1 /*
2 * Copyright (c) 2015, 2019, Red Hat, 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 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 #ifndef SHARE_GC_SHENANDOAH_SHENANDOAHHEAP_INLINE_HPP
26 #define SHARE_GC_SHENANDOAH_SHENANDOAHHEAP_INLINE_HPP
27
28 #include "classfile/javaClasses.inline.hpp"
29 #include "gc/shared/markBitMap.inline.hpp"
30 #include "gc/shared/threadLocalAllocBuffer.inline.hpp"
31 #include "gc/shared/suspendibleThreadSet.hpp"
32 #include "gc/shenandoah/shenandoahAsserts.hpp"
33 #include "gc/shenandoah/shenandoahBarrierSet.inline.hpp"
34 #include "gc/shenandoah/shenandoahCollectionSet.inline.hpp"
35 #include "gc/shenandoah/shenandoahForwarding.inline.hpp"
36 #include "gc/shenandoah/shenandoahWorkGroup.hpp"
37 #include "gc/shenandoah/shenandoahHeap.hpp"
38 #include "gc/shenandoah/shenandoahHeapRegionSet.inline.hpp"
39 #include "gc/shenandoah/shenandoahHeapRegion.inline.hpp"
40 #include "gc/shenandoah/shenandoahControlThread.hpp"
41 #include "gc/shenandoah/shenandoahMarkingContext.inline.hpp"
42 #include "gc/shenandoah/shenandoahThreadLocalData.hpp"
43 #include "oops/compressedOops.inline.hpp"
44 #include "oops/oop.inline.hpp"
45 #include "runtime/atomic.hpp"
46 #include "runtime/prefetch.inline.hpp"
47 #include "runtime/thread.hpp"
48 #include "utilities/copy.hpp"
49 #include "utilities/globalDefinitions.hpp"
50
51
52 inline ShenandoahHeapRegion* ShenandoahRegionIterator::next() {
53 size_t new_index = Atomic::add(&_index, (size_t) 1);
54 // get_region() provides the bounds-check and returns NULL on OOB.
55 return _heap->get_region(new_index - 1);
56 }
57
58 inline bool ShenandoahHeap::has_forwarded_objects() const {
59 return _gc_state.is_set(HAS_FORWARDED);
60 }
61
62 inline WorkGang* ShenandoahHeap::workers() const {
63 return _workers;
64 }
65
66 inline WorkGang* ShenandoahHeap::get_safepoint_workers() {
67 return _safepoint_workers;
68 }
69
70 inline size_t ShenandoahHeap::heap_region_index_containing(const void* addr) const {
71 uintptr_t region_start = ((uintptr_t) addr);
72 uintptr_t index = (region_start - (uintptr_t) base()) >> ShenandoahHeapRegion::region_size_bytes_shift();
73 assert(index < num_regions(), "Region index is in bounds: " PTR_FORMAT, p2i(addr));
74 return index;
75 }
76
77 inline ShenandoahHeapRegion* const ShenandoahHeap::heap_region_containing(const void* addr) const {
78 size_t index = heap_region_index_containing(addr);
79 ShenandoahHeapRegion* const result = get_region(index);
80 assert(addr >= result->bottom() && addr < result->end(), "Heap region contains the address: " PTR_FORMAT, p2i(addr));
81 return result;
82 }
83
84 template <class T>
85 inline oop ShenandoahHeap::update_with_forwarded_not_null(T* p, oop obj) {
86 if (in_collection_set(obj)) {
87 shenandoah_assert_forwarded_except(p, obj, is_full_gc_in_progress() || cancelled_gc() || is_degenerated_gc_in_progress());
88 obj = ShenandoahBarrierSet::resolve_forwarded_not_null(obj);
89 RawAccess<IS_NOT_NULL>::oop_store(p, obj);
90 }
91 #ifdef ASSERT
92 else {
93 shenandoah_assert_not_forwarded(p, obj);
94 }
95 #endif
96 return obj;
97 }
98
99 template <class T>
100 inline oop ShenandoahHeap::maybe_update_with_forwarded(T* p) {
101 T o = RawAccess<>::oop_load(p);
102 if (!CompressedOops::is_null(o)) {
103 oop obj = CompressedOops::decode_not_null(o);
104 return maybe_update_with_forwarded_not_null(p, obj);
105 } else {
106 return NULL;
107 }
108 }
109
110 template <class T>
111 inline oop ShenandoahHeap::evac_update_with_forwarded(T* p) {
112 T o = RawAccess<>::oop_load(p);
113 if (!CompressedOops::is_null(o)) {
114 oop heap_oop = CompressedOops::decode_not_null(o);
115 if (in_collection_set(heap_oop)) {
116 oop forwarded_oop = ShenandoahBarrierSet::resolve_forwarded_not_null(heap_oop);
117 if (forwarded_oop == heap_oop) {
118 forwarded_oop = evacuate_object(heap_oop, Thread::current());
119 }
120 oop prev = cas_oop(forwarded_oop, p, heap_oop);
121 if (prev == heap_oop) {
122 return forwarded_oop;
123 } else {
124 return NULL;
125 }
126 }
127 return heap_oop;
128 } else {
129 return NULL;
130 }
131 }
132
133 inline oop ShenandoahHeap::cas_oop(oop n, oop* addr, oop c) {
134 assert(is_aligned(addr, HeapWordSize), "Address should be aligned: " PTR_FORMAT, p2i(addr));
135 return (oop) Atomic::cmpxchg(addr, c, n);
136 }
137
138 inline oop ShenandoahHeap::cas_oop(oop n, narrowOop* addr, narrowOop c) {
139 assert(is_aligned(addr, sizeof(narrowOop)), "Address should be aligned: " PTR_FORMAT, p2i(addr));
140 narrowOop val = CompressedOops::encode(n);
141 return CompressedOops::decode((narrowOop) Atomic::cmpxchg(addr, c, val));
142 }
143
144 inline oop ShenandoahHeap::cas_oop(oop n, narrowOop* addr, oop c) {
145 assert(is_aligned(addr, sizeof(narrowOop)), "Address should be aligned: " PTR_FORMAT, p2i(addr));
146 narrowOop cmp = CompressedOops::encode(c);
147 narrowOop val = CompressedOops::encode(n);
148 return CompressedOops::decode((narrowOop) Atomic::cmpxchg(addr, cmp, val));
149 }
150
151 template <class T>
152 inline oop ShenandoahHeap::maybe_update_with_forwarded_not_null(T* p, oop heap_oop) {
153 shenandoah_assert_not_in_cset_loc_except(p, !is_in(p) || is_full_gc_in_progress() || is_degenerated_gc_in_progress());
154 shenandoah_assert_correct(p, heap_oop);
155
156 if (in_collection_set(heap_oop)) {
157 oop forwarded_oop = ShenandoahBarrierSet::resolve_forwarded_not_null(heap_oop);
158 if (forwarded_oop == heap_oop) {
159 // E.g. during evacuation.
160 return forwarded_oop;
161 }
162
163 shenandoah_assert_forwarded_except(p, heap_oop, is_full_gc_in_progress() || is_degenerated_gc_in_progress());
164 shenandoah_assert_not_forwarded(p, forwarded_oop);
165 shenandoah_assert_not_in_cset_except(p, forwarded_oop, cancelled_gc());
166
167 // If this fails, another thread wrote to p before us, it will be logged in SATB and the
168 // reference be updated later.
169 oop witness = cas_oop(forwarded_oop, p, heap_oop);
170
171 if (witness != heap_oop) {
172 // CAS failed, someone had beat us to it. Normally, we would return the failure witness,
173 // because that would be the proper write of to-space object, enforced by strong barriers.
174 // However, there is a corner case with arraycopy. It can happen that a Java thread
175 // beats us with an arraycopy, which first copies the array, which potentially contains
176 // from-space refs, and only afterwards updates all from-space refs to to-space refs,
177 // which leaves a short window where the new array elements can be from-space.
178 // In this case, we can just resolve the result again. As we resolve, we need to consider
179 // the contended write might have been NULL.
180 oop result = ShenandoahBarrierSet::resolve_forwarded(witness);
181 shenandoah_assert_not_forwarded_except(p, result, (result == NULL));
182 shenandoah_assert_not_in_cset_except(p, result, (result == NULL) || cancelled_gc());
183 return result;
184 } else {
185 // Success! We have updated with known to-space copy. We have already asserted it is sane.
186 return forwarded_oop;
187 }
188 } else {
189 shenandoah_assert_not_forwarded(p, heap_oop);
190 return heap_oop;
191 }
192 }
193
194 inline bool ShenandoahHeap::cancelled_gc() const {
195 return _cancelled_gc.get() == CANCELLED;
196 }
197
198 inline bool ShenandoahHeap::check_cancelled_gc_and_yield(bool sts_active) {
199 if (! (sts_active && ShenandoahSuspendibleWorkers)) {
200 return cancelled_gc();
201 }
202
203 jbyte prev = _cancelled_gc.cmpxchg(NOT_CANCELLED, CANCELLABLE);
204 if (prev == CANCELLABLE || prev == NOT_CANCELLED) {
205 if (SuspendibleThreadSet::should_yield()) {
206 SuspendibleThreadSet::yield();
207 }
208
209 // Back to CANCELLABLE. The thread that poked NOT_CANCELLED first gets
210 // to restore to CANCELLABLE.
211 if (prev == CANCELLABLE) {
212 _cancelled_gc.set(CANCELLABLE);
213 }
214 return false;
215 } else {
216 return true;
217 }
218 }
219
220 inline void ShenandoahHeap::clear_cancelled_gc() {
221 _cancelled_gc.set(CANCELLABLE);
222 _oom_evac_handler.clear();
223 }
224
225 inline HeapWord* ShenandoahHeap::allocate_from_gclab(Thread* thread, size_t size) {
226 assert(UseTLAB, "TLABs should be enabled");
227
228 PLAB* gclab = ShenandoahThreadLocalData::gclab(thread);
229 if (gclab == NULL) {
230 assert(!thread->is_Java_thread() && !thread->is_Worker_thread(),
231 "Performance: thread should have GCLAB: %s", thread->name());
232 // No GCLABs in this thread, fallback to shared allocation
233 return NULL;
234 }
235 HeapWord* obj = gclab->allocate(size);
236 if (obj != NULL) {
237 return obj;
238 }
239 // Otherwise...
240 return allocate_from_gclab_slow(thread, size);
241 }
242
243 inline oop ShenandoahHeap::evacuate_object(oop p, Thread* thread) {
244 if (ShenandoahThreadLocalData::is_oom_during_evac(Thread::current())) {
245 // This thread went through the OOM during evac protocol and it is safe to return
246 // the forward pointer. It must not attempt to evacuate any more.
247 return ShenandoahBarrierSet::resolve_forwarded(p);
248 }
249
250 assert(ShenandoahThreadLocalData::is_evac_allowed(thread), "must be enclosed in oom-evac scope");
251
252 size_t size = p->size();
253
254 assert(!heap_region_containing(p)->is_humongous(), "never evacuate humongous objects");
255
256 bool alloc_from_gclab = true;
257 HeapWord* copy = NULL;
258
259 #ifdef ASSERT
260 if (ShenandoahOOMDuringEvacALot &&
261 (os::random() & 1) == 0) { // Simulate OOM every ~2nd slow-path call
262 copy = NULL;
263 } else {
264 #endif
265 if (UseTLAB) {
266 copy = allocate_from_gclab(thread, size);
267 }
268 if (copy == NULL) {
269 ShenandoahAllocRequest req = ShenandoahAllocRequest::for_shared_gc(size);
270 copy = allocate_memory(req);
271 alloc_from_gclab = false;
272 }
273 #ifdef ASSERT
274 }
275 #endif
276
277 if (copy == NULL) {
278 control_thread()->handle_alloc_failure_evac(size);
279
280 _oom_evac_handler.handle_out_of_memory_during_evacuation();
281
282 return ShenandoahBarrierSet::resolve_forwarded(p);
283 }
284
285 // Copy the object:
286 Copy::aligned_disjoint_words(cast_from_oop<HeapWord*>(p), copy, size);
287
288 // Try to install the new forwarding pointer.
289 oop copy_val = oop(copy);
290 oop result = ShenandoahForwarding::try_update_forwardee(p, copy_val);
291 if (result == copy_val) {
292 // Successfully evacuated. Our copy is now the public one!
293 shenandoah_assert_correct(NULL, copy_val);
294 return copy_val;
295 } else {
296 // Failed to evacuate. We need to deal with the object that is left behind. Since this
297 // new allocation is certainly after TAMS, it will be considered live in the next cycle.
298 // But if it happens to contain references to evacuated regions, those references would
299 // not get updated for this stale copy during this cycle, and we will crash while scanning
300 // it the next cycle.
301 //
302 // For GCLAB allocations, it is enough to rollback the allocation ptr. Either the next
303 // object will overwrite this stale copy, or the filler object on LAB retirement will
304 // do this. For non-GCLAB allocations, we have no way to retract the allocation, and
305 // have to explicitly overwrite the copy with the filler object. With that overwrite,
306 // we have to keep the fwdptr initialized and pointing to our (stale) copy.
307 if (alloc_from_gclab) {
308 ShenandoahThreadLocalData::gclab(thread)->undo_allocation(copy, size);
309 } else {
310 fill_with_object(copy, size);
311 shenandoah_assert_correct(NULL, copy_val);
312 }
313 shenandoah_assert_correct(NULL, result);
314 return result;
315 }
316 }
317
318 template<bool RESOLVE>
319 inline bool ShenandoahHeap::requires_marking(const void* entry) const {
320 oop obj = oop(entry);
321 if (RESOLVE) {
322 obj = ShenandoahBarrierSet::resolve_forwarded_not_null(obj);
323 }
324 return !_marking_context->is_marked(obj);
325 }
326
327 template <class T>
328 inline bool ShenandoahHeap::in_collection_set(T p) const {
329 HeapWord* obj = cast_from_oop<HeapWord*>(p);
330 assert(collection_set() != NULL, "Sanity");
331 assert(is_in(obj), "should be in heap");
332
333 return collection_set()->is_in(obj);
334 }
335
336 inline bool ShenandoahHeap::is_stable() const {
337 return _gc_state.is_clear();
338 }
339
340 inline bool ShenandoahHeap::is_idle() const {
341 return _gc_state.is_unset(MARKING | EVACUATION | UPDATEREFS | TRAVERSAL);
342 }
343
344 inline bool ShenandoahHeap::is_concurrent_mark_in_progress() const {
345 return _gc_state.is_set(MARKING);
346 }
347
348 inline bool ShenandoahHeap::is_concurrent_traversal_in_progress() const {
349 return _gc_state.is_set(TRAVERSAL);
350 }
351
352 inline bool ShenandoahHeap::is_evacuation_in_progress() const {
353 return _gc_state.is_set(EVACUATION);
354 }
355
356 inline bool ShenandoahHeap::is_gc_in_progress_mask(uint mask) const {
357 return _gc_state.is_set(mask);
358 }
359
360 inline bool ShenandoahHeap::is_degenerated_gc_in_progress() const {
361 return _degenerated_gc_in_progress.is_set();
362 }
363
364 inline bool ShenandoahHeap::is_full_gc_in_progress() const {
365 return _full_gc_in_progress.is_set();
366 }
367
368 inline bool ShenandoahHeap::is_full_gc_move_in_progress() const {
369 return _full_gc_move_in_progress.is_set();
370 }
371
372 inline bool ShenandoahHeap::is_update_refs_in_progress() const {
373 return _gc_state.is_set(UPDATEREFS);
374 }
375
376 inline bool ShenandoahHeap::is_stw_gc_in_progress() const {
377 return is_full_gc_in_progress() || is_degenerated_gc_in_progress();
378 }
379
380 inline bool ShenandoahHeap::is_concurrent_root_in_progress() const {
381 return _concurrent_root_in_progress.is_set();
382 }
383
384 template<class T>
385 inline void ShenandoahHeap::marked_object_iterate(ShenandoahHeapRegion* region, T* cl) {
386 marked_object_iterate(region, cl, region->top());
387 }
388
389 template<class T>
390 inline void ShenandoahHeap::marked_object_iterate(ShenandoahHeapRegion* region, T* cl, HeapWord* limit) {
391 assert(! region->is_humongous_continuation(), "no humongous continuation regions here");
392
393 ShenandoahMarkingContext* const ctx = complete_marking_context();
394 assert(ctx->is_complete(), "sanity");
395
396 MarkBitMap* mark_bit_map = ctx->mark_bit_map();
397 HeapWord* tams = ctx->top_at_mark_start(region);
398
399 size_t skip_bitmap_delta = 1;
400 HeapWord* start = region->bottom();
401 HeapWord* end = MIN2(tams, region->end());
402
403 // Step 1. Scan below the TAMS based on bitmap data.
404 HeapWord* limit_bitmap = MIN2(limit, tams);
405
406 // Try to scan the initial candidate. If the candidate is above the TAMS, it would
407 // fail the subsequent "< limit_bitmap" checks, and fall through to Step 2.
408 HeapWord* cb = mark_bit_map->get_next_marked_addr(start, end);
409
410 intx dist = ShenandoahMarkScanPrefetch;
411 if (dist > 0) {
412 // Batched scan that prefetches the oop data, anticipating the access to
413 // either header, oop field, or forwarding pointer. Not that we cannot
414 // touch anything in oop, while it still being prefetched to get enough
415 // time for prefetch to work. This is why we try to scan the bitmap linearly,
416 // disregarding the object size. However, since we know forwarding pointer
417 // preceeds the object, we can skip over it. Once we cannot trust the bitmap,
418 // there is no point for prefetching the oop contents, as oop->size() will
419 // touch it prematurely.
420
421 // No variable-length arrays in standard C++, have enough slots to fit
422 // the prefetch distance.
423 static const int SLOT_COUNT = 256;
424 guarantee(dist <= SLOT_COUNT, "adjust slot count");
425 HeapWord* slots[SLOT_COUNT];
426
427 int avail;
428 do {
429 avail = 0;
430 for (int c = 0; (c < dist) && (cb < limit_bitmap); c++) {
431 Prefetch::read(cb, oopDesc::mark_offset_in_bytes());
432 slots[avail++] = cb;
433 cb += skip_bitmap_delta;
434 if (cb < limit_bitmap) {
435 cb = mark_bit_map->get_next_marked_addr(cb, limit_bitmap);
436 }
437 }
438
439 for (int c = 0; c < avail; c++) {
440 assert (slots[c] < tams, "only objects below TAMS here: " PTR_FORMAT " (" PTR_FORMAT ")", p2i(slots[c]), p2i(tams));
441 assert (slots[c] < limit, "only objects below limit here: " PTR_FORMAT " (" PTR_FORMAT ")", p2i(slots[c]), p2i(limit));
442 oop obj = oop(slots[c]);
443 assert(oopDesc::is_oop(obj), "sanity");
444 assert(ctx->is_marked(obj), "object expected to be marked");
445 cl->do_object(obj);
446 }
447 } while (avail > 0);
448 } else {
449 while (cb < limit_bitmap) {
450 assert (cb < tams, "only objects below TAMS here: " PTR_FORMAT " (" PTR_FORMAT ")", p2i(cb), p2i(tams));
451 assert (cb < limit, "only objects below limit here: " PTR_FORMAT " (" PTR_FORMAT ")", p2i(cb), p2i(limit));
452 oop obj = oop(cb);
453 assert(oopDesc::is_oop(obj), "sanity");
454 assert(ctx->is_marked(obj), "object expected to be marked");
455 cl->do_object(obj);
456 cb += skip_bitmap_delta;
457 if (cb < limit_bitmap) {
458 cb = mark_bit_map->get_next_marked_addr(cb, limit_bitmap);
459 }
460 }
461 }
462
463 // Step 2. Accurate size-based traversal, happens past the TAMS.
464 // This restarts the scan at TAMS, which makes sure we traverse all objects,
465 // regardless of what happened at Step 1.
466 HeapWord* cs = tams;
467 while (cs < limit) {
468 assert (cs >= tams, "only objects past TAMS here: " PTR_FORMAT " (" PTR_FORMAT ")", p2i(cs), p2i(tams));
469 assert (cs < limit, "only objects below limit here: " PTR_FORMAT " (" PTR_FORMAT ")", p2i(cs), p2i(limit));
470 oop obj = oop(cs);
471 assert(oopDesc::is_oop(obj), "sanity");
472 assert(ctx->is_marked(obj), "object expected to be marked");
473 int size = obj->size();
474 cl->do_object(obj);
475 cs += size;
476 }
477 }
478
479 template <class T>
480 class ShenandoahObjectToOopClosure : public ObjectClosure {
481 T* _cl;
482 public:
483 ShenandoahObjectToOopClosure(T* cl) : _cl(cl) {}
484
485 void do_object(oop obj) {
486 obj->oop_iterate(_cl);
487 }
488 };
489
490 template <class T>
491 class ShenandoahObjectToOopBoundedClosure : public ObjectClosure {
492 T* _cl;
493 MemRegion _bounds;
494 public:
495 ShenandoahObjectToOopBoundedClosure(T* cl, HeapWord* bottom, HeapWord* top) :
496 _cl(cl), _bounds(bottom, top) {}
497
498 void do_object(oop obj) {
499 obj->oop_iterate(_cl, _bounds);
500 }
501 };
502
503 template<class T>
504 inline void ShenandoahHeap::marked_object_oop_iterate(ShenandoahHeapRegion* region, T* cl, HeapWord* top) {
505 if (region->is_humongous()) {
506 HeapWord* bottom = region->bottom();
507 if (top > bottom) {
508 region = region->humongous_start_region();
509 ShenandoahObjectToOopBoundedClosure<T> objs(cl, bottom, top);
510 marked_object_iterate(region, &objs);
511 }
512 } else {
513 ShenandoahObjectToOopClosure<T> objs(cl);
514 marked_object_iterate(region, &objs, top);
515 }
516 }
517
518 inline ShenandoahHeapRegion* const ShenandoahHeap::get_region(size_t region_idx) const {
519 if (region_idx < _num_regions) {
520 return _regions[region_idx];
521 } else {
522 return NULL;
523 }
524 }
525
526 inline void ShenandoahHeap::mark_complete_marking_context() {
527 _marking_context->mark_complete();
528 }
529
530 inline void ShenandoahHeap::mark_incomplete_marking_context() {
531 _marking_context->mark_incomplete();
532 }
533
534 inline ShenandoahMarkingContext* ShenandoahHeap::complete_marking_context() const {
535 assert (_marking_context->is_complete()," sanity");
536 return _marking_context;
537 }
538
539 inline ShenandoahMarkingContext* ShenandoahHeap::marking_context() const {
540 return _marking_context;
541 }
542
543 #endif // SHARE_GC_SHENANDOAH_SHENANDOAHHEAP_INLINE_HPP