, deoptimization can occur while throwing a NullPointerException at monitorenter,
179 // in which case bcp should point to the monitorenter since it is within the exception's range.
181 assert(*bcp != Bytecodes::_monitorenter || is_top_frame, "a _monitorenter must be a top frame");
182 // TIERED Must know the compiler of the deoptee QQQ
183 COMPILER2_PRESENT(guarantee(*bcp != Bytecodes::_monitorenter || exec_mode != Deoptimization::Unpack_exception,
184 "shouldn't get exception during monitorenter");)
186 int popframe_preserved_args_size_in_bytes = 0;
187 int popframe_preserved_args_size_in_words = 0;
188 if (is_top_frame) {
189 JvmtiThreadState *state = thread->jvmti_thread_state();
190 if (JvmtiExport::can_pop_frame() &&
191 (thread->has_pending_popframe() || thread->popframe_forcing_deopt_reexecution())) {
192 if (thread->has_pending_popframe()) {
193 // Pop top frame after deoptimization
194 #ifndef CC_INTERP
195 pc = Interpreter::remove_activation_preserving_args_entry();
196 #else
197 // Do an uncommon trap type entry. c++ interpreter will know
198 // to pop frame and preserve the args
199 pc = Interpreter::deopt_entry(vtos, 0);
200 use_next_mdp = false;
201 #endif
202 } else {
203 // Reexecute invoke in top frame
204 pc = Interpreter::deopt_entry(vtos, 0);
205 use_next_mdp = false;
206 popframe_preserved_args_size_in_bytes = in_bytes(thread->popframe_preserved_args_size());
207 // Note: the PopFrame-related extension of the expression stack size is done in
208 // Deoptimization::fetch_unroll_info_helper
209 popframe_preserved_args_size_in_words = in_words(thread->popframe_preserved_args_size_in_words());
210 }
211 } else if (JvmtiExport::can_force_early_return() && state != NULL && state->is_earlyret_pending()) {
212 // Force early return from top frame after deoptimization
213 #ifndef CC_INTERP
214 pc = Interpreter::remove_activation_early_entry(state->earlyret_tos());
215 #else
216 // TBD: Need to implement ForceEarlyReturn for CC_INTERP (ia64)
217 #endif
218 } else {
219 // Possibly override the previous pc computation of the top (youngest) frame
220 switch (exec_mode) {
221 case Deoptimization::Unpack_deopt:
222 // use what we've got
223 break;
224 case Deoptimization::Unpack_exception:
225 // exception is pending
226 pc = SharedRuntime::raw_exception_handler_for_return_address(thread, pc);
227 // [phh] We're going to end up in some handler or other, so it doesn't
228 // matter what mdp we point to. See exception_handler_for_exception()
229 // in interpreterRuntime.cpp.
230 break;
231 case Deoptimization::Unpack_uncommon_trap:
232 case Deoptimization::Unpack_reexecute:
233 // redo last byte code
234 pc = Interpreter::deopt_entry(vtos, 0);
235 use_next_mdp = false;
236 break;
237 default:
238 ShouldNotReachHere();
239 }
240 }
241 }
243 // Setup the interpreter frame
245 assert(method() != NULL, "method must exist");
246 int temps = expressions()->size();
248 int locks = monitors() == NULL ? 0 : monitors()->number_of_monitors();
250 Interpreter::layout_activation(method(),
251 temps + callee_parameters,
252 popframe_preserved_args_size_in_words,
253 locks,
254 callee_parameters,
255 callee_locals,
256 caller,
257 iframe(),
258 is_top_frame);
260 // Update the pc in the frame object and overwrite the temporary pc
261 // we placed in the skeletal frame now that we finally know the
262 // exact interpreter address we should use.
264 _frame.patch_pc(thread, pc);
266 assert (!method()->is_synchronized() || locks > 0, "synchronized methods must have monitors");
268 BasicObjectLock* top = iframe()->interpreter_frame_monitor_begin();
269 for (int index = 0; index < locks; index++) {
270 top = iframe()->previous_monitor_in_interpreter_frame(top);
271 BasicObjectLock* src = _monitors->at(index);
272 top->set_obj(src->obj());
273 src->lock()->move_to(src->obj(), top->lock());
274 }
275 if (ProfileInterpreter) {
276 iframe()->interpreter_frame_set_mdx(0); // clear out the mdp.
277 }
278 iframe()->interpreter_frame_set_bcx((intptr_t)bcp); // cannot use bcp because frame is not initialized yet
279 if (ProfileInterpreter) {
280 methodDataOop mdo = method()->method_data();
281 if (mdo != NULL) {
282 int bci = iframe()->interpreter_frame_bci();
283 if (use_next_mdp) ++bci;
284 address mdp = mdo->bci_to_dp(bci);
285 iframe()->interpreter_frame_set_mdp(mdp);
286 }
287 }
289 // Unpack expression stack
290 // If this is an intermediate frame (i.e. not top frame) then this
291 // only unpacks the part of the expression stack not used by callee
292 // as parameters. The callee parameters are unpacked as part of the
293 // callee locals.
294 int i;
295 for(i = 0; i < expressions()->size(); i++) {
296 StackValue *value = expressions()->at(i);
297 intptr_t* addr = iframe()->interpreter_frame_expression_stack_at(i);
298 switch(value->type()) {
299 case T_INT:
300 *addr = value->get_int();
301 break;
302 case T_OBJECT:
303 *addr = value->get_int(T_OBJECT);
304 break;
305 case T_CONFLICT:
306 // A dead stack slot. Initialize to null in case it is an oop.
307 *addr = NULL_WORD;
308 break;
309 default:
310 ShouldNotReachHere();
311 }
312 }
315 // Unpack the locals
316 for(i = 0; i < locals()->size(); i++) {
317 StackValue *value = locals()->at(i);
318 intptr_t* addr = iframe()->interpreter_frame_local_at(i);
319 switch(value->type()) {
320 case T_INT:
321 *addr = value->get_int();
322 break;
323 case T_OBJECT:
324 *addr = value->get_int(T_OBJECT);
325 break;
326 case T_CONFLICT:
327 // A dead location. If it is an oop then we need a NULL to prevent GC from following it
328 *addr = NULL_WORD;
329 break;
330 default:
331 ShouldNotReachHere();
332 }
333 }
335 if (is_top_frame && JvmtiExport::can_pop_frame() && thread->popframe_forcing_deopt_reexecution()) {
336 // An interpreted frame was popped but it returns to a deoptimized
337 // frame. The incoming arguments to the interpreted activation
338 // were preserved in thread-local storage by the
339 // remove_activation_preserving_args_entry in the interpreter; now
340 // we put them back into the just-unpacked interpreter frame.
341 // Note that this assumes that the locals arena grows toward lower
342 // addresses.
343 if (popframe_preserved_args_size_in_words != 0) {
344 void* saved_args = thread->popframe_preserved_args();
345 assert(saved_args != NULL, "must have been saved by interpreter");
346 #ifdef ASSERT
347 assert(popframe_preserved_args_size_in_words <=
348 iframe()->interpreter_frame_expression_stack_size()*Interpreter::stackElementWords,
349 "expression stack size should have been extended");
350 #endif // ASSERT
351 int top_element = iframe()->interpreter_frame_expression_stack_size()-1;
352 intptr_t* base;
353 if (frame::interpreter_frame_expression_stack_direction() < 0) {
354 base = iframe()->interpreter_frame_expression_stack_at(top_element);
355 } else {
356 base = iframe()->interpreter_frame_expression_stack();
357 }
358 Copy::conjoint_bytes(saved_args,
359 base,
360 popframe_preserved_args_size_in_bytes);
361 thread->popframe_free_preserved_args();
362 }
363 }
365 #ifndef PRODUCT
366 if (TraceDeoptimization && Verbose) {
367 ttyLocker ttyl;
368 tty->print_cr("[%d Interpreted Frame]", ++unpack_counter);
369 iframe()->print_on(tty);
370 RegisterMap map(thread);
371 vframe* f = vframe::new_vframe(iframe(), &map, thread);
372 f->print();
374 tty->print_cr("locals size %d", locals()->size());
375 tty->print_cr("expression size %d", expressions()->size());
377 method()->print_value();
378 tty->cr();
379 // method()->print_codes();
380 } else if (TraceDeoptimization) {
381 tty->print(" ");
382 method()->print_value();
383 Bytecodes::Code code = Bytecodes::java_code_at(bcp);
384 int bci = method()->bci_from(bcp);
385 tty->print(" - %s", Bytecodes::name(code));
386 tty->print(" @ bci %d ", bci);
387 tty->print_cr("sp = " PTR_FORMAT, iframe()->sp());
388 }
389 #endif // PRODUCT
391 // The expression stack and locals are in the resource area don't leave
392 // a dangling pointer in the vframeArray we leave around for debug
393 // purposes
395 _locals = _expressions = NULL;
397 }
399 int vframeArrayElement::on_stack_size(int callee_parameters,
400 int callee_locals,
401 bool is_top_frame,
402 int popframe_extra_stack_expression_els) const {
403 assert(method()->max_locals() == locals()->size(), "just checking");
404 int locks = monitors() == NULL ? 0 : monitors()->number_of_monitors();
405 int temps = expressions()->size();
406 return Interpreter::size_activation(method(),
407 temps + callee_parameters,
408 popframe_extra_stack_expression_els,
409 locks,
410 callee_parameters,
411 callee_locals,
412 is_top_frame);
413 }
417 vframeArray* vframeArray::allocate(JavaThread* thread, int frame_size, GrowableArray<compiledVFrame*>* chunk,
418 RegisterMap *reg_map, frame sender, frame caller, frame self) {
420 // Allocate the vframeArray
421 vframeArray * result = (vframeArray*) AllocateHeap(sizeof(vframeArray) + // fixed part
422 sizeof(vframeArrayElement) * (chunk->length() - 1), // variable part
423 "vframeArray::allocate");
424 result->_frames = chunk->length();
425 result->_owner_thread = thread;
426 result->_sender = sender;
427 result->_caller = caller;
428 result->_original = self;
429 result->set_unroll_block(NULL); // initialize it
430 result->fill_in(thread, frame_size, chunk, reg_map);
431 return result;
432 }
434 void vframeArray::fill_in(JavaThread* thread,
435 int frame_size,
436 GrowableArray<compiledVFrame*>* chunk,
437 const RegisterMap *reg_map) {
438 // Set owner first, it is used when adding monitor chunks
440 _frame_size = frame_size;
441 for(int i = 0; i < chunk->length(); i++) {
442 element(i)->fill_in(chunk->at(i));
443 }
445 // Copy registers for callee-saved registers
446 if (reg_map != NULL) {
447 for(int i = 0; i < RegisterMap::reg_count; i++) {
448 #ifdef AMD64
449 // The register map has one entry for every int (32-bit value), so
450 // 64-bit physical registers have two entries in the map, one for
451 // each half. Ignore the high halves of 64-bit registers, just like
452 // frame::oopmapreg_to_location does.
453 //
454 // [phh] FIXME: this is a temporary hack! This code *should* work
455 // correctly w/o this hack, possibly by changing RegisterMap::pd_location
456 // in frame_amd64.cpp and the values of the phantom high half registers
457 // in amd64.ad.
458 // if (VMReg::Name(i) < SharedInfo::stack0 && is_even(i)) {
459 intptr_t* src = (intptr_t*) reg_map->location(VMRegImpl::as_VMReg(i));
460 _callee_registers[i] = src != NULL ? *src : NULL_WORD;
461 // } else {
462 // jint* src = (jint*) reg_map->location(VMReg::Name(i));
463 // _callee_registers[i] = src != NULL ? *src : NULL_WORD;
464 // }
465 #else
466 jint* src = (jint*) reg_map->location(VMRegImpl::as_VMReg(i));
467 _callee_registers[i] = src != NULL ? *src : NULL_WORD;
468 #endif
469 if (src == NULL) {
470 set_location_valid(i, false);
471 } else {
472 set_location_valid(i, true);
473 jint* dst = (jint*) register_location(i);
474 *dst = *src;
475 }
476 }
477 }
478 }
480 void vframeArray::unpack_to_stack(frame &unpack_frame, int exec_mode) {
481 // stack picture
482 // unpack_frame
483 // [new interpreter frames ] (frames are skeletal but walkable)
484 // caller_frame
485 //
486 // This routine fills in the missing data for the skeletal interpreter frames
487 // in the above picture.
489 // Find the skeletal interpreter frames to unpack into
490 RegisterMap map(JavaThread::current(), false);
491 // Get the youngest frame we will unpack (last to be unpacked)
492 frame me = unpack_frame.sender(&map);
493 int index;
494 for (index = 0; index < frames(); index++ ) {
495 *element(index)->iframe() = me;
496 // Get the caller frame (possibly skeletal)
497 me = me.sender(&map);
498 }
500 frame caller_frame = me;
502 // Do the unpacking of interpreter frames; the frame at index 0 represents the top activation, so it has no callee
504 // Unpack the frames from the oldest (frames() -1) to the youngest (0)
506 for (index = frames() - 1; index >= 0 ; index--) {
507 int callee_parameters = index == 0 ? 0 : element(index-1)->method()->size_of_parameters();
508 int callee_locals = index == 0 ? 0 : element(index-1)->method()->max_locals();
509 element(index)->unpack_on_stack(callee_parameters,
510 callee_locals,
511 &caller_frame,
512 index == 0,
513 exec_mode);
514 if (index == frames() - 1) {
515 Deoptimization::unwind_callee_save_values(element(index)->iframe(), this);
516 }
517 caller_frame = *element(index)->iframe();
518 }
521 deallocate_monitor_chunks();
522 }
524 void vframeArray::deallocate_monitor_chunks() {
525 JavaThread* jt = JavaThread::current();
526 for (int index = 0; index < frames(); index++ ) {
527 element(index)->free_monitors(jt);
528 }
529 }
531 #ifndef PRODUCT
533 bool vframeArray::structural_compare(JavaThread* thread, GrowableArray<compiledVFrame*>* chunk) {
534 if (owner_thread() != thread) return false;
535 int index = 0;
536 #if 0 // FIXME can't do this comparison
538 // Compare only within vframe array.
539 for (deoptimizedVFrame* vf = deoptimizedVFrame::cast(vframe_at(first_index())); vf; vf = vf->deoptimized_sender_or_null()) {
540 if (index >= chunk->length() || !vf->structural_compare(chunk->at(index))) return false;
541 index++;
542 }
543 if (index != chunk->length()) return false;
544 #endif
546 return true;
547 }
549 #endif
551 address vframeArray::register_location(int i) const {
552 assert(0 <= i && i < RegisterMap::reg_count, "index out of bounds");
553 return (address) & _callee_registers[i];
554 }
557 #ifndef PRODUCT
559 // Printing
561 // Note: we cannot have print_on as const, as we allocate inside the method
562 void vframeArray::print_on_2(outputStream* st) {
563 st->print_cr(" - sp: " INTPTR_FORMAT, sp());
564 st->print(" - thread: ");
565 Thread::current()->print();
566 st->print_cr(" - frame size: %d", frame_size());
567 for (int index = 0; index < frames() ; index++ ) {
568 element(index)->print(st);
569 }
570 }
572 void vframeArrayElement::print(outputStream* st) {
573 st->print_cr(" - interpreter_frame -> sp: " INTPTR_FORMAT, iframe()->sp());
574 }
576 void vframeArray::print_value_on(outputStream* st) const {
577 st->print_cr("vframeArray [%d] ", frames());
578 }
581 #endif