Mon, 20 Sep 2010 14:38:38 -0700
6984287: Regularize how GC parallel workers are specified.
Summary: Associate number of GC workers with the workgang as opposed to the task.
Reviewed-by: johnc, ysr
1 /*
2 * Copyright (c) 2001, 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 */
25 # include "incls/_precompiled.incl"
26 # include "incls/_collectedHeap.cpp.incl"
29 #ifdef ASSERT
30 int CollectedHeap::_fire_out_of_memory_count = 0;
31 #endif
33 size_t CollectedHeap::_filler_array_max_size = 0;
35 // Memory state functions.
38 CollectedHeap::CollectedHeap() : _n_par_threads(0)
40 {
41 const size_t max_len = size_t(arrayOopDesc::max_array_length(T_INT));
42 const size_t elements_per_word = HeapWordSize / sizeof(jint);
43 _filler_array_max_size = align_object_size(filler_array_hdr_size() +
44 max_len * elements_per_word);
46 _barrier_set = NULL;
47 _is_gc_active = false;
48 _total_collections = _total_full_collections = 0;
49 _gc_cause = _gc_lastcause = GCCause::_no_gc;
50 NOT_PRODUCT(_promotion_failure_alot_count = 0;)
51 NOT_PRODUCT(_promotion_failure_alot_gc_number = 0;)
53 if (UsePerfData) {
54 EXCEPTION_MARK;
56 // create the gc cause jvmstat counters
57 _perf_gc_cause = PerfDataManager::create_string_variable(SUN_GC, "cause",
58 80, GCCause::to_string(_gc_cause), CHECK);
60 _perf_gc_lastcause =
61 PerfDataManager::create_string_variable(SUN_GC, "lastCause",
62 80, GCCause::to_string(_gc_lastcause), CHECK);
63 }
64 _defer_initial_card_mark = false; // strengthened by subclass in pre_initialize() below.
65 }
67 void CollectedHeap::pre_initialize() {
68 // Used for ReduceInitialCardMarks (when COMPILER2 is used);
69 // otherwise remains unused.
70 #ifdef COMPILER2
71 _defer_initial_card_mark = ReduceInitialCardMarks && can_elide_tlab_store_barriers()
72 && (DeferInitialCardMark || card_mark_must_follow_store());
73 #else
74 assert(_defer_initial_card_mark == false, "Who would set it?");
75 #endif
76 }
78 #ifndef PRODUCT
79 void CollectedHeap::check_for_bad_heap_word_value(HeapWord* addr, size_t size) {
80 if (CheckMemoryInitialization && ZapUnusedHeapArea) {
81 for (size_t slot = 0; slot < size; slot += 1) {
82 assert((*(intptr_t*) (addr + slot)) != ((intptr_t) badHeapWordVal),
83 "Found badHeapWordValue in post-allocation check");
84 }
85 }
86 }
88 void CollectedHeap::check_for_non_bad_heap_word_value(HeapWord* addr, size_t size)
89 {
90 if (CheckMemoryInitialization && ZapUnusedHeapArea) {
91 for (size_t slot = 0; slot < size; slot += 1) {
92 assert((*(intptr_t*) (addr + slot)) == ((intptr_t) badHeapWordVal),
93 "Found non badHeapWordValue in pre-allocation check");
94 }
95 }
96 }
97 #endif // PRODUCT
99 #ifdef ASSERT
100 void CollectedHeap::check_for_valid_allocation_state() {
101 Thread *thread = Thread::current();
102 // How to choose between a pending exception and a potential
103 // OutOfMemoryError? Don't allow pending exceptions.
104 // This is a VM policy failure, so how do we exhaustively test it?
105 assert(!thread->has_pending_exception(),
106 "shouldn't be allocating with pending exception");
107 if (StrictSafepointChecks) {
108 assert(thread->allow_allocation(),
109 "Allocation done by thread for which allocation is blocked "
110 "by No_Allocation_Verifier!");
111 // Allocation of an oop can always invoke a safepoint,
112 // hence, the true argument
113 thread->check_for_valid_safepoint_state(true);
114 }
115 }
116 #endif
118 HeapWord* CollectedHeap::allocate_from_tlab_slow(Thread* thread, size_t size) {
120 // Retain tlab and allocate object in shared space if
121 // the amount free in the tlab is too large to discard.
122 if (thread->tlab().free() > thread->tlab().refill_waste_limit()) {
123 thread->tlab().record_slow_allocation(size);
124 return NULL;
125 }
127 // Discard tlab and allocate a new one.
128 // To minimize fragmentation, the last TLAB may be smaller than the rest.
129 size_t new_tlab_size = thread->tlab().compute_size(size);
131 thread->tlab().clear_before_allocation();
133 if (new_tlab_size == 0) {
134 return NULL;
135 }
137 // Allocate a new TLAB...
138 HeapWord* obj = Universe::heap()->allocate_new_tlab(new_tlab_size);
139 if (obj == NULL) {
140 return NULL;
141 }
142 if (ZeroTLAB) {
143 // ..and clear it.
144 Copy::zero_to_words(obj, new_tlab_size);
145 } else {
146 // ...and clear just the allocated object.
147 Copy::zero_to_words(obj, size);
148 }
149 thread->tlab().fill(obj, obj + size, new_tlab_size);
150 return obj;
151 }
153 void CollectedHeap::flush_deferred_store_barrier(JavaThread* thread) {
154 MemRegion deferred = thread->deferred_card_mark();
155 if (!deferred.is_empty()) {
156 assert(_defer_initial_card_mark, "Otherwise should be empty");
157 {
158 // Verify that the storage points to a parsable object in heap
159 DEBUG_ONLY(oop old_obj = oop(deferred.start());)
160 assert(is_in(old_obj), "Not in allocated heap");
161 assert(!can_elide_initializing_store_barrier(old_obj),
162 "Else should have been filtered in new_store_pre_barrier()");
163 assert(!is_in_permanent(old_obj), "Sanity: not expected");
164 assert(old_obj->is_oop(true), "Not an oop");
165 assert(old_obj->is_parsable(), "Will not be concurrently parsable");
166 assert(deferred.word_size() == (size_t)(old_obj->size()),
167 "Mismatch: multiple objects?");
168 }
169 BarrierSet* bs = barrier_set();
170 assert(bs->has_write_region_opt(), "No write_region() on BarrierSet");
171 bs->write_region(deferred);
172 // "Clear" the deferred_card_mark field
173 thread->set_deferred_card_mark(MemRegion());
174 }
175 assert(thread->deferred_card_mark().is_empty(), "invariant");
176 }
178 // Helper for ReduceInitialCardMarks. For performance,
179 // compiled code may elide card-marks for initializing stores
180 // to a newly allocated object along the fast-path. We
181 // compensate for such elided card-marks as follows:
182 // (a) Generational, non-concurrent collectors, such as
183 // GenCollectedHeap(ParNew,DefNew,Tenured) and
184 // ParallelScavengeHeap(ParallelGC, ParallelOldGC)
185 // need the card-mark if and only if the region is
186 // in the old gen, and do not care if the card-mark
187 // succeeds or precedes the initializing stores themselves,
188 // so long as the card-mark is completed before the next
189 // scavenge. For all these cases, we can do a card mark
190 // at the point at which we do a slow path allocation
191 // in the old gen, i.e. in this call.
192 // (b) GenCollectedHeap(ConcurrentMarkSweepGeneration) requires
193 // in addition that the card-mark for an old gen allocated
194 // object strictly follow any associated initializing stores.
195 // In these cases, the memRegion remembered below is
196 // used to card-mark the entire region either just before the next
197 // slow-path allocation by this thread or just before the next scavenge or
198 // CMS-associated safepoint, whichever of these events happens first.
199 // (The implicit assumption is that the object has been fully
200 // initialized by this point, a fact that we assert when doing the
201 // card-mark.)
202 // (c) G1CollectedHeap(G1) uses two kinds of write barriers. When a
203 // G1 concurrent marking is in progress an SATB (pre-write-)barrier is
204 // is used to remember the pre-value of any store. Initializing
205 // stores will not need this barrier, so we need not worry about
206 // compensating for the missing pre-barrier here. Turning now
207 // to the post-barrier, we note that G1 needs a RS update barrier
208 // which simply enqueues a (sequence of) dirty cards which may
209 // optionally be refined by the concurrent update threads. Note
210 // that this barrier need only be applied to a non-young write,
211 // but, like in CMS, because of the presence of concurrent refinement
212 // (much like CMS' precleaning), must strictly follow the oop-store.
213 // Thus, using the same protocol for maintaining the intended
214 // invariants turns out, serendepitously, to be the same for both
215 // G1 and CMS.
216 //
217 // For any future collector, this code should be reexamined with
218 // that specific collector in mind, and the documentation above suitably
219 // extended and updated.
220 oop CollectedHeap::new_store_pre_barrier(JavaThread* thread, oop new_obj) {
221 // If a previous card-mark was deferred, flush it now.
222 flush_deferred_store_barrier(thread);
223 if (can_elide_initializing_store_barrier(new_obj)) {
224 // The deferred_card_mark region should be empty
225 // following the flush above.
226 assert(thread->deferred_card_mark().is_empty(), "Error");
227 } else {
228 MemRegion mr((HeapWord*)new_obj, new_obj->size());
229 assert(!mr.is_empty(), "Error");
230 if (_defer_initial_card_mark) {
231 // Defer the card mark
232 thread->set_deferred_card_mark(mr);
233 } else {
234 // Do the card mark
235 BarrierSet* bs = barrier_set();
236 assert(bs->has_write_region_opt(), "No write_region() on BarrierSet");
237 bs->write_region(mr);
238 }
239 }
240 return new_obj;
241 }
243 size_t CollectedHeap::filler_array_hdr_size() {
244 return size_t(align_object_offset(arrayOopDesc::header_size(T_INT))); // align to Long
245 }
247 size_t CollectedHeap::filler_array_min_size() {
248 return align_object_size(filler_array_hdr_size()); // align to MinObjAlignment
249 }
251 size_t CollectedHeap::filler_array_max_size() {
252 return _filler_array_max_size;
253 }
255 #ifdef ASSERT
256 void CollectedHeap::fill_args_check(HeapWord* start, size_t words)
257 {
258 assert(words >= min_fill_size(), "too small to fill");
259 assert(words % MinObjAlignment == 0, "unaligned size");
260 assert(Universe::heap()->is_in_reserved(start), "not in heap");
261 assert(Universe::heap()->is_in_reserved(start + words - 1), "not in heap");
262 }
264 void CollectedHeap::zap_filler_array(HeapWord* start, size_t words, bool zap)
265 {
266 if (ZapFillerObjects && zap) {
267 Copy::fill_to_words(start + filler_array_hdr_size(),
268 words - filler_array_hdr_size(), 0XDEAFBABE);
269 }
270 }
271 #endif // ASSERT
273 void
274 CollectedHeap::fill_with_array(HeapWord* start, size_t words, bool zap)
275 {
276 assert(words >= filler_array_min_size(), "too small for an array");
277 assert(words <= filler_array_max_size(), "too big for a single object");
279 const size_t payload_size = words - filler_array_hdr_size();
280 const size_t len = payload_size * HeapWordSize / sizeof(jint);
282 // Set the length first for concurrent GC.
283 ((arrayOop)start)->set_length((int)len);
284 post_allocation_setup_common(Universe::intArrayKlassObj(), start, words);
285 DEBUG_ONLY(zap_filler_array(start, words, zap);)
286 }
288 void
289 CollectedHeap::fill_with_object_impl(HeapWord* start, size_t words, bool zap)
290 {
291 assert(words <= filler_array_max_size(), "too big for a single object");
293 if (words >= filler_array_min_size()) {
294 fill_with_array(start, words, zap);
295 } else if (words > 0) {
296 assert(words == min_fill_size(), "unaligned size");
297 post_allocation_setup_common(SystemDictionary::Object_klass(), start,
298 words);
299 }
300 }
302 void CollectedHeap::fill_with_object(HeapWord* start, size_t words, bool zap)
303 {
304 DEBUG_ONLY(fill_args_check(start, words);)
305 HandleMark hm; // Free handles before leaving.
306 fill_with_object_impl(start, words, zap);
307 }
309 void CollectedHeap::fill_with_objects(HeapWord* start, size_t words, bool zap)
310 {
311 DEBUG_ONLY(fill_args_check(start, words);)
312 HandleMark hm; // Free handles before leaving.
314 #ifdef _LP64
315 // A single array can fill ~8G, so multiple objects are needed only in 64-bit.
316 // First fill with arrays, ensuring that any remaining space is big enough to
317 // fill. The remainder is filled with a single object.
318 const size_t min = min_fill_size();
319 const size_t max = filler_array_max_size();
320 while (words > max) {
321 const size_t cur = words - max >= min ? max : max - min;
322 fill_with_array(start, cur, zap);
323 start += cur;
324 words -= cur;
325 }
326 #endif
328 fill_with_object_impl(start, words, zap);
329 }
331 HeapWord* CollectedHeap::allocate_new_tlab(size_t size) {
332 guarantee(false, "thread-local allocation buffers not supported");
333 return NULL;
334 }
336 void CollectedHeap::ensure_parsability(bool retire_tlabs) {
337 // The second disjunct in the assertion below makes a concession
338 // for the start-up verification done while the VM is being
339 // created. Callers be careful that you know that mutators
340 // aren't going to interfere -- for instance, this is permissible
341 // if we are still single-threaded and have either not yet
342 // started allocating (nothing much to verify) or we have
343 // started allocating but are now a full-fledged JavaThread
344 // (and have thus made our TLAB's) available for filling.
345 assert(SafepointSynchronize::is_at_safepoint() ||
346 !is_init_completed(),
347 "Should only be called at a safepoint or at start-up"
348 " otherwise concurrent mutator activity may make heap "
349 " unparsable again");
350 const bool use_tlab = UseTLAB;
351 const bool deferred = _defer_initial_card_mark;
352 // The main thread starts allocating via a TLAB even before it
353 // has added itself to the threads list at vm boot-up.
354 assert(!use_tlab || Threads::first() != NULL,
355 "Attempt to fill tlabs before main thread has been added"
356 " to threads list is doomed to failure!");
357 for (JavaThread *thread = Threads::first(); thread; thread = thread->next()) {
358 if (use_tlab) thread->tlab().make_parsable(retire_tlabs);
359 #ifdef COMPILER2
360 // The deferred store barriers must all have been flushed to the
361 // card-table (or other remembered set structure) before GC starts
362 // processing the card-table (or other remembered set).
363 if (deferred) flush_deferred_store_barrier(thread);
364 #else
365 assert(!deferred, "Should be false");
366 assert(thread->deferred_card_mark().is_empty(), "Should be empty");
367 #endif
368 }
369 }
371 void CollectedHeap::accumulate_statistics_all_tlabs() {
372 if (UseTLAB) {
373 assert(SafepointSynchronize::is_at_safepoint() ||
374 !is_init_completed(),
375 "should only accumulate statistics on tlabs at safepoint");
377 ThreadLocalAllocBuffer::accumulate_statistics_before_gc();
378 }
379 }
381 void CollectedHeap::resize_all_tlabs() {
382 if (UseTLAB) {
383 assert(SafepointSynchronize::is_at_safepoint() ||
384 !is_init_completed(),
385 "should only resize tlabs at safepoint");
387 ThreadLocalAllocBuffer::resize_all_tlabs();
388 }
389 }
391 void CollectedHeap::pre_full_gc_dump() {
392 if (HeapDumpBeforeFullGC) {
393 TraceTime tt("Heap Dump: ", PrintGCDetails, false, gclog_or_tty);
394 // We are doing a "major" collection and a heap dump before
395 // major collection has been requested.
396 HeapDumper::dump_heap();
397 }
398 if (PrintClassHistogramBeforeFullGC) {
399 TraceTime tt("Class Histogram: ", PrintGCDetails, true, gclog_or_tty);
400 VM_GC_HeapInspection inspector(gclog_or_tty, false /* ! full gc */, false /* ! prologue */);
401 inspector.doit();
402 }
403 }
405 void CollectedHeap::post_full_gc_dump() {
406 if (HeapDumpAfterFullGC) {
407 TraceTime tt("Heap Dump", PrintGCDetails, false, gclog_or_tty);
408 HeapDumper::dump_heap();
409 }
410 if (PrintClassHistogramAfterFullGC) {
411 TraceTime tt("Class Histogram", PrintGCDetails, true, gclog_or_tty);
412 VM_GC_HeapInspection inspector(gclog_or_tty, false /* ! full gc */, false /* ! prologue */);
413 inspector.doit();
414 }
415 }