src/heap.cc

/*
 * Copyright (C) 2011 The Android Open Source Project
 *
 * Licensed under the Apache License, Version 2.0 (the "License");
 * you may not use this file except in compliance with the License.
 * You may obtain a copy of the License at
 *
 *      http://www.apache.org/licenses/LICENSE-2.0
 *
 * Unless required by applicable law or agreed to in writing, software
 * distributed under the License is distributed on an "AS IS" BASIS,
 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 * See the License for the specific language governing permissions and
 * limitations under the License.
 */

#include "heap.h"

#include <sys/types.h>
#include <sys/wait.h>

#include <limits>
#include <vector>

#include "card_table.h"
#include "debugger.h"
#include "image.h"
#include "mark_sweep.h"
#include "object.h"
#include "object_utils.h"
#include "os.h"
#include "scoped_heap_lock.h"
#include "scoped_thread_list_lock_releaser.h"
#include "ScopedLocalRef.h"
#include "space.h"
#include "stl_util.h"
#include "thread_list.h"
#include "timing_logger.h"
#include "UniquePtr.h"
#include "well_known_classes.h"

namespace art {

static void UpdateFirstAndLastSpace(Space** first_space, Space** last_space, Space* space) {
  if (*first_space == NULL) {
    *first_space = space;
    *last_space = space;
  } else {
    if ((*first_space)->Begin() > space->Begin()) {
      *first_space = space;
    } else if (space->Begin() > (*last_space)->Begin()) {
      *last_space = space;
    }
  }
}

static bool GenerateImage(const std::string& image_file_name) {
  const std::string boot_class_path_string(Runtime::Current()->GetBootClassPathString());
  std::vector<std::string> boot_class_path;
  Split(boot_class_path_string, ':', boot_class_path);
  if (boot_class_path.empty()) {
    LOG(FATAL) << "Failed to generate image because no boot class path specified";
  }

  std::vector<char*> arg_vector;

  std::string dex2oat_string(GetAndroidRoot());
  dex2oat_string += (kIsDebugBuild ? "/bin/dex2oatd" : "/bin/dex2oat");
  const char* dex2oat = dex2oat_string.c_str();
  arg_vector.push_back(strdup(dex2oat));

  std::string image_option_string("--image=");
  image_option_string += image_file_name;
  const char* image_option = image_option_string.c_str();
  arg_vector.push_back(strdup(image_option));

  arg_vector.push_back(strdup("--runtime-arg"));
  arg_vector.push_back(strdup("-Xms64m"));

  arg_vector.push_back(strdup("--runtime-arg"));
  arg_vector.push_back(strdup("-Xmx64m"));

  for (size_t i = 0; i < boot_class_path.size(); i++) {
    std::string dex_file_option_string("--dex-file=");
    dex_file_option_string += boot_class_path[i];
    const char* dex_file_option = dex_file_option_string.c_str();
    arg_vector.push_back(strdup(dex_file_option));
  }

  std::string oat_file_option_string("--oat-file=");
  oat_file_option_string += image_file_name;
  oat_file_option_string.erase(oat_file_option_string.size() - 3);
  oat_file_option_string += "oat";
  const char* oat_file_option = oat_file_option_string.c_str();
  arg_vector.push_back(strdup(oat_file_option));

  arg_vector.push_back(strdup("--base=0x60000000"));

  std::string command_line(Join(arg_vector, ' '));
  LOG(INFO) << command_line;

  arg_vector.push_back(NULL);
  char** argv = &arg_vector[0];

  // fork and exec dex2oat
  pid_t pid = fork();
  if (pid == 0) {
    // no allocation allowed between fork and exec

    // change process groups, so we don't get reaped by ProcessManager
    setpgid(0, 0);

    execv(dex2oat, argv);

    PLOG(FATAL) << "execv(" << dex2oat << ") failed";
    return false;
  } else {
    STLDeleteElements(&arg_vector);

    // wait for dex2oat to finish
    int status;
    pid_t got_pid = TEMP_FAILURE_RETRY(waitpid(pid, &status, 0));
    if (got_pid != pid) {
      PLOG(ERROR) << "waitpid failed: wanted " << pid << ", got " << got_pid;
      return false;
    }
    if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) {
      LOG(ERROR) << dex2oat << " failed: " << command_line;
      return false;
    }
  }
  return true;
}

Heap::Heap(size_t initial_size, size_t growth_limit, size_t capacity,
           const std::string& original_image_file_name)
    : lock_(NULL),
      image_space_(NULL),
      alloc_space_(NULL),
      mark_bitmap_(NULL),
      live_bitmap_(NULL),
      card_table_(NULL),
      card_marking_disabled_(false),
      is_gc_running_(false),
      concurrent_start_size_(128 * KB),
      concurrent_min_free_(256 * KB),
      try_running_gc_(false),
      requesting_gc_(false),
      num_bytes_allocated_(0),
      num_objects_allocated_(0),
      last_trim_time_(0),
      reference_referent_offset_(0),
      reference_queue_offset_(0),
      reference_queueNext_offset_(0),
      reference_pendingNext_offset_(0),
      finalizer_reference_zombie_offset_(0),
      target_utilization_(0.5),
      verify_objects_(false) {
  if (VLOG_IS_ON(heap) || VLOG_IS_ON(startup)) {
    LOG(INFO) << "Heap() entering";
  }

  // Compute the bounds of all spaces for allocating live and mark bitmaps
  // there will be at least one space (the alloc space)
  Space* first_space = NULL;
  Space* last_space = NULL;

  // Requested begin for the alloc space, to follow the mapped image and oat files
  byte* requested_begin = NULL;
  std::string image_file_name(original_image_file_name);
  if (!image_file_name.empty()) {
    if (OS::FileExists(image_file_name.c_str())) {
      // If the /system file exists, it should be up-to-date, don't try to generate
      image_space_ = Space::CreateImageSpace(image_file_name);
    } else {
      // If the /system file didn't exist, we need to use one from the art-cache.
      // If the cache file exists, try to open, but if it fails, regenerate.
      // If it does not exist, generate.
      image_file_name = GetArtCacheFilenameOrDie(image_file_name);
      if (OS::FileExists(image_file_name.c_str())) {
        image_space_ = Space::CreateImageSpace(image_file_name);
      }
      if (image_space_ == NULL) {
        if (!GenerateImage(image_file_name)) {
          LOG(FATAL) << "Failed to generate image: " << image_file_name;
        }
        image_space_ = Space::CreateImageSpace(image_file_name);
      }
    }
    if (image_space_ == NULL) {
      LOG(FATAL) << "Failed to create space from " << image_file_name;
    }

    AddSpace(image_space_);
    UpdateFirstAndLastSpace(&first_space, &last_space, image_space_);
    // Oat files referenced by image files immediately follow them in memory, ensure alloc space
    // isn't going to get in the middle
    byte* oat_end_addr = image_space_->GetImageHeader().GetOatEnd();
    CHECK(oat_end_addr > image_space_->End());
    if (oat_end_addr > requested_begin) {
      requested_begin = reinterpret_cast<byte*>(RoundUp(reinterpret_cast<uintptr_t>(oat_end_addr),
                                                        kPageSize));
    }
  }

  alloc_space_ = Space::CreateAllocSpace("alloc space", initial_size, growth_limit, capacity,
                                         requested_begin);
  if (alloc_space_ == NULL) {
    LOG(FATAL) << "Failed to create alloc space";
  }
  AddSpace(alloc_space_);
  UpdateFirstAndLastSpace(&first_space, &last_space, alloc_space_);
  byte* heap_begin = first_space->Begin();
  size_t heap_capacity = (last_space->Begin() - first_space->Begin()) + last_space->NonGrowthLimitCapacity();

  // Allocate the initial live bitmap.
  UniquePtr<HeapBitmap> live_bitmap(HeapBitmap::Create("dalvik-bitmap-1", heap_begin, heap_capacity));
  if (live_bitmap.get() == NULL) {
    LOG(FATAL) << "Failed to create live bitmap";
  }

  // Mark image objects in the live bitmap
  for (size_t i = 0; i < spaces_.size(); ++i) {
    Space* space = spaces_[i];
    if (space->IsImageSpace()) {
      space->AsImageSpace()->RecordImageAllocations(live_bitmap.get());
    }
  }

  // Allocate the initial mark bitmap.
  UniquePtr<HeapBitmap> mark_bitmap(HeapBitmap::Create("dalvik-bitmap-2", heap_begin, heap_capacity));
  if (mark_bitmap.get() == NULL) {
    LOG(FATAL) << "Failed to create mark bitmap";
  }

  // Allocate the card table.
  UniquePtr<CardTable> card_table(CardTable::Create(heap_begin, heap_capacity));
  if (card_table.get() == NULL) {
    LOG(FATAL) << "Failed to create card table";
  }

  live_bitmap_ = live_bitmap.release();
  mark_bitmap_ = mark_bitmap.release();
  card_table_ = card_table.release();

  num_bytes_allocated_ = 0;
  num_objects_allocated_ = 0;

  mark_stack_ = MarkStack::Create();

  // It's still too early to take a lock because there are no threads yet,
  // but we can create the heap lock now. We don't create it earlier to
  // make it clear that you can't use locks during heap initialization.
  lock_ = new Mutex("Heap lock", kHeapLock);
  condition_ = new ConditionVariable("Heap condition variable");

  concurrent_start_bytes_ = std::numeric_limits<size_t>::max();

  if (VLOG_IS_ON(heap) || VLOG_IS_ON(startup)) {
    LOG(INFO) << "Heap() exiting";
  }
}

void Heap::AddSpace(Space* space) {
  spaces_.push_back(space);
}

Heap::~Heap() {
  VLOG(heap) << "~Heap()";
  // We can't take the heap lock here because there might be a daemon thread suspended with the
  // heap lock held. We know though that no non-daemon threads are executing, and we know that
  // all daemon threads are suspended, and we also know that the threads list have been deleted, so
  // those threads can't resume. We're the only running thread, and we can do whatever we like...
  STLDeleteElements(&spaces_);
  delete mark_bitmap_;
  delete live_bitmap_;
  delete card_table_;
  delete mark_stack_;
  delete condition_;
  delete lock_;
}

static void MSpaceChunkCallback(void* start, void* end, size_t used_bytes, void* arg) {
  size_t& max_contiguous_allocation = *reinterpret_cast<size_t*>(arg);

  size_t chunk_size = static_cast<size_t>(reinterpret_cast<uint8_t*>(end) - reinterpret_cast<uint8_t*>(start));
  size_t chunk_free_bytes = 0;
  if (used_bytes < chunk_size) {
    chunk_free_bytes = chunk_size - used_bytes;
  }

  if (chunk_free_bytes > max_contiguous_allocation) {
    max_contiguous_allocation = chunk_free_bytes;
  }
}

Object* Heap::AllocObject(Class* c, size_t byte_count) {
  // Used in the detail message if we throw an OOME.
  int64_t total_bytes_free;
  size_t max_contiguous_allocation;

  {
    ScopedHeapLock heap_lock;
    DCHECK(c == NULL || (c->IsClassClass() && byte_count >= sizeof(Class)) ||
           (c->IsVariableSize() || c->GetObjectSize() == byte_count) ||
           strlen(ClassHelper(c).GetDescriptor()) == 0);
    DCHECK_GE(byte_count, sizeof(Object));
    Object* obj = AllocateLocked(byte_count);
    if (obj != NULL) {
      obj->SetClass(c);
      if (Dbg::IsAllocTrackingEnabled()) {
        Dbg::RecordAllocation(c, byte_count);
      }

      if (!is_gc_running_ && num_bytes_allocated_ >= concurrent_start_bytes_) {
        // The SirtRef is necessary since the calls in RequestConcurrentGC
        // are a safepoint.
        SirtRef<Object> ref(obj);
        RequestConcurrentGC();
      }
      VerifyObject(obj);
      return obj;
    }
    total_bytes_free = GetFreeMemory();
    max_contiguous_allocation = 0;
    GetAllocSpace()->Walk(MSpaceChunkCallback, &max_contiguous_allocation);
  }

  std::string msg(StringPrintf("Failed to allocate a %zd-byte %s (%lld total bytes free; largest possible contiguous allocation %zd bytes)",
                               byte_count,
                               PrettyDescriptor(c).c_str(),
                               total_bytes_free, max_contiguous_allocation));
  Thread::Current()->ThrowOutOfMemoryError(msg.c_str());
  return NULL;
}

bool Heap::IsHeapAddress(const Object* obj) {
  // Note: we deliberately don't take the lock here, and mustn't test anything that would
  // require taking the lock.
  if (obj == NULL) {
    return true;
  }
  if (!IsAligned<kObjectAlignment>(obj)) {
    return false;
  }
  for (size_t i = 0; i < spaces_.size(); ++i) {
    if (spaces_[i]->Contains(obj)) {
      return true;
    }
  }
  return false;
}

bool Heap::IsLiveObjectLocked(const Object* obj) {
  lock_->AssertHeld();
  return IsHeapAddress(obj) && live_bitmap_->Test(obj);
}

#if VERIFY_OBJECT_ENABLED
void Heap::VerifyObject(const Object* obj) {
  if (this == NULL || !verify_objects_ || Runtime::Current()->IsShuttingDown() ||
      Thread::Current() == NULL ||
      Runtime::Current()->GetThreadList()->GetLockOwner() == Thread::Current()->GetTid()) {
    return;
  }
  ScopedHeapLock heap_lock;
  Heap::VerifyObjectLocked(obj);
}
#endif

void Heap::VerifyObjectLocked(const Object* obj) {
  lock_->AssertHeld();
  if (obj != NULL) {
    if (!IsAligned<kObjectAlignment>(obj)) {
      LOG(FATAL) << "Object isn't aligned: " << obj;
    } else if (!live_bitmap_->Test(obj)) {
      LOG(FATAL) << "Object is dead: " << obj;
    }
    // Ignore early dawn of the universe verifications
    if (num_objects_allocated_ > 10) {
      const byte* raw_addr = reinterpret_cast<const byte*>(obj) +
          Object::ClassOffset().Int32Value();
      const Class* c = *reinterpret_cast<Class* const *>(raw_addr);
      if (c == NULL) {
        LOG(FATAL) << "Null class in object: " << obj;
      } else if (!IsAligned<kObjectAlignment>(c)) {
        LOG(FATAL) << "Class isn't aligned: " << c << " in object: " << obj;
      } else if (!live_bitmap_->Test(c)) {
        LOG(FATAL) << "Class of object is dead: " << c << " in object: " << obj;
      }
      // Check obj.getClass().getClass() == obj.getClass().getClass().getClass()
      // Note: we don't use the accessors here as they have internal sanity checks
      // that we don't want to run
      raw_addr = reinterpret_cast<const byte*>(c) + Object::ClassOffset().Int32Value();
      const Class* c_c = *reinterpret_cast<Class* const *>(raw_addr);
      raw_addr = reinterpret_cast<const byte*>(c_c) + Object::ClassOffset().Int32Value();
      const Class* c_c_c = *reinterpret_cast<Class* const *>(raw_addr);
      CHECK_EQ(c_c, c_c_c);
    }
  }
}

void Heap::VerificationCallback(Object* obj, void* arg) {
  DCHECK(obj != NULL);
  reinterpret_cast<Heap*>(arg)->VerifyObjectLocked(obj);
}

void Heap::VerifyHeap() {
  ScopedHeapLock heap_lock;
  live_bitmap_->Walk(Heap::VerificationCallback, this);
}

void Heap::RecordAllocationLocked(AllocSpace* space, const Object* obj) {
#ifndef NDEBUG
  if (Runtime::Current()->IsStarted()) {
    lock_->AssertHeld();
  }
#endif
  size_t size = space->AllocationSize(obj);
  DCHECK_GT(size, 0u);
  num_bytes_allocated_ += size;
  num_objects_allocated_ += 1;

  if (Runtime::Current()->HasStatsEnabled()) {
    RuntimeStats* global_stats = Runtime::Current()->GetStats();
    RuntimeStats* thread_stats = Thread::Current()->GetStats();
    ++global_stats->allocated_objects;
    ++thread_stats->allocated_objects;
    global_stats->allocated_bytes += size;
    thread_stats->allocated_bytes += size;
  }

  live_bitmap_->Set(obj);
}

void Heap::RecordFreeLocked(size_t freed_objects, size_t freed_bytes) {
  lock_->AssertHeld();

  if (freed_objects < num_objects_allocated_) {
    num_objects_allocated_ -= freed_objects;
  } else {
    num_objects_allocated_ = 0;
  }
  if (freed_bytes < num_bytes_allocated_) {
    num_bytes_allocated_ -= freed_bytes;
  } else {
    num_bytes_allocated_ = 0;
  }

  if (Runtime::Current()->HasStatsEnabled()) {
    RuntimeStats* global_stats = Runtime::Current()->GetStats();
    RuntimeStats* thread_stats = Thread::Current()->GetStats();
    ++global_stats->freed_objects;
    ++thread_stats->freed_objects;
    global_stats->freed_bytes += freed_bytes;
    thread_stats->freed_bytes += freed_bytes;
  }
}

Object* Heap::AllocateLocked(size_t size) {
  lock_->AssertHeld();
  DCHECK(alloc_space_ != NULL);
  AllocSpace* space = alloc_space_;
  Object* obj = AllocateLocked(space, size);
  if (obj != NULL) {
    RecordAllocationLocked(space, obj);
  }
  return obj;
}

Object* Heap::AllocateLocked(AllocSpace* space, size_t alloc_size) {
  lock_->AssertHeld();

  // Since allocation can cause a GC which will need to SuspendAll, make sure all allocations are
  // done in the runnable state where suspension is expected.
  DCHECK_EQ(Thread::Current()->GetState(), kRunnable);
  Thread::Current()->AssertThreadSuspensionIsAllowable();

  // Fail impossible allocations
  if (alloc_size > space->Capacity()) {
    // On failure collect soft references
    WaitForConcurrentGcToComplete();
    CollectGarbageInternal(false, true);
    return NULL;
  }

  Object* ptr = space->AllocWithoutGrowth(alloc_size);
  if (ptr != NULL) {
    return ptr;
  }

  // The allocation failed.  If the GC is running, block until it completes and retry.
  if (is_gc_running_) {
    // The GC is concurrently tracing the heap.  Release the heap lock, wait for the GC to
    // complete, and retrying allocating.
    WaitForConcurrentGcToComplete();
    ptr = space->AllocWithoutGrowth(alloc_size);
    if (ptr != NULL) {
      return ptr;
    }
  }

  // Another failure.  Our thread was starved or there may be too many
  // live objects.  Try a foreground GC.  This will have no effect if
  // the concurrent GC is already running.
  if (Runtime::Current()->HasStatsEnabled()) {
    ++Runtime::Current()->GetStats()->gc_for_alloc_count;
    ++Thread::Current()->GetStats()->gc_for_alloc_count;
  }
  // We don't need a WaitForConcurrentGcToComplete here since we checked
  // is_gc_running_ earlier and we are in a heap lock.
  CollectGarbageInternal(false, false);
  ptr = space->AllocWithoutGrowth(alloc_size);
  if (ptr != NULL) {
    return ptr;
  }

  // Even that didn't work;  this is an exceptional state.
  // Try harder, growing the heap if necessary.
  ptr = space->AllocWithGrowth(alloc_size);
  if (ptr != NULL) {
    size_t new_footprint = space->GetFootprintLimit();
    // OLD-TODO: may want to grow a little bit more so that the amount of
    //       free space is equal to the old free space + the
    //       utilization slop for the new allocation.
    VLOG(gc) << "Grow heap (frag case) to " << PrettySize(new_footprint)
             << " for a " << PrettySize(alloc_size) << " allocation";
    return ptr;
  }

  // Most allocations should have succeeded by now, so the heap is really full, really fragmented,
  // or the requested size is really big. Do another GC, collecting SoftReferences this time. The
  // VM spec requires that all SoftReferences have been collected and cleared before throwing OOME.

  // OLD-TODO: wait for the finalizers from the previous GC to finish
  VLOG(gc) << "Forcing collection of SoftReferences for " << PrettySize(alloc_size) << " allocation";
  // We don't need a WaitForConcurrentGcToComplete here either.
  CollectGarbageInternal(false, true);
  ptr = space->AllocWithGrowth(alloc_size);
  if (ptr != NULL) {
    return ptr;
  }

  return NULL;
}

int64_t Heap::GetMaxMemory() {
  return alloc_space_->Capacity();
}

int64_t Heap::GetTotalMemory() {
  return alloc_space_->Capacity();
}

int64_t Heap::GetFreeMemory() {
  return alloc_space_->Capacity() - num_bytes_allocated_;
}

class InstanceCounter {
 public:
  InstanceCounter(Class* c, bool count_assignable)
      : class_(c), count_assignable_(count_assignable), count_(0) {
  }

  size_t GetCount() {
    return count_;
  }

  static void Callback(Object* o, void* arg) {
    reinterpret_cast<InstanceCounter*>(arg)->VisitInstance(o);
  }

 private:
  void VisitInstance(Object* o) {
    Class* instance_class = o->GetClass();
    if (count_assignable_) {
      if (instance_class == class_) {
        ++count_;
      }
    } else {
      if (instance_class != NULL && class_->IsAssignableFrom(instance_class)) {
        ++count_;
      }
    }
  }

  Class* class_;
  bool count_assignable_;
  size_t count_;
};

int64_t Heap::CountInstances(Class* c, bool count_assignable) {
  ScopedHeapLock heap_lock;
  InstanceCounter counter(c, count_assignable);
  live_bitmap_->Walk(InstanceCounter::Callback, &counter);
  return counter.GetCount();
}

void Heap::CollectGarbage(bool clear_soft_references) {
  ScopedHeapLock heap_lock;
  // If we just waited for a GC to complete then we do not need to do another
  // GC unless we clear soft references.
  if (!WaitForConcurrentGcToComplete() || clear_soft_references) {
    CollectGarbageInternal(false, clear_soft_references);
  }
}

void Heap::CollectGarbageInternal(bool concurrent, bool clear_soft_references) {
  lock_->AssertHeld();

  DCHECK(!is_gc_running_);
  is_gc_running_ = true;

  TimingLogger timings("CollectGarbageInternal");
  uint64_t t0 = NanoTime(), root_end = 0, dirty_begin = 0, dirty_end = 0;

  ThreadList* thread_list = Runtime::Current()->GetThreadList();
  thread_list->SuspendAll();
  timings.AddSplit("SuspendAll");

  size_t initial_size = num_bytes_allocated_;
  Object* cleared_references = NULL;
  {
    MarkSweep mark_sweep(mark_stack_);
    timings.AddSplit("ctor");

    mark_sweep.Init();
    timings.AddSplit("Init");

    if (concurrent) {
      card_table_->ClearNonImageSpaceCards(this);
    }

    mark_sweep.MarkRoots();
    timings.AddSplit("MarkRoots");

    if (!concurrent) {
      mark_sweep.ScanDirtyImageRoots();
      timings.AddSplit("ScanDirtyImageRoots");
    }

    // Roots are marked on the bitmap and the mark_stack is empty
    DCHECK(mark_sweep.IsMarkStackEmpty());

    if (concurrent) {
      // We need to resume before unlocking or else a thread waiting for the
      // heap lock would re-suspend since we have not yet called ResumeAll.
      thread_list->ResumeAll();
      Unlock();
      root_end = NanoTime();
      timings.AddSplit("RootEnd");
    }

    // Recursively mark all bits set in the non-image mark bitmap
    mark_sweep.RecursiveMark();
    timings.AddSplit("RecursiveMark");

    if (concurrent) {
      dirty_begin = NanoTime();
      Lock();
      thread_list->SuspendAll();
      timings.AddSplit("ReSuspend");

      // Re-mark root set.
      mark_sweep.ReMarkRoots();
      timings.AddSplit("ReMarkRoots");

      // Scan dirty objects, this is required even if we are not doing a
      // concurrent GC since we use the card table to locate image roots.
      mark_sweep.RecursiveMarkDirtyObjects();
      timings.AddSplit("RecursiveMarkDirtyObjects");
    }

    mark_sweep.ProcessReferences(clear_soft_references);
    timings.AddSplit("ProcessReferences");

    // TODO: swap live and marked bitmaps
    // Note: Need to be careful about image spaces if we do this since not
    // everything image space will be marked, resulting in things not being
    // marked as live anymore.

    // Verify that we only reach marked objects from the image space
    mark_sweep.VerifyImageRoots();
    timings.AddSplit("VerifyImageRoots");

    mark_sweep.Sweep();
    timings.AddSplit("Sweep");

    cleared_references = mark_sweep.GetClearedReferences();
  }

  GrowForUtilization();
  timings.AddSplit("GrowForUtilization");
  thread_list->ResumeAll();
  dirty_end = NanoTime();

  EnqueueClearedReferences(&cleared_references);
  RequestHeapTrim();
  timings.AddSplit("Finish");

  uint64_t t1 = NanoTime();
  uint64_t duration_ns = t1 - t0;
  bool gc_was_particularly_slow = duration_ns > MsToNs(50); // TODO: crank this down for concurrent.
  if (VLOG_IS_ON(gc) || gc_was_particularly_slow) {
    // TODO: somehow make the specific GC implementation (here MarkSweep) responsible for logging.
    // Reason: For CMS sometimes initial_size < num_bytes_allocated_ results in overflow (3GB freed message).
    size_t bytes_freed = initial_size - num_bytes_allocated_;
    // lose low nanoseconds in duration. TODO: make this part of PrettyDuration
    duration_ns = (duration_ns / 1000) * 1000;
    if (concurrent) {
      uint64_t pause_roots_time = (root_end - t0) / 1000 * 1000;
      uint64_t pause_dirty_time = (dirty_end - dirty_begin) / 1000 * 1000;
      LOG(INFO) << "GC freed " << PrettySize(bytes_freed) << ", " << GetPercentFree() << "% free, "
                << PrettySize(num_bytes_allocated_) << "/" << PrettySize(GetTotalMemory()) << ", "
                << "paused " << PrettyDuration(pause_roots_time) << "+" << PrettyDuration(pause_dirty_time)
                << ", total " << PrettyDuration(duration_ns);
    } else {
      uint64_t markSweepTime = (dirty_end - t0) / 1000 * 1000;
      LOG(INFO) << "GC freed " << PrettySize(bytes_freed) << ", " << GetPercentFree() << "% free, "
                << PrettySize(num_bytes_allocated_) << "/" << PrettySize(GetTotalMemory()) << ", "
                << "paused " << PrettyDuration(markSweepTime)
                << ", total " << PrettyDuration(duration_ns);
    }
  }
  Dbg::GcDidFinish();
  if (VLOG_IS_ON(heap)) {
    timings.Dump();
  }

  is_gc_running_ = false;

  // Wake anyone who may have been waiting for the GC to complete.
  condition_->Broadcast();
}

bool Heap::WaitForConcurrentGcToComplete() {
  lock_->AssertHeld();

  // Busy wait for GC to finish
  if (is_gc_running_) {
    uint64_t wait_start = NanoTime();

    do {
      ScopedThreadStateChange tsc(Thread::Current(), kVmWait);
      ScopedThreadListLockReleaser list_lock_releaser;
      condition_->Wait(*lock_);
    } while (is_gc_running_);
    uint64_t wait_time = NanoTime() - wait_start;
    if (wait_time > MsToNs(5)) {
      LOG(INFO) << "WaitForConcurrentGcToComplete blocked for " << PrettyDuration(wait_time);
    }
    DCHECK(!is_gc_running_);
    return true;
  }
  return false;
}

void Heap::DumpForSigQuit(std::ostream& os) {
  os << "Heap: " << GetPercentFree() << "% free, "
     << PrettySize(num_bytes_allocated_) << "/" << PrettySize(GetTotalMemory())
     << "; " << num_objects_allocated_ << " objects\n";
}

size_t Heap::GetPercentFree() {
  size_t total = GetTotalMemory();
  return 100 - static_cast<size_t>(100.0f * static_cast<float>(num_bytes_allocated_) / total);
}

void Heap::SetIdealFootprint(size_t max_allowed_footprint) {
  size_t alloc_space_capacity = alloc_space_->Capacity();
  if (max_allowed_footprint > alloc_space_capacity) {
    VLOG(gc) << "Clamp target GC heap from " << PrettySize(max_allowed_footprint)
             << " to " << PrettySize(alloc_space_capacity);
    max_allowed_footprint = alloc_space_capacity;
  }
  alloc_space_->SetFootprintLimit(max_allowed_footprint);
}

// kHeapIdealFree is the ideal maximum free size, when we grow the heap for utilization.
static const size_t kHeapIdealFree = 2 * MB;
// kHeapMinFree guarantees that you always have at least 512 KB free, when you grow for utilization,
// regardless of target utilization ratio.
static const size_t kHeapMinFree = kHeapIdealFree / 4;

void Heap::GrowForUtilization() {
  lock_->AssertHeld();

  // We know what our utilization is at this moment.
  // This doesn't actually resize any memory. It just lets the heap grow more
  // when necessary.
  size_t target_size(num_bytes_allocated_ / Heap::GetTargetHeapUtilization());

  if (target_size > num_bytes_allocated_ + kHeapIdealFree) {
    target_size = num_bytes_allocated_ + kHeapIdealFree;
  } else if (target_size < num_bytes_allocated_ + kHeapMinFree) {
    target_size = num_bytes_allocated_ + kHeapMinFree;
  }

  // Calculate when to perform the next ConcurrentGC.
  if (GetTotalMemory() - num_bytes_allocated_ < concurrent_min_free_) {
    // Not enough free memory to perform concurrent GC.
    concurrent_start_bytes_ = std::numeric_limits<size_t>::max();
  } else {
    concurrent_start_bytes_ = alloc_space_->GetFootprintLimit() - concurrent_start_size_;
  }

  SetIdealFootprint(target_size);
}

void Heap::ClearGrowthLimit() {
  ScopedHeapLock heap_lock;
  WaitForConcurrentGcToComplete();
  alloc_space_->ClearGrowthLimit();
}

pid_t Heap::GetLockOwner() {
  return lock_->GetOwner();
}

void Heap::Lock() {
  // Grab the lock, but put ourselves into kVmWait if it looks
  // like we're going to have to wait on the mutex. This prevents
  // deadlock if another thread is calling CollectGarbageInternal,
  // since they will have the heap lock and be waiting for mutators to
  // suspend.
  if (!lock_->TryLock()) {
    ScopedThreadStateChange tsc(Thread::Current(), kVmWait);
    lock_->Lock();
  }
}

void Heap::Unlock() {
  lock_->Unlock();
}

void Heap::SetReferenceOffsets(MemberOffset reference_referent_offset,
    MemberOffset reference_queue_offset,
    MemberOffset reference_queueNext_offset,
    MemberOffset reference_pendingNext_offset,
    MemberOffset finalizer_reference_zombie_offset) {
  reference_referent_offset_ = reference_referent_offset;
  reference_queue_offset_ = reference_queue_offset;
  reference_queueNext_offset_ = reference_queueNext_offset;
  reference_pendingNext_offset_ = reference_pendingNext_offset;
  finalizer_reference_zombie_offset_ = finalizer_reference_zombie_offset;
  CHECK_NE(reference_referent_offset_.Uint32Value(), 0U);
  CHECK_NE(reference_queue_offset_.Uint32Value(), 0U);
  CHECK_NE(reference_queueNext_offset_.Uint32Value(), 0U);
  CHECK_NE(reference_pendingNext_offset_.Uint32Value(), 0U);
  CHECK_NE(finalizer_reference_zombie_offset_.Uint32Value(), 0U);
}

Object* Heap::GetReferenceReferent(Object* reference) {
  DCHECK(reference != NULL);
  DCHECK_NE(reference_referent_offset_.Uint32Value(), 0U);
  return reference->GetFieldObject<Object*>(reference_referent_offset_, true);
}

void Heap::ClearReferenceReferent(Object* reference) {
  DCHECK(reference != NULL);
  DCHECK_NE(reference_referent_offset_.Uint32Value(), 0U);
  reference->SetFieldObject(reference_referent_offset_, NULL, true);
}

// Returns true if the reference object has not yet been enqueued.
bool Heap::IsEnqueuable(const Object* ref) {
  DCHECK(ref != NULL);
  const Object* queue = ref->GetFieldObject<Object*>(reference_queue_offset_, false);
  const Object* queue_next = ref->GetFieldObject<Object*>(reference_queueNext_offset_, false);
  return (queue != NULL) && (queue_next == NULL);
}

void Heap::EnqueueReference(Object* ref, Object** cleared_reference_list) {
  DCHECK(ref != NULL);
  CHECK(ref->GetFieldObject<Object*>(reference_queue_offset_, false) != NULL);
  CHECK(ref->GetFieldObject<Object*>(reference_queueNext_offset_, false) == NULL);
  EnqueuePendingReference(ref, cleared_reference_list);
}

void Heap::EnqueuePendingReference(Object* ref, Object** list) {
  DCHECK(ref != NULL);
  DCHECK(list != NULL);

  if (*list == NULL) {
    ref->SetFieldObject(reference_pendingNext_offset_, ref, false);
    *list = ref;
  } else {
    Object* head = (*list)->GetFieldObject<Object*>(reference_pendingNext_offset_, false);
    ref->SetFieldObject(reference_pendingNext_offset_, head, false);
    (*list)->SetFieldObject(reference_pendingNext_offset_, ref, false);
  }
}

Object* Heap::DequeuePendingReference(Object** list) {
  DCHECK(list != NULL);
  DCHECK(*list != NULL);
  Object* head = (*list)->GetFieldObject<Object*>(reference_pendingNext_offset_, false);
  Object* ref;
  if (*list == head) {
    ref = *list;
    *list = NULL;
  } else {
    Object* next = head->GetFieldObject<Object*>(reference_pendingNext_offset_, false);
    (*list)->SetFieldObject(reference_pendingNext_offset_, next, false);
    ref = head;
  }
  ref->SetFieldObject(reference_pendingNext_offset_, NULL, false);
  return ref;
}

void Heap::AddFinalizerReference(Thread* self, Object* object) {
  ScopedThreadStateChange tsc(self, kRunnable);
  JValue args[1];
  args[0].SetL(object);
  DecodeMethod(WellKnownClasses::java_lang_ref_FinalizerReference_add)->Invoke(self, NULL, args, NULL);
}

void Heap::EnqueueClearedReferences(Object** cleared) {
  DCHECK(cleared != NULL);
  if (*cleared != NULL) {
    Thread* self = Thread::Current();
    ScopedThreadStateChange tsc(self, kRunnable);
    JValue args[1];
    args[0].SetL(*cleared);
    DecodeMethod(WellKnownClasses::java_lang_ref_ReferenceQueue_add)->Invoke(self, NULL, args, NULL);
    *cleared = NULL;
  }
}

void Heap::RequestConcurrentGC() {
  // Make sure that we can do a concurrent GC.
  if (requesting_gc_ ||
      !Runtime::Current()->IsFinishedStarting() ||
      Runtime::Current()->IsShuttingDown() ||
      !Runtime::Current()->IsConcurrentGcEnabled()) {
    return;
  }

  requesting_gc_ = true;
  JNIEnv* env = Thread::Current()->GetJniEnv();
  DCHECK(WellKnownClasses::java_lang_Daemons != NULL);
  DCHECK(WellKnownClasses::java_lang_Daemons_requestGC != NULL);
  env->CallStaticVoidMethod(WellKnownClasses::java_lang_Daemons, WellKnownClasses::java_lang_Daemons_requestGC);
  CHECK(!env->ExceptionCheck());
  requesting_gc_ = false;
}

void Heap::ConcurrentGC() {
  ScopedHeapLock heap_lock;
  // We shouldn't need a WaitForConcurrentGcToComplete here since only
  // concurrent GC resumes threads before the GC is completed and this function
  // is only called within the GC daemon thread.
  CHECK(!is_gc_running_);
  // Current thread needs to be runnable or else we can't suspend all threads.
  ScopedThreadStateChange tsc(Thread::Current(), kRunnable);
  CollectGarbageInternal(true, false);
}

void Heap::Trim() {
  lock_->AssertHeld();
  WaitForConcurrentGcToComplete();
  GetAllocSpace()->Trim();
}

void Heap::RequestHeapTrim() {
  // We don't have a good measure of how worthwhile a trim might be. We can't use the live bitmap
  // because that only marks object heads, so a large array looks like lots of empty space. We
  // don't just call dlmalloc all the time, because the cost of an _attempted_ trim is proportional
  // to utilization (which is probably inversely proportional to how much benefit we can expect).
  // We could try mincore(2) but that's only a measure of how many pages we haven't given away,
  // not how much use we're making of those pages.
  float utilization = static_cast<float>(num_bytes_allocated_) / alloc_space_->Size();
  uint64_t ms_time = NsToMs(NanoTime());
  if (utilization > 0.75f || ms_time - last_trim_time_ < 2 * 1000) {
    // Don't bother trimming the heap if it's more than 75% utilized, or if a
    // heap trim occurred in the last two seconds.
    return;
  }
  if (!Runtime::Current()->IsFinishedStarting() || Runtime::Current()->IsShuttingDown()) {
    // Heap trimming isn't supported without a Java runtime or Daemons (such as at dex2oat time)
    // Also: we do not wish to start a heap trim if the runtime is shutting down.
    return;
  }
  last_trim_time_ = ms_time;
  JNIEnv* env = Thread::Current()->GetJniEnv();
  DCHECK(WellKnownClasses::java_lang_Daemons != NULL);
  DCHECK(WellKnownClasses::java_lang_Daemons_requestHeapTrim != NULL);
  env->CallStaticVoidMethod(WellKnownClasses::java_lang_Daemons, WellKnownClasses::java_lang_Daemons_requestHeapTrim);
  CHECK(!env->ExceptionCheck());
}

}  // namespace art