1 /*
2 * Copyright (C) 2011 The Android Open Source Project
3 *
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at
7 *
8 * http://www.apache.org/licenses/LICENSE-2.0
9 *
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
15 */
16
17 #include "heap.h"
18
19 #define ATRACE_TAG ATRACE_TAG_DALVIK
20 #include <cutils/trace.h>
21
22 #include <limits>
23 #include <vector>
24 #include <valgrind.h>
25
26 #include "base/stl_util.h"
27 #include "common_throws.h"
28 #include "cutils/sched_policy.h"
29 #include "debugger.h"
30 #include "gc/accounting/atomic_stack.h"
31 #include "gc/accounting/card_table-inl.h"
32 #include "gc/accounting/heap_bitmap-inl.h"
33 #include "gc/accounting/mod_union_table-inl.h"
34 #include "gc/accounting/space_bitmap-inl.h"
35 #include "gc/collector/mark_sweep-inl.h"
36 #include "gc/collector/partial_mark_sweep.h"
37 #include "gc/collector/sticky_mark_sweep.h"
38 #include "gc/space/dlmalloc_space-inl.h"
39 #include "gc/space/image_space.h"
40 #include "gc/space/large_object_space.h"
41 #include "gc/space/space-inl.h"
42 #include "image.h"
43 #include "invoke_arg_array_builder.h"
44 #include "mirror/art_field-inl.h"
45 #include "mirror/class-inl.h"
46 #include "mirror/object.h"
47 #include "mirror/object-inl.h"
48 #include "mirror/object_array-inl.h"
49 #include "object_utils.h"
50 #include "os.h"
51 #include "ScopedLocalRef.h"
52 #include "scoped_thread_state_change.h"
53 #include "sirt_ref.h"
54 #include "thread_list.h"
55 #include "UniquePtr.h"
56 #include "well_known_classes.h"
57
58 namespace art {
59 namespace gc {
60
61 static constexpr bool kGCALotMode = false;
62 static constexpr size_t kGcAlotInterval = KB;
63 static constexpr bool kDumpGcPerformanceOnShutdown = false;
64 // Minimum amount of remaining bytes before a concurrent GC is triggered.
65 static constexpr size_t kMinConcurrentRemainingBytes = 128 * KB;
66 // If true, measure the total allocation time.
67 static constexpr bool kMeasureAllocationTime = false;
68
Heap(size_t initial_size,size_t growth_limit,size_t min_free,size_t max_free,double target_utilization,size_t capacity,const std::string & original_image_file_name,bool concurrent_gc,size_t parallel_gc_threads,size_t conc_gc_threads,bool low_memory_mode,size_t long_pause_log_threshold,size_t long_gc_log_threshold,bool ignore_max_footprint)69 Heap::Heap(size_t initial_size, size_t growth_limit, size_t min_free, size_t max_free,
70 double target_utilization, size_t capacity, const std::string& original_image_file_name,
71 bool concurrent_gc, size_t parallel_gc_threads, size_t conc_gc_threads,
72 bool low_memory_mode, size_t long_pause_log_threshold, size_t long_gc_log_threshold,
73 bool ignore_max_footprint)
74 : alloc_space_(NULL),
75 card_table_(NULL),
76 concurrent_gc_(concurrent_gc),
77 parallel_gc_threads_(parallel_gc_threads),
78 conc_gc_threads_(conc_gc_threads),
79 low_memory_mode_(low_memory_mode),
80 long_pause_log_threshold_(long_pause_log_threshold),
81 long_gc_log_threshold_(long_gc_log_threshold),
82 ignore_max_footprint_(ignore_max_footprint),
83 have_zygote_space_(false),
84 soft_ref_queue_lock_(NULL),
85 weak_ref_queue_lock_(NULL),
86 finalizer_ref_queue_lock_(NULL),
87 phantom_ref_queue_lock_(NULL),
88 is_gc_running_(false),
89 last_gc_type_(collector::kGcTypeNone),
90 next_gc_type_(collector::kGcTypePartial),
91 capacity_(capacity),
92 growth_limit_(growth_limit),
93 max_allowed_footprint_(initial_size),
94 native_footprint_gc_watermark_(initial_size),
95 native_footprint_limit_(2 * initial_size),
96 activity_thread_class_(NULL),
97 application_thread_class_(NULL),
98 activity_thread_(NULL),
99 application_thread_(NULL),
100 last_process_state_id_(NULL),
101 // Initially care about pauses in case we never get notified of process states, or if the JNI
102 // code becomes broken.
103 care_about_pause_times_(true),
104 concurrent_start_bytes_(concurrent_gc_ ? initial_size - kMinConcurrentRemainingBytes
105 : std::numeric_limits<size_t>::max()),
106 total_bytes_freed_ever_(0),
107 total_objects_freed_ever_(0),
108 large_object_threshold_(3 * kPageSize),
109 num_bytes_allocated_(0),
110 native_bytes_allocated_(0),
111 gc_memory_overhead_(0),
112 verify_missing_card_marks_(false),
113 verify_system_weaks_(false),
114 verify_pre_gc_heap_(false),
115 verify_post_gc_heap_(false),
116 verify_mod_union_table_(false),
117 min_alloc_space_size_for_sticky_gc_(2 * MB),
118 min_remaining_space_for_sticky_gc_(1 * MB),
119 last_trim_time_ms_(0),
120 allocation_rate_(0),
121 /* For GC a lot mode, we limit the allocations stacks to be kGcAlotInterval allocations. This
122 * causes a lot of GC since we do a GC for alloc whenever the stack is full. When heap
123 * verification is enabled, we limit the size of allocation stacks to speed up their
124 * searching.
125 */
126 max_allocation_stack_size_(kGCALotMode ? kGcAlotInterval
127 : (kDesiredHeapVerification > kNoHeapVerification) ? KB : MB),
128 reference_referent_offset_(0),
129 reference_queue_offset_(0),
130 reference_queueNext_offset_(0),
131 reference_pendingNext_offset_(0),
132 finalizer_reference_zombie_offset_(0),
133 min_free_(min_free),
134 max_free_(max_free),
135 target_utilization_(target_utilization),
136 total_wait_time_(0),
137 total_allocation_time_(0),
138 verify_object_mode_(kHeapVerificationNotPermitted),
139 running_on_valgrind_(RUNNING_ON_VALGRIND) {
140 if (VLOG_IS_ON(heap) || VLOG_IS_ON(startup)) {
141 LOG(INFO) << "Heap() entering";
142 }
143
144 live_bitmap_.reset(new accounting::HeapBitmap(this));
145 mark_bitmap_.reset(new accounting::HeapBitmap(this));
146
147 // Requested begin for the alloc space, to follow the mapped image and oat files
148 byte* requested_alloc_space_begin = NULL;
149 std::string image_file_name(original_image_file_name);
150 if (!image_file_name.empty()) {
151 space::ImageSpace* image_space = space::ImageSpace::Create(image_file_name);
152 CHECK(image_space != NULL) << "Failed to create space for " << image_file_name;
153 AddContinuousSpace(image_space);
154 // Oat files referenced by image files immediately follow them in memory, ensure alloc space
155 // isn't going to get in the middle
156 byte* oat_file_end_addr = image_space->GetImageHeader().GetOatFileEnd();
157 CHECK_GT(oat_file_end_addr, image_space->End());
158 if (oat_file_end_addr > requested_alloc_space_begin) {
159 requested_alloc_space_begin =
160 reinterpret_cast<byte*>(RoundUp(reinterpret_cast<uintptr_t>(oat_file_end_addr),
161 kPageSize));
162 }
163 }
164
165 alloc_space_ = space::DlMallocSpace::Create(Runtime::Current()->IsZygote() ? "zygote space" : "alloc space",
166 initial_size,
167 growth_limit, capacity,
168 requested_alloc_space_begin);
169 CHECK(alloc_space_ != NULL) << "Failed to create alloc space";
170 alloc_space_->SetFootprintLimit(alloc_space_->Capacity());
171 AddContinuousSpace(alloc_space_);
172
173 // Allocate the large object space.
174 const bool kUseFreeListSpaceForLOS = false;
175 if (kUseFreeListSpaceForLOS) {
176 large_object_space_ = space::FreeListSpace::Create("large object space", NULL, capacity);
177 } else {
178 large_object_space_ = space::LargeObjectMapSpace::Create("large object space");
179 }
180 CHECK(large_object_space_ != NULL) << "Failed to create large object space";
181 AddDiscontinuousSpace(large_object_space_);
182
183 // Compute heap capacity. Continuous spaces are sorted in order of Begin().
184 byte* heap_begin = continuous_spaces_.front()->Begin();
185 size_t heap_capacity = continuous_spaces_.back()->End() - continuous_spaces_.front()->Begin();
186 if (continuous_spaces_.back()->IsDlMallocSpace()) {
187 heap_capacity += continuous_spaces_.back()->AsDlMallocSpace()->NonGrowthLimitCapacity();
188 }
189
190 // Allocate the card table.
191 card_table_.reset(accounting::CardTable::Create(heap_begin, heap_capacity));
192 CHECK(card_table_.get() != NULL) << "Failed to create card table";
193
194 image_mod_union_table_.reset(new accounting::ModUnionTableToZygoteAllocspace(this));
195 CHECK(image_mod_union_table_.get() != NULL) << "Failed to create image mod-union table";
196
197 zygote_mod_union_table_.reset(new accounting::ModUnionTableCardCache(this));
198 CHECK(zygote_mod_union_table_.get() != NULL) << "Failed to create Zygote mod-union table";
199
200 // TODO: Count objects in the image space here.
201 num_bytes_allocated_ = 0;
202
203 // Default mark stack size in bytes.
204 static const size_t default_mark_stack_size = 64 * KB;
205 mark_stack_.reset(accounting::ObjectStack::Create("mark stack", default_mark_stack_size));
206 allocation_stack_.reset(accounting::ObjectStack::Create("allocation stack",
207 max_allocation_stack_size_));
208 live_stack_.reset(accounting::ObjectStack::Create("live stack",
209 max_allocation_stack_size_));
210
211 // It's still too early to take a lock because there are no threads yet, but we can create locks
212 // now. We don't create it earlier to make it clear that you can't use locks during heap
213 // initialization.
214 gc_complete_lock_ = new Mutex("GC complete lock");
215 gc_complete_cond_.reset(new ConditionVariable("GC complete condition variable",
216 *gc_complete_lock_));
217
218 // Create the reference queue locks, this is required so for parallel object scanning in the GC.
219 soft_ref_queue_lock_ = new Mutex("Soft reference queue lock");
220 weak_ref_queue_lock_ = new Mutex("Weak reference queue lock");
221 finalizer_ref_queue_lock_ = new Mutex("Finalizer reference queue lock");
222 phantom_ref_queue_lock_ = new Mutex("Phantom reference queue lock");
223
224 last_gc_time_ns_ = NanoTime();
225 last_gc_size_ = GetBytesAllocated();
226
227 if (ignore_max_footprint_) {
228 SetIdealFootprint(std::numeric_limits<size_t>::max());
229 concurrent_start_bytes_ = max_allowed_footprint_;
230 }
231
232 // Create our garbage collectors.
233 for (size_t i = 0; i < 2; ++i) {
234 const bool concurrent = i != 0;
235 mark_sweep_collectors_.push_back(new collector::MarkSweep(this, concurrent));
236 mark_sweep_collectors_.push_back(new collector::PartialMarkSweep(this, concurrent));
237 mark_sweep_collectors_.push_back(new collector::StickyMarkSweep(this, concurrent));
238 }
239
240 CHECK_NE(max_allowed_footprint_, 0U);
241 if (VLOG_IS_ON(heap) || VLOG_IS_ON(startup)) {
242 LOG(INFO) << "Heap() exiting";
243 }
244 }
245
CreateThreadPool()246 void Heap::CreateThreadPool() {
247 const size_t num_threads = std::max(parallel_gc_threads_, conc_gc_threads_);
248 if (num_threads != 0) {
249 thread_pool_.reset(new ThreadPool(num_threads));
250 }
251 }
252
DeleteThreadPool()253 void Heap::DeleteThreadPool() {
254 thread_pool_.reset(nullptr);
255 }
256
ReadStaticInt(JNIEnvExt * env,jclass clz,const char * name,int * out_value)257 static bool ReadStaticInt(JNIEnvExt* env, jclass clz, const char* name, int* out_value) {
258 CHECK(out_value != NULL);
259 jfieldID field = env->GetStaticFieldID(clz, name, "I");
260 if (field == NULL) {
261 env->ExceptionClear();
262 return false;
263 }
264 *out_value = env->GetStaticIntField(clz, field);
265 return true;
266 }
267
ListenForProcessStateChange()268 void Heap::ListenForProcessStateChange() {
269 VLOG(heap) << "Heap notified of process state change";
270
271 Thread* self = Thread::Current();
272 JNIEnvExt* env = self->GetJniEnv();
273
274 if (!have_zygote_space_) {
275 return;
276 }
277
278 if (activity_thread_class_ == NULL) {
279 jclass clz = env->FindClass("android/app/ActivityThread");
280 if (clz == NULL) {
281 env->ExceptionClear();
282 LOG(WARNING) << "Could not find activity thread class in process state change";
283 return;
284 }
285 activity_thread_class_ = reinterpret_cast<jclass>(env->NewGlobalRef(clz));
286 }
287
288 if (activity_thread_class_ != NULL && activity_thread_ == NULL) {
289 jmethodID current_activity_method = env->GetStaticMethodID(activity_thread_class_,
290 "currentActivityThread",
291 "()Landroid/app/ActivityThread;");
292 if (current_activity_method == NULL) {
293 env->ExceptionClear();
294 LOG(WARNING) << "Could not get method for currentActivityThread";
295 return;
296 }
297
298 jobject obj = env->CallStaticObjectMethod(activity_thread_class_, current_activity_method);
299 if (obj == NULL) {
300 env->ExceptionClear();
301 LOG(WARNING) << "Could not get current activity";
302 return;
303 }
304 activity_thread_ = env->NewGlobalRef(obj);
305 }
306
307 if (process_state_cares_about_pause_time_.empty()) {
308 // Just attempt to do this the first time.
309 jclass clz = env->FindClass("android/app/ActivityManager");
310 if (clz == NULL) {
311 LOG(WARNING) << "Activity manager class is null";
312 return;
313 }
314 ScopedLocalRef<jclass> activity_manager(env, clz);
315 std::vector<const char*> care_about_pauses;
316 care_about_pauses.push_back("PROCESS_STATE_TOP");
317 care_about_pauses.push_back("PROCESS_STATE_IMPORTANT_BACKGROUND");
318 // Attempt to read the constants and classify them as whether or not we care about pause times.
319 for (size_t i = 0; i < care_about_pauses.size(); ++i) {
320 int process_state = 0;
321 if (ReadStaticInt(env, activity_manager.get(), care_about_pauses[i], &process_state)) {
322 process_state_cares_about_pause_time_.insert(process_state);
323 VLOG(heap) << "Adding process state " << process_state
324 << " to set of states which care about pause time";
325 }
326 }
327 }
328
329 if (application_thread_class_ == NULL) {
330 jclass clz = env->FindClass("android/app/ActivityThread$ApplicationThread");
331 if (clz == NULL) {
332 env->ExceptionClear();
333 LOG(WARNING) << "Could not get application thread class";
334 return;
335 }
336 application_thread_class_ = reinterpret_cast<jclass>(env->NewGlobalRef(clz));
337 last_process_state_id_ = env->GetFieldID(application_thread_class_, "mLastProcessState", "I");
338 if (last_process_state_id_ == NULL) {
339 env->ExceptionClear();
340 LOG(WARNING) << "Could not get last process state member";
341 return;
342 }
343 }
344
345 if (application_thread_class_ != NULL && application_thread_ == NULL) {
346 jmethodID get_application_thread =
347 env->GetMethodID(activity_thread_class_, "getApplicationThread",
348 "()Landroid/app/ActivityThread$ApplicationThread;");
349 if (get_application_thread == NULL) {
350 LOG(WARNING) << "Could not get method ID for get application thread";
351 return;
352 }
353
354 jobject obj = env->CallObjectMethod(activity_thread_, get_application_thread);
355 if (obj == NULL) {
356 LOG(WARNING) << "Could not get application thread";
357 return;
358 }
359
360 application_thread_ = env->NewGlobalRef(obj);
361 }
362
363 if (application_thread_ != NULL && last_process_state_id_ != NULL) {
364 int process_state = env->GetIntField(application_thread_, last_process_state_id_);
365 env->ExceptionClear();
366
367 care_about_pause_times_ = process_state_cares_about_pause_time_.find(process_state) !=
368 process_state_cares_about_pause_time_.end();
369
370 VLOG(heap) << "New process state " << process_state
371 << " care about pauses " << care_about_pause_times_;
372 }
373 }
374
AddContinuousSpace(space::ContinuousSpace * space)375 void Heap::AddContinuousSpace(space::ContinuousSpace* space) {
376 WriterMutexLock mu(Thread::Current(), *Locks::heap_bitmap_lock_);
377 DCHECK(space != NULL);
378 DCHECK(space->GetLiveBitmap() != NULL);
379 live_bitmap_->AddContinuousSpaceBitmap(space->GetLiveBitmap());
380 DCHECK(space->GetMarkBitmap() != NULL);
381 mark_bitmap_->AddContinuousSpaceBitmap(space->GetMarkBitmap());
382 continuous_spaces_.push_back(space);
383 if (space->IsDlMallocSpace() && !space->IsLargeObjectSpace()) {
384 alloc_space_ = space->AsDlMallocSpace();
385 }
386
387 // Ensure that spaces remain sorted in increasing order of start address (required for CMS finger)
388 std::sort(continuous_spaces_.begin(), continuous_spaces_.end(),
389 [](const space::ContinuousSpace* a, const space::ContinuousSpace* b) {
390 return a->Begin() < b->Begin();
391 });
392
393 // Ensure that ImageSpaces < ZygoteSpaces < AllocSpaces so that we can do address based checks to
394 // avoid redundant marking.
395 bool seen_zygote = false, seen_alloc = false;
396 for (const auto& space : continuous_spaces_) {
397 if (space->IsImageSpace()) {
398 DCHECK(!seen_zygote);
399 DCHECK(!seen_alloc);
400 } else if (space->IsZygoteSpace()) {
401 DCHECK(!seen_alloc);
402 seen_zygote = true;
403 } else if (space->IsDlMallocSpace()) {
404 seen_alloc = true;
405 }
406 }
407 }
408
RegisterGCAllocation(size_t bytes)409 void Heap::RegisterGCAllocation(size_t bytes) {
410 if (this != NULL) {
411 gc_memory_overhead_.fetch_add(bytes);
412 }
413 }
414
RegisterGCDeAllocation(size_t bytes)415 void Heap::RegisterGCDeAllocation(size_t bytes) {
416 if (this != NULL) {
417 gc_memory_overhead_.fetch_sub(bytes);
418 }
419 }
420
AddDiscontinuousSpace(space::DiscontinuousSpace * space)421 void Heap::AddDiscontinuousSpace(space::DiscontinuousSpace* space) {
422 WriterMutexLock mu(Thread::Current(), *Locks::heap_bitmap_lock_);
423 DCHECK(space != NULL);
424 DCHECK(space->GetLiveObjects() != NULL);
425 live_bitmap_->AddDiscontinuousObjectSet(space->GetLiveObjects());
426 DCHECK(space->GetMarkObjects() != NULL);
427 mark_bitmap_->AddDiscontinuousObjectSet(space->GetMarkObjects());
428 discontinuous_spaces_.push_back(space);
429 }
430
DumpGcPerformanceInfo(std::ostream & os)431 void Heap::DumpGcPerformanceInfo(std::ostream& os) {
432 // Dump cumulative timings.
433 os << "Dumping cumulative Gc timings\n";
434 uint64_t total_duration = 0;
435
436 // Dump cumulative loggers for each GC type.
437 uint64_t total_paused_time = 0;
438 for (const auto& collector : mark_sweep_collectors_) {
439 CumulativeLogger& logger = collector->GetCumulativeTimings();
440 if (logger.GetTotalNs() != 0) {
441 os << Dumpable<CumulativeLogger>(logger);
442 const uint64_t total_ns = logger.GetTotalNs();
443 const uint64_t total_pause_ns = collector->GetTotalPausedTimeNs();
444 double seconds = NsToMs(logger.GetTotalNs()) / 1000.0;
445 const uint64_t freed_bytes = collector->GetTotalFreedBytes();
446 const uint64_t freed_objects = collector->GetTotalFreedObjects();
447 os << collector->GetName() << " total time: " << PrettyDuration(total_ns) << "\n"
448 << collector->GetName() << " paused time: " << PrettyDuration(total_pause_ns) << "\n"
449 << collector->GetName() << " freed: " << freed_objects
450 << " objects with total size " << PrettySize(freed_bytes) << "\n"
451 << collector->GetName() << " throughput: " << freed_objects / seconds << "/s / "
452 << PrettySize(freed_bytes / seconds) << "/s\n";
453 total_duration += total_ns;
454 total_paused_time += total_pause_ns;
455 }
456 }
457 uint64_t allocation_time = static_cast<uint64_t>(total_allocation_time_) * kTimeAdjust;
458 size_t total_objects_allocated = GetObjectsAllocatedEver();
459 size_t total_bytes_allocated = GetBytesAllocatedEver();
460 if (total_duration != 0) {
461 const double total_seconds = static_cast<double>(total_duration / 1000) / 1000000.0;
462 os << "Total time spent in GC: " << PrettyDuration(total_duration) << "\n";
463 os << "Mean GC size throughput: "
464 << PrettySize(GetBytesFreedEver() / total_seconds) << "/s\n";
465 os << "Mean GC object throughput: "
466 << (GetObjectsFreedEver() / total_seconds) << " objects/s\n";
467 }
468 os << "Total number of allocations: " << total_objects_allocated << "\n";
469 os << "Total bytes allocated " << PrettySize(total_bytes_allocated) << "\n";
470 if (kMeasureAllocationTime) {
471 os << "Total time spent allocating: " << PrettyDuration(allocation_time) << "\n";
472 os << "Mean allocation time: " << PrettyDuration(allocation_time / total_objects_allocated)
473 << "\n";
474 }
475 os << "Total mutator paused time: " << PrettyDuration(total_paused_time) << "\n";
476 os << "Total time waiting for GC to complete: " << PrettyDuration(total_wait_time_) << "\n";
477 os << "Approximate GC data structures memory overhead: " << gc_memory_overhead_;
478 }
479
~Heap()480 Heap::~Heap() {
481 if (kDumpGcPerformanceOnShutdown) {
482 DumpGcPerformanceInfo(LOG(INFO));
483 }
484
485 STLDeleteElements(&mark_sweep_collectors_);
486
487 // If we don't reset then the mark stack complains in it's destructor.
488 allocation_stack_->Reset();
489 live_stack_->Reset();
490
491 VLOG(heap) << "~Heap()";
492 // We can't take the heap lock here because there might be a daemon thread suspended with the
493 // heap lock held. We know though that no non-daemon threads are executing, and we know that
494 // all daemon threads are suspended, and we also know that the threads list have been deleted, so
495 // those threads can't resume. We're the only running thread, and we can do whatever we like...
496 STLDeleteElements(&continuous_spaces_);
497 STLDeleteElements(&discontinuous_spaces_);
498 delete gc_complete_lock_;
499 delete soft_ref_queue_lock_;
500 delete weak_ref_queue_lock_;
501 delete finalizer_ref_queue_lock_;
502 delete phantom_ref_queue_lock_;
503 }
504
FindContinuousSpaceFromObject(const mirror::Object * obj,bool fail_ok) const505 space::ContinuousSpace* Heap::FindContinuousSpaceFromObject(const mirror::Object* obj,
506 bool fail_ok) const {
507 for (const auto& space : continuous_spaces_) {
508 if (space->Contains(obj)) {
509 return space;
510 }
511 }
512 if (!fail_ok) {
513 LOG(FATAL) << "object " << reinterpret_cast<const void*>(obj) << " not inside any spaces!";
514 }
515 return NULL;
516 }
517
FindDiscontinuousSpaceFromObject(const mirror::Object * obj,bool fail_ok) const518 space::DiscontinuousSpace* Heap::FindDiscontinuousSpaceFromObject(const mirror::Object* obj,
519 bool fail_ok) const {
520 for (const auto& space : discontinuous_spaces_) {
521 if (space->Contains(obj)) {
522 return space;
523 }
524 }
525 if (!fail_ok) {
526 LOG(FATAL) << "object " << reinterpret_cast<const void*>(obj) << " not inside any spaces!";
527 }
528 return NULL;
529 }
530
FindSpaceFromObject(const mirror::Object * obj,bool fail_ok) const531 space::Space* Heap::FindSpaceFromObject(const mirror::Object* obj, bool fail_ok) const {
532 space::Space* result = FindContinuousSpaceFromObject(obj, true);
533 if (result != NULL) {
534 return result;
535 }
536 return FindDiscontinuousSpaceFromObject(obj, true);
537 }
538
GetImageSpace() const539 space::ImageSpace* Heap::GetImageSpace() const {
540 for (const auto& space : continuous_spaces_) {
541 if (space->IsImageSpace()) {
542 return space->AsImageSpace();
543 }
544 }
545 return NULL;
546 }
547
MSpaceChunkCallback(void * start,void * end,size_t used_bytes,void * arg)548 static void MSpaceChunkCallback(void* start, void* end, size_t used_bytes, void* arg) {
549 size_t chunk_size = reinterpret_cast<uint8_t*>(end) - reinterpret_cast<uint8_t*>(start);
550 if (used_bytes < chunk_size) {
551 size_t chunk_free_bytes = chunk_size - used_bytes;
552 size_t& max_contiguous_allocation = *reinterpret_cast<size_t*>(arg);
553 max_contiguous_allocation = std::max(max_contiguous_allocation, chunk_free_bytes);
554 }
555 }
556
AllocObject(Thread * self,mirror::Class * c,size_t byte_count)557 mirror::Object* Heap::AllocObject(Thread* self, mirror::Class* c, size_t byte_count) {
558 DCHECK(c == NULL || (c->IsClassClass() && byte_count >= sizeof(mirror::Class)) ||
559 (c->IsVariableSize() || c->GetObjectSize() == byte_count) ||
560 strlen(ClassHelper(c).GetDescriptor()) == 0);
561 DCHECK_GE(byte_count, sizeof(mirror::Object));
562
563 mirror::Object* obj = NULL;
564 size_t bytes_allocated = 0;
565 uint64_t allocation_start = 0;
566 if (UNLIKELY(kMeasureAllocationTime)) {
567 allocation_start = NanoTime() / kTimeAdjust;
568 }
569
570 // We need to have a zygote space or else our newly allocated large object can end up in the
571 // Zygote resulting in it being prematurely freed.
572 // We can only do this for primitive objects since large objects will not be within the card table
573 // range. This also means that we rely on SetClass not dirtying the object's card.
574 bool large_object_allocation =
575 byte_count >= large_object_threshold_ && have_zygote_space_ && c->IsPrimitiveArray();
576 if (UNLIKELY(large_object_allocation)) {
577 obj = Allocate(self, large_object_space_, byte_count, &bytes_allocated);
578 // Make sure that our large object didn't get placed anywhere within the space interval or else
579 // it breaks the immune range.
580 DCHECK(obj == NULL ||
581 reinterpret_cast<byte*>(obj) < continuous_spaces_.front()->Begin() ||
582 reinterpret_cast<byte*>(obj) >= continuous_spaces_.back()->End());
583 } else {
584 obj = Allocate(self, alloc_space_, byte_count, &bytes_allocated);
585 // Ensure that we did not allocate into a zygote space.
586 DCHECK(obj == NULL || !have_zygote_space_ || !FindSpaceFromObject(obj, false)->IsZygoteSpace());
587 }
588
589 if (LIKELY(obj != NULL)) {
590 obj->SetClass(c);
591
592 // Record allocation after since we want to use the atomic add for the atomic fence to guard
593 // the SetClass since we do not want the class to appear NULL in another thread.
594 RecordAllocation(bytes_allocated, obj);
595
596 if (Dbg::IsAllocTrackingEnabled()) {
597 Dbg::RecordAllocation(c, byte_count);
598 }
599 if (UNLIKELY(static_cast<size_t>(num_bytes_allocated_) >= concurrent_start_bytes_)) {
600 // The SirtRef is necessary since the calls in RequestConcurrentGC are a safepoint.
601 SirtRef<mirror::Object> ref(self, obj);
602 RequestConcurrentGC(self);
603 }
604 if (kDesiredHeapVerification > kNoHeapVerification) {
605 VerifyObject(obj);
606 }
607
608 if (UNLIKELY(kMeasureAllocationTime)) {
609 total_allocation_time_.fetch_add(NanoTime() / kTimeAdjust - allocation_start);
610 }
611
612 return obj;
613 } else {
614 std::ostringstream oss;
615 int64_t total_bytes_free = GetFreeMemory();
616 oss << "Failed to allocate a " << byte_count << " byte allocation with " << total_bytes_free
617 << " free bytes";
618 // If the allocation failed due to fragmentation, print out the largest continuous allocation.
619 if (!large_object_allocation && total_bytes_free >= byte_count) {
620 size_t max_contiguous_allocation = 0;
621 for (const auto& space : continuous_spaces_) {
622 if (space->IsDlMallocSpace()) {
623 space->AsDlMallocSpace()->Walk(MSpaceChunkCallback, &max_contiguous_allocation);
624 }
625 }
626 oss << "; failed due to fragmentation (largest possible contiguous allocation "
627 << max_contiguous_allocation << " bytes)";
628 }
629 self->ThrowOutOfMemoryError(oss.str().c_str());
630 return NULL;
631 }
632 }
633
IsHeapAddress(const mirror::Object * obj)634 bool Heap::IsHeapAddress(const mirror::Object* obj) {
635 // Note: we deliberately don't take the lock here, and mustn't test anything that would
636 // require taking the lock.
637 if (obj == NULL) {
638 return true;
639 }
640 if (UNLIKELY(!IsAligned<kObjectAlignment>(obj))) {
641 return false;
642 }
643 return FindSpaceFromObject(obj, true) != NULL;
644 }
645
IsLiveObjectLocked(const mirror::Object * obj,bool search_allocation_stack,bool search_live_stack,bool sorted)646 bool Heap::IsLiveObjectLocked(const mirror::Object* obj, bool search_allocation_stack,
647 bool search_live_stack, bool sorted) {
648 // Locks::heap_bitmap_lock_->AssertReaderHeld(Thread::Current());
649 if (obj == NULL || UNLIKELY(!IsAligned<kObjectAlignment>(obj))) {
650 return false;
651 }
652 space::ContinuousSpace* c_space = FindContinuousSpaceFromObject(obj, true);
653 space::DiscontinuousSpace* d_space = NULL;
654 if (c_space != NULL) {
655 if (c_space->GetLiveBitmap()->Test(obj)) {
656 return true;
657 }
658 } else {
659 d_space = FindDiscontinuousSpaceFromObject(obj, true);
660 if (d_space != NULL) {
661 if (d_space->GetLiveObjects()->Test(obj)) {
662 return true;
663 }
664 }
665 }
666 // This is covering the allocation/live stack swapping that is done without mutators suspended.
667 for (size_t i = 0; i < (sorted ? 1 : 5); ++i) {
668 if (i > 0) {
669 NanoSleep(MsToNs(10));
670 }
671
672 if (search_allocation_stack) {
673 if (sorted) {
674 if (allocation_stack_->ContainsSorted(const_cast<mirror::Object*>(obj))) {
675 return true;
676 }
677 } else if (allocation_stack_->Contains(const_cast<mirror::Object*>(obj))) {
678 return true;
679 }
680 }
681
682 if (search_live_stack) {
683 if (sorted) {
684 if (live_stack_->ContainsSorted(const_cast<mirror::Object*>(obj))) {
685 return true;
686 }
687 } else if (live_stack_->Contains(const_cast<mirror::Object*>(obj))) {
688 return true;
689 }
690 }
691 }
692 // We need to check the bitmaps again since there is a race where we mark something as live and
693 // then clear the stack containing it.
694 if (c_space != NULL) {
695 if (c_space->GetLiveBitmap()->Test(obj)) {
696 return true;
697 }
698 } else {
699 d_space = FindDiscontinuousSpaceFromObject(obj, true);
700 if (d_space != NULL && d_space->GetLiveObjects()->Test(obj)) {
701 return true;
702 }
703 }
704 return false;
705 }
706
VerifyObjectImpl(const mirror::Object * obj)707 void Heap::VerifyObjectImpl(const mirror::Object* obj) {
708 if (Thread::Current() == NULL ||
709 Runtime::Current()->GetThreadList()->GetLockOwner() == Thread::Current()->GetTid()) {
710 return;
711 }
712 VerifyObjectBody(obj);
713 }
714
DumpSpaces()715 void Heap::DumpSpaces() {
716 for (const auto& space : continuous_spaces_) {
717 accounting::SpaceBitmap* live_bitmap = space->GetLiveBitmap();
718 accounting::SpaceBitmap* mark_bitmap = space->GetMarkBitmap();
719 LOG(INFO) << space << " " << *space << "\n"
720 << live_bitmap << " " << *live_bitmap << "\n"
721 << mark_bitmap << " " << *mark_bitmap;
722 }
723 for (const auto& space : discontinuous_spaces_) {
724 LOG(INFO) << space << " " << *space << "\n";
725 }
726 }
727
VerifyObjectBody(const mirror::Object * obj)728 void Heap::VerifyObjectBody(const mirror::Object* obj) {
729 CHECK(IsAligned<kObjectAlignment>(obj)) << "Object isn't aligned: " << obj;
730 // Ignore early dawn of the universe verifications.
731 if (UNLIKELY(static_cast<size_t>(num_bytes_allocated_.load()) < 10 * KB)) {
732 return;
733 }
734 const byte* raw_addr = reinterpret_cast<const byte*>(obj) +
735 mirror::Object::ClassOffset().Int32Value();
736 const mirror::Class* c = *reinterpret_cast<mirror::Class* const *>(raw_addr);
737 if (UNLIKELY(c == NULL)) {
738 LOG(FATAL) << "Null class in object: " << obj;
739 } else if (UNLIKELY(!IsAligned<kObjectAlignment>(c))) {
740 LOG(FATAL) << "Class isn't aligned: " << c << " in object: " << obj;
741 }
742 // Check obj.getClass().getClass() == obj.getClass().getClass().getClass()
743 // Note: we don't use the accessors here as they have internal sanity checks
744 // that we don't want to run
745 raw_addr = reinterpret_cast<const byte*>(c) + mirror::Object::ClassOffset().Int32Value();
746 const mirror::Class* c_c = *reinterpret_cast<mirror::Class* const *>(raw_addr);
747 raw_addr = reinterpret_cast<const byte*>(c_c) + mirror::Object::ClassOffset().Int32Value();
748 const mirror::Class* c_c_c = *reinterpret_cast<mirror::Class* const *>(raw_addr);
749 CHECK_EQ(c_c, c_c_c);
750
751 if (verify_object_mode_ != kVerifyAllFast) {
752 // TODO: the bitmap tests below are racy if VerifyObjectBody is called without the
753 // heap_bitmap_lock_.
754 if (!IsLiveObjectLocked(obj)) {
755 DumpSpaces();
756 LOG(FATAL) << "Object is dead: " << obj;
757 }
758 if (!IsLiveObjectLocked(c)) {
759 LOG(FATAL) << "Class of object is dead: " << c << " in object: " << obj;
760 }
761 }
762 }
763
VerificationCallback(mirror::Object * obj,void * arg)764 void Heap::VerificationCallback(mirror::Object* obj, void* arg) {
765 DCHECK(obj != NULL);
766 reinterpret_cast<Heap*>(arg)->VerifyObjectBody(obj);
767 }
768
VerifyHeap()769 void Heap::VerifyHeap() {
770 ReaderMutexLock mu(Thread::Current(), *Locks::heap_bitmap_lock_);
771 GetLiveBitmap()->Walk(Heap::VerificationCallback, this);
772 }
773
RecordAllocation(size_t size,mirror::Object * obj)774 inline void Heap::RecordAllocation(size_t size, mirror::Object* obj) {
775 DCHECK(obj != NULL);
776 DCHECK_GT(size, 0u);
777 num_bytes_allocated_.fetch_add(size);
778
779 if (Runtime::Current()->HasStatsEnabled()) {
780 RuntimeStats* thread_stats = Thread::Current()->GetStats();
781 ++thread_stats->allocated_objects;
782 thread_stats->allocated_bytes += size;
783
784 // TODO: Update these atomically.
785 RuntimeStats* global_stats = Runtime::Current()->GetStats();
786 ++global_stats->allocated_objects;
787 global_stats->allocated_bytes += size;
788 }
789
790 // This is safe to do since the GC will never free objects which are neither in the allocation
791 // stack or the live bitmap.
792 while (!allocation_stack_->AtomicPushBack(obj)) {
793 CollectGarbageInternal(collector::kGcTypeSticky, kGcCauseForAlloc, false);
794 }
795 }
796
RecordFree(size_t freed_objects,size_t freed_bytes)797 void Heap::RecordFree(size_t freed_objects, size_t freed_bytes) {
798 DCHECK_LE(freed_bytes, static_cast<size_t>(num_bytes_allocated_));
799 num_bytes_allocated_.fetch_sub(freed_bytes);
800
801 if (Runtime::Current()->HasStatsEnabled()) {
802 RuntimeStats* thread_stats = Thread::Current()->GetStats();
803 thread_stats->freed_objects += freed_objects;
804 thread_stats->freed_bytes += freed_bytes;
805
806 // TODO: Do this concurrently.
807 RuntimeStats* global_stats = Runtime::Current()->GetStats();
808 global_stats->freed_objects += freed_objects;
809 global_stats->freed_bytes += freed_bytes;
810 }
811 }
812
IsOutOfMemoryOnAllocation(size_t alloc_size,bool grow)813 inline bool Heap::IsOutOfMemoryOnAllocation(size_t alloc_size, bool grow) {
814 size_t new_footprint = num_bytes_allocated_ + alloc_size;
815 if (UNLIKELY(new_footprint > max_allowed_footprint_)) {
816 if (UNLIKELY(new_footprint > growth_limit_)) {
817 return true;
818 }
819 if (!concurrent_gc_) {
820 if (!grow) {
821 return true;
822 } else {
823 max_allowed_footprint_ = new_footprint;
824 }
825 }
826 }
827 return false;
828 }
829
TryToAllocate(Thread * self,space::AllocSpace * space,size_t alloc_size,bool grow,size_t * bytes_allocated)830 inline mirror::Object* Heap::TryToAllocate(Thread* self, space::AllocSpace* space, size_t alloc_size,
831 bool grow, size_t* bytes_allocated) {
832 if (UNLIKELY(IsOutOfMemoryOnAllocation(alloc_size, grow))) {
833 return NULL;
834 }
835 return space->Alloc(self, alloc_size, bytes_allocated);
836 }
837
838 // DlMallocSpace-specific version.
TryToAllocate(Thread * self,space::DlMallocSpace * space,size_t alloc_size,bool grow,size_t * bytes_allocated)839 inline mirror::Object* Heap::TryToAllocate(Thread* self, space::DlMallocSpace* space, size_t alloc_size,
840 bool grow, size_t* bytes_allocated) {
841 if (UNLIKELY(IsOutOfMemoryOnAllocation(alloc_size, grow))) {
842 return NULL;
843 }
844 if (LIKELY(!running_on_valgrind_)) {
845 return space->AllocNonvirtual(self, alloc_size, bytes_allocated);
846 } else {
847 return space->Alloc(self, alloc_size, bytes_allocated);
848 }
849 }
850
851 template <class T>
Allocate(Thread * self,T * space,size_t alloc_size,size_t * bytes_allocated)852 inline mirror::Object* Heap::Allocate(Thread* self, T* space, size_t alloc_size,
853 size_t* bytes_allocated) {
854 // Since allocation can cause a GC which will need to SuspendAll, make sure all allocations are
855 // done in the runnable state where suspension is expected.
856 DCHECK_EQ(self->GetState(), kRunnable);
857 self->AssertThreadSuspensionIsAllowable();
858
859 mirror::Object* ptr = TryToAllocate(self, space, alloc_size, false, bytes_allocated);
860 if (ptr != NULL) {
861 return ptr;
862 }
863 return AllocateInternalWithGc(self, space, alloc_size, bytes_allocated);
864 }
865
AllocateInternalWithGc(Thread * self,space::AllocSpace * space,size_t alloc_size,size_t * bytes_allocated)866 mirror::Object* Heap::AllocateInternalWithGc(Thread* self, space::AllocSpace* space,
867 size_t alloc_size, size_t* bytes_allocated) {
868 mirror::Object* ptr;
869
870 // The allocation failed. If the GC is running, block until it completes, and then retry the
871 // allocation.
872 collector::GcType last_gc = WaitForConcurrentGcToComplete(self);
873 if (last_gc != collector::kGcTypeNone) {
874 // A GC was in progress and we blocked, retry allocation now that memory has been freed.
875 ptr = TryToAllocate(self, space, alloc_size, false, bytes_allocated);
876 if (ptr != NULL) {
877 return ptr;
878 }
879 }
880
881 // Loop through our different Gc types and try to Gc until we get enough free memory.
882 for (size_t i = static_cast<size_t>(last_gc) + 1;
883 i < static_cast<size_t>(collector::kGcTypeMax); ++i) {
884 bool run_gc = false;
885 collector::GcType gc_type = static_cast<collector::GcType>(i);
886 switch (gc_type) {
887 case collector::kGcTypeSticky: {
888 const size_t alloc_space_size = alloc_space_->Size();
889 run_gc = alloc_space_size > min_alloc_space_size_for_sticky_gc_ &&
890 alloc_space_->Capacity() - alloc_space_size >= min_remaining_space_for_sticky_gc_;
891 break;
892 }
893 case collector::kGcTypePartial:
894 run_gc = have_zygote_space_;
895 break;
896 case collector::kGcTypeFull:
897 run_gc = true;
898 break;
899 default:
900 break;
901 }
902
903 if (run_gc) {
904 // If we actually ran a different type of Gc than requested, we can skip the index forwards.
905 collector::GcType gc_type_ran = CollectGarbageInternal(gc_type, kGcCauseForAlloc, false);
906 DCHECK_GE(static_cast<size_t>(gc_type_ran), i);
907 i = static_cast<size_t>(gc_type_ran);
908
909 // Did we free sufficient memory for the allocation to succeed?
910 ptr = TryToAllocate(self, space, alloc_size, false, bytes_allocated);
911 if (ptr != NULL) {
912 return ptr;
913 }
914 }
915 }
916
917 // Allocations have failed after GCs; this is an exceptional state.
918 // Try harder, growing the heap if necessary.
919 ptr = TryToAllocate(self, space, alloc_size, true, bytes_allocated);
920 if (ptr != NULL) {
921 return ptr;
922 }
923
924 // Most allocations should have succeeded by now, so the heap is really full, really fragmented,
925 // or the requested size is really big. Do another GC, collecting SoftReferences this time. The
926 // VM spec requires that all SoftReferences have been collected and cleared before throwing OOME.
927
928 // OLD-TODO: wait for the finalizers from the previous GC to finish
929 VLOG(gc) << "Forcing collection of SoftReferences for " << PrettySize(alloc_size)
930 << " allocation";
931
932 // We don't need a WaitForConcurrentGcToComplete here either.
933 CollectGarbageInternal(collector::kGcTypeFull, kGcCauseForAlloc, true);
934 return TryToAllocate(self, space, alloc_size, true, bytes_allocated);
935 }
936
SetTargetHeapUtilization(float target)937 void Heap::SetTargetHeapUtilization(float target) {
938 DCHECK_GT(target, 0.0f); // asserted in Java code
939 DCHECK_LT(target, 1.0f);
940 target_utilization_ = target;
941 }
942
GetObjectsAllocated() const943 size_t Heap::GetObjectsAllocated() const {
944 size_t total = 0;
945 typedef std::vector<space::ContinuousSpace*>::const_iterator It;
946 for (It it = continuous_spaces_.begin(), end = continuous_spaces_.end(); it != end; ++it) {
947 space::ContinuousSpace* space = *it;
948 if (space->IsDlMallocSpace()) {
949 total += space->AsDlMallocSpace()->GetObjectsAllocated();
950 }
951 }
952 typedef std::vector<space::DiscontinuousSpace*>::const_iterator It2;
953 for (It2 it = discontinuous_spaces_.begin(), end = discontinuous_spaces_.end(); it != end; ++it) {
954 space::DiscontinuousSpace* space = *it;
955 total += space->AsLargeObjectSpace()->GetObjectsAllocated();
956 }
957 return total;
958 }
959
GetObjectsAllocatedEver() const960 size_t Heap::GetObjectsAllocatedEver() const {
961 size_t total = 0;
962 typedef std::vector<space::ContinuousSpace*>::const_iterator It;
963 for (It it = continuous_spaces_.begin(), end = continuous_spaces_.end(); it != end; ++it) {
964 space::ContinuousSpace* space = *it;
965 if (space->IsDlMallocSpace()) {
966 total += space->AsDlMallocSpace()->GetTotalObjectsAllocated();
967 }
968 }
969 typedef std::vector<space::DiscontinuousSpace*>::const_iterator It2;
970 for (It2 it = discontinuous_spaces_.begin(), end = discontinuous_spaces_.end(); it != end; ++it) {
971 space::DiscontinuousSpace* space = *it;
972 total += space->AsLargeObjectSpace()->GetTotalObjectsAllocated();
973 }
974 return total;
975 }
976
GetBytesAllocatedEver() const977 size_t Heap::GetBytesAllocatedEver() const {
978 size_t total = 0;
979 typedef std::vector<space::ContinuousSpace*>::const_iterator It;
980 for (It it = continuous_spaces_.begin(), end = continuous_spaces_.end(); it != end; ++it) {
981 space::ContinuousSpace* space = *it;
982 if (space->IsDlMallocSpace()) {
983 total += space->AsDlMallocSpace()->GetTotalBytesAllocated();
984 }
985 }
986 typedef std::vector<space::DiscontinuousSpace*>::const_iterator It2;
987 for (It2 it = discontinuous_spaces_.begin(), end = discontinuous_spaces_.end(); it != end; ++it) {
988 space::DiscontinuousSpace* space = *it;
989 total += space->AsLargeObjectSpace()->GetTotalBytesAllocated();
990 }
991 return total;
992 }
993
994 class InstanceCounter {
995 public:
InstanceCounter(const std::vector<mirror::Class * > & classes,bool use_is_assignable_from,uint64_t * counts)996 InstanceCounter(const std::vector<mirror::Class*>& classes, bool use_is_assignable_from, uint64_t* counts)
997 SHARED_LOCKS_REQUIRED(Locks::mutator_lock_)
998 : classes_(classes), use_is_assignable_from_(use_is_assignable_from), counts_(counts) {
999 }
1000
operator ()(const mirror::Object * o) const1001 void operator()(const mirror::Object* o) const SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
1002 for (size_t i = 0; i < classes_.size(); ++i) {
1003 const mirror::Class* instance_class = o->GetClass();
1004 if (use_is_assignable_from_) {
1005 if (instance_class != NULL && classes_[i]->IsAssignableFrom(instance_class)) {
1006 ++counts_[i];
1007 }
1008 } else {
1009 if (instance_class == classes_[i]) {
1010 ++counts_[i];
1011 }
1012 }
1013 }
1014 }
1015
1016 private:
1017 const std::vector<mirror::Class*>& classes_;
1018 bool use_is_assignable_from_;
1019 uint64_t* const counts_;
1020
1021 DISALLOW_COPY_AND_ASSIGN(InstanceCounter);
1022 };
1023
CountInstances(const std::vector<mirror::Class * > & classes,bool use_is_assignable_from,uint64_t * counts)1024 void Heap::CountInstances(const std::vector<mirror::Class*>& classes, bool use_is_assignable_from,
1025 uint64_t* counts) {
1026 // We only want reachable instances, so do a GC. This also ensures that the alloc stack
1027 // is empty, so the live bitmap is the only place we need to look.
1028 Thread* self = Thread::Current();
1029 self->TransitionFromRunnableToSuspended(kNative);
1030 CollectGarbage(false);
1031 self->TransitionFromSuspendedToRunnable();
1032
1033 InstanceCounter counter(classes, use_is_assignable_from, counts);
1034 ReaderMutexLock mu(self, *Locks::heap_bitmap_lock_);
1035 GetLiveBitmap()->Visit(counter);
1036 }
1037
1038 class InstanceCollector {
1039 public:
InstanceCollector(mirror::Class * c,int32_t max_count,std::vector<mirror::Object * > & instances)1040 InstanceCollector(mirror::Class* c, int32_t max_count, std::vector<mirror::Object*>& instances)
1041 SHARED_LOCKS_REQUIRED(Locks::mutator_lock_)
1042 : class_(c), max_count_(max_count), instances_(instances) {
1043 }
1044
operator ()(const mirror::Object * o) const1045 void operator()(const mirror::Object* o) const SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
1046 const mirror::Class* instance_class = o->GetClass();
1047 if (instance_class == class_) {
1048 if (max_count_ == 0 || instances_.size() < max_count_) {
1049 instances_.push_back(const_cast<mirror::Object*>(o));
1050 }
1051 }
1052 }
1053
1054 private:
1055 mirror::Class* class_;
1056 uint32_t max_count_;
1057 std::vector<mirror::Object*>& instances_;
1058
1059 DISALLOW_COPY_AND_ASSIGN(InstanceCollector);
1060 };
1061
GetInstances(mirror::Class * c,int32_t max_count,std::vector<mirror::Object * > & instances)1062 void Heap::GetInstances(mirror::Class* c, int32_t max_count,
1063 std::vector<mirror::Object*>& instances) {
1064 // We only want reachable instances, so do a GC. This also ensures that the alloc stack
1065 // is empty, so the live bitmap is the only place we need to look.
1066 Thread* self = Thread::Current();
1067 self->TransitionFromRunnableToSuspended(kNative);
1068 CollectGarbage(false);
1069 self->TransitionFromSuspendedToRunnable();
1070
1071 InstanceCollector collector(c, max_count, instances);
1072 ReaderMutexLock mu(self, *Locks::heap_bitmap_lock_);
1073 GetLiveBitmap()->Visit(collector);
1074 }
1075
1076 class ReferringObjectsFinder {
1077 public:
ReferringObjectsFinder(mirror::Object * object,int32_t max_count,std::vector<mirror::Object * > & referring_objects)1078 ReferringObjectsFinder(mirror::Object* object, int32_t max_count,
1079 std::vector<mirror::Object*>& referring_objects)
1080 SHARED_LOCKS_REQUIRED(Locks::mutator_lock_)
1081 : object_(object), max_count_(max_count), referring_objects_(referring_objects) {
1082 }
1083
1084 // For bitmap Visit.
1085 // TODO: Fix lock analysis to not use NO_THREAD_SAFETY_ANALYSIS, requires support for
1086 // annotalysis on visitors.
operator ()(const mirror::Object * o) const1087 void operator()(const mirror::Object* o) const NO_THREAD_SAFETY_ANALYSIS {
1088 collector::MarkSweep::VisitObjectReferences(o, *this);
1089 }
1090
1091 // For MarkSweep::VisitObjectReferences.
operator ()(const mirror::Object * referrer,const mirror::Object * object,const MemberOffset &,bool) const1092 void operator()(const mirror::Object* referrer, const mirror::Object* object,
1093 const MemberOffset&, bool) const {
1094 if (object == object_ && (max_count_ == 0 || referring_objects_.size() < max_count_)) {
1095 referring_objects_.push_back(const_cast<mirror::Object*>(referrer));
1096 }
1097 }
1098
1099 private:
1100 mirror::Object* object_;
1101 uint32_t max_count_;
1102 std::vector<mirror::Object*>& referring_objects_;
1103
1104 DISALLOW_COPY_AND_ASSIGN(ReferringObjectsFinder);
1105 };
1106
GetReferringObjects(mirror::Object * o,int32_t max_count,std::vector<mirror::Object * > & referring_objects)1107 void Heap::GetReferringObjects(mirror::Object* o, int32_t max_count,
1108 std::vector<mirror::Object*>& referring_objects) {
1109 // We only want reachable instances, so do a GC. This also ensures that the alloc stack
1110 // is empty, so the live bitmap is the only place we need to look.
1111 Thread* self = Thread::Current();
1112 self->TransitionFromRunnableToSuspended(kNative);
1113 CollectGarbage(false);
1114 self->TransitionFromSuspendedToRunnable();
1115
1116 ReferringObjectsFinder finder(o, max_count, referring_objects);
1117 ReaderMutexLock mu(self, *Locks::heap_bitmap_lock_);
1118 GetLiveBitmap()->Visit(finder);
1119 }
1120
CollectGarbage(bool clear_soft_references)1121 void Heap::CollectGarbage(bool clear_soft_references) {
1122 // Even if we waited for a GC we still need to do another GC since weaks allocated during the
1123 // last GC will not have necessarily been cleared.
1124 Thread* self = Thread::Current();
1125 WaitForConcurrentGcToComplete(self);
1126 CollectGarbageInternal(collector::kGcTypeFull, kGcCauseExplicit, clear_soft_references);
1127 }
1128
PreZygoteFork()1129 void Heap::PreZygoteFork() {
1130 static Mutex zygote_creation_lock_("zygote creation lock", kZygoteCreationLock);
1131 // Do this before acquiring the zygote creation lock so that we don't get lock order violations.
1132 CollectGarbage(false);
1133 Thread* self = Thread::Current();
1134 MutexLock mu(self, zygote_creation_lock_);
1135
1136 // Try to see if we have any Zygote spaces.
1137 if (have_zygote_space_) {
1138 return;
1139 }
1140
1141 VLOG(heap) << "Starting PreZygoteFork with alloc space size " << PrettySize(alloc_space_->Size());
1142
1143 {
1144 // Flush the alloc stack.
1145 WriterMutexLock mu(self, *Locks::heap_bitmap_lock_);
1146 FlushAllocStack();
1147 }
1148
1149 // Turns the current alloc space into a Zygote space and obtain the new alloc space composed
1150 // of the remaining available heap memory.
1151 space::DlMallocSpace* zygote_space = alloc_space_;
1152 alloc_space_ = zygote_space->CreateZygoteSpace("alloc space");
1153 alloc_space_->SetFootprintLimit(alloc_space_->Capacity());
1154
1155 // Change the GC retention policy of the zygote space to only collect when full.
1156 zygote_space->SetGcRetentionPolicy(space::kGcRetentionPolicyFullCollect);
1157 AddContinuousSpace(alloc_space_);
1158 have_zygote_space_ = true;
1159
1160 // Reset the cumulative loggers since we now have a few additional timing phases.
1161 for (const auto& collector : mark_sweep_collectors_) {
1162 collector->ResetCumulativeStatistics();
1163 }
1164 }
1165
FlushAllocStack()1166 void Heap::FlushAllocStack() {
1167 MarkAllocStack(alloc_space_->GetLiveBitmap(), large_object_space_->GetLiveObjects(),
1168 allocation_stack_.get());
1169 allocation_stack_->Reset();
1170 }
1171
MarkAllocStack(accounting::SpaceBitmap * bitmap,accounting::SpaceSetMap * large_objects,accounting::ObjectStack * stack)1172 void Heap::MarkAllocStack(accounting::SpaceBitmap* bitmap, accounting::SpaceSetMap* large_objects,
1173 accounting::ObjectStack* stack) {
1174 mirror::Object** limit = stack->End();
1175 for (mirror::Object** it = stack->Begin(); it != limit; ++it) {
1176 const mirror::Object* obj = *it;
1177 DCHECK(obj != NULL);
1178 if (LIKELY(bitmap->HasAddress(obj))) {
1179 bitmap->Set(obj);
1180 } else {
1181 large_objects->Set(obj);
1182 }
1183 }
1184 }
1185
1186
1187 const char* gc_cause_and_type_strings[3][4] = {
1188 {"", "GC Alloc Sticky", "GC Alloc Partial", "GC Alloc Full"},
1189 {"", "GC Background Sticky", "GC Background Partial", "GC Background Full"},
1190 {"", "GC Explicit Sticky", "GC Explicit Partial", "GC Explicit Full"}};
1191
CollectGarbageInternal(collector::GcType gc_type,GcCause gc_cause,bool clear_soft_references)1192 collector::GcType Heap::CollectGarbageInternal(collector::GcType gc_type, GcCause gc_cause,
1193 bool clear_soft_references) {
1194 Thread* self = Thread::Current();
1195
1196 ScopedThreadStateChange tsc(self, kWaitingPerformingGc);
1197 Locks::mutator_lock_->AssertNotHeld(self);
1198
1199 if (self->IsHandlingStackOverflow()) {
1200 LOG(WARNING) << "Performing GC on a thread that is handling a stack overflow.";
1201 }
1202
1203 // Ensure there is only one GC at a time.
1204 bool start_collect = false;
1205 while (!start_collect) {
1206 {
1207 MutexLock mu(self, *gc_complete_lock_);
1208 if (!is_gc_running_) {
1209 is_gc_running_ = true;
1210 start_collect = true;
1211 }
1212 }
1213 if (!start_collect) {
1214 // TODO: timinglog this.
1215 WaitForConcurrentGcToComplete(self);
1216
1217 // TODO: if another thread beat this one to do the GC, perhaps we should just return here?
1218 // Not doing at the moment to ensure soft references are cleared.
1219 }
1220 }
1221 gc_complete_lock_->AssertNotHeld(self);
1222
1223 if (gc_cause == kGcCauseForAlloc && Runtime::Current()->HasStatsEnabled()) {
1224 ++Runtime::Current()->GetStats()->gc_for_alloc_count;
1225 ++Thread::Current()->GetStats()->gc_for_alloc_count;
1226 }
1227
1228 uint64_t gc_start_time_ns = NanoTime();
1229 uint64_t gc_start_size = GetBytesAllocated();
1230 // Approximate allocation rate in bytes / second.
1231 if (UNLIKELY(gc_start_time_ns == last_gc_time_ns_)) {
1232 LOG(WARNING) << "Timers are broken (gc_start_time == last_gc_time_).";
1233 }
1234 uint64_t ms_delta = NsToMs(gc_start_time_ns - last_gc_time_ns_);
1235 if (ms_delta != 0) {
1236 allocation_rate_ = ((gc_start_size - last_gc_size_) * 1000) / ms_delta;
1237 VLOG(heap) << "Allocation rate: " << PrettySize(allocation_rate_) << "/s";
1238 }
1239
1240 if (gc_type == collector::kGcTypeSticky &&
1241 alloc_space_->Size() < min_alloc_space_size_for_sticky_gc_) {
1242 gc_type = collector::kGcTypePartial;
1243 }
1244
1245 DCHECK_LT(gc_type, collector::kGcTypeMax);
1246 DCHECK_NE(gc_type, collector::kGcTypeNone);
1247 DCHECK_LE(gc_cause, kGcCauseExplicit);
1248
1249 ATRACE_BEGIN(gc_cause_and_type_strings[gc_cause][gc_type]);
1250
1251 collector::MarkSweep* collector = NULL;
1252 for (const auto& cur_collector : mark_sweep_collectors_) {
1253 if (cur_collector->IsConcurrent() == concurrent_gc_ && cur_collector->GetGcType() == gc_type) {
1254 collector = cur_collector;
1255 break;
1256 }
1257 }
1258 CHECK(collector != NULL)
1259 << "Could not find garbage collector with concurrent=" << concurrent_gc_
1260 << " and type=" << gc_type;
1261
1262 collector->clear_soft_references_ = clear_soft_references;
1263 collector->Run();
1264 total_objects_freed_ever_ += collector->GetFreedObjects();
1265 total_bytes_freed_ever_ += collector->GetFreedBytes();
1266 if (care_about_pause_times_) {
1267 const size_t duration = collector->GetDurationNs();
1268 std::vector<uint64_t> pauses = collector->GetPauseTimes();
1269 // GC for alloc pauses the allocating thread, so consider it as a pause.
1270 bool was_slow = duration > long_gc_log_threshold_ ||
1271 (gc_cause == kGcCauseForAlloc && duration > long_pause_log_threshold_);
1272 if (!was_slow) {
1273 for (uint64_t pause : pauses) {
1274 was_slow = was_slow || pause > long_pause_log_threshold_;
1275 }
1276 }
1277
1278 if (was_slow) {
1279 const size_t percent_free = GetPercentFree();
1280 const size_t current_heap_size = GetBytesAllocated();
1281 const size_t total_memory = GetTotalMemory();
1282 std::ostringstream pause_string;
1283 for (size_t i = 0; i < pauses.size(); ++i) {
1284 pause_string << PrettyDuration((pauses[i] / 1000) * 1000)
1285 << ((i != pauses.size() - 1) ? ", " : "");
1286 }
1287 LOG(INFO) << gc_cause << " " << collector->GetName()
1288 << " GC freed " << collector->GetFreedObjects() << "("
1289 << PrettySize(collector->GetFreedBytes()) << ") AllocSpace objects, "
1290 << collector->GetFreedLargeObjects() << "("
1291 << PrettySize(collector->GetFreedLargeObjectBytes()) << ") LOS objects, "
1292 << percent_free << "% free, " << PrettySize(current_heap_size) << "/"
1293 << PrettySize(total_memory) << ", " << "paused " << pause_string.str()
1294 << " total " << PrettyDuration((duration / 1000) * 1000);
1295 if (VLOG_IS_ON(heap)) {
1296 LOG(INFO) << Dumpable<base::TimingLogger>(collector->GetTimings());
1297 }
1298 }
1299 }
1300
1301 {
1302 MutexLock mu(self, *gc_complete_lock_);
1303 is_gc_running_ = false;
1304 last_gc_type_ = gc_type;
1305 // Wake anyone who may have been waiting for the GC to complete.
1306 gc_complete_cond_->Broadcast(self);
1307 }
1308
1309 ATRACE_END();
1310
1311 // Inform DDMS that a GC completed.
1312 Dbg::GcDidFinish();
1313 return gc_type;
1314 }
1315
UpdateAndMarkModUnion(collector::MarkSweep * mark_sweep,base::TimingLogger & timings,collector::GcType gc_type)1316 void Heap::UpdateAndMarkModUnion(collector::MarkSweep* mark_sweep, base::TimingLogger& timings,
1317 collector::GcType gc_type) {
1318 if (gc_type == collector::kGcTypeSticky) {
1319 // Don't need to do anything for mod union table in this case since we are only scanning dirty
1320 // cards.
1321 return;
1322 }
1323
1324 base::TimingLogger::ScopedSplit split("UpdateModUnionTable", &timings);
1325 // Update zygote mod union table.
1326 if (gc_type == collector::kGcTypePartial) {
1327 base::TimingLogger::ScopedSplit split("UpdateZygoteModUnionTable", &timings);
1328 zygote_mod_union_table_->Update();
1329
1330 timings.NewSplit("ZygoteMarkReferences");
1331 zygote_mod_union_table_->MarkReferences(mark_sweep);
1332 }
1333
1334 // Processes the cards we cleared earlier and adds their objects into the mod-union table.
1335 timings.NewSplit("UpdateModUnionTable");
1336 image_mod_union_table_->Update();
1337
1338 // Scans all objects in the mod-union table.
1339 timings.NewSplit("MarkImageToAllocSpaceReferences");
1340 image_mod_union_table_->MarkReferences(mark_sweep);
1341 }
1342
RootMatchesObjectVisitor(const mirror::Object * root,void * arg)1343 static void RootMatchesObjectVisitor(const mirror::Object* root, void* arg) {
1344 mirror::Object* obj = reinterpret_cast<mirror::Object*>(arg);
1345 if (root == obj) {
1346 LOG(INFO) << "Object " << obj << " is a root";
1347 }
1348 }
1349
1350 class ScanVisitor {
1351 public:
operator ()(const mirror::Object * obj) const1352 void operator()(const mirror::Object* obj) const {
1353 LOG(ERROR) << "Would have rescanned object " << obj;
1354 }
1355 };
1356
1357 // Verify a reference from an object.
1358 class VerifyReferenceVisitor {
1359 public:
1360 explicit VerifyReferenceVisitor(Heap* heap)
SHARED_LOCKS_REQUIRED(Locks::mutator_lock_,Locks::heap_bitmap_lock_)1361 SHARED_LOCKS_REQUIRED(Locks::mutator_lock_, Locks::heap_bitmap_lock_)
1362 : heap_(heap), failed_(false) {}
1363
Failed() const1364 bool Failed() const {
1365 return failed_;
1366 }
1367
1368 // TODO: Fix lock analysis to not use NO_THREAD_SAFETY_ANALYSIS, requires support for smarter
1369 // analysis on visitors.
operator ()(const mirror::Object * obj,const mirror::Object * ref,const MemberOffset & offset,bool) const1370 void operator()(const mirror::Object* obj, const mirror::Object* ref,
1371 const MemberOffset& offset, bool /* is_static */) const
1372 NO_THREAD_SAFETY_ANALYSIS {
1373 // Verify that the reference is live.
1374 if (UNLIKELY(ref != NULL && !IsLive(ref))) {
1375 accounting::CardTable* card_table = heap_->GetCardTable();
1376 accounting::ObjectStack* alloc_stack = heap_->allocation_stack_.get();
1377 accounting::ObjectStack* live_stack = heap_->live_stack_.get();
1378
1379 if (!failed_) {
1380 // Print message on only on first failure to prevent spam.
1381 LOG(ERROR) << "!!!!!!!!!!!!!!Heap corruption detected!!!!!!!!!!!!!!!!!!!";
1382 failed_ = true;
1383 }
1384 if (obj != nullptr) {
1385 byte* card_addr = card_table->CardFromAddr(obj);
1386 LOG(ERROR) << "Object " << obj << " references dead object " << ref << " at offset "
1387 << offset << "\n card value = " << static_cast<int>(*card_addr);
1388 if (heap_->IsHeapAddress(obj->GetClass())) {
1389 LOG(ERROR) << "Obj type " << PrettyTypeOf(obj);
1390 } else {
1391 LOG(ERROR) << "Object " << obj << " class(" << obj->GetClass() << ") not a heap address";
1392 }
1393
1394 // Attmept to find the class inside of the recently freed objects.
1395 space::ContinuousSpace* ref_space = heap_->FindContinuousSpaceFromObject(ref, true);
1396 if (ref_space->IsDlMallocSpace()) {
1397 space::DlMallocSpace* space = ref_space->AsDlMallocSpace();
1398 mirror::Class* ref_class = space->FindRecentFreedObject(ref);
1399 if (ref_class != nullptr) {
1400 LOG(ERROR) << "Reference " << ref << " found as a recently freed object with class "
1401 << PrettyClass(ref_class);
1402 } else {
1403 LOG(ERROR) << "Reference " << ref << " not found as a recently freed object";
1404 }
1405 }
1406
1407 if (ref->GetClass() != nullptr && heap_->IsHeapAddress(ref->GetClass()) &&
1408 ref->GetClass()->IsClass()) {
1409 LOG(ERROR) << "Ref type " << PrettyTypeOf(ref);
1410 } else {
1411 LOG(ERROR) << "Ref " << ref << " class(" << ref->GetClass()
1412 << ") is not a valid heap address";
1413 }
1414
1415 card_table->CheckAddrIsInCardTable(reinterpret_cast<const byte*>(obj));
1416 void* cover_begin = card_table->AddrFromCard(card_addr);
1417 void* cover_end = reinterpret_cast<void*>(reinterpret_cast<size_t>(cover_begin) +
1418 accounting::CardTable::kCardSize);
1419 LOG(ERROR) << "Card " << reinterpret_cast<void*>(card_addr) << " covers " << cover_begin
1420 << "-" << cover_end;
1421 accounting::SpaceBitmap* bitmap = heap_->GetLiveBitmap()->GetContinuousSpaceBitmap(obj);
1422
1423 // Print out how the object is live.
1424 if (bitmap != NULL && bitmap->Test(obj)) {
1425 LOG(ERROR) << "Object " << obj << " found in live bitmap";
1426 }
1427 if (alloc_stack->Contains(const_cast<mirror::Object*>(obj))) {
1428 LOG(ERROR) << "Object " << obj << " found in allocation stack";
1429 }
1430 if (live_stack->Contains(const_cast<mirror::Object*>(obj))) {
1431 LOG(ERROR) << "Object " << obj << " found in live stack";
1432 }
1433 if (alloc_stack->Contains(const_cast<mirror::Object*>(ref))) {
1434 LOG(ERROR) << "Ref " << ref << " found in allocation stack";
1435 }
1436 if (live_stack->Contains(const_cast<mirror::Object*>(ref))) {
1437 LOG(ERROR) << "Ref " << ref << " found in live stack";
1438 }
1439 // Attempt to see if the card table missed the reference.
1440 ScanVisitor scan_visitor;
1441 byte* byte_cover_begin = reinterpret_cast<byte*>(card_table->AddrFromCard(card_addr));
1442 card_table->Scan(bitmap, byte_cover_begin,
1443 byte_cover_begin + accounting::CardTable::kCardSize, scan_visitor);
1444
1445 // Search to see if any of the roots reference our object.
1446 void* arg = const_cast<void*>(reinterpret_cast<const void*>(obj));
1447 Runtime::Current()->VisitRoots(&RootMatchesObjectVisitor, arg, false, false);
1448
1449 // Search to see if any of the roots reference our reference.
1450 arg = const_cast<void*>(reinterpret_cast<const void*>(ref));
1451 Runtime::Current()->VisitRoots(&RootMatchesObjectVisitor, arg, false, false);
1452 } else {
1453 LOG(ERROR) << "Root references dead object " << ref << "\nRef type " << PrettyTypeOf(ref);
1454 }
1455 }
1456 }
1457
IsLive(const mirror::Object * obj) const1458 bool IsLive(const mirror::Object* obj) const NO_THREAD_SAFETY_ANALYSIS {
1459 return heap_->IsLiveObjectLocked(obj, true, false, true);
1460 }
1461
VerifyRoots(const mirror::Object * root,void * arg)1462 static void VerifyRoots(const mirror::Object* root, void* arg) {
1463 VerifyReferenceVisitor* visitor = reinterpret_cast<VerifyReferenceVisitor*>(arg);
1464 (*visitor)(NULL, root, MemberOffset(0), true);
1465 }
1466
1467 private:
1468 Heap* const heap_;
1469 mutable bool failed_;
1470 };
1471
1472 // Verify all references within an object, for use with HeapBitmap::Visit.
1473 class VerifyObjectVisitor {
1474 public:
VerifyObjectVisitor(Heap * heap)1475 explicit VerifyObjectVisitor(Heap* heap) : heap_(heap), failed_(false) {}
1476
operator ()(const mirror::Object * obj) const1477 void operator()(const mirror::Object* obj) const
1478 SHARED_LOCKS_REQUIRED(Locks::mutator_lock_, Locks::heap_bitmap_lock_) {
1479 // Note: we are verifying the references in obj but not obj itself, this is because obj must
1480 // be live or else how did we find it in the live bitmap?
1481 VerifyReferenceVisitor visitor(heap_);
1482 // The class doesn't count as a reference but we should verify it anyways.
1483 visitor(obj, obj->GetClass(), MemberOffset(0), false);
1484 collector::MarkSweep::VisitObjectReferences(obj, visitor);
1485 failed_ = failed_ || visitor.Failed();
1486 }
1487
Failed() const1488 bool Failed() const {
1489 return failed_;
1490 }
1491
1492 private:
1493 Heap* const heap_;
1494 mutable bool failed_;
1495 };
1496
1497 // Must do this with mutators suspended since we are directly accessing the allocation stacks.
VerifyHeapReferences()1498 bool Heap::VerifyHeapReferences() {
1499 Locks::mutator_lock_->AssertExclusiveHeld(Thread::Current());
1500 // Lets sort our allocation stacks so that we can efficiently binary search them.
1501 allocation_stack_->Sort();
1502 live_stack_->Sort();
1503 // Perform the verification.
1504 VerifyObjectVisitor visitor(this);
1505 Runtime::Current()->VisitRoots(VerifyReferenceVisitor::VerifyRoots, &visitor, false, false);
1506 GetLiveBitmap()->Visit(visitor);
1507 // Verify objects in the allocation stack since these will be objects which were:
1508 // 1. Allocated prior to the GC (pre GC verification).
1509 // 2. Allocated during the GC (pre sweep GC verification).
1510 for (mirror::Object** it = allocation_stack_->Begin(); it != allocation_stack_->End(); ++it) {
1511 visitor(*it);
1512 }
1513 // We don't want to verify the objects in the live stack since they themselves may be
1514 // pointing to dead objects if they are not reachable.
1515 if (visitor.Failed()) {
1516 // Dump mod-union tables.
1517 image_mod_union_table_->Dump(LOG(ERROR) << "Image mod-union table: ");
1518 zygote_mod_union_table_->Dump(LOG(ERROR) << "Zygote mod-union table: ");
1519 DumpSpaces();
1520 return false;
1521 }
1522 return true;
1523 }
1524
1525 class VerifyReferenceCardVisitor {
1526 public:
VerifyReferenceCardVisitor(Heap * heap,bool * failed)1527 VerifyReferenceCardVisitor(Heap* heap, bool* failed)
1528 SHARED_LOCKS_REQUIRED(Locks::mutator_lock_,
1529 Locks::heap_bitmap_lock_)
1530 : heap_(heap), failed_(failed) {
1531 }
1532
1533 // TODO: Fix lock analysis to not use NO_THREAD_SAFETY_ANALYSIS, requires support for
1534 // annotalysis on visitors.
operator ()(const mirror::Object * obj,const mirror::Object * ref,const MemberOffset & offset,bool is_static) const1535 void operator()(const mirror::Object* obj, const mirror::Object* ref, const MemberOffset& offset,
1536 bool is_static) const NO_THREAD_SAFETY_ANALYSIS {
1537 // Filter out class references since changing an object's class does not mark the card as dirty.
1538 // Also handles large objects, since the only reference they hold is a class reference.
1539 if (ref != NULL && !ref->IsClass()) {
1540 accounting::CardTable* card_table = heap_->GetCardTable();
1541 // If the object is not dirty and it is referencing something in the live stack other than
1542 // class, then it must be on a dirty card.
1543 if (!card_table->AddrIsInCardTable(obj)) {
1544 LOG(ERROR) << "Object " << obj << " is not in the address range of the card table";
1545 *failed_ = true;
1546 } else if (!card_table->IsDirty(obj)) {
1547 // Card should be either kCardDirty if it got re-dirtied after we aged it, or
1548 // kCardDirty - 1 if it didnt get touched since we aged it.
1549 accounting::ObjectStack* live_stack = heap_->live_stack_.get();
1550 if (live_stack->ContainsSorted(const_cast<mirror::Object*>(ref))) {
1551 if (live_stack->ContainsSorted(const_cast<mirror::Object*>(obj))) {
1552 LOG(ERROR) << "Object " << obj << " found in live stack";
1553 }
1554 if (heap_->GetLiveBitmap()->Test(obj)) {
1555 LOG(ERROR) << "Object " << obj << " found in live bitmap";
1556 }
1557 LOG(ERROR) << "Object " << obj << " " << PrettyTypeOf(obj)
1558 << " references " << ref << " " << PrettyTypeOf(ref) << " in live stack";
1559
1560 // Print which field of the object is dead.
1561 if (!obj->IsObjectArray()) {
1562 const mirror::Class* klass = is_static ? obj->AsClass() : obj->GetClass();
1563 CHECK(klass != NULL);
1564 const mirror::ObjectArray<mirror::ArtField>* fields = is_static ? klass->GetSFields()
1565 : klass->GetIFields();
1566 CHECK(fields != NULL);
1567 for (int32_t i = 0; i < fields->GetLength(); ++i) {
1568 const mirror::ArtField* cur = fields->Get(i);
1569 if (cur->GetOffset().Int32Value() == offset.Int32Value()) {
1570 LOG(ERROR) << (is_static ? "Static " : "") << "field in the live stack is "
1571 << PrettyField(cur);
1572 break;
1573 }
1574 }
1575 } else {
1576 const mirror::ObjectArray<mirror::Object>* object_array =
1577 obj->AsObjectArray<mirror::Object>();
1578 for (int32_t i = 0; i < object_array->GetLength(); ++i) {
1579 if (object_array->Get(i) == ref) {
1580 LOG(ERROR) << (is_static ? "Static " : "") << "obj[" << i << "] = ref";
1581 }
1582 }
1583 }
1584
1585 *failed_ = true;
1586 }
1587 }
1588 }
1589 }
1590
1591 private:
1592 Heap* const heap_;
1593 bool* const failed_;
1594 };
1595
1596 class VerifyLiveStackReferences {
1597 public:
VerifyLiveStackReferences(Heap * heap)1598 explicit VerifyLiveStackReferences(Heap* heap)
1599 : heap_(heap),
1600 failed_(false) {}
1601
operator ()(const mirror::Object * obj) const1602 void operator()(const mirror::Object* obj) const
1603 SHARED_LOCKS_REQUIRED(Locks::mutator_lock_, Locks::heap_bitmap_lock_) {
1604 VerifyReferenceCardVisitor visitor(heap_, const_cast<bool*>(&failed_));
1605 collector::MarkSweep::VisitObjectReferences(obj, visitor);
1606 }
1607
Failed() const1608 bool Failed() const {
1609 return failed_;
1610 }
1611
1612 private:
1613 Heap* const heap_;
1614 bool failed_;
1615 };
1616
VerifyMissingCardMarks()1617 bool Heap::VerifyMissingCardMarks() {
1618 Locks::mutator_lock_->AssertExclusiveHeld(Thread::Current());
1619
1620 // We need to sort the live stack since we binary search it.
1621 live_stack_->Sort();
1622 VerifyLiveStackReferences visitor(this);
1623 GetLiveBitmap()->Visit(visitor);
1624
1625 // We can verify objects in the live stack since none of these should reference dead objects.
1626 for (mirror::Object** it = live_stack_->Begin(); it != live_stack_->End(); ++it) {
1627 visitor(*it);
1628 }
1629
1630 if (visitor.Failed()) {
1631 DumpSpaces();
1632 return false;
1633 }
1634 return true;
1635 }
1636
SwapStacks()1637 void Heap::SwapStacks() {
1638 allocation_stack_.swap(live_stack_);
1639 }
1640
ProcessCards(base::TimingLogger & timings)1641 void Heap::ProcessCards(base::TimingLogger& timings) {
1642 // Clear cards and keep track of cards cleared in the mod-union table.
1643 for (const auto& space : continuous_spaces_) {
1644 if (space->IsImageSpace()) {
1645 base::TimingLogger::ScopedSplit split("ImageModUnionClearCards", &timings);
1646 image_mod_union_table_->ClearCards(space);
1647 } else if (space->IsZygoteSpace()) {
1648 base::TimingLogger::ScopedSplit split("ZygoteModUnionClearCards", &timings);
1649 zygote_mod_union_table_->ClearCards(space);
1650 } else {
1651 base::TimingLogger::ScopedSplit split("AllocSpaceClearCards", &timings);
1652 // No mod union table for the AllocSpace. Age the cards so that the GC knows that these cards
1653 // were dirty before the GC started.
1654 card_table_->ModifyCardsAtomic(space->Begin(), space->End(), AgeCardVisitor(), VoidFunctor());
1655 }
1656 }
1657 }
1658
PreGcVerification(collector::GarbageCollector * gc)1659 void Heap::PreGcVerification(collector::GarbageCollector* gc) {
1660 ThreadList* thread_list = Runtime::Current()->GetThreadList();
1661 Thread* self = Thread::Current();
1662
1663 if (verify_pre_gc_heap_) {
1664 thread_list->SuspendAll();
1665 {
1666 ReaderMutexLock mu(self, *Locks::heap_bitmap_lock_);
1667 if (!VerifyHeapReferences()) {
1668 LOG(FATAL) << "Pre " << gc->GetName() << " heap verification failed";
1669 }
1670 }
1671 thread_list->ResumeAll();
1672 }
1673
1674 // Check that all objects which reference things in the live stack are on dirty cards.
1675 if (verify_missing_card_marks_) {
1676 thread_list->SuspendAll();
1677 {
1678 ReaderMutexLock mu(self, *Locks::heap_bitmap_lock_);
1679 SwapStacks();
1680 // Sort the live stack so that we can quickly binary search it later.
1681 if (!VerifyMissingCardMarks()) {
1682 LOG(FATAL) << "Pre " << gc->GetName() << " missing card mark verification failed";
1683 }
1684 SwapStacks();
1685 }
1686 thread_list->ResumeAll();
1687 }
1688
1689 if (verify_mod_union_table_) {
1690 thread_list->SuspendAll();
1691 ReaderMutexLock reader_lock(self, *Locks::heap_bitmap_lock_);
1692 zygote_mod_union_table_->Update();
1693 zygote_mod_union_table_->Verify();
1694 image_mod_union_table_->Update();
1695 image_mod_union_table_->Verify();
1696 thread_list->ResumeAll();
1697 }
1698 }
1699
PreSweepingGcVerification(collector::GarbageCollector * gc)1700 void Heap::PreSweepingGcVerification(collector::GarbageCollector* gc) {
1701 // Called before sweeping occurs since we want to make sure we are not going so reclaim any
1702 // reachable objects.
1703 if (verify_post_gc_heap_) {
1704 Thread* self = Thread::Current();
1705 CHECK_NE(self->GetState(), kRunnable);
1706 {
1707 WriterMutexLock mu(self, *Locks::heap_bitmap_lock_);
1708 // Swapping bound bitmaps does nothing.
1709 gc->SwapBitmaps();
1710 if (!VerifyHeapReferences()) {
1711 LOG(FATAL) << "Pre sweeping " << gc->GetName() << " GC verification failed";
1712 }
1713 gc->SwapBitmaps();
1714 }
1715 }
1716 }
1717
PostGcVerification(collector::GarbageCollector * gc)1718 void Heap::PostGcVerification(collector::GarbageCollector* gc) {
1719 if (verify_system_weaks_) {
1720 Thread* self = Thread::Current();
1721 ReaderMutexLock mu(self, *Locks::heap_bitmap_lock_);
1722 collector::MarkSweep* mark_sweep = down_cast<collector::MarkSweep*>(gc);
1723 mark_sweep->VerifySystemWeaks();
1724 }
1725 }
1726
WaitForConcurrentGcToComplete(Thread * self)1727 collector::GcType Heap::WaitForConcurrentGcToComplete(Thread* self) {
1728 collector::GcType last_gc_type = collector::kGcTypeNone;
1729 if (concurrent_gc_) {
1730 ATRACE_BEGIN("GC: Wait For Concurrent");
1731 bool do_wait;
1732 uint64_t wait_start = NanoTime();
1733 {
1734 // Check if GC is running holding gc_complete_lock_.
1735 MutexLock mu(self, *gc_complete_lock_);
1736 do_wait = is_gc_running_;
1737 }
1738 if (do_wait) {
1739 uint64_t wait_time;
1740 // We must wait, change thread state then sleep on gc_complete_cond_;
1741 ScopedThreadStateChange tsc(Thread::Current(), kWaitingForGcToComplete);
1742 {
1743 MutexLock mu(self, *gc_complete_lock_);
1744 while (is_gc_running_) {
1745 gc_complete_cond_->Wait(self);
1746 }
1747 last_gc_type = last_gc_type_;
1748 wait_time = NanoTime() - wait_start;
1749 total_wait_time_ += wait_time;
1750 }
1751 if (wait_time > long_pause_log_threshold_) {
1752 LOG(INFO) << "WaitForConcurrentGcToComplete blocked for " << PrettyDuration(wait_time);
1753 }
1754 }
1755 ATRACE_END();
1756 }
1757 return last_gc_type;
1758 }
1759
DumpForSigQuit(std::ostream & os)1760 void Heap::DumpForSigQuit(std::ostream& os) {
1761 os << "Heap: " << GetPercentFree() << "% free, " << PrettySize(GetBytesAllocated()) << "/"
1762 << PrettySize(GetTotalMemory()) << "; " << GetObjectsAllocated() << " objects\n";
1763 DumpGcPerformanceInfo(os);
1764 }
1765
GetPercentFree()1766 size_t Heap::GetPercentFree() {
1767 return static_cast<size_t>(100.0f * static_cast<float>(GetFreeMemory()) / GetTotalMemory());
1768 }
1769
SetIdealFootprint(size_t max_allowed_footprint)1770 void Heap::SetIdealFootprint(size_t max_allowed_footprint) {
1771 if (max_allowed_footprint > GetMaxMemory()) {
1772 VLOG(gc) << "Clamp target GC heap from " << PrettySize(max_allowed_footprint) << " to "
1773 << PrettySize(GetMaxMemory());
1774 max_allowed_footprint = GetMaxMemory();
1775 }
1776 max_allowed_footprint_ = max_allowed_footprint;
1777 }
1778
UpdateMaxNativeFootprint()1779 void Heap::UpdateMaxNativeFootprint() {
1780 size_t native_size = native_bytes_allocated_;
1781 // TODO: Tune the native heap utilization to be a value other than the java heap utilization.
1782 size_t target_size = native_size / GetTargetHeapUtilization();
1783 if (target_size > native_size + max_free_) {
1784 target_size = native_size + max_free_;
1785 } else if (target_size < native_size + min_free_) {
1786 target_size = native_size + min_free_;
1787 }
1788 native_footprint_gc_watermark_ = target_size;
1789 native_footprint_limit_ = 2 * target_size - native_size;
1790 }
1791
GrowForUtilization(collector::GcType gc_type,uint64_t gc_duration)1792 void Heap::GrowForUtilization(collector::GcType gc_type, uint64_t gc_duration) {
1793 // We know what our utilization is at this moment.
1794 // This doesn't actually resize any memory. It just lets the heap grow more when necessary.
1795 const size_t bytes_allocated = GetBytesAllocated();
1796 last_gc_size_ = bytes_allocated;
1797 last_gc_time_ns_ = NanoTime();
1798
1799 size_t target_size;
1800 if (gc_type != collector::kGcTypeSticky) {
1801 // Grow the heap for non sticky GC.
1802 target_size = bytes_allocated / GetTargetHeapUtilization();
1803 if (target_size > bytes_allocated + max_free_) {
1804 target_size = bytes_allocated + max_free_;
1805 } else if (target_size < bytes_allocated + min_free_) {
1806 target_size = bytes_allocated + min_free_;
1807 }
1808 next_gc_type_ = collector::kGcTypeSticky;
1809 } else {
1810 // Based on how close the current heap size is to the target size, decide
1811 // whether or not to do a partial or sticky GC next.
1812 if (bytes_allocated + min_free_ <= max_allowed_footprint_) {
1813 next_gc_type_ = collector::kGcTypeSticky;
1814 } else {
1815 next_gc_type_ = collector::kGcTypePartial;
1816 }
1817
1818 // If we have freed enough memory, shrink the heap back down.
1819 if (bytes_allocated + max_free_ < max_allowed_footprint_) {
1820 target_size = bytes_allocated + max_free_;
1821 } else {
1822 target_size = std::max(bytes_allocated, max_allowed_footprint_);
1823 }
1824 }
1825
1826 if (!ignore_max_footprint_) {
1827 SetIdealFootprint(target_size);
1828
1829 if (concurrent_gc_) {
1830 // Calculate when to perform the next ConcurrentGC.
1831
1832 // Calculate the estimated GC duration.
1833 double gc_duration_seconds = NsToMs(gc_duration) / 1000.0;
1834 // Estimate how many remaining bytes we will have when we need to start the next GC.
1835 size_t remaining_bytes = allocation_rate_ * gc_duration_seconds;
1836 remaining_bytes = std::max(remaining_bytes, kMinConcurrentRemainingBytes);
1837 if (UNLIKELY(remaining_bytes > max_allowed_footprint_)) {
1838 // A never going to happen situation that from the estimated allocation rate we will exceed
1839 // the applications entire footprint with the given estimated allocation rate. Schedule
1840 // another GC straight away.
1841 concurrent_start_bytes_ = bytes_allocated;
1842 } else {
1843 // Start a concurrent GC when we get close to the estimated remaining bytes. When the
1844 // allocation rate is very high, remaining_bytes could tell us that we should start a GC
1845 // right away.
1846 concurrent_start_bytes_ = std::max(max_allowed_footprint_ - remaining_bytes, bytes_allocated);
1847 }
1848 DCHECK_LE(concurrent_start_bytes_, max_allowed_footprint_);
1849 DCHECK_LE(max_allowed_footprint_, growth_limit_);
1850 }
1851 }
1852
1853 UpdateMaxNativeFootprint();
1854 }
1855
ClearGrowthLimit()1856 void Heap::ClearGrowthLimit() {
1857 growth_limit_ = capacity_;
1858 alloc_space_->ClearGrowthLimit();
1859 }
1860
SetReferenceOffsets(MemberOffset reference_referent_offset,MemberOffset reference_queue_offset,MemberOffset reference_queueNext_offset,MemberOffset reference_pendingNext_offset,MemberOffset finalizer_reference_zombie_offset)1861 void Heap::SetReferenceOffsets(MemberOffset reference_referent_offset,
1862 MemberOffset reference_queue_offset,
1863 MemberOffset reference_queueNext_offset,
1864 MemberOffset reference_pendingNext_offset,
1865 MemberOffset finalizer_reference_zombie_offset) {
1866 reference_referent_offset_ = reference_referent_offset;
1867 reference_queue_offset_ = reference_queue_offset;
1868 reference_queueNext_offset_ = reference_queueNext_offset;
1869 reference_pendingNext_offset_ = reference_pendingNext_offset;
1870 finalizer_reference_zombie_offset_ = finalizer_reference_zombie_offset;
1871 CHECK_NE(reference_referent_offset_.Uint32Value(), 0U);
1872 CHECK_NE(reference_queue_offset_.Uint32Value(), 0U);
1873 CHECK_NE(reference_queueNext_offset_.Uint32Value(), 0U);
1874 CHECK_NE(reference_pendingNext_offset_.Uint32Value(), 0U);
1875 CHECK_NE(finalizer_reference_zombie_offset_.Uint32Value(), 0U);
1876 }
1877
GetReferenceReferent(mirror::Object * reference)1878 mirror::Object* Heap::GetReferenceReferent(mirror::Object* reference) {
1879 DCHECK(reference != NULL);
1880 DCHECK_NE(reference_referent_offset_.Uint32Value(), 0U);
1881 return reference->GetFieldObject<mirror::Object*>(reference_referent_offset_, true);
1882 }
1883
ClearReferenceReferent(mirror::Object * reference)1884 void Heap::ClearReferenceReferent(mirror::Object* reference) {
1885 DCHECK(reference != NULL);
1886 DCHECK_NE(reference_referent_offset_.Uint32Value(), 0U);
1887 reference->SetFieldObject(reference_referent_offset_, NULL, true);
1888 }
1889
1890 // Returns true if the reference object has not yet been enqueued.
IsEnqueuable(const mirror::Object * ref)1891 bool Heap::IsEnqueuable(const mirror::Object* ref) {
1892 DCHECK(ref != NULL);
1893 const mirror::Object* queue =
1894 ref->GetFieldObject<mirror::Object*>(reference_queue_offset_, false);
1895 const mirror::Object* queue_next =
1896 ref->GetFieldObject<mirror::Object*>(reference_queueNext_offset_, false);
1897 return (queue != NULL) && (queue_next == NULL);
1898 }
1899
EnqueueReference(mirror::Object * ref,mirror::Object ** cleared_reference_list)1900 void Heap::EnqueueReference(mirror::Object* ref, mirror::Object** cleared_reference_list) {
1901 DCHECK(ref != NULL);
1902 CHECK(ref->GetFieldObject<mirror::Object*>(reference_queue_offset_, false) != NULL);
1903 CHECK(ref->GetFieldObject<mirror::Object*>(reference_queueNext_offset_, false) == NULL);
1904 EnqueuePendingReference(ref, cleared_reference_list);
1905 }
1906
IsEnqueued(mirror::Object * ref)1907 bool Heap::IsEnqueued(mirror::Object* ref) {
1908 // Since the references are stored as cyclic lists it means that once enqueued, the pending next
1909 // will always be non-null.
1910 return ref->GetFieldObject<mirror::Object*>(GetReferencePendingNextOffset(), false) != nullptr;
1911 }
1912
EnqueuePendingReference(mirror::Object * ref,mirror::Object ** list)1913 void Heap::EnqueuePendingReference(mirror::Object* ref, mirror::Object** list) {
1914 DCHECK(ref != NULL);
1915 DCHECK(list != NULL);
1916 if (*list == NULL) {
1917 // 1 element cyclic queue, ie: Reference ref = ..; ref.pendingNext = ref;
1918 ref->SetFieldObject(reference_pendingNext_offset_, ref, false);
1919 *list = ref;
1920 } else {
1921 mirror::Object* head =
1922 (*list)->GetFieldObject<mirror::Object*>(reference_pendingNext_offset_, false);
1923 ref->SetFieldObject(reference_pendingNext_offset_, head, false);
1924 (*list)->SetFieldObject(reference_pendingNext_offset_, ref, false);
1925 }
1926 }
1927
DequeuePendingReference(mirror::Object ** list)1928 mirror::Object* Heap::DequeuePendingReference(mirror::Object** list) {
1929 DCHECK(list != NULL);
1930 DCHECK(*list != NULL);
1931 mirror::Object* head = (*list)->GetFieldObject<mirror::Object*>(reference_pendingNext_offset_,
1932 false);
1933 mirror::Object* ref;
1934
1935 // Note: the following code is thread-safe because it is only called from ProcessReferences which
1936 // is single threaded.
1937 if (*list == head) {
1938 ref = *list;
1939 *list = NULL;
1940 } else {
1941 mirror::Object* next = head->GetFieldObject<mirror::Object*>(reference_pendingNext_offset_,
1942 false);
1943 (*list)->SetFieldObject(reference_pendingNext_offset_, next, false);
1944 ref = head;
1945 }
1946 ref->SetFieldObject(reference_pendingNext_offset_, NULL, false);
1947 return ref;
1948 }
1949
AddFinalizerReference(Thread * self,mirror::Object * object)1950 void Heap::AddFinalizerReference(Thread* self, mirror::Object* object) {
1951 ScopedObjectAccess soa(self);
1952 JValue result;
1953 ArgArray arg_array(NULL, 0);
1954 arg_array.Append(reinterpret_cast<uint32_t>(object));
1955 soa.DecodeMethod(WellKnownClasses::java_lang_ref_FinalizerReference_add)->Invoke(self,
1956 arg_array.GetArray(), arg_array.GetNumBytes(), &result, 'V');
1957 }
1958
EnqueueClearedReferences(mirror::Object ** cleared)1959 void Heap::EnqueueClearedReferences(mirror::Object** cleared) {
1960 DCHECK(cleared != NULL);
1961 if (*cleared != NULL) {
1962 // When a runtime isn't started there are no reference queues to care about so ignore.
1963 if (LIKELY(Runtime::Current()->IsStarted())) {
1964 ScopedObjectAccess soa(Thread::Current());
1965 JValue result;
1966 ArgArray arg_array(NULL, 0);
1967 arg_array.Append(reinterpret_cast<uint32_t>(*cleared));
1968 soa.DecodeMethod(WellKnownClasses::java_lang_ref_ReferenceQueue_add)->Invoke(soa.Self(),
1969 arg_array.GetArray(), arg_array.GetNumBytes(), &result, 'V');
1970 }
1971 *cleared = NULL;
1972 }
1973 }
1974
RequestConcurrentGC(Thread * self)1975 void Heap::RequestConcurrentGC(Thread* self) {
1976 // Make sure that we can do a concurrent GC.
1977 Runtime* runtime = Runtime::Current();
1978 DCHECK(concurrent_gc_);
1979 if (runtime == NULL || !runtime->IsFinishedStarting() ||
1980 !runtime->IsConcurrentGcEnabled()) {
1981 return;
1982 }
1983 {
1984 MutexLock mu(self, *Locks::runtime_shutdown_lock_);
1985 if (runtime->IsShuttingDown()) {
1986 return;
1987 }
1988 }
1989 if (self->IsHandlingStackOverflow()) {
1990 return;
1991 }
1992
1993 // We already have a request pending, no reason to start more until we update
1994 // concurrent_start_bytes_.
1995 concurrent_start_bytes_ = std::numeric_limits<size_t>::max();
1996
1997 JNIEnv* env = self->GetJniEnv();
1998 DCHECK(WellKnownClasses::java_lang_Daemons != NULL);
1999 DCHECK(WellKnownClasses::java_lang_Daemons_requestGC != NULL);
2000 env->CallStaticVoidMethod(WellKnownClasses::java_lang_Daemons,
2001 WellKnownClasses::java_lang_Daemons_requestGC);
2002 CHECK(!env->ExceptionCheck());
2003 }
2004
ConcurrentGC(Thread * self)2005 void Heap::ConcurrentGC(Thread* self) {
2006 {
2007 MutexLock mu(self, *Locks::runtime_shutdown_lock_);
2008 if (Runtime::Current()->IsShuttingDown()) {
2009 return;
2010 }
2011 }
2012
2013 // Wait for any GCs currently running to finish.
2014 if (WaitForConcurrentGcToComplete(self) == collector::kGcTypeNone) {
2015 CollectGarbageInternal(next_gc_type_, kGcCauseBackground, false);
2016 }
2017 }
2018
RequestHeapTrim()2019 void Heap::RequestHeapTrim() {
2020 // GC completed and now we must decide whether to request a heap trim (advising pages back to the
2021 // kernel) or not. Issuing a request will also cause trimming of the libc heap. As a trim scans
2022 // a space it will hold its lock and can become a cause of jank.
2023 // Note, the large object space self trims and the Zygote space was trimmed and unchanging since
2024 // forking.
2025
2026 // We don't have a good measure of how worthwhile a trim might be. We can't use the live bitmap
2027 // because that only marks object heads, so a large array looks like lots of empty space. We
2028 // don't just call dlmalloc all the time, because the cost of an _attempted_ trim is proportional
2029 // to utilization (which is probably inversely proportional to how much benefit we can expect).
2030 // We could try mincore(2) but that's only a measure of how many pages we haven't given away,
2031 // not how much use we're making of those pages.
2032 uint64_t ms_time = MilliTime();
2033 float utilization =
2034 static_cast<float>(alloc_space_->GetBytesAllocated()) / alloc_space_->Size();
2035 if ((utilization > 0.75f && !IsLowMemoryMode()) || ((ms_time - last_trim_time_ms_) < 2 * 1000)) {
2036 // Don't bother trimming the alloc space if it's more than 75% utilized and low memory mode is
2037 // not enabled, or if a heap trim occurred in the last two seconds.
2038 return;
2039 }
2040
2041 Thread* self = Thread::Current();
2042 {
2043 MutexLock mu(self, *Locks::runtime_shutdown_lock_);
2044 Runtime* runtime = Runtime::Current();
2045 if (runtime == NULL || !runtime->IsFinishedStarting() || runtime->IsShuttingDown()) {
2046 // Heap trimming isn't supported without a Java runtime or Daemons (such as at dex2oat time)
2047 // Also: we do not wish to start a heap trim if the runtime is shutting down (a racy check
2048 // as we don't hold the lock while requesting the trim).
2049 return;
2050 }
2051 }
2052
2053 last_trim_time_ms_ = ms_time;
2054 ListenForProcessStateChange();
2055
2056 // Trim only if we do not currently care about pause times.
2057 if (!care_about_pause_times_) {
2058 JNIEnv* env = self->GetJniEnv();
2059 DCHECK(WellKnownClasses::java_lang_Daemons != NULL);
2060 DCHECK(WellKnownClasses::java_lang_Daemons_requestHeapTrim != NULL);
2061 env->CallStaticVoidMethod(WellKnownClasses::java_lang_Daemons,
2062 WellKnownClasses::java_lang_Daemons_requestHeapTrim);
2063 CHECK(!env->ExceptionCheck());
2064 }
2065 }
2066
Trim()2067 size_t Heap::Trim() {
2068 // Handle a requested heap trim on a thread outside of the main GC thread.
2069 return alloc_space_->Trim();
2070 }
2071
IsGCRequestPending() const2072 bool Heap::IsGCRequestPending() const {
2073 return concurrent_start_bytes_ != std::numeric_limits<size_t>::max();
2074 }
2075
RegisterNativeAllocation(int bytes)2076 void Heap::RegisterNativeAllocation(int bytes) {
2077 // Total number of native bytes allocated.
2078 native_bytes_allocated_.fetch_add(bytes);
2079 Thread* self = Thread::Current();
2080 if (static_cast<size_t>(native_bytes_allocated_) > native_footprint_gc_watermark_) {
2081 // The second watermark is higher than the gc watermark. If you hit this it means you are
2082 // allocating native objects faster than the GC can keep up with.
2083 if (static_cast<size_t>(native_bytes_allocated_) > native_footprint_limit_) {
2084 JNIEnv* env = self->GetJniEnv();
2085 // Can't do this in WellKnownClasses::Init since System is not properly set up at that
2086 // point.
2087 if (WellKnownClasses::java_lang_System_runFinalization == NULL) {
2088 DCHECK(WellKnownClasses::java_lang_System != NULL);
2089 WellKnownClasses::java_lang_System_runFinalization =
2090 CacheMethod(env, WellKnownClasses::java_lang_System, true, "runFinalization", "()V");
2091 assert(WellKnownClasses::java_lang_System_runFinalization != NULL);
2092 }
2093 if (WaitForConcurrentGcToComplete(self) != collector::kGcTypeNone) {
2094 // Just finished a GC, attempt to run finalizers.
2095 env->CallStaticVoidMethod(WellKnownClasses::java_lang_System,
2096 WellKnownClasses::java_lang_System_runFinalization);
2097 CHECK(!env->ExceptionCheck());
2098 }
2099
2100 // If we still are over the watermark, attempt a GC for alloc and run finalizers.
2101 if (static_cast<size_t>(native_bytes_allocated_) > native_footprint_limit_) {
2102 CollectGarbageInternal(collector::kGcTypePartial, kGcCauseForAlloc, false);
2103 env->CallStaticVoidMethod(WellKnownClasses::java_lang_System,
2104 WellKnownClasses::java_lang_System_runFinalization);
2105 CHECK(!env->ExceptionCheck());
2106 }
2107 // We have just run finalizers, update the native watermark since it is very likely that
2108 // finalizers released native managed allocations.
2109 UpdateMaxNativeFootprint();
2110 } else {
2111 if (!IsGCRequestPending()) {
2112 RequestConcurrentGC(self);
2113 }
2114 }
2115 }
2116 }
2117
RegisterNativeFree(int bytes)2118 void Heap::RegisterNativeFree(int bytes) {
2119 int expected_size, new_size;
2120 do {
2121 expected_size = native_bytes_allocated_.load();
2122 new_size = expected_size - bytes;
2123 if (new_size < 0) {
2124 ThrowRuntimeException("attempted to free %d native bytes with only %d native bytes registered as allocated",
2125 bytes, expected_size);
2126 break;
2127 }
2128 } while (!native_bytes_allocated_.compare_and_swap(expected_size, new_size));
2129 }
2130
GetTotalMemory() const2131 int64_t Heap::GetTotalMemory() const {
2132 int64_t ret = 0;
2133 for (const auto& space : continuous_spaces_) {
2134 if (space->IsImageSpace()) {
2135 // Currently don't include the image space.
2136 } else if (space->IsDlMallocSpace()) {
2137 // Zygote or alloc space
2138 ret += space->AsDlMallocSpace()->GetFootprint();
2139 }
2140 }
2141 for (const auto& space : discontinuous_spaces_) {
2142 if (space->IsLargeObjectSpace()) {
2143 ret += space->AsLargeObjectSpace()->GetBytesAllocated();
2144 }
2145 }
2146 return ret;
2147 }
2148
2149 } // namespace gc
2150 } // namespace art
2151