• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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