1 /*
  2  * Copyright (c) 2015, 2022, Red Hat, Inc. All rights reserved.
  3  * Copyright Amazon.com Inc. or its affiliates. All Rights Reserved.
  4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  5  *
  6  * This code is free software; you can redistribute it and/or modify it
  7  * under the terms of the GNU General Public License version 2 only, as
  8  * published by the Free Software Foundation.
  9  *
 10  * This code is distributed in the hope that it will be useful, but WITHOUT
 11  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
 12  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
 13  * version 2 for more details (a copy is included in the LICENSE file that
 14  * accompanied this code).
 15  *
 16  * You should have received a copy of the GNU General Public License version
 17  * 2 along with this work; if not, write to the Free Software Foundation,
 18  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
 19  *
 20  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
 21  * or visit www.oracle.com if you need additional information or have any
 22  * questions.
 23  *
 24  */
 25 
 26 #ifndef SHARE_GC_SHENANDOAH_SHENANDOAHMARK_INLINE_HPP
 27 #define SHARE_GC_SHENANDOAH_SHENANDOAHMARK_INLINE_HPP
 28 
 29 #include "gc/shenandoah/shenandoahMark.hpp"
 30 
 31 #include "gc/shared/continuationGCSupport.inline.hpp"
 32 #include "gc/shenandoah/shenandoahAsserts.hpp"
 33 #include "gc/shenandoah/shenandoahBarrierSet.inline.hpp"
 34 #include "gc/shenandoah/shenandoahHeap.inline.hpp"
 35 #include "gc/shenandoah/shenandoahMarkingContext.inline.hpp"
 36 #include "gc/shenandoah/shenandoahStringDedup.inline.hpp"
 37 #include "gc/shenandoah/shenandoahTaskqueue.inline.hpp"
 38 #include "gc/shenandoah/shenandoahUtils.hpp"
 39 #include "memory/iterator.inline.hpp"
 40 #include "oops/compressedOops.inline.hpp"
 41 #include "oops/oop.inline.hpp"
 42 #include "runtime/prefetch.inline.hpp"
 43 #include "utilities/devirtualizer.inline.hpp"
 44 #include "utilities/powerOfTwo.hpp"
 45 
 46 template <StringDedupMode STRING_DEDUP>
 47 void ShenandoahMark::dedup_string(oop obj, StringDedup::Requests* const req) {
 48   if (STRING_DEDUP == ENQUEUE_DEDUP) {
 49     if (ShenandoahStringDedup::is_candidate(obj)) {
 50       req->add(obj);
 51     }
 52   } else if (STRING_DEDUP == ALWAYS_DEDUP) {
 53     if (ShenandoahStringDedup::is_string_candidate(obj) &&
 54         !ShenandoahStringDedup::dedup_requested(obj)) {
 55         req->add(obj);
 56     }
 57   }
 58 }
 59 
 60 template <class T, ShenandoahGenerationType GENERATION, StringDedupMode STRING_DEDUP>
 61 void ShenandoahMark::do_task(ShenandoahObjToScanQueue* q, T* cl, ShenandoahLiveData* live_data, StringDedup::Requests* const req, ShenandoahMarkTask* task, uint worker_id) {
 62   oop obj = task->obj();
 63 
 64   // TODO: This will push array chunks into the mark queue with no regard for
 65   // generations. I don't think it will break anything, but the young generation
 66   // scan might end up processing some old generation array chunks.
 67 
 68   shenandoah_assert_not_forwarded(nullptr, obj);
 69   shenandoah_assert_marked(nullptr, obj);
 70   shenandoah_assert_not_in_cset_except(nullptr, obj, ShenandoahHeap::heap()->cancelled_gc());
 71 
 72   // Are we in weak subgraph scan?
 73   bool weak = task->is_weak();
 74   cl->set_weak(weak);
 75 
 76   if (task->is_not_chunked()) {
 77     if (obj->is_instance()) {
 78       // Case 1: Normal oop, process as usual.
 79       if (ContinuationGCSupport::relativize_stack_chunk(obj)) {
 80           // Loom doesn't support mixing of weak marking and strong marking of
 81           // stack chunks.
 82           cl->set_weak(false);
 83       }
 84 
 85       obj->oop_iterate(cl);
 86       dedup_string<STRING_DEDUP>(obj, req);
 87     } else if (obj->is_objArray()) {
 88       // Case 2: Object array instance and no chunk is set. Must be the first
 89       // time we visit it, start the chunked processing.
 90       do_chunked_array_start<T>(q, cl, obj, weak);
 91     } else {
 92       // Case 3: Primitive array. Do nothing, no oops there. We use the same
 93       // performance tweak TypeArrayKlass::oop_oop_iterate_impl is using:
 94       // We skip iterating over the klass pointer since we know that
 95       // Universe::TypeArrayKlass never moves.
 96       assert (obj->is_typeArray(), "should be type array");
 97     }
 98     // Count liveness the last: push the outstanding work to the queues first
 99     // Avoid double-counting objects that are visited twice due to upgrade
100     // from final- to strong mark.
101     if (task->count_liveness()) {
102       count_liveness<GENERATION>(live_data, obj, worker_id);
103     }
104   } else {
105     // Case 4: Array chunk, has sensible chunk id. Process it.
106     do_chunked_array<T>(q, cl, obj, task->chunk(), task->pow(), weak);
107   }
108 }
109 
110 template <ShenandoahGenerationType GENERATION>
111 inline void ShenandoahMark::count_liveness(ShenandoahLiveData* live_data, oop obj, uint worker_id) {
112   const ShenandoahHeap* const heap = ShenandoahHeap::heap();
113   const size_t region_idx = heap->heap_region_index_containing(obj);
114   ShenandoahHeapRegion* const region = heap->get_region(region_idx);
115   const size_t size = obj->size();
116 
117   // Age census for objects in the young generation
118   if (GENERATION == YOUNG || (GENERATION == GLOBAL_GEN && region->is_young())) {
119     assert(heap->mode()->is_generational(), "Only if generational");
120     if (ShenandoahGenerationalAdaptiveTenuring && !ShenandoahGenerationalCensusAtEvac) {
121       assert(region->is_young(), "Only for young objects");
122       uint age = ShenandoahHeap::get_object_age_concurrent(obj);
123       CENSUS_NOISE(heap->age_census()->add(age, region->age(), region->youth(), size, worker_id);)
124       NO_CENSUS_NOISE(heap->age_census()->add(age, region->age(), size, worker_id);)
125     }
126   }
127 
128   if (!region->is_humongous_start()) {
129     assert(!region->is_humongous(), "Cannot have continuations here");
130     assert(region->is_affiliated(), "Do not count live data within Free Regular Region " SIZE_FORMAT, region_idx);
131     ShenandoahLiveData cur = live_data[region_idx];
132     size_t new_val = size + cur;
133     if (new_val >= SHENANDOAH_LIVEDATA_MAX) {
134       // overflow, flush to region data
135       region->increase_live_data_gc_words(new_val);
136       live_data[region_idx] = 0;
137     } else {
138       // still good, remember in locals
139       live_data[region_idx] = (ShenandoahLiveData) new_val;
140     }
141   } else {
142     shenandoah_assert_in_correct_region(nullptr, obj);
143     size_t num_regions = ShenandoahHeapRegion::required_regions(size * HeapWordSize);
144 
145     assert(region->is_affiliated(), "Do not count live data within FREE Humongous Start Region " SIZE_FORMAT, region_idx);
146     for (size_t i = region_idx; i < region_idx + num_regions; i++) {
147       ShenandoahHeapRegion* chain_reg = heap->get_region(i);
148       assert(chain_reg->is_humongous(), "Expecting a humongous region");
149       assert(chain_reg->is_affiliated(), "Do not count live data within FREE Humongous Continuation Region " SIZE_FORMAT, i);
150       chain_reg->increase_live_data_gc_words(chain_reg->used() >> LogHeapWordSize);
151     }
152   }
153 }
154 
155 template <class T>
156 inline void ShenandoahMark::do_chunked_array_start(ShenandoahObjToScanQueue* q, T* cl, oop obj, bool weak) {
157   assert(obj->is_objArray(), "expect object array");
158   objArrayOop array = objArrayOop(obj);
159   int len = array->length();
160 
161   // Mark objArray klass metadata
162   if (Devirtualizer::do_metadata(cl)) {
163     Devirtualizer::do_klass(cl, array->klass());
164   }
165 
166   if (len <= (int) ObjArrayMarkingStride*2) {
167     // A few slices only, process directly
168     array->oop_iterate_range(cl, 0, len);
169   } else {
170     int bits = log2i_graceful(len);
171     // Compensate for non-power-of-two arrays, cover the array in excess:
172     if (len != (1 << bits)) bits++;
173 
174     // Only allow full chunks on the queue. This frees do_chunked_array() from checking from/to
175     // boundaries against array->length(), touching the array header on every chunk.
176     //
177     // To do this, we cut the prefix in full-sized chunks, and submit them on the queue.
178     // If the array is not divided in chunk sizes, then there would be an irregular tail,
179     // which we will process separately.
180 
181     int last_idx = 0;
182 
183     int chunk = 1;
184     int pow = bits;
185 
186     // Handle overflow
187     if (pow >= 31) {
188       assert (pow == 31, "sanity");
189       pow--;
190       chunk = 2;
191       last_idx = (1 << pow);
192       bool pushed = q->push(ShenandoahMarkTask(array, true, weak, 1, pow));
193       assert(pushed, "overflow queue should always succeed pushing");
194     }
195 
196     // Split out tasks, as suggested in ShenandoahMarkTask docs. Record the last
197     // successful right boundary to figure out the irregular tail.
198     while ((1 << pow) > (int)ObjArrayMarkingStride &&
199            (chunk*2 < ShenandoahMarkTask::chunk_size())) {
200       pow--;
201       int left_chunk = chunk*2 - 1;
202       int right_chunk = chunk*2;
203       int left_chunk_end = left_chunk * (1 << pow);
204       if (left_chunk_end < len) {
205         bool pushed = q->push(ShenandoahMarkTask(array, true, weak, left_chunk, pow));
206         assert(pushed, "overflow queue should always succeed pushing");
207         chunk = right_chunk;
208         last_idx = left_chunk_end;
209       } else {
210         chunk = left_chunk;
211       }
212     }
213 
214     // Process the irregular tail, if present
215     int from = last_idx;
216     if (from < len) {
217       array->oop_iterate_range(cl, from, len);
218     }
219   }
220 }
221 
222 template <class T>
223 inline void ShenandoahMark::do_chunked_array(ShenandoahObjToScanQueue* q, T* cl, oop obj, int chunk, int pow, bool weak) {
224   assert(obj->is_objArray(), "expect object array");
225   objArrayOop array = objArrayOop(obj);
226 
227   assert (ObjArrayMarkingStride > 0, "sanity");
228 
229   // Split out tasks, as suggested in ShenandoahMarkTask docs. Avoid pushing tasks that
230   // are known to start beyond the array.
231   while ((1 << pow) > (int)ObjArrayMarkingStride && (chunk*2 < ShenandoahMarkTask::chunk_size())) {
232     pow--;
233     chunk *= 2;
234     bool pushed = q->push(ShenandoahMarkTask(array, true, weak, chunk - 1, pow));
235     assert(pushed, "overflow queue should always succeed pushing");
236   }
237 
238   int chunk_size = 1 << pow;
239 
240   int from = (chunk - 1) * chunk_size;
241   int to = chunk * chunk_size;
242 
243 #ifdef ASSERT
244   int len = array->length();
245   assert (0 <= from && from < len, "from is sane: %d/%d", from, len);
246   assert (0 < to && to <= len, "to is sane: %d/%d", to, len);
247 #endif
248 
249   array->oop_iterate_range(cl, from, to);
250 }
251 
252 template <ShenandoahGenerationType GENERATION>
253 class ShenandoahSATBBufferClosure : public SATBBufferClosure {
254 private:
255   ShenandoahObjToScanQueue* _queue;
256   ShenandoahObjToScanQueue* _old_queue;
257   ShenandoahHeap* _heap;
258   ShenandoahMarkingContext* const _mark_context;
259 public:
260   ShenandoahSATBBufferClosure(ShenandoahObjToScanQueue* q, ShenandoahObjToScanQueue* old_q) :
261     _queue(q),
262     _old_queue(old_q),
263     _heap(ShenandoahHeap::heap()),
264     _mark_context(_heap->marking_context())
265   {
266   }
267 
268   void do_buffer(void **buffer, size_t size) {
269     assert(size == 0 || !_heap->has_forwarded_objects() || _heap->is_concurrent_old_mark_in_progress(), "Forwarded objects are not expected here");
270     for (size_t i = 0; i < size; ++i) {
271       oop *p = (oop *) &buffer[i];
272       ShenandoahMark::mark_through_ref<oop, GENERATION>(p, _queue, _old_queue, _mark_context, false);
273     }
274   }
275 };
276 
277 template<ShenandoahGenerationType GENERATION>
278 bool ShenandoahMark::in_generation(ShenandoahHeap* const heap, oop obj) {
279   // Each in-line expansion of in_generation() resolves GENERATION at compile time.
280   if (GENERATION == YOUNG) {
281     return heap->is_in_young(obj);
282   } else if (GENERATION == OLD) {
283     return heap->is_in_old(obj);
284   } else if (GENERATION == GLOBAL_GEN || GENERATION == GLOBAL_NON_GEN) {
285     return true;
286   } else {
287     return false;
288   }
289 }
290 
291 template<class T, ShenandoahGenerationType GENERATION>
292 inline void ShenandoahMark::mark_through_ref(T *p, ShenandoahObjToScanQueue* q, ShenandoahObjToScanQueue* old_q, ShenandoahMarkingContext* const mark_context, bool weak) {
293   // Note: This is a very hot code path, so the code should be conditional on GENERATION template
294   // parameter where possible, in order to generate the most efficient code.
295 
296   T o = RawAccess<>::oop_load(p);
297   if (!CompressedOops::is_null(o)) {
298     oop obj = CompressedOops::decode_not_null(o);
299 
300     ShenandoahHeap* heap = ShenandoahHeap::heap();
301     shenandoah_assert_not_forwarded(p, obj);
302     shenandoah_assert_not_in_cset_except(p, obj, heap->cancelled_gc());
303     if (in_generation<GENERATION>(heap, obj)) {
304       mark_ref(q, mark_context, weak, obj);
305       shenandoah_assert_marked(p, obj);
306       // TODO: As implemented herein, GLOBAL_GEN collections reconstruct the card table during GLOBAL_GEN concurrent
307       // marking. Note that the card table is cleaned at init_mark time so it needs to be reconstructed to support
308       // future young-gen collections.  It might be better to reconstruct card table in
309       // ShenandoahHeapRegion::global_oop_iterate_and_fill_dead.  We could either mark all live memory as dirty, or could
310       // use the GLOBAL update-refs scanning of pointers to determine precisely which cards to flag as dirty.
311       if (GENERATION == YOUNG && heap->is_in_old(p)) {
312         // Mark card as dirty because remembered set scanning still finds interesting pointer.
313         heap->mark_card_as_dirty((HeapWord*)p);
314       } else if (GENERATION == GLOBAL_GEN && heap->is_in_old(p) && heap->is_in_young(obj)) {
315         // Mark card as dirty because GLOBAL marking finds interesting pointer.
316         heap->mark_card_as_dirty((HeapWord*)p);
317       }
318     } else if (old_q != nullptr) {
319       // Young mark, bootstrapping old_q or concurrent with old_q marking.
320       mark_ref(old_q, mark_context, weak, obj);
321       shenandoah_assert_marked(p, obj);
322     } else if (GENERATION == OLD) {
323       // Old mark, found a young pointer.
324       // TODO:  Rethink this: may be redundant with dirtying of cards identified during young-gen remembered set scanning
325       // and by mutator write barriers.  Assert
326       if (heap->is_in(p)) {
327         assert(heap->is_in_young(obj), "Expected young object.");
328         heap->mark_card_as_dirty(p);
329       }
330     }
331   }
332 }
333 
334 inline void ShenandoahMark::mark_ref(ShenandoahObjToScanQueue* q,
335                               ShenandoahMarkingContext* const mark_context,
336                               bool weak, oop obj) {
337   bool skip_live = false;
338   bool marked;
339   if (weak) {
340     marked = mark_context->mark_weak(obj);
341   } else {
342     marked = mark_context->mark_strong(obj, /* was_upgraded = */ skip_live);
343   }
344   if (marked) {
345     bool pushed = q->push(ShenandoahMarkTask(obj, skip_live, weak));
346     assert(pushed, "overflow queue should always succeed pushing");
347   }
348 }
349 
350 ShenandoahObjToScanQueueSet* ShenandoahMark::task_queues() const {
351   return _task_queues;
352 }
353 
354 ShenandoahObjToScanQueue* ShenandoahMark::get_queue(uint index) const {
355   return _task_queues->queue(index);
356 }
357 
358 ShenandoahObjToScanQueue* ShenandoahMark::get_old_queue(uint index) const {
359   if (_old_gen_task_queues != nullptr) {
360     return _old_gen_task_queues->queue(index);
361   }
362   return nullptr;
363 }
364 
365 #endif // SHARE_GC_SHENANDOAH_SHENANDOAHMARK_INLINE_HPP