heap.cc revision 161a8e0fcbf26922c97654c00b68082be71eeb50
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
25#include "base/stl_util.h"
26#include "common_throws.h"
27#include "cutils/sched_policy.h"
28#include "debugger.h"
29#include "gc/accounting/atomic_stack.h"
30#include "gc/accounting/card_table-inl.h"
31#include "gc/accounting/heap_bitmap-inl.h"
32#include "gc/accounting/mod_union_table-inl.h"
33#include "gc/accounting/space_bitmap-inl.h"
34#include "gc/collector/mark_sweep-inl.h"
35#include "gc/collector/partial_mark_sweep.h"
36#include "gc/collector/sticky_mark_sweep.h"
37#include "gc/space/image_space.h"
38#include "gc/space/large_object_space.h"
39#include "gc/space/space-inl.h"
40#include "image.h"
41#include "invoke_arg_array_builder.h"
42#include "mirror/class-inl.h"
43#include "mirror/field-inl.h"
44#include "mirror/object.h"
45#include "mirror/object-inl.h"
46#include "mirror/object_array-inl.h"
47#include "object_utils.h"
48#include "os.h"
49#include "ScopedLocalRef.h"
50#include "scoped_thread_state_change.h"
51#include "sirt_ref.h"
52#include "thread_list.h"
53#include "UniquePtr.h"
54#include "well_known_classes.h"
55
56namespace art {
57namespace gc {
58
59// When to create a log message about a slow GC, 100ms.
60static const uint64_t kSlowGcThreshold = MsToNs(100);
61// When to create a log message about a long pause, 5ms.
62static const uint64_t kLongGcPauseThreshold = MsToNs(5);
63static const bool kGCALotMode = false;
64static const size_t kGcAlotInterval = KB;
65static const bool kDumpGcPerformanceOnShutdown = false;
66// Minimum amount of remaining bytes before a concurrent GC is triggered.
67static const size_t kMinConcurrentRemainingBytes = 128 * KB;
68const double Heap::kDefaultTargetUtilization = 0.5;
69
70Heap::Heap(size_t initial_size, size_t growth_limit, size_t min_free, size_t max_free,
71           double target_utilization, size_t capacity,
72           const std::string& original_image_file_name, bool concurrent_gc, size_t num_gc_threads)
73    : alloc_space_(NULL),
74      card_table_(NULL),
75      concurrent_gc_(concurrent_gc),
76      num_gc_threads_(num_gc_threads),
77      have_zygote_space_(false),
78      reference_queue_lock_(NULL),
79      is_gc_running_(false),
80      last_gc_type_(collector::kGcTypeNone),
81      next_gc_type_(collector::kGcTypePartial),
82      capacity_(capacity),
83      growth_limit_(growth_limit),
84      max_allowed_footprint_(initial_size),
85      native_footprint_gc_watermark_(initial_size),
86      native_footprint_limit_(2 * initial_size),
87      concurrent_start_bytes_(concurrent_gc ? initial_size - (kMinConcurrentRemainingBytes)
88                                            :  std::numeric_limits<size_t>::max()),
89      total_bytes_freed_ever_(0),
90      total_objects_freed_ever_(0),
91      large_object_threshold_(3 * kPageSize),
92      num_bytes_allocated_(0),
93      native_bytes_allocated_(0),
94      process_state_(PROCESS_STATE_TOP),
95      gc_memory_overhead_(0),
96      verify_missing_card_marks_(false),
97      verify_system_weaks_(false),
98      verify_pre_gc_heap_(false),
99      verify_post_gc_heap_(false),
100      verify_mod_union_table_(false),
101      min_alloc_space_size_for_sticky_gc_(2 * MB),
102      min_remaining_space_for_sticky_gc_(1 * MB),
103      last_trim_time_ms_(0),
104      allocation_rate_(0),
105      /* For GC a lot mode, we limit the allocations stacks to be kGcAlotInterval allocations. This
106       * causes a lot of GC since we do a GC for alloc whenever the stack is full. When heap
107       * verification is enabled, we limit the size of allocation stacks to speed up their
108       * searching.
109       */
110      max_allocation_stack_size_(kGCALotMode ? kGcAlotInterval
111          : (kDesiredHeapVerification > kNoHeapVerification) ? KB : MB),
112      reference_referent_offset_(0),
113      reference_queue_offset_(0),
114      reference_queueNext_offset_(0),
115      reference_pendingNext_offset_(0),
116      finalizer_reference_zombie_offset_(0),
117      min_free_(min_free),
118      max_free_(max_free),
119      target_utilization_(target_utilization),
120      total_wait_time_(0),
121      measure_allocation_time_(false),
122      total_allocation_time_(0),
123      verify_object_mode_(kHeapVerificationNotPermitted) {
124  if (VLOG_IS_ON(heap) || VLOG_IS_ON(startup)) {
125    LOG(INFO) << "Heap() entering";
126  }
127
128  live_bitmap_.reset(new accounting::HeapBitmap(this));
129  mark_bitmap_.reset(new accounting::HeapBitmap(this));
130
131  // Requested begin for the alloc space, to follow the mapped image and oat files
132  byte* requested_alloc_space_begin = NULL;
133  std::string image_file_name(original_image_file_name);
134  if (!image_file_name.empty()) {
135    space::ImageSpace* image_space = space::ImageSpace::Create(image_file_name);
136    CHECK(image_space != NULL) << "Failed to create space for " << image_file_name;
137    AddContinuousSpace(image_space);
138    // Oat files referenced by image files immediately follow them in memory, ensure alloc space
139    // isn't going to get in the middle
140    byte* oat_file_end_addr = image_space->GetImageHeader().GetOatFileEnd();
141    CHECK_GT(oat_file_end_addr, image_space->End());
142    if (oat_file_end_addr > requested_alloc_space_begin) {
143      requested_alloc_space_begin =
144          reinterpret_cast<byte*>(RoundUp(reinterpret_cast<uintptr_t>(oat_file_end_addr),
145                                          kPageSize));
146    }
147  }
148
149  // Allocate the large object space.
150  const bool kUseFreeListSpaceForLOS  = false;
151  if (kUseFreeListSpaceForLOS) {
152    large_object_space_ = space::FreeListSpace::Create("large object space", NULL, capacity);
153  } else {
154    large_object_space_ = space::LargeObjectMapSpace::Create("large object space");
155  }
156  CHECK(large_object_space_ != NULL) << "Failed to create large object space";
157  AddDiscontinuousSpace(large_object_space_);
158
159  alloc_space_ = space::DlMallocSpace::Create(Runtime::Current()->IsZygote() ? "zygote space" : "alloc space",
160                                              initial_size,
161                                              growth_limit, capacity,
162                                              requested_alloc_space_begin);
163  CHECK(alloc_space_ != NULL) << "Failed to create alloc space";
164  alloc_space_->SetFootprintLimit(alloc_space_->Capacity());
165  AddContinuousSpace(alloc_space_);
166
167  // Compute heap capacity. Continuous spaces are sorted in order of Begin().
168  byte* heap_begin = continuous_spaces_.front()->Begin();
169  size_t heap_capacity = continuous_spaces_.back()->End() - continuous_spaces_.front()->Begin();
170  if (continuous_spaces_.back()->IsDlMallocSpace()) {
171    heap_capacity += continuous_spaces_.back()->AsDlMallocSpace()->NonGrowthLimitCapacity();
172  }
173
174  // Mark image objects in the live bitmap
175  // TODO: C++0x
176  typedef std::vector<space::ContinuousSpace*>::iterator It;
177  for (It it = continuous_spaces_.begin(); it != continuous_spaces_.end(); ++it) {
178    space::ContinuousSpace* space = *it;
179    if (space->IsImageSpace()) {
180      space::ImageSpace* image_space = space->AsImageSpace();
181      image_space->RecordImageAllocations(image_space->GetLiveBitmap());
182    }
183  }
184
185  // Allocate the card table.
186  card_table_.reset(accounting::CardTable::Create(heap_begin, heap_capacity));
187  CHECK(card_table_.get() != NULL) << "Failed to create card table";
188
189  image_mod_union_table_.reset(new accounting::ModUnionTableToZygoteAllocspace(this));
190  CHECK(image_mod_union_table_.get() != NULL) << "Failed to create image mod-union table";
191
192  zygote_mod_union_table_.reset(new accounting::ModUnionTableCardCache(this));
193  CHECK(zygote_mod_union_table_.get() != NULL) << "Failed to create Zygote mod-union table";
194
195  // TODO: Count objects in the image space here.
196  num_bytes_allocated_ = 0;
197
198  // Default mark stack size in bytes.
199  static const size_t default_mark_stack_size = 64 * KB;
200  mark_stack_.reset(accounting::ObjectStack::Create("mark stack", default_mark_stack_size));
201  allocation_stack_.reset(accounting::ObjectStack::Create("allocation stack",
202                                                          max_allocation_stack_size_));
203  live_stack_.reset(accounting::ObjectStack::Create("live stack",
204                                                    max_allocation_stack_size_));
205
206  // It's still too early to take a lock because there are no threads yet, but we can create locks
207  // now. We don't create it earlier to make it clear that you can't use locks during heap
208  // initialization.
209  gc_complete_lock_ = new Mutex("GC complete lock");
210  gc_complete_cond_.reset(new ConditionVariable("GC complete condition variable",
211                                                *gc_complete_lock_));
212
213  // Create the reference queue lock, this is required so for parallel object scanning in the GC.
214  reference_queue_lock_ = new Mutex("reference queue lock");
215
216  last_gc_time_ns_ = NanoTime();
217  last_gc_size_ = GetBytesAllocated();
218
219  // Create our garbage collectors.
220  for (size_t i = 0; i < 2; ++i) {
221    const bool concurrent = i != 0;
222    mark_sweep_collectors_.push_back(new collector::MarkSweep(this, concurrent));
223    mark_sweep_collectors_.push_back(new collector::PartialMarkSweep(this, concurrent));
224    mark_sweep_collectors_.push_back(new collector::StickyMarkSweep(this, concurrent));
225  }
226
227  CHECK_NE(max_allowed_footprint_, 0U);
228  if (VLOG_IS_ON(heap) || VLOG_IS_ON(startup)) {
229    LOG(INFO) << "Heap() exiting";
230  }
231}
232
233void Heap::CreateThreadPool() {
234  thread_pool_.reset(new ThreadPool(num_gc_threads_));
235}
236
237void Heap::DeleteThreadPool() {
238  thread_pool_.reset(NULL);
239}
240
241// Sort spaces based on begin address
242struct ContinuousSpaceSorter {
243  bool operator()(const space::ContinuousSpace* a, const space::ContinuousSpace* b) const {
244    return a->Begin() < b->Begin();
245  }
246};
247
248void Heap::UpdateProcessState(ProcessState process_state) {
249  process_state_ = process_state;
250}
251
252void Heap::AddContinuousSpace(space::ContinuousSpace* space) {
253  WriterMutexLock mu(Thread::Current(), *Locks::heap_bitmap_lock_);
254  DCHECK(space != NULL);
255  DCHECK(space->GetLiveBitmap() != NULL);
256  live_bitmap_->AddContinuousSpaceBitmap(space->GetLiveBitmap());
257  DCHECK(space->GetMarkBitmap() != NULL);
258  mark_bitmap_->AddContinuousSpaceBitmap(space->GetMarkBitmap());
259  continuous_spaces_.push_back(space);
260  if (space->IsDlMallocSpace() && !space->IsLargeObjectSpace()) {
261    alloc_space_ = space->AsDlMallocSpace();
262  }
263
264  // Ensure that spaces remain sorted in increasing order of start address (required for CMS finger)
265  std::sort(continuous_spaces_.begin(), continuous_spaces_.end(), ContinuousSpaceSorter());
266
267  // Ensure that ImageSpaces < ZygoteSpaces < AllocSpaces so that we can do address based checks to
268  // avoid redundant marking.
269  bool seen_zygote = false, seen_alloc = false;
270  typedef std::vector<space::ContinuousSpace*>::const_iterator It;
271  for (It it = continuous_spaces_.begin(); it != continuous_spaces_.end(); ++it) {
272    space::ContinuousSpace* space = *it;
273    if (space->IsImageSpace()) {
274      DCHECK(!seen_zygote);
275      DCHECK(!seen_alloc);
276    } else if (space->IsZygoteSpace()) {
277      DCHECK(!seen_alloc);
278      seen_zygote = true;
279    } else if (space->IsDlMallocSpace()) {
280      seen_alloc = true;
281    }
282  }
283}
284
285void Heap::RegisterGCAllocation(size_t bytes) {
286  if (this != NULL) {
287    gc_memory_overhead_.fetch_add(bytes);
288  }
289}
290
291void Heap::RegisterGCDeAllocation(size_t bytes) {
292  if (this != NULL) {
293    gc_memory_overhead_.fetch_sub(bytes);
294  }
295}
296
297void Heap::AddDiscontinuousSpace(space::DiscontinuousSpace* space) {
298  WriterMutexLock mu(Thread::Current(), *Locks::heap_bitmap_lock_);
299  DCHECK(space != NULL);
300  DCHECK(space->GetLiveObjects() != NULL);
301  live_bitmap_->AddDiscontinuousObjectSet(space->GetLiveObjects());
302  DCHECK(space->GetMarkObjects() != NULL);
303  mark_bitmap_->AddDiscontinuousObjectSet(space->GetMarkObjects());
304  discontinuous_spaces_.push_back(space);
305}
306
307void Heap::DumpGcPerformanceInfo(std::ostream& os) {
308  // Dump cumulative timings.
309  os << "Dumping cumulative Gc timings\n";
310  uint64_t total_duration = 0;
311
312  // Dump cumulative loggers for each GC type.
313  // TODO: C++0x
314  uint64_t total_paused_time = 0;
315  typedef std::vector<collector::MarkSweep*>::const_iterator It;
316  for (It it = mark_sweep_collectors_.begin();
317       it != mark_sweep_collectors_.end(); ++it) {
318    collector::MarkSweep* collector = *it;
319    CumulativeLogger& logger = collector->GetCumulativeTimings();
320    if (logger.GetTotalNs() != 0) {
321      os << Dumpable<CumulativeLogger>(logger);
322      const uint64_t total_ns = logger.GetTotalNs();
323      const uint64_t total_pause_ns = (*it)->GetTotalPausedTimeNs();
324      double seconds = NsToMs(logger.GetTotalNs()) / 1000.0;
325      const uint64_t freed_bytes = collector->GetTotalFreedBytes();
326      const uint64_t freed_objects = collector->GetTotalFreedObjects();
327      os << collector->GetName() << " total time: " << PrettyDuration(total_ns) << "\n"
328         << collector->GetName() << " paused time: " << PrettyDuration(total_pause_ns) << "\n"
329         << collector->GetName() << " freed: " << freed_objects
330         << " objects with total size " << PrettySize(freed_bytes) << "\n"
331         << collector->GetName() << " throughput: " << freed_objects / seconds << "/s / "
332         << PrettySize(freed_bytes / seconds) << "/s\n";
333      total_duration += total_ns;
334      total_paused_time += total_pause_ns;
335    }
336  }
337  uint64_t allocation_time = static_cast<uint64_t>(total_allocation_time_) * kTimeAdjust;
338  size_t total_objects_allocated = GetObjectsAllocatedEver();
339  size_t total_bytes_allocated = GetBytesAllocatedEver();
340  if (total_duration != 0) {
341    const double total_seconds = static_cast<double>(total_duration / 1000) / 1000000.0;
342    os << "Total time spent in GC: " << PrettyDuration(total_duration) << "\n";
343    os << "Mean GC size throughput: "
344       << PrettySize(GetBytesFreedEver() / total_seconds) << "/s\n";
345    os << "Mean GC object throughput: "
346       << (GetObjectsFreedEver() / total_seconds) << " objects/s\n";
347  }
348  os << "Total number of allocations: " << total_objects_allocated << "\n";
349  os << "Total bytes allocated " << PrettySize(total_bytes_allocated) << "\n";
350  if (measure_allocation_time_) {
351    os << "Total time spent allocating: " << PrettyDuration(allocation_time) << "\n";
352    os << "Mean allocation time: " << PrettyDuration(allocation_time / total_objects_allocated)
353       << "\n";
354  }
355  os << "Total mutator paused time: " << PrettyDuration(total_paused_time) << "\n";
356  os << "Total time waiting for GC to complete: " << PrettyDuration(total_wait_time_) << "\n";
357  os << "Approximate GC data structures memory overhead: " << gc_memory_overhead_;
358}
359
360Heap::~Heap() {
361  if (kDumpGcPerformanceOnShutdown) {
362    DumpGcPerformanceInfo(LOG(INFO));
363  }
364
365  STLDeleteElements(&mark_sweep_collectors_);
366
367  // If we don't reset then the mark stack complains in it's destructor.
368  allocation_stack_->Reset();
369  live_stack_->Reset();
370
371  VLOG(heap) << "~Heap()";
372  // We can't take the heap lock here because there might be a daemon thread suspended with the
373  // heap lock held. We know though that no non-daemon threads are executing, and we know that
374  // all daemon threads are suspended, and we also know that the threads list have been deleted, so
375  // those threads can't resume. We're the only running thread, and we can do whatever we like...
376  STLDeleteElements(&continuous_spaces_);
377  STLDeleteElements(&discontinuous_spaces_);
378  delete gc_complete_lock_;
379  delete reference_queue_lock_;
380}
381
382space::ContinuousSpace* Heap::FindContinuousSpaceFromObject(const mirror::Object* obj,
383                                                            bool fail_ok) const {
384  // TODO: C++0x auto
385  typedef std::vector<space::ContinuousSpace*>::const_iterator It;
386  for (It it = continuous_spaces_.begin(), end = continuous_spaces_.end(); it != end; ++it) {
387    if ((*it)->Contains(obj)) {
388      return *it;
389    }
390  }
391  if (!fail_ok) {
392    LOG(FATAL) << "object " << reinterpret_cast<const void*>(obj) << " not inside any spaces!";
393  }
394  return NULL;
395}
396
397space::DiscontinuousSpace* Heap::FindDiscontinuousSpaceFromObject(const mirror::Object* obj,
398                                                                  bool fail_ok) const {
399  // TODO: C++0x auto
400  typedef std::vector<space::DiscontinuousSpace*>::const_iterator It;
401  for (It it = discontinuous_spaces_.begin(), end = discontinuous_spaces_.end(); it != end; ++it) {
402    if ((*it)->Contains(obj)) {
403      return *it;
404    }
405  }
406  if (!fail_ok) {
407    LOG(FATAL) << "object " << reinterpret_cast<const void*>(obj) << " not inside any spaces!";
408  }
409  return NULL;
410}
411
412space::Space* Heap::FindSpaceFromObject(const mirror::Object* obj, bool fail_ok) const {
413  space::Space* result = FindContinuousSpaceFromObject(obj, true);
414  if (result != NULL) {
415    return result;
416  }
417  return FindDiscontinuousSpaceFromObject(obj, true);
418}
419
420space::ImageSpace* Heap::GetImageSpace() const {
421  // TODO: C++0x auto
422  typedef std::vector<space::ContinuousSpace*>::const_iterator It;
423  for (It it = continuous_spaces_.begin(), end = continuous_spaces_.end(); it != end; ++it) {
424    if ((*it)->IsImageSpace()) {
425      return (*it)->AsImageSpace();
426    }
427  }
428  return NULL;
429}
430
431static void MSpaceChunkCallback(void* start, void* end, size_t used_bytes, void* arg) {
432  size_t chunk_size = reinterpret_cast<uint8_t*>(end) - reinterpret_cast<uint8_t*>(start);
433  if (used_bytes < chunk_size) {
434    size_t chunk_free_bytes = chunk_size - used_bytes;
435    size_t& max_contiguous_allocation = *reinterpret_cast<size_t*>(arg);
436    max_contiguous_allocation = std::max(max_contiguous_allocation, chunk_free_bytes);
437  }
438}
439
440mirror::Object* Heap::AllocObject(Thread* self, mirror::Class* c, size_t byte_count) {
441  DCHECK(c == NULL || (c->IsClassClass() && byte_count >= sizeof(mirror::Class)) ||
442         (c->IsVariableSize() || c->GetObjectSize() == byte_count) ||
443         strlen(ClassHelper(c).GetDescriptor()) == 0);
444  DCHECK_GE(byte_count, sizeof(mirror::Object));
445
446  mirror::Object* obj = NULL;
447  size_t size = 0;
448  uint64_t allocation_start = 0;
449  if (UNLIKELY(measure_allocation_time_)) {
450    allocation_start = NanoTime() / kTimeAdjust;
451  }
452
453  // We need to have a zygote space or else our newly allocated large object can end up in the
454  // Zygote resulting in it being prematurely freed.
455  // We can only do this for primive objects since large objects will not be within the card table
456  // range. This also means that we rely on SetClass not dirtying the object's card.
457  bool large_object_allocation =
458      byte_count >= large_object_threshold_ && have_zygote_space_ && c->IsPrimitiveArray();
459  if (UNLIKELY(large_object_allocation)) {
460    size = RoundUp(byte_count, kPageSize);
461    obj = Allocate(self, large_object_space_, size);
462    // Make sure that our large object didn't get placed anywhere within the space interval or else
463    // it breaks the immune range.
464    DCHECK(obj == NULL ||
465           reinterpret_cast<byte*>(obj) < continuous_spaces_.front()->Begin() ||
466           reinterpret_cast<byte*>(obj) >= continuous_spaces_.back()->End());
467  } else {
468    obj = Allocate(self, alloc_space_, byte_count);
469
470    // Ensure that we did not allocate into a zygote space.
471    DCHECK(obj == NULL || !have_zygote_space_ || !FindSpaceFromObject(obj, false)->IsZygoteSpace());
472    size = alloc_space_->AllocationSize(obj);
473  }
474
475  if (LIKELY(obj != NULL)) {
476    obj->SetClass(c);
477
478    // Record allocation after since we want to use the atomic add for the atomic fence to guard
479    // the SetClass since we do not want the class to appear NULL in another thread.
480    RecordAllocation(size, obj);
481
482    if (Dbg::IsAllocTrackingEnabled()) {
483      Dbg::RecordAllocation(c, byte_count);
484    }
485    if (static_cast<size_t>(num_bytes_allocated_) >= concurrent_start_bytes_) {
486      // The SirtRef is necessary since the calls in RequestConcurrentGC are a safepoint.
487      SirtRef<mirror::Object> ref(self, obj);
488      RequestConcurrentGC(self);
489    }
490    VerifyObject(obj);
491
492    if (UNLIKELY(measure_allocation_time_)) {
493      total_allocation_time_.fetch_add(NanoTime() / kTimeAdjust - allocation_start);
494    }
495
496    return obj;
497  } else {
498    std::ostringstream oss;
499    int64_t total_bytes_free = GetFreeMemory();
500    oss << "Failed to allocate a " << byte_count << " byte allocation with " << total_bytes_free
501        << " free bytes";
502    // If the allocation failed due to fragmentation, print out the largest continuous allocation.
503    if (!large_object_allocation && total_bytes_free >= byte_count) {
504      size_t max_contiguous_allocation = 0;
505      // TODO: C++0x auto
506      typedef std::vector<space::ContinuousSpace*>::const_iterator It;
507      for (It it = continuous_spaces_.begin(), end = continuous_spaces_.end(); it != end; ++it) {
508        space::ContinuousSpace* space = *it;
509        if (space->IsDlMallocSpace()) {
510          space->AsDlMallocSpace()->Walk(MSpaceChunkCallback, &max_contiguous_allocation);
511        }
512      }
513      oss << "; failed due to fragmentation (largest possible contiguous allocation "
514          <<  max_contiguous_allocation << " bytes)";
515    }
516    self->ThrowOutOfMemoryError(oss.str().c_str());
517    return NULL;
518  }
519}
520
521bool Heap::IsHeapAddress(const mirror::Object* obj) {
522  // Note: we deliberately don't take the lock here, and mustn't test anything that would
523  // require taking the lock.
524  if (obj == NULL) {
525    return true;
526  }
527  if (UNLIKELY(!IsAligned<kObjectAlignment>(obj))) {
528    return false;
529  }
530  return FindSpaceFromObject(obj, true) != NULL;
531}
532
533bool Heap::IsLiveObjectLocked(const mirror::Object* obj) {
534  // Locks::heap_bitmap_lock_->AssertReaderHeld(Thread::Current());
535  if (obj == NULL || UNLIKELY(!IsAligned<kObjectAlignment>(obj))) {
536    return false;
537  }
538  space::ContinuousSpace* c_space = FindContinuousSpaceFromObject(obj, true);
539  space::DiscontinuousSpace* d_space = NULL;
540  if (c_space != NULL) {
541    if (c_space->GetLiveBitmap()->Test(obj)) {
542      return true;
543    }
544  } else {
545    d_space = FindDiscontinuousSpaceFromObject(obj, true);
546    if (d_space != NULL) {
547      if (d_space->GetLiveObjects()->Test(obj)) {
548        return true;
549      }
550    }
551  }
552  // This is covering the allocation/live stack swapping that is done without mutators suspended.
553  for (size_t i = 0; i < 5; ++i) {
554    if (allocation_stack_->Contains(const_cast<mirror::Object*>(obj)) ||
555        live_stack_->Contains(const_cast<mirror::Object*>(obj))) {
556      return true;
557    }
558    NanoSleep(MsToNs(10));
559  }
560  // We need to check the bitmaps again since there is a race where we mark something as live and
561  // then clear the stack containing it.
562  if (c_space != NULL) {
563    if (c_space->GetLiveBitmap()->Test(obj)) {
564      return true;
565    }
566  } else {
567    d_space = FindDiscontinuousSpaceFromObject(obj, true);
568    if (d_space != NULL && d_space->GetLiveObjects()->Test(obj)) {
569      return true;
570    }
571  }
572  return false;
573}
574
575void Heap::VerifyObjectImpl(const mirror::Object* obj) {
576  if (Thread::Current() == NULL ||
577      Runtime::Current()->GetThreadList()->GetLockOwner() == Thread::Current()->GetTid()) {
578    return;
579  }
580  VerifyObjectBody(obj);
581}
582
583void Heap::DumpSpaces() {
584  // TODO: C++0x auto
585  typedef std::vector<space::ContinuousSpace*>::const_iterator It;
586  for (It it = continuous_spaces_.begin(), end = continuous_spaces_.end(); it != end; ++it) {
587    space::ContinuousSpace* space = *it;
588    accounting::SpaceBitmap* live_bitmap = space->GetLiveBitmap();
589    accounting::SpaceBitmap* mark_bitmap = space->GetMarkBitmap();
590    LOG(INFO) << space << " " << *space << "\n"
591              << live_bitmap << " " << *live_bitmap << "\n"
592              << mark_bitmap << " " << *mark_bitmap;
593  }
594  typedef std::vector<space::DiscontinuousSpace*>::const_iterator It2;
595  for (It2 it = discontinuous_spaces_.begin(), end = discontinuous_spaces_.end(); it != end; ++it) {
596    space::DiscontinuousSpace* space = *it;
597    LOG(INFO) << space << " " << *space << "\n";
598  }
599}
600
601void Heap::VerifyObjectBody(const mirror::Object* obj) {
602  if (UNLIKELY(!IsAligned<kObjectAlignment>(obj))) {
603    LOG(FATAL) << "Object isn't aligned: " << obj;
604  }
605  if (UNLIKELY(GetObjectsAllocated() <= 10)) {  // Ignore early dawn of the universe verifications.
606    return;
607  }
608  const byte* raw_addr = reinterpret_cast<const byte*>(obj) +
609      mirror::Object::ClassOffset().Int32Value();
610  const mirror::Class* c = *reinterpret_cast<mirror::Class* const *>(raw_addr);
611  if (UNLIKELY(c == NULL)) {
612    LOG(FATAL) << "Null class in object: " << obj;
613  } else if (UNLIKELY(!IsAligned<kObjectAlignment>(c))) {
614    LOG(FATAL) << "Class isn't aligned: " << c << " in object: " << obj;
615  }
616  // Check obj.getClass().getClass() == obj.getClass().getClass().getClass()
617  // Note: we don't use the accessors here as they have internal sanity checks
618  // that we don't want to run
619  raw_addr = reinterpret_cast<const byte*>(c) + mirror::Object::ClassOffset().Int32Value();
620  const mirror::Class* c_c = *reinterpret_cast<mirror::Class* const *>(raw_addr);
621  raw_addr = reinterpret_cast<const byte*>(c_c) + mirror::Object::ClassOffset().Int32Value();
622  const mirror::Class* c_c_c = *reinterpret_cast<mirror::Class* const *>(raw_addr);
623  CHECK_EQ(c_c, c_c_c);
624
625  if (verify_object_mode_ != kVerifyAllFast) {
626    // TODO: the bitmap tests below are racy if VerifyObjectBody is called without the
627    //       heap_bitmap_lock_.
628    if (!IsLiveObjectLocked(obj)) {
629      DumpSpaces();
630      LOG(FATAL) << "Object is dead: " << obj;
631    }
632    if (!IsLiveObjectLocked(c)) {
633      LOG(FATAL) << "Class of object is dead: " << c << " in object: " << obj;
634    }
635  }
636}
637
638void Heap::VerificationCallback(mirror::Object* obj, void* arg) {
639  DCHECK(obj != NULL);
640  reinterpret_cast<Heap*>(arg)->VerifyObjectBody(obj);
641}
642
643void Heap::VerifyHeap() {
644  ReaderMutexLock mu(Thread::Current(), *Locks::heap_bitmap_lock_);
645  GetLiveBitmap()->Walk(Heap::VerificationCallback, this);
646}
647
648void Heap::RecordAllocation(size_t size, mirror::Object* obj) {
649  DCHECK(obj != NULL);
650  DCHECK_GT(size, 0u);
651  num_bytes_allocated_.fetch_add(size);
652
653  if (Runtime::Current()->HasStatsEnabled()) {
654    RuntimeStats* thread_stats = Thread::Current()->GetStats();
655    ++thread_stats->allocated_objects;
656    thread_stats->allocated_bytes += size;
657
658    // TODO: Update these atomically.
659    RuntimeStats* global_stats = Runtime::Current()->GetStats();
660    ++global_stats->allocated_objects;
661    global_stats->allocated_bytes += size;
662  }
663
664  // This is safe to do since the GC will never free objects which are neither in the allocation
665  // stack or the live bitmap.
666  while (!allocation_stack_->AtomicPushBack(obj)) {
667    CollectGarbageInternal(collector::kGcTypeSticky, kGcCauseForAlloc, false);
668  }
669}
670
671void Heap::RecordFree(size_t freed_objects, size_t freed_bytes) {
672  DCHECK_LE(freed_bytes, static_cast<size_t>(num_bytes_allocated_));
673  num_bytes_allocated_.fetch_sub(freed_bytes);
674
675  if (Runtime::Current()->HasStatsEnabled()) {
676    RuntimeStats* thread_stats = Thread::Current()->GetStats();
677    thread_stats->freed_objects += freed_objects;
678    thread_stats->freed_bytes += freed_bytes;
679
680    // TODO: Do this concurrently.
681    RuntimeStats* global_stats = Runtime::Current()->GetStats();
682    global_stats->freed_objects += freed_objects;
683    global_stats->freed_bytes += freed_bytes;
684  }
685}
686
687mirror::Object* Heap::TryToAllocate(Thread* self, space::AllocSpace* space, size_t alloc_size,
688                                    bool grow) {
689  // Should we try to use a CAS here and fix up num_bytes_allocated_ later with AllocationSize?
690  if (num_bytes_allocated_ + alloc_size > max_allowed_footprint_) {
691    // max_allowed_footprint_ <= growth_limit_ so it is safe to check in here.
692    if (num_bytes_allocated_ + alloc_size > growth_limit_) {
693      // Completely out of memory.
694      return NULL;
695    }
696  }
697
698  return space->Alloc(self, alloc_size);
699}
700
701mirror::Object* Heap::Allocate(Thread* self, space::AllocSpace* space, size_t alloc_size) {
702  // Since allocation can cause a GC which will need to SuspendAll, make sure all allocations are
703  // done in the runnable state where suspension is expected.
704  DCHECK_EQ(self->GetState(), kRunnable);
705  self->AssertThreadSuspensionIsAllowable();
706
707  mirror::Object* ptr = TryToAllocate(self, space, alloc_size, false);
708  if (ptr != NULL) {
709    return ptr;
710  }
711
712  // The allocation failed. If the GC is running, block until it completes, and then retry the
713  // allocation.
714  collector::GcType last_gc = WaitForConcurrentGcToComplete(self);
715  if (last_gc != collector::kGcTypeNone) {
716    // A GC was in progress and we blocked, retry allocation now that memory has been freed.
717    ptr = TryToAllocate(self, space, alloc_size, false);
718    if (ptr != NULL) {
719      return ptr;
720    }
721  }
722
723  // Loop through our different Gc types and try to Gc until we get enough free memory.
724  for (size_t i = static_cast<size_t>(last_gc) + 1;
725      i < static_cast<size_t>(collector::kGcTypeMax); ++i) {
726    bool run_gc = false;
727    collector::GcType gc_type = static_cast<collector::GcType>(i);
728    switch (gc_type) {
729      case collector::kGcTypeSticky: {
730          const size_t alloc_space_size = alloc_space_->Size();
731          run_gc = alloc_space_size > min_alloc_space_size_for_sticky_gc_ &&
732              alloc_space_->Capacity() - alloc_space_size >= min_remaining_space_for_sticky_gc_;
733          break;
734        }
735      case collector::kGcTypePartial:
736        run_gc = have_zygote_space_;
737        break;
738      case collector::kGcTypeFull:
739        run_gc = true;
740        break;
741      default:
742        break;
743    }
744
745    if (run_gc) {
746      // If we actually ran a different type of Gc than requested, we can skip the index forwards.
747      collector::GcType gc_type_ran = CollectGarbageInternal(gc_type, kGcCauseForAlloc, false);
748      DCHECK_GE(static_cast<size_t>(gc_type_ran), i);
749      i = static_cast<size_t>(gc_type_ran);
750
751      // Did we free sufficient memory for the allocation to succeed?
752      ptr = TryToAllocate(self, space, alloc_size, false);
753      if (ptr != NULL) {
754        return ptr;
755      }
756    }
757  }
758
759  // Allocations have failed after GCs;  this is an exceptional state.
760  // Try harder, growing the heap if necessary.
761  ptr = TryToAllocate(self, space, alloc_size, true);
762  if (ptr != NULL) {
763    return ptr;
764  }
765
766  // Most allocations should have succeeded by now, so the heap is really full, really fragmented,
767  // or the requested size is really big. Do another GC, collecting SoftReferences this time. The
768  // VM spec requires that all SoftReferences have been collected and cleared before throwing OOME.
769
770  // OLD-TODO: wait for the finalizers from the previous GC to finish
771  VLOG(gc) << "Forcing collection of SoftReferences for " << PrettySize(alloc_size)
772           << " allocation";
773
774  // We don't need a WaitForConcurrentGcToComplete here either.
775  CollectGarbageInternal(collector::kGcTypeFull, kGcCauseForAlloc, true);
776  return TryToAllocate(self, space, alloc_size, true);
777}
778
779void Heap::SetTargetHeapUtilization(float target) {
780  DCHECK_GT(target, 0.0f);  // asserted in Java code
781  DCHECK_LT(target, 1.0f);
782  target_utilization_ = target;
783}
784
785size_t Heap::GetObjectsAllocated() const {
786  size_t total = 0;
787  typedef std::vector<space::ContinuousSpace*>::const_iterator It;
788  for (It it = continuous_spaces_.begin(), end = continuous_spaces_.end(); it != end; ++it) {
789    space::ContinuousSpace* space = *it;
790    if (space->IsDlMallocSpace()) {
791      total += space->AsDlMallocSpace()->GetObjectsAllocated();
792    }
793  }
794  typedef std::vector<space::DiscontinuousSpace*>::const_iterator It2;
795  for (It2 it = discontinuous_spaces_.begin(), end = discontinuous_spaces_.end(); it != end; ++it) {
796    space::DiscontinuousSpace* space = *it;
797    total += space->AsLargeObjectSpace()->GetObjectsAllocated();
798  }
799  return total;
800}
801
802size_t Heap::GetObjectsAllocatedEver() const {
803  size_t total = 0;
804  typedef std::vector<space::ContinuousSpace*>::const_iterator It;
805  for (It it = continuous_spaces_.begin(), end = continuous_spaces_.end(); it != end; ++it) {
806    space::ContinuousSpace* space = *it;
807    if (space->IsDlMallocSpace()) {
808      total += space->AsDlMallocSpace()->GetTotalObjectsAllocated();
809    }
810  }
811  typedef std::vector<space::DiscontinuousSpace*>::const_iterator It2;
812  for (It2 it = discontinuous_spaces_.begin(), end = discontinuous_spaces_.end(); it != end; ++it) {
813    space::DiscontinuousSpace* space = *it;
814    total += space->AsLargeObjectSpace()->GetTotalObjectsAllocated();
815  }
816  return total;
817}
818
819size_t Heap::GetBytesAllocatedEver() const {
820  size_t total = 0;
821  typedef std::vector<space::ContinuousSpace*>::const_iterator It;
822  for (It it = continuous_spaces_.begin(), end = continuous_spaces_.end(); it != end; ++it) {
823    space::ContinuousSpace* space = *it;
824    if (space->IsDlMallocSpace()) {
825      total += space->AsDlMallocSpace()->GetTotalBytesAllocated();
826    }
827  }
828  typedef std::vector<space::DiscontinuousSpace*>::const_iterator It2;
829  for (It2 it = discontinuous_spaces_.begin(), end = discontinuous_spaces_.end(); it != end; ++it) {
830    space::DiscontinuousSpace* space = *it;
831    total += space->AsLargeObjectSpace()->GetTotalBytesAllocated();
832  }
833  return total;
834}
835
836class InstanceCounter {
837 public:
838  InstanceCounter(const std::vector<mirror::Class*>& classes, bool use_is_assignable_from, uint64_t* counts)
839      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_)
840      : classes_(classes), use_is_assignable_from_(use_is_assignable_from), counts_(counts) {
841  }
842
843  void operator()(const mirror::Object* o) const SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
844    for (size_t i = 0; i < classes_.size(); ++i) {
845      const mirror::Class* instance_class = o->GetClass();
846      if (use_is_assignable_from_) {
847        if (instance_class != NULL && classes_[i]->IsAssignableFrom(instance_class)) {
848          ++counts_[i];
849        }
850      } else {
851        if (instance_class == classes_[i]) {
852          ++counts_[i];
853        }
854      }
855    }
856  }
857
858 private:
859  const std::vector<mirror::Class*>& classes_;
860  bool use_is_assignable_from_;
861  uint64_t* const counts_;
862
863  DISALLOW_COPY_AND_ASSIGN(InstanceCounter);
864};
865
866void Heap::CountInstances(const std::vector<mirror::Class*>& classes, bool use_is_assignable_from,
867                          uint64_t* counts) {
868  // We only want reachable instances, so do a GC. This also ensures that the alloc stack
869  // is empty, so the live bitmap is the only place we need to look.
870  Thread* self = Thread::Current();
871  self->TransitionFromRunnableToSuspended(kNative);
872  CollectGarbage(false);
873  self->TransitionFromSuspendedToRunnable();
874
875  InstanceCounter counter(classes, use_is_assignable_from, counts);
876  ReaderMutexLock mu(self, *Locks::heap_bitmap_lock_);
877  GetLiveBitmap()->Visit(counter);
878}
879
880class InstanceCollector {
881 public:
882  InstanceCollector(mirror::Class* c, int32_t max_count, std::vector<mirror::Object*>& instances)
883      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_)
884      : class_(c), max_count_(max_count), instances_(instances) {
885  }
886
887  void operator()(const mirror::Object* o) const SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
888    const mirror::Class* instance_class = o->GetClass();
889    if (instance_class == class_) {
890      if (max_count_ == 0 || instances_.size() < max_count_) {
891        instances_.push_back(const_cast<mirror::Object*>(o));
892      }
893    }
894  }
895
896 private:
897  mirror::Class* class_;
898  uint32_t max_count_;
899  std::vector<mirror::Object*>& instances_;
900
901  DISALLOW_COPY_AND_ASSIGN(InstanceCollector);
902};
903
904void Heap::GetInstances(mirror::Class* c, int32_t max_count,
905                        std::vector<mirror::Object*>& instances) {
906  // We only want reachable instances, so do a GC. This also ensures that the alloc stack
907  // is empty, so the live bitmap is the only place we need to look.
908  Thread* self = Thread::Current();
909  self->TransitionFromRunnableToSuspended(kNative);
910  CollectGarbage(false);
911  self->TransitionFromSuspendedToRunnable();
912
913  InstanceCollector collector(c, max_count, instances);
914  ReaderMutexLock mu(self, *Locks::heap_bitmap_lock_);
915  GetLiveBitmap()->Visit(collector);
916}
917
918class ReferringObjectsFinder {
919 public:
920  ReferringObjectsFinder(mirror::Object* object, int32_t max_count,
921                         std::vector<mirror::Object*>& referring_objects)
922      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_)
923      : object_(object), max_count_(max_count), referring_objects_(referring_objects) {
924  }
925
926  // For bitmap Visit.
927  // TODO: Fix lock analysis to not use NO_THREAD_SAFETY_ANALYSIS, requires support for
928  // annotalysis on visitors.
929  void operator()(const mirror::Object* o) const NO_THREAD_SAFETY_ANALYSIS {
930    collector::MarkSweep::VisitObjectReferences(o, *this);
931  }
932
933  // For MarkSweep::VisitObjectReferences.
934  void operator()(const mirror::Object* referrer, const mirror::Object* object,
935                  const MemberOffset&, bool) const {
936    if (object == object_ && (max_count_ == 0 || referring_objects_.size() < max_count_)) {
937      referring_objects_.push_back(const_cast<mirror::Object*>(referrer));
938    }
939  }
940
941 private:
942  mirror::Object* object_;
943  uint32_t max_count_;
944  std::vector<mirror::Object*>& referring_objects_;
945
946  DISALLOW_COPY_AND_ASSIGN(ReferringObjectsFinder);
947};
948
949void Heap::GetReferringObjects(mirror::Object* o, int32_t max_count,
950                               std::vector<mirror::Object*>& referring_objects) {
951  // We only want reachable instances, so do a GC. This also ensures that the alloc stack
952  // is empty, so the live bitmap is the only place we need to look.
953  Thread* self = Thread::Current();
954  self->TransitionFromRunnableToSuspended(kNative);
955  CollectGarbage(false);
956  self->TransitionFromSuspendedToRunnable();
957
958  ReferringObjectsFinder finder(o, max_count, referring_objects);
959  ReaderMutexLock mu(self, *Locks::heap_bitmap_lock_);
960  GetLiveBitmap()->Visit(finder);
961}
962
963void Heap::CollectGarbage(bool clear_soft_references) {
964  // Even if we waited for a GC we still need to do another GC since weaks allocated during the
965  // last GC will not have necessarily been cleared.
966  Thread* self = Thread::Current();
967  WaitForConcurrentGcToComplete(self);
968  CollectGarbageInternal(collector::kGcTypeFull, kGcCauseExplicit, clear_soft_references);
969}
970
971void Heap::PreZygoteFork() {
972  static Mutex zygote_creation_lock_("zygote creation lock", kZygoteCreationLock);
973  // Do this before acquiring the zygote creation lock so that we don't get lock order violations.
974  CollectGarbage(false);
975  Thread* self = Thread::Current();
976  MutexLock mu(self, zygote_creation_lock_);
977
978  // Try to see if we have any Zygote spaces.
979  if (have_zygote_space_) {
980    return;
981  }
982
983  VLOG(heap) << "Starting PreZygoteFork with alloc space size " << PrettySize(alloc_space_->Size());
984
985  {
986    // Flush the alloc stack.
987    WriterMutexLock mu(self, *Locks::heap_bitmap_lock_);
988    FlushAllocStack();
989  }
990
991  // Turns the current alloc space into a Zygote space and obtain the new alloc space composed
992  // of the remaining available heap memory.
993  space::DlMallocSpace* zygote_space = alloc_space_;
994  alloc_space_ = zygote_space->CreateZygoteSpace("alloc space");
995  alloc_space_->SetFootprintLimit(alloc_space_->Capacity());
996
997  // Change the GC retention policy of the zygote space to only collect when full.
998  zygote_space->SetGcRetentionPolicy(space::kGcRetentionPolicyFullCollect);
999  AddContinuousSpace(alloc_space_);
1000  have_zygote_space_ = true;
1001
1002  // Reset the cumulative loggers since we now have a few additional timing phases.
1003  // TODO: C++0x
1004  typedef std::vector<collector::MarkSweep*>::const_iterator It;
1005  for (It it = mark_sweep_collectors_.begin(), end = mark_sweep_collectors_.end();
1006      it != end; ++it) {
1007    (*it)->ResetCumulativeStatistics();
1008  }
1009}
1010
1011void Heap::FlushAllocStack() {
1012  MarkAllocStack(alloc_space_->GetLiveBitmap(), large_object_space_->GetLiveObjects(),
1013                 allocation_stack_.get());
1014  allocation_stack_->Reset();
1015}
1016
1017void Heap::MarkAllocStack(accounting::SpaceBitmap* bitmap, accounting::SpaceSetMap* large_objects,
1018                          accounting::ObjectStack* stack) {
1019  mirror::Object** limit = stack->End();
1020  for (mirror::Object** it = stack->Begin(); it != limit; ++it) {
1021    const mirror::Object* obj = *it;
1022    DCHECK(obj != NULL);
1023    if (LIKELY(bitmap->HasAddress(obj))) {
1024      bitmap->Set(obj);
1025    } else {
1026      large_objects->Set(obj);
1027    }
1028  }
1029}
1030
1031void Heap::UnMarkAllocStack(accounting::SpaceBitmap* bitmap, accounting::SpaceSetMap* large_objects,
1032                            accounting::ObjectStack* stack) {
1033  mirror::Object** limit = stack->End();
1034  for (mirror::Object** it = stack->Begin(); it != limit; ++it) {
1035    const mirror::Object* obj = *it;
1036    DCHECK(obj != NULL);
1037    if (LIKELY(bitmap->HasAddress(obj))) {
1038      bitmap->Clear(obj);
1039    } else {
1040      large_objects->Clear(obj);
1041    }
1042  }
1043}
1044
1045collector::GcType Heap::CollectGarbageInternal(collector::GcType gc_type, GcCause gc_cause,
1046                                               bool clear_soft_references) {
1047  Thread* self = Thread::Current();
1048
1049  switch (gc_cause) {
1050    case kGcCauseForAlloc:
1051      ATRACE_BEGIN("GC (alloc)");
1052      break;
1053    case kGcCauseBackground:
1054      ATRACE_BEGIN("GC (background)");
1055      break;
1056    case kGcCauseExplicit:
1057      ATRACE_BEGIN("GC (explicit)");
1058      break;
1059  }
1060
1061  ScopedThreadStateChange tsc(self, kWaitingPerformingGc);
1062  Locks::mutator_lock_->AssertNotHeld(self);
1063
1064  if (self->IsHandlingStackOverflow()) {
1065    LOG(WARNING) << "Performing GC on a thread that is handling a stack overflow.";
1066  }
1067
1068  // Ensure there is only one GC at a time.
1069  bool start_collect = false;
1070  while (!start_collect) {
1071    {
1072      MutexLock mu(self, *gc_complete_lock_);
1073      if (!is_gc_running_) {
1074        is_gc_running_ = true;
1075        start_collect = true;
1076      }
1077    }
1078    if (!start_collect) {
1079      WaitForConcurrentGcToComplete(self);
1080      // TODO: if another thread beat this one to do the GC, perhaps we should just return here?
1081      //       Not doing at the moment to ensure soft references are cleared.
1082    }
1083  }
1084  gc_complete_lock_->AssertNotHeld(self);
1085
1086  if (gc_cause == kGcCauseForAlloc && Runtime::Current()->HasStatsEnabled()) {
1087    ++Runtime::Current()->GetStats()->gc_for_alloc_count;
1088    ++Thread::Current()->GetStats()->gc_for_alloc_count;
1089  }
1090
1091  uint64_t gc_start_time_ns = NanoTime();
1092  uint64_t gc_start_size = GetBytesAllocated();
1093  // Approximate allocation rate in bytes / second.
1094  if (UNLIKELY(gc_start_time_ns == last_gc_time_ns_)) {
1095    LOG(WARNING) << "Timers are broken (gc_start_time == last_gc_time_).";
1096  }
1097  uint64_t ms_delta = NsToMs(gc_start_time_ns - last_gc_time_ns_);
1098  if (ms_delta != 0) {
1099    allocation_rate_ = ((gc_start_size - last_gc_size_) * 1000) / ms_delta;
1100    VLOG(heap) << "Allocation rate: " << PrettySize(allocation_rate_) << "/s";
1101  }
1102
1103  if (gc_type == collector::kGcTypeSticky &&
1104      alloc_space_->Size() < min_alloc_space_size_for_sticky_gc_) {
1105    gc_type = collector::kGcTypePartial;
1106  }
1107
1108  DCHECK_LT(gc_type, collector::kGcTypeMax);
1109  DCHECK_NE(gc_type, collector::kGcTypeNone);
1110  collector::MarkSweep* collector = NULL;
1111  typedef std::vector<collector::MarkSweep*>::iterator It;
1112  for (It it = mark_sweep_collectors_.begin(), end = mark_sweep_collectors_.end();
1113      it != end; ++it) {
1114    collector::MarkSweep* cur_collector = *it;
1115    if (cur_collector->IsConcurrent() == concurrent_gc_ && cur_collector->GetGcType() == gc_type) {
1116      collector = cur_collector;
1117      break;
1118    }
1119  }
1120  CHECK(collector != NULL)
1121      << "Could not find garbage collector with concurrent=" << concurrent_gc_
1122      << " and type=" << gc_type;
1123  collector->clear_soft_references_ = clear_soft_references;
1124  collector->Run();
1125  total_objects_freed_ever_ += collector->GetFreedObjects();
1126  total_bytes_freed_ever_ += collector->GetFreedBytes();
1127
1128  const size_t duration = collector->GetDurationNs();
1129  std::vector<uint64_t> pauses = collector->GetPauseTimes();
1130  bool was_slow = duration > kSlowGcThreshold ||
1131      (gc_cause == kGcCauseForAlloc && duration > kLongGcPauseThreshold);
1132  for (size_t i = 0; i < pauses.size(); ++i) {
1133    if (pauses[i] > kLongGcPauseThreshold) {
1134      was_slow = true;
1135    }
1136  }
1137
1138  if (was_slow) {
1139    const size_t percent_free = GetPercentFree();
1140    const size_t current_heap_size = GetBytesAllocated();
1141    const size_t total_memory = GetTotalMemory();
1142    std::ostringstream pause_string;
1143    for (size_t i = 0; i < pauses.size(); ++i) {
1144      pause_string << PrettyDuration((pauses[i] / 1000) * 1000)
1145                   << ((i != pauses.size() - 1) ? ", " : "");
1146    }
1147    LOG(INFO) << gc_cause << " " << collector->GetName()
1148              << "GC freed " << PrettySize(collector->GetFreedBytes()) << ", "
1149              << percent_free << "% free, " << PrettySize(current_heap_size) << "/"
1150              << PrettySize(total_memory) << ", " << "paused " << pause_string.str()
1151              << " total " << PrettyDuration((duration / 1000) * 1000);
1152    if (VLOG_IS_ON(heap)) {
1153      LOG(INFO) << Dumpable<base::TimingLogger>(collector->GetTimings());
1154    }
1155  }
1156
1157  {
1158    MutexLock mu(self, *gc_complete_lock_);
1159    is_gc_running_ = false;
1160    last_gc_type_ = gc_type;
1161    // Wake anyone who may have been waiting for the GC to complete.
1162    gc_complete_cond_->Broadcast(self);
1163  }
1164
1165  // Inform DDMS that a GC completed.
1166  ATRACE_END();
1167  Dbg::GcDidFinish();
1168  return gc_type;
1169}
1170
1171void Heap::UpdateAndMarkModUnion(collector::MarkSweep* mark_sweep, base::TimingLogger& timings,
1172                                 collector::GcType gc_type) {
1173  if (gc_type == collector::kGcTypeSticky) {
1174    // Don't need to do anything for mod union table in this case since we are only scanning dirty
1175    // cards.
1176    return;
1177  }
1178
1179  // Update zygote mod union table.
1180  if (gc_type == collector::kGcTypePartial) {
1181    timings.NewSplit("UpdateZygoteModUnionTable");
1182    zygote_mod_union_table_->Update();
1183
1184    timings.NewSplit("ZygoteMarkReferences");
1185    zygote_mod_union_table_->MarkReferences(mark_sweep);
1186  }
1187
1188  // Processes the cards we cleared earlier and adds their objects into the mod-union table.
1189  timings.NewSplit("UpdateModUnionTable");
1190  image_mod_union_table_->Update();
1191
1192  // Scans all objects in the mod-union table.
1193  timings.NewSplit("MarkImageToAllocSpaceReferences");
1194  image_mod_union_table_->MarkReferences(mark_sweep);
1195}
1196
1197static void RootMatchesObjectVisitor(const mirror::Object* root, void* arg) {
1198  mirror::Object* obj = reinterpret_cast<mirror::Object*>(arg);
1199  if (root == obj) {
1200    LOG(INFO) << "Object " << obj << " is a root";
1201  }
1202}
1203
1204class ScanVisitor {
1205 public:
1206  void operator()(const mirror::Object* obj) const {
1207    LOG(INFO) << "Would have rescanned object " << obj;
1208  }
1209};
1210
1211// Verify a reference from an object.
1212class VerifyReferenceVisitor {
1213 public:
1214  explicit VerifyReferenceVisitor(Heap* heap)
1215      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_, Locks::heap_bitmap_lock_)
1216      : heap_(heap), failed_(false) {}
1217
1218  bool Failed() const {
1219    return failed_;
1220  }
1221
1222  // TODO: Fix lock analysis to not use NO_THREAD_SAFETY_ANALYSIS, requires support for smarter
1223  // analysis on visitors.
1224  void operator()(const mirror::Object* obj, const mirror::Object* ref,
1225                  const MemberOffset& offset, bool /* is_static */) const
1226      NO_THREAD_SAFETY_ANALYSIS {
1227    // Verify that the reference is live.
1228    if (UNLIKELY(ref != NULL && !IsLive(ref))) {
1229      accounting::CardTable* card_table = heap_->GetCardTable();
1230      accounting::ObjectStack* alloc_stack = heap_->allocation_stack_.get();
1231      accounting::ObjectStack* live_stack = heap_->live_stack_.get();
1232
1233      if (obj != NULL) {
1234        byte* card_addr = card_table->CardFromAddr(obj);
1235        LOG(ERROR) << "Object " << obj << " references dead object " << ref << " at offset " << offset
1236                   << "\nIsDirty = " << (*card_addr == accounting::CardTable::kCardDirty)
1237                   << "\nObj type " << PrettyTypeOf(obj)
1238                   << "\nRef type " << PrettyTypeOf(ref);
1239        card_table->CheckAddrIsInCardTable(reinterpret_cast<const byte*>(obj));
1240        void* cover_begin = card_table->AddrFromCard(card_addr);
1241        void* cover_end = reinterpret_cast<void*>(reinterpret_cast<size_t>(cover_begin) +
1242            accounting::CardTable::kCardSize);
1243        LOG(ERROR) << "Card " << reinterpret_cast<void*>(card_addr) << " covers " << cover_begin
1244            << "-" << cover_end;
1245        accounting::SpaceBitmap* bitmap = heap_->GetLiveBitmap()->GetContinuousSpaceBitmap(obj);
1246
1247        // Print out how the object is live.
1248        if (bitmap != NULL && bitmap->Test(obj)) {
1249          LOG(ERROR) << "Object " << obj << " found in live bitmap";
1250        }
1251        if (alloc_stack->ContainsSorted(const_cast<mirror::Object*>(obj))) {
1252          LOG(ERROR) << "Object " << obj << " found in allocation stack";
1253        }
1254        if (live_stack->ContainsSorted(const_cast<mirror::Object*>(obj))) {
1255          LOG(ERROR) << "Object " << obj << " found in live stack";
1256        }
1257        // Attempt to see if the card table missed the reference.
1258        ScanVisitor scan_visitor;
1259        byte* byte_cover_begin = reinterpret_cast<byte*>(card_table->AddrFromCard(card_addr));
1260        card_table->Scan(bitmap, byte_cover_begin,
1261                         byte_cover_begin + accounting::CardTable::kCardSize,
1262                         scan_visitor, VoidFunctor());
1263
1264        // Search to see if any of the roots reference our object.
1265        void* arg = const_cast<void*>(reinterpret_cast<const void*>(obj));
1266        Runtime::Current()->VisitRoots(&RootMatchesObjectVisitor, arg, false, false);
1267
1268        // Search to see if any of the roots reference our reference.
1269        arg = const_cast<void*>(reinterpret_cast<const void*>(ref));
1270        Runtime::Current()->VisitRoots(&RootMatchesObjectVisitor, arg, false, false);
1271      } else {
1272        LOG(ERROR) << "Root references dead object " << ref << "\nRef type " << PrettyTypeOf(ref);
1273      }
1274      if (alloc_stack->ContainsSorted(const_cast<mirror::Object*>(ref))) {
1275        LOG(ERROR) << "Reference " << ref << " found in allocation stack!";
1276      }
1277      if (live_stack->ContainsSorted(const_cast<mirror::Object*>(ref))) {
1278        LOG(ERROR) << "Reference " << ref << " found in live stack!";
1279      }
1280      heap_->image_mod_union_table_->Dump(LOG(ERROR) << "Image mod-union table: ");
1281      heap_->zygote_mod_union_table_->Dump(LOG(ERROR) << "Zygote mod-union table: ");
1282      failed_ = true;
1283    }
1284  }
1285
1286  bool IsLive(const mirror::Object* obj) const NO_THREAD_SAFETY_ANALYSIS {
1287    return heap_->IsLiveObjectLocked(obj);
1288  }
1289
1290  static void VerifyRoots(const mirror::Object* root, void* arg) {
1291    VerifyReferenceVisitor* visitor = reinterpret_cast<VerifyReferenceVisitor*>(arg);
1292    (*visitor)(NULL, root, MemberOffset(0), true);
1293  }
1294
1295 private:
1296  Heap* const heap_;
1297  mutable bool failed_;
1298};
1299
1300// Verify all references within an object, for use with HeapBitmap::Visit.
1301class VerifyObjectVisitor {
1302 public:
1303  explicit VerifyObjectVisitor(Heap* heap) : heap_(heap), failed_(false) {}
1304
1305  void operator()(const mirror::Object* obj) const
1306      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_, Locks::heap_bitmap_lock_) {
1307    // Note: we are verifying the references in obj but not obj itself, this is because obj must
1308    // be live or else how did we find it in the live bitmap?
1309    VerifyReferenceVisitor visitor(heap_);
1310    collector::MarkSweep::VisitObjectReferences(obj, visitor);
1311    failed_ = failed_ || visitor.Failed();
1312  }
1313
1314  bool Failed() const {
1315    return failed_;
1316  }
1317
1318 private:
1319  Heap* const heap_;
1320  mutable bool failed_;
1321};
1322
1323// Must do this with mutators suspended since we are directly accessing the allocation stacks.
1324bool Heap::VerifyHeapReferences() {
1325  Locks::mutator_lock_->AssertExclusiveHeld(Thread::Current());
1326  // Lets sort our allocation stacks so that we can efficiently binary search them.
1327  allocation_stack_->Sort();
1328  live_stack_->Sort();
1329  // Perform the verification.
1330  VerifyObjectVisitor visitor(this);
1331  Runtime::Current()->VisitRoots(VerifyReferenceVisitor::VerifyRoots, &visitor, false, false);
1332  GetLiveBitmap()->Visit(visitor);
1333  // We don't want to verify the objects in the allocation stack since they themselves may be
1334  // pointing to dead objects if they are not reachable.
1335  if (visitor.Failed()) {
1336    DumpSpaces();
1337    return false;
1338  }
1339  return true;
1340}
1341
1342class VerifyReferenceCardVisitor {
1343 public:
1344  VerifyReferenceCardVisitor(Heap* heap, bool* failed)
1345      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_,
1346                            Locks::heap_bitmap_lock_)
1347      : heap_(heap), failed_(failed) {
1348  }
1349
1350  // TODO: Fix lock analysis to not use NO_THREAD_SAFETY_ANALYSIS, requires support for
1351  // annotalysis on visitors.
1352  void operator()(const mirror::Object* obj, const mirror::Object* ref, const MemberOffset& offset,
1353                  bool is_static) const NO_THREAD_SAFETY_ANALYSIS {
1354    // Filter out class references since changing an object's class does not mark the card as dirty.
1355    // Also handles large objects, since the only reference they hold is a class reference.
1356    if (ref != NULL && !ref->IsClass()) {
1357      accounting::CardTable* card_table = heap_->GetCardTable();
1358      // If the object is not dirty and it is referencing something in the live stack other than
1359      // class, then it must be on a dirty card.
1360      if (!card_table->AddrIsInCardTable(obj)) {
1361        LOG(ERROR) << "Object " << obj << " is not in the address range of the card table";
1362        *failed_ = true;
1363      } else if (!card_table->IsDirty(obj)) {
1364        // Card should be either kCardDirty if it got re-dirtied after we aged it, or
1365        // kCardDirty - 1 if it didnt get touched since we aged it.
1366        accounting::ObjectStack* live_stack = heap_->live_stack_.get();
1367        if (live_stack->ContainsSorted(const_cast<mirror::Object*>(ref))) {
1368          if (live_stack->ContainsSorted(const_cast<mirror::Object*>(obj))) {
1369            LOG(ERROR) << "Object " << obj << " found in live stack";
1370          }
1371          if (heap_->GetLiveBitmap()->Test(obj)) {
1372            LOG(ERROR) << "Object " << obj << " found in live bitmap";
1373          }
1374          LOG(ERROR) << "Object " << obj << " " << PrettyTypeOf(obj)
1375                    << " references " << ref << " " << PrettyTypeOf(ref) << " in live stack";
1376
1377          // Print which field of the object is dead.
1378          if (!obj->IsObjectArray()) {
1379            const mirror::Class* klass = is_static ? obj->AsClass() : obj->GetClass();
1380            CHECK(klass != NULL);
1381            const mirror::ObjectArray<mirror::Field>* fields = is_static ? klass->GetSFields()
1382                                                                         : klass->GetIFields();
1383            CHECK(fields != NULL);
1384            for (int32_t i = 0; i < fields->GetLength(); ++i) {
1385              const mirror::Field* cur = fields->Get(i);
1386              if (cur->GetOffset().Int32Value() == offset.Int32Value()) {
1387                LOG(ERROR) << (is_static ? "Static " : "") << "field in the live stack is "
1388                          << PrettyField(cur);
1389                break;
1390              }
1391            }
1392          } else {
1393            const mirror::ObjectArray<mirror::Object>* object_array =
1394                obj->AsObjectArray<mirror::Object>();
1395            for (int32_t i = 0; i < object_array->GetLength(); ++i) {
1396              if (object_array->Get(i) == ref) {
1397                LOG(ERROR) << (is_static ? "Static " : "") << "obj[" << i << "] = ref";
1398              }
1399            }
1400          }
1401
1402          *failed_ = true;
1403        }
1404      }
1405    }
1406  }
1407
1408 private:
1409  Heap* const heap_;
1410  bool* const failed_;
1411};
1412
1413class VerifyLiveStackReferences {
1414 public:
1415  explicit VerifyLiveStackReferences(Heap* heap)
1416      : heap_(heap),
1417        failed_(false) {}
1418
1419  void operator()(const mirror::Object* obj) const
1420      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_, Locks::heap_bitmap_lock_) {
1421    VerifyReferenceCardVisitor visitor(heap_, const_cast<bool*>(&failed_));
1422    collector::MarkSweep::VisitObjectReferences(obj, visitor);
1423  }
1424
1425  bool Failed() const {
1426    return failed_;
1427  }
1428
1429 private:
1430  Heap* const heap_;
1431  bool failed_;
1432};
1433
1434bool Heap::VerifyMissingCardMarks() {
1435  Locks::mutator_lock_->AssertExclusiveHeld(Thread::Current());
1436
1437  // We need to sort the live stack since we binary search it.
1438  live_stack_->Sort();
1439  VerifyLiveStackReferences visitor(this);
1440  GetLiveBitmap()->Visit(visitor);
1441
1442  // We can verify objects in the live stack since none of these should reference dead objects.
1443  for (mirror::Object** it = live_stack_->Begin(); it != live_stack_->End(); ++it) {
1444    visitor(*it);
1445  }
1446
1447  if (visitor.Failed()) {
1448    DumpSpaces();
1449    return false;
1450  }
1451  return true;
1452}
1453
1454void Heap::SwapStacks() {
1455  allocation_stack_.swap(live_stack_);
1456
1457  // Sort the live stack so that we can quickly binary search it later.
1458  if (verify_object_mode_ > kNoHeapVerification) {
1459    live_stack_->Sort();
1460  }
1461}
1462
1463void Heap::ProcessCards(base::TimingLogger& timings) {
1464  // Clear cards and keep track of cards cleared in the mod-union table.
1465  typedef std::vector<space::ContinuousSpace*>::iterator It;
1466  for (It it = continuous_spaces_.begin(), end = continuous_spaces_.end(); it != end; ++it) {
1467    space::ContinuousSpace* space = *it;
1468    if (space->IsImageSpace()) {
1469      timings.NewSplit("ModUnionClearCards");
1470      image_mod_union_table_->ClearCards(space);
1471    } else if (space->IsZygoteSpace()) {
1472      timings.NewSplit("ZygoteModUnionClearCards");
1473      zygote_mod_union_table_->ClearCards(space);
1474    } else {
1475      // No mod union table for the AllocSpace. Age the cards so that the GC knows that these cards
1476      // were dirty before the GC started.
1477      timings.NewSplit("AllocSpaceClearCards");
1478      card_table_->ModifyCardsAtomic(space->Begin(), space->End(), AgeCardVisitor(), VoidFunctor());
1479    }
1480  }
1481}
1482
1483void Heap::PreGcVerification(collector::GarbageCollector* gc) {
1484  ThreadList* thread_list = Runtime::Current()->GetThreadList();
1485  Thread* self = Thread::Current();
1486
1487  if (verify_pre_gc_heap_) {
1488    thread_list->SuspendAll();
1489    {
1490      ReaderMutexLock mu(self, *Locks::heap_bitmap_lock_);
1491      if (!VerifyHeapReferences()) {
1492        LOG(FATAL) << "Pre " << gc->GetName() << " heap verification failed";
1493      }
1494    }
1495    thread_list->ResumeAll();
1496  }
1497
1498  // Check that all objects which reference things in the live stack are on dirty cards.
1499  if (verify_missing_card_marks_) {
1500    thread_list->SuspendAll();
1501    {
1502      ReaderMutexLock mu(self, *Locks::heap_bitmap_lock_);
1503      SwapStacks();
1504      // Sort the live stack so that we can quickly binary search it later.
1505      if (!VerifyMissingCardMarks()) {
1506        LOG(FATAL) << "Pre " << gc->GetName() << " missing card mark verification failed";
1507      }
1508      SwapStacks();
1509    }
1510    thread_list->ResumeAll();
1511  }
1512
1513  if (verify_mod_union_table_) {
1514    thread_list->SuspendAll();
1515    ReaderMutexLock reader_lock(self, *Locks::heap_bitmap_lock_);
1516    zygote_mod_union_table_->Update();
1517    zygote_mod_union_table_->Verify();
1518    image_mod_union_table_->Update();
1519    image_mod_union_table_->Verify();
1520    thread_list->ResumeAll();
1521  }
1522}
1523
1524void Heap::PreSweepingGcVerification(collector::GarbageCollector* gc) {
1525  ThreadList* thread_list = Runtime::Current()->GetThreadList();
1526
1527  // Called before sweeping occurs since we want to make sure we are not going so reclaim any
1528  // reachable objects.
1529  if (verify_post_gc_heap_) {
1530    Thread* self = Thread::Current();
1531    CHECK_NE(self->GetState(), kRunnable);
1532    Locks::mutator_lock_->SharedUnlock(self);
1533    thread_list->SuspendAll();
1534    {
1535      WriterMutexLock mu(self, *Locks::heap_bitmap_lock_);
1536      // Swapping bound bitmaps does nothing.
1537      gc->SwapBitmaps();
1538      if (!VerifyHeapReferences()) {
1539        LOG(FATAL) << "Post " << gc->GetName() << "GC verification failed";
1540      }
1541      gc->SwapBitmaps();
1542    }
1543    thread_list->ResumeAll();
1544    Locks::mutator_lock_->SharedLock(self);
1545  }
1546}
1547
1548void Heap::PostGcVerification(collector::GarbageCollector* gc) {
1549  Thread* self = Thread::Current();
1550
1551  if (verify_system_weaks_) {
1552    ReaderMutexLock mu(self, *Locks::heap_bitmap_lock_);
1553    collector::MarkSweep* mark_sweep = down_cast<collector::MarkSweep*>(gc);
1554    mark_sweep->VerifySystemWeaks();
1555  }
1556}
1557
1558collector::GcType Heap::WaitForConcurrentGcToComplete(Thread* self) {
1559  collector::GcType last_gc_type = collector::kGcTypeNone;
1560  if (concurrent_gc_) {
1561    ATRACE_BEGIN("GC: Wait For Concurrent");
1562    bool do_wait;
1563    uint64_t wait_start = NanoTime();
1564    {
1565      // Check if GC is running holding gc_complete_lock_.
1566      MutexLock mu(self, *gc_complete_lock_);
1567      do_wait = is_gc_running_;
1568    }
1569    if (do_wait) {
1570      uint64_t wait_time;
1571      // We must wait, change thread state then sleep on gc_complete_cond_;
1572      ScopedThreadStateChange tsc(Thread::Current(), kWaitingForGcToComplete);
1573      {
1574        MutexLock mu(self, *gc_complete_lock_);
1575        while (is_gc_running_) {
1576          gc_complete_cond_->Wait(self);
1577        }
1578        last_gc_type = last_gc_type_;
1579        wait_time = NanoTime() - wait_start;
1580        total_wait_time_ += wait_time;
1581      }
1582      if (wait_time > kLongGcPauseThreshold) {
1583        LOG(INFO) << "WaitForConcurrentGcToComplete blocked for " << PrettyDuration(wait_time);
1584      }
1585    }
1586    ATRACE_END();
1587  }
1588  return last_gc_type;
1589}
1590
1591void Heap::DumpForSigQuit(std::ostream& os) {
1592  os << "Heap: " << GetPercentFree() << "% free, " << PrettySize(GetBytesAllocated()) << "/"
1593     << PrettySize(GetTotalMemory()) << "; " << GetObjectsAllocated() << " objects\n";
1594  DumpGcPerformanceInfo(os);
1595}
1596
1597size_t Heap::GetPercentFree() {
1598  return static_cast<size_t>(100.0f * static_cast<float>(GetFreeMemory()) / GetTotalMemory());
1599}
1600
1601void Heap::SetIdealFootprint(size_t max_allowed_footprint) {
1602  if (max_allowed_footprint > GetMaxMemory()) {
1603    VLOG(gc) << "Clamp target GC heap from " << PrettySize(max_allowed_footprint) << " to "
1604             << PrettySize(GetMaxMemory());
1605    max_allowed_footprint = GetMaxMemory();
1606  }
1607  max_allowed_footprint_ = max_allowed_footprint;
1608}
1609
1610void Heap::UpdateMaxNativeFootprint() {
1611  size_t native_size = native_bytes_allocated_;
1612  // TODO: Tune the native heap utilization to be a value other than the java heap utilization.
1613  size_t target_size = native_size / GetTargetHeapUtilization();
1614  if (target_size > native_size + max_free_) {
1615    target_size = native_size + max_free_;
1616  } else if (target_size < native_size + min_free_) {
1617    target_size = native_size + min_free_;
1618  }
1619  native_footprint_gc_watermark_ = target_size;
1620  native_footprint_limit_ = 2 * target_size - native_size;
1621}
1622
1623void Heap::GrowForUtilization(collector::GcType gc_type, uint64_t gc_duration) {
1624  // We know what our utilization is at this moment.
1625  // This doesn't actually resize any memory. It just lets the heap grow more when necessary.
1626  const size_t bytes_allocated = GetBytesAllocated();
1627  last_gc_size_ = bytes_allocated;
1628  last_gc_time_ns_ = NanoTime();
1629
1630  size_t target_size;
1631  if (gc_type != collector::kGcTypeSticky) {
1632    // Grow the heap for non sticky GC.
1633    target_size = bytes_allocated / GetTargetHeapUtilization();
1634    if (target_size > bytes_allocated + max_free_) {
1635      target_size = bytes_allocated + max_free_;
1636    } else if (target_size < bytes_allocated + min_free_) {
1637      target_size = bytes_allocated + min_free_;
1638    }
1639    next_gc_type_ = collector::kGcTypeSticky;
1640  } else {
1641    // Based on how close the current heap size is to the target size, decide
1642    // whether or not to do a partial or sticky GC next.
1643    if (bytes_allocated + min_free_ <= max_allowed_footprint_) {
1644      next_gc_type_ = collector::kGcTypeSticky;
1645    } else {
1646      next_gc_type_ = collector::kGcTypePartial;
1647    }
1648
1649    // If we have freed enough memory, shrink the heap back down.
1650    if (bytes_allocated + max_free_ < max_allowed_footprint_) {
1651      target_size = bytes_allocated + max_free_;
1652    } else {
1653      target_size = std::max(bytes_allocated, max_allowed_footprint_);
1654    }
1655  }
1656  SetIdealFootprint(target_size);
1657
1658  // Calculate when to perform the next ConcurrentGC.
1659  if (concurrent_gc_) {
1660    // Calculate the estimated GC duration.
1661    double gc_duration_seconds = NsToMs(gc_duration) / 1000.0;
1662    // Estimate how many remaining bytes we will have when we need to start the next GC.
1663    size_t remaining_bytes = allocation_rate_ * gc_duration_seconds;
1664    remaining_bytes = std::max(remaining_bytes, kMinConcurrentRemainingBytes);
1665    if (UNLIKELY(remaining_bytes > max_allowed_footprint_)) {
1666      // A never going to happen situation that from the estimated allocation rate we will exceed
1667      // the applications entire footprint with the given estimated allocation rate. Schedule
1668      // another GC straight away.
1669      concurrent_start_bytes_ = bytes_allocated;
1670    } else {
1671      // Start a concurrent GC when we get close to the estimated remaining bytes. When the
1672      // allocation rate is very high, remaining_bytes could tell us that we should start a GC
1673      // right away.
1674      concurrent_start_bytes_ = std::max(max_allowed_footprint_ - remaining_bytes, bytes_allocated);
1675    }
1676    DCHECK_LE(concurrent_start_bytes_, max_allowed_footprint_);
1677    DCHECK_LE(max_allowed_footprint_, growth_limit_);
1678  }
1679
1680  UpdateMaxNativeFootprint();
1681}
1682
1683void Heap::ClearGrowthLimit() {
1684  growth_limit_ = capacity_;
1685  alloc_space_->ClearGrowthLimit();
1686}
1687
1688void Heap::SetReferenceOffsets(MemberOffset reference_referent_offset,
1689                                MemberOffset reference_queue_offset,
1690                                MemberOffset reference_queueNext_offset,
1691                                MemberOffset reference_pendingNext_offset,
1692                                MemberOffset finalizer_reference_zombie_offset) {
1693  reference_referent_offset_ = reference_referent_offset;
1694  reference_queue_offset_ = reference_queue_offset;
1695  reference_queueNext_offset_ = reference_queueNext_offset;
1696  reference_pendingNext_offset_ = reference_pendingNext_offset;
1697  finalizer_reference_zombie_offset_ = finalizer_reference_zombie_offset;
1698  CHECK_NE(reference_referent_offset_.Uint32Value(), 0U);
1699  CHECK_NE(reference_queue_offset_.Uint32Value(), 0U);
1700  CHECK_NE(reference_queueNext_offset_.Uint32Value(), 0U);
1701  CHECK_NE(reference_pendingNext_offset_.Uint32Value(), 0U);
1702  CHECK_NE(finalizer_reference_zombie_offset_.Uint32Value(), 0U);
1703}
1704
1705mirror::Object* Heap::GetReferenceReferent(mirror::Object* reference) {
1706  DCHECK(reference != NULL);
1707  DCHECK_NE(reference_referent_offset_.Uint32Value(), 0U);
1708  return reference->GetFieldObject<mirror::Object*>(reference_referent_offset_, true);
1709}
1710
1711void Heap::ClearReferenceReferent(mirror::Object* reference) {
1712  DCHECK(reference != NULL);
1713  DCHECK_NE(reference_referent_offset_.Uint32Value(), 0U);
1714  reference->SetFieldObject(reference_referent_offset_, NULL, true);
1715}
1716
1717// Returns true if the reference object has not yet been enqueued.
1718bool Heap::IsEnqueuable(const mirror::Object* ref) {
1719  DCHECK(ref != NULL);
1720  const mirror::Object* queue =
1721      ref->GetFieldObject<mirror::Object*>(reference_queue_offset_, false);
1722  const mirror::Object* queue_next =
1723      ref->GetFieldObject<mirror::Object*>(reference_queueNext_offset_, false);
1724  return (queue != NULL) && (queue_next == NULL);
1725}
1726
1727void Heap::EnqueueReference(mirror::Object* ref, mirror::Object** cleared_reference_list) {
1728  DCHECK(ref != NULL);
1729  CHECK(ref->GetFieldObject<mirror::Object*>(reference_queue_offset_, false) != NULL);
1730  CHECK(ref->GetFieldObject<mirror::Object*>(reference_queueNext_offset_, false) == NULL);
1731  EnqueuePendingReference(ref, cleared_reference_list);
1732}
1733
1734void Heap::EnqueuePendingReference(mirror::Object* ref, mirror::Object** list) {
1735  DCHECK(ref != NULL);
1736  DCHECK(list != NULL);
1737
1738  // TODO: Remove this lock, use atomic stacks for storing references.
1739  MutexLock mu(Thread::Current(), *reference_queue_lock_);
1740  if (*list == NULL) {
1741    ref->SetFieldObject(reference_pendingNext_offset_, ref, false);
1742    *list = ref;
1743  } else {
1744    mirror::Object* head =
1745        (*list)->GetFieldObject<mirror::Object*>(reference_pendingNext_offset_, false);
1746    ref->SetFieldObject(reference_pendingNext_offset_, head, false);
1747    (*list)->SetFieldObject(reference_pendingNext_offset_, ref, false);
1748  }
1749}
1750
1751mirror::Object* Heap::DequeuePendingReference(mirror::Object** list) {
1752  DCHECK(list != NULL);
1753  DCHECK(*list != NULL);
1754  mirror::Object* head = (*list)->GetFieldObject<mirror::Object*>(reference_pendingNext_offset_,
1755                                                                  false);
1756  mirror::Object* ref;
1757
1758  // Note: the following code is thread-safe because it is only called from ProcessReferences which
1759  // is single threaded.
1760  if (*list == head) {
1761    ref = *list;
1762    *list = NULL;
1763  } else {
1764    mirror::Object* next = head->GetFieldObject<mirror::Object*>(reference_pendingNext_offset_,
1765                                                                 false);
1766    (*list)->SetFieldObject(reference_pendingNext_offset_, next, false);
1767    ref = head;
1768  }
1769  ref->SetFieldObject(reference_pendingNext_offset_, NULL, false);
1770  return ref;
1771}
1772
1773void Heap::AddFinalizerReference(Thread* self, mirror::Object* object) {
1774  ScopedObjectAccess soa(self);
1775  JValue result;
1776  ArgArray arg_array(NULL, 0);
1777  arg_array.Append(reinterpret_cast<uint32_t>(object));
1778  soa.DecodeMethod(WellKnownClasses::java_lang_ref_FinalizerReference_add)->Invoke(self,
1779      arg_array.GetArray(), arg_array.GetNumBytes(), &result, 'V');
1780}
1781
1782void Heap::EnqueueClearedReferences(mirror::Object** cleared) {
1783  DCHECK(cleared != NULL);
1784  if (*cleared != NULL) {
1785    // When a runtime isn't started there are no reference queues to care about so ignore.
1786    if (LIKELY(Runtime::Current()->IsStarted())) {
1787      ScopedObjectAccess soa(Thread::Current());
1788      JValue result;
1789      ArgArray arg_array(NULL, 0);
1790      arg_array.Append(reinterpret_cast<uint32_t>(*cleared));
1791      soa.DecodeMethod(WellKnownClasses::java_lang_ref_ReferenceQueue_add)->Invoke(soa.Self(),
1792          arg_array.GetArray(), arg_array.GetNumBytes(), &result, 'V');
1793    }
1794    *cleared = NULL;
1795  }
1796}
1797
1798void Heap::RequestConcurrentGC(Thread* self) {
1799  // Make sure that we can do a concurrent GC.
1800  Runtime* runtime = Runtime::Current();
1801  DCHECK(concurrent_gc_);
1802  if (runtime == NULL || !runtime->IsFinishedStarting() ||
1803      !runtime->IsConcurrentGcEnabled()) {
1804    return;
1805  }
1806  {
1807    MutexLock mu(self, *Locks::runtime_shutdown_lock_);
1808    if (runtime->IsShuttingDown()) {
1809      return;
1810    }
1811  }
1812  if (self->IsHandlingStackOverflow()) {
1813    return;
1814  }
1815
1816  // We already have a request pending, no reason to start more until we update
1817  // concurrent_start_bytes_.
1818  concurrent_start_bytes_ = std::numeric_limits<size_t>::max();
1819
1820  JNIEnv* env = self->GetJniEnv();
1821  DCHECK(WellKnownClasses::java_lang_Daemons != NULL);
1822  DCHECK(WellKnownClasses::java_lang_Daemons_requestGC != NULL);
1823  env->CallStaticVoidMethod(WellKnownClasses::java_lang_Daemons,
1824                            WellKnownClasses::java_lang_Daemons_requestGC);
1825  CHECK(!env->ExceptionCheck());
1826}
1827
1828void Heap::ConcurrentGC(Thread* self) {
1829  {
1830    MutexLock mu(self, *Locks::runtime_shutdown_lock_);
1831    if (Runtime::Current()->IsShuttingDown()) {
1832      return;
1833    }
1834  }
1835
1836  // Wait for any GCs currently running to finish.
1837  if (WaitForConcurrentGcToComplete(self) == collector::kGcTypeNone) {
1838    CollectGarbageInternal(next_gc_type_, kGcCauseBackground, false);
1839  }
1840}
1841
1842void Heap::RequestHeapTrim() {
1843  // GC completed and now we must decide whether to request a heap trim (advising pages back to the
1844  // kernel) or not. Issuing a request will also cause trimming of the libc heap. As a trim scans
1845  // a space it will hold its lock and can become a cause of jank.
1846  // Note, the large object space self trims and the Zygote space was trimmed and unchanging since
1847  // forking.
1848
1849  // We don't have a good measure of how worthwhile a trim might be. We can't use the live bitmap
1850  // because that only marks object heads, so a large array looks like lots of empty space. We
1851  // don't just call dlmalloc all the time, because the cost of an _attempted_ trim is proportional
1852  // to utilization (which is probably inversely proportional to how much benefit we can expect).
1853  // We could try mincore(2) but that's only a measure of how many pages we haven't given away,
1854  // not how much use we're making of those pages.
1855  uint64_t ms_time = MilliTime();
1856  float utilization =
1857      static_cast<float>(alloc_space_->GetBytesAllocated()) / alloc_space_->Size();
1858  if ((utilization > 0.75f) || ((ms_time - last_trim_time_ms_) < 2 * 1000)) {
1859    // Don't bother trimming the alloc space if it's more than 75% utilized, or if a
1860    // heap trim occurred in the last two seconds.
1861    return;
1862  }
1863
1864  Thread* self = Thread::Current();
1865  {
1866    MutexLock mu(self, *Locks::runtime_shutdown_lock_);
1867    Runtime* runtime = Runtime::Current();
1868    if (runtime == NULL || !runtime->IsFinishedStarting() || runtime->IsShuttingDown()) {
1869      // Heap trimming isn't supported without a Java runtime or Daemons (such as at dex2oat time)
1870      // Also: we do not wish to start a heap trim if the runtime is shutting down (a racy check
1871      // as we don't hold the lock while requesting the trim).
1872      return;
1873    }
1874  }
1875
1876  SchedPolicy policy;
1877  get_sched_policy(self->GetTid(), &policy);
1878  if (policy == SP_FOREGROUND || policy == SP_AUDIO_APP) {
1879    // Don't trim the heap if we are a foreground or audio app.
1880    return;
1881  }
1882
1883  last_trim_time_ms_ = ms_time;
1884  JNIEnv* env = self->GetJniEnv();
1885  DCHECK(WellKnownClasses::java_lang_Daemons != NULL);
1886  DCHECK(WellKnownClasses::java_lang_Daemons_requestHeapTrim != NULL);
1887  env->CallStaticVoidMethod(WellKnownClasses::java_lang_Daemons,
1888                            WellKnownClasses::java_lang_Daemons_requestHeapTrim);
1889  CHECK(!env->ExceptionCheck());
1890}
1891
1892size_t Heap::Trim() {
1893  // Handle a requested heap trim on a thread outside of the main GC thread.
1894  return alloc_space_->Trim();
1895}
1896
1897bool Heap::IsGCRequestPending() const {
1898  return concurrent_start_bytes_ != std::numeric_limits<size_t>::max();
1899}
1900
1901void Heap::RegisterNativeAllocation(int bytes) {
1902  // Total number of native bytes allocated.
1903  native_bytes_allocated_.fetch_add(bytes);
1904  Thread* self = Thread::Current();
1905  if (static_cast<size_t>(native_bytes_allocated_) > native_footprint_gc_watermark_) {
1906    // The second watermark is higher than the gc watermark. If you hit this it means you are
1907    // allocating native objects faster than the GC can keep up with.
1908    if (static_cast<size_t>(native_bytes_allocated_) > native_footprint_limit_) {
1909        JNIEnv* env = self->GetJniEnv();
1910        // Can't do this in WellKnownClasses::Init since System is not properly set up at that
1911        // point.
1912        if (WellKnownClasses::java_lang_System_runFinalization == NULL) {
1913          DCHECK(WellKnownClasses::java_lang_System != NULL);
1914          WellKnownClasses::java_lang_System_runFinalization =
1915              CacheMethod(env, WellKnownClasses::java_lang_System, true, "runFinalization", "()V");
1916          assert(WellKnownClasses::java_lang_System_runFinalization != NULL);
1917        }
1918        if (WaitForConcurrentGcToComplete(self) != collector::kGcTypeNone) {
1919          // Just finished a GC, attempt to run finalizers.
1920          env->CallStaticVoidMethod(WellKnownClasses::java_lang_System,
1921                                    WellKnownClasses::java_lang_System_runFinalization);
1922          CHECK(!env->ExceptionCheck());
1923        }
1924
1925        // If we still are over the watermark, attempt a GC for alloc and run finalizers.
1926        if (static_cast<size_t>(native_bytes_allocated_) > native_footprint_limit_) {
1927          CollectGarbageInternal(collector::kGcTypePartial, kGcCauseForAlloc, false);
1928          env->CallStaticVoidMethod(WellKnownClasses::java_lang_System,
1929                                    WellKnownClasses::java_lang_System_runFinalization);
1930          CHECK(!env->ExceptionCheck());
1931        }
1932        // We have just run finalizers, update the native watermark since it is very likely that
1933        // finalizers released native managed allocations.
1934        UpdateMaxNativeFootprint();
1935    } else {
1936      if (!IsGCRequestPending()) {
1937        RequestConcurrentGC(self);
1938      }
1939    }
1940  }
1941}
1942
1943void Heap::RegisterNativeFree(int bytes) {
1944  int expected_size, new_size;
1945  do {
1946      expected_size = native_bytes_allocated_.load();
1947      new_size = expected_size - bytes;
1948      if (new_size < 0) {
1949        ThrowRuntimeException("attempted to free %d native bytes with only %d native bytes registered as allocated",
1950                              bytes, expected_size);
1951        break;
1952      }
1953  } while (!native_bytes_allocated_.compare_and_swap(expected_size, new_size));
1954}
1955
1956int64_t Heap::GetTotalMemory() const {
1957  int64_t ret = 0;
1958  typedef std::vector<space::ContinuousSpace*>::const_iterator It;
1959  for (It it = continuous_spaces_.begin(), end = continuous_spaces_.end(); it != end; ++it) {
1960    space::ContinuousSpace* space = *it;
1961    if (space->IsImageSpace()) {
1962      // Currently don't include the image space.
1963    } else if (space->IsDlMallocSpace()) {
1964      // Zygote or alloc space
1965      ret += space->AsDlMallocSpace()->GetFootprint();
1966    }
1967  }
1968  typedef std::vector<space::DiscontinuousSpace*>::const_iterator It2;
1969  for (It2 it = discontinuous_spaces_.begin(), end = discontinuous_spaces_.end(); it != end; ++it) {
1970    space::DiscontinuousSpace* space = *it;
1971    if (space->IsLargeObjectSpace()) {
1972      ret += space->AsLargeObjectSpace()->GetBytesAllocated();
1973    }
1974  }
1975  return ret;
1976}
1977
1978}  // namespace gc
1979}  // namespace art
1980